Research ArticleLoad Balancing Opportunistic Routing for CognitiveRadio Ad Hoc Networks

Wenxuan Duan1 Xing Tang 1 Junwei Zhou 1 Jing Wang2 and Guosheng Zhou1

1School of Computer Science and Technology Wuhan University of Technology Wuhan China2School of Computer Science Wuhan University Wuhan China

Correspondence should be addressed to Xing Tang tangxingwhuteducn

Received 16 June 2018 Revised 4 October 2018 Accepted 8 November 2018 Published 2 December 2018

Academic Editor Ioannis Chatzigiannakis

Copyright copy 2018 WenxuanDuan et alThis is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Recent research activities have shown that opportunistic routing can achieve considerable performance gains in Cognitive RadioAd hocNetworks (CRAHNs)Most of these studies focused on designing appropriatemetrics to select and prioritize the forwardingcandidates However in multiple-flow networks a small number of nodesmay always be with the higher priority order for differentflowsThus some nodesmay easily become overloaded with toomuch traffic and be severely congested To overcome this problemwe propose a load balancing opportunistic routing (LBOR) scheme to maximize the total throughput of the whole networkWe first formulate the problem of maximizing the total throughput of the network as a linear programming problem Thenwe develop heuristic load balancing candidate forwarder sorting and selection algorithms Simulation results and comparisonsdemonstrate that our proposed LBOR scheme outperforms existing opportunistic routing protocols with nonload balancingmethods in CRAHNs

1 Introduction

Cognitive Radio Networks (CRNs) [1] have been consideredas a promising solution to improve the efficiency of spectrumusage and hence alleviate the spectrum scarcity problem InCRNs secondary users (SUs) can opportunistically access thespectrum as long as the primary users (PUs) do not occupythe licensed spectrum at a particular time and a specificgeographic area

Cognitive Radio (CR) technology can be combined withad hoc networks which is called cognitive radio ad hocnetworks (CRAHNs) in which wireless devices can dynam-ically establish networks using the vacant spectrum bandsallocated to PUs without the need of fixed infrastructuresDue to the changing spectrum availability and dynamicnetwork topology routing introduces a significant challengein CRAHNs

In recent years several routing protocols [2ndash8] havebeen proposed for CRAHNs However most of these stud-ies focus on how to design a proper metric to measurethe quality of a preselected route from the source to thedestination Unfortunately owing to the unpredictable and

intermittent nature of links between nodes a prefixed pathwould incur a lot of retransmissions in CRAHNs To alleviatethis problem opportunistic routing [9] was proposed toexploit the broadcast nature of wireless transmissions thatone transmission can be overheard by multiple neighbors Inopportunistic routing the decision of the next relay node canchange dynamically following the network conditions Thusopportunistic routing can better fit the need with unreliablewireless links in CRAHNs

Recently some initial work for opportunistic routing canbe found in [10ndash17] Most of them are designed based ona predefined performance metric such as the geographicaldistance [10ndash14] the packet delivery ratio [15 16] andthe expected transmission time (ETT) [17] Indeed theseschemes can improve the performance of opportunisticroutingHowever adopting onemetric to select and prioritizethe forwarding candidates for multiple paths some nodesmay easily become overloaded with too much traffic Forinstancemultiple end-to-end pathsmay share some forward-ing candidates Adopting the same metric a small number ofnodes may always be with higher priority order for differentpaths Thus the nodes may be severely congested and have

HindawiWireless Communications and Mobile ComputingVolume 2018 Article ID 9412782 16 pageshttpsdoiorg10115520189412782

2 Wireless Communications and Mobile Computing

to drop a large number of packets due to buffer limitationsTherefore load balancing in opportunistic routing becomes acritical problem that may affect the throughput of the wholenetwork

In this paper we propose a load balancing opportunisticrouting (LBOR) scheme to maximize the total throughput ofthe whole network To the best of our knowledge there hasbeen no paper which jointly considers load balancing andopportunistic routing for CRAHNs In summary the majorcontributions are listed as follows

(i) We analyze the throughput bound of one opportunis-tic routing module and formulate the problem ofmaximizing the total throughput of the network as alinear programming problem

(ii) We present a newmetric for selecting and prioritizingthe forwarder with considering the traffic load anddevelop load balancing candidate forwarder sortingand selection algorithms

(iii) We conduct simulations to show that our algorithmcan provide better throughput performance than thestate-of-the-art opportunistic routing approaches forCRAHNs

The rest of this paper is organized as follows Section 2briefly overviews the related work Section 3 introduces thesystemmodelWe analyze the throughput of one-hop oppor-tunistic routing and formulate the maximum throughputproblem in Section 4 In Section 5 we propose algorithmsto solve the problem Simulation results are presented inSection 6 Finally the paper is concluded in Section 7

2 Related Work

Owing to the frequent dynamic changes in the CRNstraditional routing are not well adapted to be applied toCRNs Nowadays scholars have proposed many routingprotocols for CRNs To address the routing problem in CRNscaused by dynamic channel availability Saleem et al [2ndash4]presented a cluster-based routing scheme called SMARTwhich consisted of clustering mechanisms and an artificialintelligence approach Zareei et al [5] proposed a novel on-demand cluster-based hybrid routing protocol for CRAHNsThey firstly introduced a novel spectrum-aware clusteringmechanism which divided nodes into clusters based on thespectrum availability power level and stability Then theydeveloped a routing algorithm to minimize the delay whileachieving acceptable delivery ratio To provide secure andreliable routing in CRNs SEARCH [6] implemented a secureand reliable routing based on distributed Boltzmann-Gibbslearning algorithm The author considered the trust valueas well as the total delay for the successful and reliabletransmission of the packet Guirguis et al [7] presented amultihop routing protocol for CRNs in which they integratedthe collaborative beamforming technique with routing Luet al [8] first derived the actual spectrum accessible prob-ability of SUs from the perspectives of social activities andthen proposed a greedy spectrum-aware routing algorithmHowever most of these studies focus on how to design a

proper metric to measure the quality of a preselected routefrom the source to the destination Unfortunately owing tothe unpredictable and intermittent nature of links betweennodes a prefixed path would incur a lot of retransmissions inCRAHNs

Opportunistic routing was proposed to exploit the broad-cast nature of wireless transmissions that one transmissioncan be overheard by multiple neighbors Instead of adoptinga fixed relay path a source node broadcasts packets toneighboring nodes and selects a relay based on the receivedresponses under current link conditions Coutinho et al [10]proposed the GEDAR routing protocol for underwater wire-less sensor networks which utilized the location informationof the neighbor nodes and some known sonobuoys to select anext hop forwarder set of neighbors Rahman et al [11] pro-posed an energy-efficient cooperative opportunistic routing(EECOR) protocol for underwater acoustic sensor networksTang et al [12] proposed a novel distributed protocol thatdivides the whole path from the source to the destinationinto several smaller opportunistic route segments In eachopportunistic route segment there is a temporary sourcenode a temporary destination node and a set of potentialrelay nodes The packets are transmitted through a numberof intermediate destination nodes step-by-step until the ulti-mate destination is reached Since their scheme utilizes onlylocal spectrum opportunities topology information andgeometric conditions to compute the forwarding set for eachshort-term opportunistic route segment it can better adaptto dynamic spectrum environments and changing networktopologies in CR ad hoc networks Furthermore they discussa geographical opportunistic routing scheme combined withnetwork coding for CRAHNs [13 14] Wang et al [15]implemented a spectrum-aware any-path routing (SAAR)scheme with consideration of both the salient spectrumuncertainty feature of CRNs and the unreliable transmissioncharacteristics of the wireless medium SAAR significantlyincreases the packet delivery ratio and reduces the end-to-end delay with low communication and computationoverhead which enables it suitable and scalable to be usedin CRAHNs Sanchez-Iborra et al [16] introduced a novelopportunistic routing protocol called JOKER which takesinto account both the packet delivery reliability of the linksand the distance-progress towards the final destination Cuiet al [17] proposed a DCSS-OCR protocol for CRAHNsto discover stable routing opportunities with novel metricsreferred to as the path access probability and ETT Zhonget al [18] jointly considered energy trust and social featureand they designed a secure OR called ETOR However theproblem that the misbehavior node attacks the normal nodeis still unsolved Tang et al [19] presented a centralized and adistributed opportunistic routing relay algorithm to achievethe optimal system throughput in multihop WLANs BenFradj et al [20] described the energy-efficient opportunisticrouting which focused on the selection of the forwardinglist to minimize the energy consumption Zikria et al [21]proposed a heuristic forwarder selection scheme for wirelesssensor networks called HASORF Most of these studiesmentioned above only consider the performance of routingwhen designing the candidate nodes priority metrics and fail

Wireless Communications and Mobile Computing 3

to take the problem of load balancing into account whichcauses the network prone to block

For the design of load balancing in wireless networksmany scholars have made some achievements The OELR[22] algorithm is used to solve the problem of networkservice life due to the excessive load of some nodes in mobilead hoc networks Although the algorithm can reduce theenergy consumption and ensure the reliability of the datatransmission it can only achieve load balancing for the localnetwork The ORPL-LB [23] algorithm is used to solve theinfluence of node active time on network load and energyconsumption in a wireless sensor network To achieve thebalance of energy consumption the algorithm firstly adjuststhe active time of the node according to the energy andthe flow Then it regulates the current working cycle of thenode based on the working period of the target The LBIA[24] algorithm is used to solve the problem of networkcongestion caused by the excessive load of some nodes Inthis algorithm in addition to load balancing the interdomaindata flow interference and domain data flow interferenceare considered So et al [25] proposed a load balancingopportunistic routing algorithm for wireless sensor networksThis opportunistic routing scheme designs a new forwardnode selectionmetric based on the residual energy of networknodes Xu et al [26] proposed a dynamic resource allocationmethod named DRAM which is designed through staticresource allocation and dynamic service migration to achievethe load balance for the fog computing systems Howeverthe existing network load balancing algorithms cannot bedirectly applied in CRAHNs

To the best of our knowledge jointly considering load bal-ancing and opportunistic routing is novel in CRAHNs Firstthe existing load balancing schemes [22ndash26] are designed fortraditional wireless networks and cannot be directly appliedin CRAHNS due to the varying availability of spectrumsSecond the forwarder set selection metric of existing oppor-tunistic routing schemes [12ndash18] only consider the routingparameters such as the end-to-end delay the throughput andthe number of transmissions Different from these worksour metric is load-dependent and takes care of the linkquality as well Most importantly our candidate forwarderselection and ordering algorithms can bypass blocked linksand distribute the excessive loads of high loaded nodes tonearby nodes

3 System Model

31 Network Model We consider a dense-scaling secondarynetwork (SN) coexisting with a primary network (PN)deployed in a specific area The PN consists of M Poissondistributed primary users We assume that the interferencerange of PU is 120588119875 If a PU is active and its transmissionfrequency overlaps an SUrsquos channel the SU which is insidethe interference range of the PU is not permitted to transmitat this time Meanwhile the transmission range of PU isassumed to be 119877119875 (without loss of generality 120588119875 gt 119877119875)The SN is composed of N SUs who know their locationsWe assume SUs are static or quasi-static Similar to PUsrsquonetwork the transmission and interference ranges of SUs

are assumed to be 119877119878 and 120588119878 respectively (without loss ofgenerality 120588119878 gt 119877119878) In our proposed scheme we need tocompute the distance between neighboring nodes and thedestination Thus it is essential to assume that all SUs canacquire their own location information by equipping an on-board localization device whose accuracy is around 5meters

In our system multiple PUs and SUs share a set oforthogonal channels denoted by 119862119867 = 119888ℎ1 119888ℎ2 119888ℎ119898SUs can exchange messages over a common control channel(CCC) Each SU is equipped with two radios one half-duplex cognitive radio that can switch among CH for datatransmissions and the other half-duplex normal radio inCCCfor exchanging control messages Note that PUs have therights of accessing the licensed bands for communicationswhile SUs can only opportunistically access the spectrum forad hoc device-to-device transmissions We model the occu-pation time of PUs in each data channel as an independentand identically distributed alternating ON (PU is active) andOFF (PU is inactive) process For ease of simulation andexplanation we assume that the mean time spent on bothON and OFF states is 119879119874119873119888ℎ and 119879119874119865119865119888ℎ respectively and theyfollow an exponential distribution with expectation 1120582119874119873119888ℎand 1120582119874119865119865119888ℎ Then we can obtain the probability of channelch to be busy (ON) and idle (OFF) respectively as follows

119875119888ℎ119887119906119904119910 = 120582119874119865119865119888ℎ120582119874119873119888ℎ

+ 120582119874119865119865119888ℎ

(1)

119875119888ℎ119894119889119897119890 = 120582119874119873119888ℎ120582119874119873119888ℎ

+ 120582119874119865119865119888ℎ

(2)

32 Traffic Model In this model each node 119899119894 can sense itsenvironment and find a set of available frequency bands (iethose bands that are not currently used by primary users)Weuse 119862119867119894119895 = 119862119867119894 cap 119862119867119895 to denote the set of channels thatare overlapped between node 119899119894 and node 119899119895 which can beused for communications Besides we use 119889119894119895 to represent theEuclidean distance between two nodes We assume that SUscan communicate with other nodes within their transmissionranges when they have common channels (ie there exists alinkedge 119897 between SUs) We use 119901119894119895 to represent the packetreception ratio (PRR) of the link 119897119894119895 between 119899119894 and 119899119895 We usean indicator 120595119894119895 to determine the state of the link 119897119894119895 between119899119894 and 119899119895 There exists a link between 119899119894 and 119899119895 if 119862119867119894119895 = 0and 119889119894119895 lt 119877119878 (ie 120595119894119895 = 1 otherwise 120595119894119895 = 0)

120595119894119895 = 1 119862119867119894119895 = 0 119886119899119889 119889119894119895 lt 1198771198780 otherwise (3)

Thus the SN can bemodeled as a graphG = (VE) whereV is the set of nodes (SUs) and E is the set of all the possiblelinks formed by nodes in V We divide the network time intoseveral time slotswith the length of120596Weuse C119862119897119906 and119862119897119886 torepresent the maximum capacity the used capacity and theavailable capacity of the link 119897 at the beginning of each timeslot119862119897119906 can be obtained through the exchange of informationbetween nodesMeanwhile119862119897119886 can be calculated according to

4 Wireless Communications and Mobile Computing

the used capacity (119862119897119906) and the maximum capacity (C) whichis shown as follows

119862119897119886 = C minus 119862119897119906 119862119897119906 lt C

0 119862119897119880 ge C (4)

where the first equation of (4) is adopted when the link is notoverloaded and the second is applied to the case that the linkis overloaded The available capacity of a whole link set canbe derived according to the available capacity of each link inthe set For a link set LS we denote the available capacity ofLS as 119862119871119878119886 Then we have

119862119871119878119886 = sum119897isin119871119878

119862119897119886 (5)

wheresum119897isin119871119878 119862119897119886 represents the sum of the available capacity ofeach link

33 Flow Model We assume that a data flow 119891 is composedof all the packets sent by a node at a time slot where 120578119898 isthe maximum bits of the data flow The average size of 119891 isV119891 (V119891120598(0 120578119898])We use ϝ to represent the data flow set whichincludes all the end-to-end data flows in the network For adata flow119891 (119891120598ϝ) we use s(119891) and d(119891) to represent its sourcenode and destination node respectively For a link 119897 (119897 isin 119864)we use V119897119891 to represent the average size of the data flow 119891transmitted among the link 119897 at a time slot Then accordingto the flow conservation for each node n (n isin V) we haveV119891 + sum

119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891 119894119891 119899 = 119904 (119891) 119886119899119889 119891 isin 119865sum

119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119894119891 119899 = 119904 (119891) 119889 (119891) 119886119899119889 119891 isin 119865

(6)

where 119871119868(119899) and 119871119874(119899) are the input link set and output linkset of node n respectively The first equation in (6) representsthe flow conservation formula of the source node whereV119891 is the data sent and sum119897isin119871119868(119899)

V119897119891 is the data successfullyreceived Similarly the second equation represents the flowconservation of the relay node Note that all the symbols ondata in this paper represent the data sentreceived within atime slot

According to (4) and (5) we can obtain the followingproperty

Property 1 (the data requirement) The total amount of datatransmitted link 119897 (119897 isin 119871119878) denoted by 119881119897119891 should meet therequirement 119881119897119891 lt 119862119897119886 The total amount of data transmittedon a set of links should meet the requirement sum119897isin119871119878119881119897119891 lt 119874119865

119886

A summary of major notations used in the paper is givenin Table 1 for easy reference

Table 1 Summary of notations

Symbol Definition120588P120588S The interference range of PUSURP RS The transmission range of PUSUCH=chj Channel set j=12 m119879119874119873119888ℎ 119879119874119865119865119888ℎ

Themean time that PU spent in ONOFFstates1120582119874119873119888ℎ The expectation of 119879119874119873119888ℎ1120582119874119865119865119888ℎ The expectation of 119879119874119865119865119888ℎ119875119888ℎ119887119906119904119910119875119888ℎ119894119889119897119890 The probability of channel ch to be busyidle

119862119867119894119895 = 119862119867119894 cap 119862119867119895

The set of common channels between node119899119894 and 119899119895119889119894119895 The Euclidean distance between node 119899119894 and119899119895119897119894119895 The link between node 119899119894 and 119899119895119901119894119895 The packet reception ratio of 119897119894119895120595119894119895 The state of 119897119894119895C119862119897119906119862119897119886 Themaximumusedavailable capacity of

link l119862119871119878119886 The available capacity of the link set LSV119891 The average size of the data flow 119891120578119898 Themaximum bits of the data flow 119891F The data flow set119881119897119891 The data flowing through link l119871119868(119899)119871119874(119899) The inputoutput link set of node n119889119903119890119897119886119910(119899119894 119899119895) The relay distance between node 119899119894 and 119899119895119873119894 The one-hop neighboring set of node 119899119894119865119894 The forwarding candidate set of node 119899119894119902119894 The data transfer rate of node 119899119894120588119898 The priority order of node 119899119898120591 The sum of 120582119874119865119865119888 and 120582119874119873119888120576 The state of channels120585 The effective forwarding rate

119875119903119890119897119886119910 (119899119895) The probability of the candidate node 119899119895successfully receiving the packet

119862119879119879119888ℎ119894 The cognitive transport throughput of theopportunistic module (119899119894 119865119894)119879119903119890119897119886119910 The time required for one-hop relayΘ The load score

4 Problem Formulation

In this section based on the system model we first describethe working process of one-hop opportunistic routing inCRAHNs and then analyze its throughput

41 Opportunistic Routing Primer We use one-hop oppor-tunistic routing to represent the process that sender node119899119894 (119899119894 isin V) sends data to its candidate node set For any node119899119895 (119899119895 isin 119865119894) in the candidate node set it should meet the

Wireless Communications and Mobile Computing 5

requirement that 119889119903119890119897119886119910(119899119894 119899119895) ge 0 where 119889119903119890119897119886119910(119899119894 119899119895) can becomputed as follows

119889119903119890119897119886119910 (119899119894 119899119895) = 119889119899119894119863 minus 119889119899119895119863 (7)

where D is the destination node 119889119899119894119863 is the Euclideandistance between 119899119894 and D and 119889119899119895119863 is the Euclidean distancebetween 119899119895 and D

Thenwe describe how to define anopportunisticmoduleFor any node 119899119894 (119899119894 isin V) we use119873119894 to represent its one-hopneighboring set Every node in 119873119894 must meet the followingconditions (1)The node operates on the same channel as thatfor node 119899119894 (2)The node is in the transmission range of node119899119894 Then a subset 119865119894 of 119873119894 can be selected as the forwardingcandidate set of node 119899119894 We use (119899119894 119873119894) and (119899119894 119865119894) to denotethe neighboring module and the opportunistic module ofnode 119899119894 respectively

Then we will use an example to illustrate the conceptof one opportunistic module The node S and D representsthe source node and the destination node respectively Thenode 119899 is the forward node and the neighbor set of 119899 isN = 1198991 1198992 1198993 1198994 1198995 In the neighboring set N nodes 11989911198992 and 1198993 meet the conditions of 119889119903119890119897119886119910(119899 119899119895) ge 0 (1 ≪j ≪ 3) Then the forwarding candidate node set of 119899 isF = 1198991 1198992 1198993 Therefore the neighboring set of 119899 is (119899119873)and the opportunistic module is (119899 119865) (see Figure 1)

In the opportunistic module every node has its relaypriority After the sender node broadcasts the packet to itsforwarding candidate set one of the candidate nodes contin-ues the forwarding process based on their relay priority Aforwarding candidate will send the message only when all thenodes with higher priorities fail to transmit Only when noneof the forwarding candidates has successfully received thepacket the senderwill retransmit the packetsThe forwardingprocess reiterates until all the packet is delivered to thedestination

In CRAHNs the process of one-hop opportunistic rout-ing is shown in Figure 2 The period of 1199051 to 1199052 is for channeldetection and the period of 1199052 to 1199053 is for data transmission

In the channel sensing step similarly to [27] the sendersearches for a temporarily unoccupied channel in collabora-tion with its neighbors using the energy detection techniqueBefore detecting the data channel the sender broadcasts ashort message (ie sensing invitation (SNSINV) in the CCC)to inform neighboring nodes of the selected data channelthe location information of the sender and the destinationand the used capacity of the output link The transmissionof SNSINV message in the CCC follows the CSMACAmechanism as specified in IEEE 80211 MAC After receivingthe SNSINV neighboring SUs set the selected data channel tobe the nonaccessible state so that no SU can transmit in thechosen data channel during the sensing period of the senderUsing the location information in SNSINV the neighboringSUs evaluate whether they are eligible relay candidates (iewhether the relay distance is nonnegative) Eligible relaycandidates will collaborate with the sender in channel sensingand data transmission In this situation other SUs cannottransmit in the selected data channel during the reservedperiod specified in SNSINV (the time for EnergyDetection in

S Dn

n5n1

n2

n3n4

Figure 1 An example of one opportunistic module (119899 119865)

Figure 2)When the channel is sensed idle (ie no PUactivityis detected) the sender and its candidate nodes will forwardthe data Otherwise the sender selects another channel andrepeats the channel sensing process

In the data transmission step the sender broadcasts thepacket to its candidate nodes The candidate nodes transmitACK messages in priority order after a shorter back-offwindow (SIFS) A candidate SU keeps listening to the datachannel until it overhears an ACK or it prepares to transmitthe ACK The waiting time required for each candidate nodeis 119906 If the sender receives no ACK message it will repeat thechannel sensing and data transmission steps

42 Throughput Bound of One Opportunistic Module In thissubsection we study the capacity region of one opportunisticmodule (119899119894 119865119894) This capacity region will serve as a bound ofthroughput corresponding to the links in the opportunisticmodule

Assume that the data transfer rate of the sender node 119899119894is 119902119894 and its candidate forwarding set is 119865119894 = 1198991 119899119898 inwhich the priority of nodes satisfies 1205881 gt gt 120588119898 We use119871119878119894 = 1198971198941 119897119894119898 to represent its forwarding candidate linkset For a candidate node 119899119895(1 le 119895 le 119898) its correspondingcandidate link is 119897119894119895 whose link state and PRR are 120595119894119895 and119901119894119895 respectively Assume that the probability of channel119888ℎ (119888ℎ120598119862) to be idle at 1199050 is 119875119888ℎ119894119889119897119890 Then we can estimate theidle probability of channel 119888ℎ at 1199051(1199051 gt 1199050) according to itsstate (busy or idle) at 1199050 which is shown as follows

119875119888ℎ119894119889119897119890 (1199051) = 119875119888ℎ119894119889119897119890 + (120576 minus 119875119888ℎ119894119889119897119890) 119890minus120591(1199051minus1199050) (8)

where 120591 = 120582119874119865119865119888 + 120582119874119873119888 and 120576 is a 0-1 variable The variable120576 indicates the state of channels where 120576 = 1 indicates thechannel 119888ℎ is idle at 1199051 and 120576 = 0 indicates the channel 119888ℎ isbusy at 1199051 Then the probability of channelrsquos idle state in theprocess of channel sensing denoted by119875119878119888ℎ119894119889119897119890 can be obtainedby

119875119878119888ℎ119894119889119897119890 = 119875119888ℎ119894119889119897119890 (1199051) ∙ 119875119888ℎ119894119889119897119890 (1199051 1199052) (9)

where119875119888ℎ119894119889119897119890(1199051 1199052) is the probability that channel stays idle overthe time (1199051 1199052) According to [16] 119875119888ℎ119894119889119897119890(1199051 1199052) can be derivedby

119875119888ℎ119894119889119897119890 (1199051 1199052) = intinfin1199052minus1199051

119865119888ℎ119874119865119865 (119909)119864 [119879119888ℎ119874119865119865]119889119909 (10)

6 Wireless Communications and Mobile Computing

One-hop relay

CCC

ch

ch

tt2t1t0

middot middot middot

PUActive

Energydetecti

on

data u ACK SNSIV SIFS

Figure 2 The process of one-hop opportunistic routing for CRAHNs

where 119865119888ℎ119874119865119865119864[119879119888ℎ119874119865119865] is the probability density function(PDF) of the residual time of the channel 119888ℎ remains idleMore specifically 119865119888ℎ119874119865119865(119909) is the cumulative distributionfunction (CDF) of the duration of the OFF state and 119864[119879119888ℎ119874119865119865]is the expected channel OFF time of the channel 119888ℎ

In the data transmission step due to the interferencesome candidate nodes cannot receive the packet successfullyWhen the SU node is disrupted by PUsrsquo appearance the linkbetween it and the sender becomes unusable Therefore thethroughput is directly related to the success rate of the linksin the opportunistic module According to the analysis resultabove we introduce the metric of effective forwarding rate(EFR) which indicates the success rate of data transmissionof each candidate link in the opportunistic module denotedby 120585 Thus for the candidate link 119897119894119895 of the candidate node119899119895(1 le 119895 le 119898) in the opportunistic module (119899119894 119865119894) theeffective forwarding rate can be denoted by 120585(119897119894119895) which isdefined in the following equation

120585 (119897119894119895) = 119902119894 ∙ 119901119894119895 ∙ 119895minus1prod119896=0

(1 minus 120595119894119896 ∙ 119901119894119896) (11)

where prod119895minus1

119896=0(1 minus 120595119894119896 ∙ 119901119894119896) is the probability that 119899119895 received

the packet successfully when all candidate nodes with higherpriority than node 119899119895 are failed to receive the packet

According to the (11) accumulating all EFR values oflinks in the candidate link set we can obtain the EFR of theopportunistic module as follows

120585 (119871119878119894) = 119898sum119895=1

120585 (119897119894119895) = 119902119894 ∙ (1 minus 119898prod119895=0

(1 minus 120593119894119895 ∙ 119901119894119895)) (12)

where prod119898119895=0(1 minus 120593119894119895 ∙ 119901119894119895) means that none of the candidate

nodes have received the packetAssume that the sender node 119899119894 chooses the channel119888ℎ as the data transmission channel Then according to

(9)-(11) we can deduce the equation for the probability of thecandidate node 119899119895 successfully receiving the packet denotedas 119875119903119890119897119886119910(119899119895) which is shown as follows

119875119903119890119897119886119910 (119899119895) = 120585 (119897119894119895) ∙ 119875119878119888ℎ119894119889119897119890 ∙ 119875119888ℎ119894119889119897119890 (1199052 1199053) (13)

According to the conclusions in [5] we can statethat when the link state is stable the channel remainsidle throughout the data forwarding process (ie 119875119878119888ℎ119894119889119897119890 ∙119875119888ℎ119894119889119897119890(1199052 1199053) = 1)

Based on (13) we can compute the cognitive transportthroughput (CTT) of one opportunistic module (119899119894 119865119894) inchannel 119888ℎ as follows

119862119879119879119888ℎ119894 = 119898sum119895=1

119875119888ℎ119903119890119897119886119910 (119899119895) ∙ ( 119871119879119903119890119897119886119910) (14)

where sum119898119895=1 119875119888ℎ119903119890119897119886119910(119899119895) is the data transmission success rate of

thewhole opportunisticmodule L is the data size transmittedby the opportunistic module and 119879119903119890119897119886119910 is the time requiredfor the whole process To facilitate the subsequent researchwe assume that the transmission time of all opportunisticmodules in the network is the same Therefore (14) can besimplified as follows

119862119879119879119888ℎ119894 = 119898sum119895=1

119875119888ℎ119903119890119897119886119910 (119899119895) ∙ 119892119871 (15)

Wireless Communications and Mobile Computing 7

43 Maximize the Total Throughput of the Network In thissubsection we formulate the optimization problem of net-work throughput as linear programming (LP) problem

P max sum119897isin119864119891isin119865

V119897119891 (16)

st V119891 + sum119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119899 = 119904 (119891) 119891 isin 119865 forall119899 isin 119881

(17)

sum119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119899 = 119904 (119891) 119899 = 119889 (119891) 119891 isin 119865 forall119899 isin 119881

(18)

sum119897isin119871119868(119899)

V119897119891 ge 0 forall119891 isin 119865 forall119899 isin 119881 (19)

sum119897isin1198710(119899)

V119897119891 = 0 119905 (119897) = 119889 (119891) forall119899 isin 119881 (20)

V119897119891 le 119862119897119886 119891 isin 119865 forall119897 isin 119864 (21)

sum119897isin119871119865

V119897119891 le 119862119871119878119886 119891 isin 119865 forall119897 isin 119871119878 (22)

sum119897isin119871119865

V119897119891 le 119862119879119879119888ℎ119894 forall119899 isin 119881 (23)

In the problem P we aim to maximize the throughput ofthe network as shown in (16) Equation (17) represents theflow conservation condition of the source and (18) representsthe flow conservation conditions of other nodes At eachnode except for the source and the destination the amountof incoming flow is equal to the amount of outgoing flowEquation (19) indicates that the amount of flow receivedby each node is nonnegative Equation (20) ensures thatthe outgoing flow from the destination of each data flowis 0 where 119905(119897) is the end node of link 119897 The physicalmeaning of (21) is that the actual flow delivered on each linkis constrained by the total amount of flow residing in thenetwork Equations (22) and (23) indicate that the amountof traffic passed through the opportunistic module is lessthan the available capacity and the throughput of the modulerespectively

In cognitive radio networks the state of the link willfrequently change due to the changing states (busy or idle)of the channel The dynamic link state may also lead tothe network topology change and make the candidate setbecome unavailable and nonoptimal Thus we cannot verifythe accuracy and effectiveness of the load balancing algo-rithm which is designed based on local network topologyinformation Moreover we cannot guarantee that the packetsare always transmitted through the links with lower trafficload Therefore a possible practical solution for this problemis distributing the traffic according to the load and thedata forwarding capacity of the link By this way we canimprove the network throughput while guaranteeing thecommunication quality

5 Opportunistic Routing

In this section we first propose a new metric to measurethe priority of candidate nodes and provide the sorting algo-rithm Then we design the load balancing based candidateforwarder selection algorithm under the stable condition ofthe link

51 Load Balancing Candidate Forwarder Sorting AlgorithmFirst we take the metric of linkrsquos reliability as an example toanalyze the reason for the uneven distribution of link loadIn the algorithms which are based on the link reliabilitywith the rise of the delivery rate of the candidate link thepriority of the corresponding candidate node will ascendBased on opportunistic routing the nodewith higher prioritywill receive more data As a result for links in the candidatelink set with the rise of the priority the load on the linkwill elevate As the amount of transmitted data increase thehigher priority linkrsquos traffic load also increases This case maycause the link blocked In the meantime the amount of dataallocated to the link with lower priority is small which maylead to the waste of link resources Therefore with the rise ofthe delivery rate of the candidate link the available capacityof the link will decline

511 Metric Design According to the consideration abovewe redesign a metric called Load Score to measure thepriority of candidate nodes based on the delivery rate andavailable capacity of each candidate link of an opportunisticmodule denoted asΘ In the opportunisticmodule (n F) F =1198991 119899119898 is the candidate node set and LS = 1198971 119897119898 isthe candidate link set For any candidate node 119899119890(1 le 119890 le 119898)the Load Score can be derived as follows

Θ119890 = 119901119890 ∙ 119862119897119890119886 (24)

The physical meaning of the metric is the maximumamount of data transmitted in a time slot when the linkhas the highest priority This metric considers both thetransmission quality of the link and load of the link Weestimate the prioritization of candidate nodes based on theLoad Score Thus it can achieve the balanced distributionof traffic load and reduce the number of packet drops andqueuing delays Meanwhile it ensures that the link transfersthe maximum amount of data Next we will analyze thespecial situations that may occur in the process of sortingcandidate nodes

According to the condition that multiple candidate nodesmay have the same Load Score we propose the followingproposition

Proposition 2 When multiple candidate nodes have thesame Load Score a higher priority candidate node is alwaysassociated with a lower packet delivery rate

Proof We assume there are 119911 candidate nodes 1198991 119899119911 inthe opportunistic module (nF) and assume they have thesame value of Load Scores denoted asΘ The correspondinglinks of nodes are denoted by 1198971 119897119911 We use 119866 to represent

8 Wireless Communications and Mobile Computing

the maximum throughput of link set 1198971 119897119911 in a steadystate Then we have

G = q ∙ 1199011 ∙ 1198741198971119886 + q ∙ 1199012 ∙ 1198621198972119886 (1 minus 1199011) + + q ∙ 119901119911

∙ 119862119897119911119886 119911minus1sum119896=0

(1 minus 119901119896)= 119902 ∙ Θ ∙ (1 + (1 minus 1199011) + + 119911minus1sum

119896=0

(1 minus 119901119896))(25)

In (25) 119902 and Θ are constant values Then the value of 119866 isdecided by 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) For conveniencewe use the notation Υ to denote the formula 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) Then the maximum value of 119866 can beachieved when Υ has the maximum value Obviously Υ isthe sum of the polynomials As a result the necessary andsufficient condition forΥ to attain the maximum value is thateach term ofΥmust reach the maximum valueThat is to say(1minus1199011) sum119911minus1

119896=0(1minus119901119896)must take the maximum value at thesame time Intuitively these items are inversely proportionalto the delivery rate of the candidate link Therefore we canconclude that the maximum Υ can be achieved when 1199011 lt1199012 lt lt 119901119911 Accordingly the condition 1198621198971119886 gt 1198621198972119886 gt gt119862119897119911119886 is valid Then Proposition 2 holds

512 Load BalancingCandidate Forwarder Sorting AlgorithmIn this subsection we will explain how to sort the candidatenodes in LS In the opportunistic module (n F) F =1198991 119899119898 is the candidate node set FL = 1198971 119897119898 isthe candidate link set and PR = 1199011 119901119898 is a PRR setEach PRR corresponds to a candidate link Based on (24) wecan calculate the Load Score of each candidate node which isdenoted as H = Θ1 Θ119898 We develop a simple selectionsorting algorithm to sort H = Θ1 Θ119898 which is shownas in Algorithm 1

In Algorithm 1 the inputs are the candidate node set119865 the PRR set 119875119877 and the LS set 119867 The algorithm sortscandidate nodes based on their LS In the algorithm line(4) is used to mark the highest priority nodes for eachiteration and lines (6) to (12) indicate the priority orderingprocess According to the process of the algorithm the timecomplexity of this algorithm is O(m2) where m representsthe number of candidate nodes

For the ordered candidate node set 1198651015840 we can calcu-late the maximum throughput of the opportunistic module(n 1198651015840) according to (11) which is shown as follows

CTT = q ∙ Θ1 + q ∙ Θ2 (1 minus 1199011) + + q

∙ Θ119898

119898minus1sum119896=0

(1 minus 119901119897119896) (26)

52 Load Balancing Based Relay Selection Algorithm In thissection we further study the load balancing solution InCRAHNs the packet will be forwarded to the destinationthrough several opportunistic modules until the packet

Input F = 1198991 119899119898 PR = 1199011 119901119898 H = Θ1 Θ119898Output An ordered candidate node set 1198651015840(1) Initialize 1198651015840 = (2) Initialize a b k(3) for eachΘ119886120598H and 1 le a le m do(4) blarr997888 a(5) for eachΘ119896120598H and a + 1 le k le m do(6) if Θ119887 lt Θ119896 then(7) blarr997888 k(8) else if Θ119887 = Θ119896 then(9) if 119875119886 gt 119875119896 then(10) blarr997888 k(11) end if(12) end if(13) end for(14) 1198651015840119886 larr997888 119865119887(15) end for(16) output 1198651015840(17) end

Algorithm 1 Load Balancing Based Candidate Forwarder SortingAlgorithm

arrives to its destination In each opportunistic module theselection of the last hop is limited by the forwarding capabilityof the next hop To achieve the global optimal it is necessaryto predict the forwarding capability of the opportunisticmodule which is the final hop Then we determine thecandidate node set of the first hop opportunistic modulenamely the optimal candidate node set for the first hopopportunistic module However this method of selectingthe optimal opportunistic module is not appropriate Inthe cognitive radio network the link status is dynamic dueto the dynamic activities of PUs Hence we cannot verifythat the prediction of the forwarding ability of the last hopopportunistic module is convergent Thus we cannot provethe optimal opportunistic module is accurate In [16] theauthor chose an intermediate destinationnode to alleviate theinfluence of the dynamic change of link state Inspired by thismethod we adopt an opportunistic module iteration methodto determine the optimal candidate node set of the currentopportunistic module Since this method is based on theforwarding ability of the next hop opportunistic module it isnecessary to predict the occupied capacity of each candidatelink in this opportunistic module

In this paper we use Minimum Mean Square Error(MMSE) [28] based flow prediction model to estimate theoccupied capacity of the link in the next time slot Theformula of the model is as follows

119862119897119906 = MMSE (119883119905 | t = 0 1 2 ) (27)

where 119883119905|t = 0 1 2 is the current flow and historicalflow statistics In [28] the author describes the predictionscheme in detail and readers can refer to it for more details

For the opportunistic module (n F) the candidate nodeset is F = 1198991 119899119898 the candidate link set is LS =1198971 119897119898 and the corresponding LS set isH = Θ1 Θ119898

Wireless Communications and Mobile Computing 9

Input F = n1 nm H = Θ1 ΘmOutput The optimal node set Foptimal(1) Initialize arrays of H1015840 F1015840 F10158401015840 Foptimal and LS1015840(2) Initialize values of o1 o2 and ctt(3) for each ne120598F and 1 le e le m do(4) Search the opportunistic module(ne Fe) of ne(5) F1015840larr997888the candidate node set of (ne Fe)(6) LS1015840larr997888the candidate link set of (ne Fe)(7) for each n1015840k120598F and 1 le k le |Fe| do(8) o1 larr997888 the used capacity of l1015840k(9) o2 larr997888 the available capacity of l1015840k(10) H1015840

k larr997888 the LS of n1015840k(11) end for(12) F10158401015840 larr997888 sorting F1015840 according to Algorithm 1(13) cttlarr997888 the throughput of (ne F10158401015840)(14) if Θe gt ctt do(15) Θe larr997888 ctt(16) end if(17) end for(18) Foptimal larr997888 sorting F according to Algorithm 1(19) output Foptimal(20) end

Algorithm 2 Load Balancing Relay Selection Algorithm

The selection steps of the optimal opportunistic module aredescribed as follows

(1) For any candidate node 119899119890 isin 119865 we first determineits opportunistic module (119899119890 119865119890) according to relay distancesThen we perform the following steps (1) predicting theoccupied capacity of each link in candidate link set 119865119871119890 basedon (27) (2) calculating the available capacity of each link in119865119871119890 based on (14) (3) computing the LS of each node in thecandidate node set 119865119890 based on (24) (4) sorting the candidatenode set 119865119890 and obtaining an ordered candidate node set 1198651015840119890according to Algorithm 1 and (5) calculating the maximumthroughput CTT((119899119890 1198651015840119890)) of the opportunistic module basedon (26)

(2)We compare the LS valueΘ119890 of candidate node 119899119890 withCTT((119899119890 1198651015840119890)) and assign a smaller value to Θ119890

(3) We sort the candidate node set 119865 based on Algo-rithm 1 and the ordered set of candidate nodes is the optimalcandidate node set

The details of the Load Balancing Relay Selection Algo-rithm are shown in Algorithm 2 In Algorithm 2 lines (1)and (2) define the relevant parameters and their initializedvalues Lines (3) to (18) show the selection process for theoptimal candidate forwarding node According to the processof Algorithm 2 and the time complexity of Algorithm 1 thetime complexity of Algorithm 2 is O(m|1198651015840|2) where m isthe number of candidate nodes of the opportunistic module(nF) and |1198651015840| is the maximum number of candidate nodesin all the next hop opportunistic modules We can selectthe optimal opportunistic module for each hop according toAlgorithm 2 It balances the traffic load of each candidate linkwhile ensuring the quality of communication Meanwhile itimproves the throughput of the local network

6 Performance Evaluation

In this section we evaluate the performance of our proposedscheme with two well-known routing protocols for CRAHNsand present the simulation results on the SU-PU interferenceratio the packet delivery ratio the throughput and the end-to-end delay under different number of flows PUsrsquo activitiesnumber of channels and number of PUs

61 Simulation Settings The network parameter settings aresummarized in Table 2 In the simulations multiple SUsand PUs are randomly deployed in a 1000m times 1000marea and they are assumed to be stationary throughout thesimulation As stated in Section 31 the PU activities of eachchannel are modeled as an exponential ON-OFF processThe status of each channel is associated with two parametersone is the expected channel OFF time 119864[119879119874119865119865119888ℎ ] and theother is the probability of the idle state 119875119888ℎ119894119889119897119890 The channelstatus is estimated by utilizing periodic spectrum sensingand on-demand sensing before data transmissions In thesimulations 30 constant bit rate (CBR) flows are generatedbetween randomly chosen source-destination pairs of SUsand they are associated with the packet size of 512 Bytesand flow rate of 5Kbps Notice that the most representativeopportunistic routing scheme ExOR [9] adopted the methodin [29] to obtain the packet reception ratio between twonodes Accordingly in this simulation the packet receptionratios of the links are also based on the distance-to-deliveryratio relationshipmeasured in [29]Thus we require the posi-tion information of different nodes to compute the distancebetween neighboring nodes and obtain the packet receptionratios of the links We evaluate our scheme through NS-2simulation [30] In NS-2 our LBOR scheme is implementedas the routing agent In this work we only focus on thenetwork layer without considering MAC and physical layerissues Since the IEEE 80211 MAC protocol is customized forCR support we use the modified 80211cr as the CSMAMACagent in our simulation script We also apply CRN patch [31]to our simulations that can support multiple channels at thephysical layer

In the simulations we compare our LBOR scheme withSAAR scheme [15] and SAORprotocol [27]TheSAORproto-col is a well-knownmultipath opportunistic routing protocolcoupled with spectrum sensing and sharing in multichannelCRAHNs The SAAR scheme is a recently published any-path opportunistic routing scheme with consideration of theunreliable transmission feature and the uncertain spectrumavailability characteristic of CRAHNs

The performance metrics for our experiments are definedas follows

(i) The throughput is defined as the total amount oftraffic (in bits per second) an SU receiver receivesfrom the sender divided by the time it takes for thereceiver to obtain the last packet A routing schemewith higher throughput is desirable for CRAHNs

(ii) The end-to-end delay is defined as the average delaywhich is calculated by summing up the end-to-enddelays of all packets received by all SU destination

10 Wireless Communications and Mobile Computing

Table 2 Simulation parameters

Parameter valueNumber of SUs 100Number of PUs 25Transmission ranges of SUsand PUs 250m

Number of channels 6Channelsrsquo idle stateprobabilities1198751198881119894119889119897119890 1198751198886119894119889119897119890 03 03 05 05 07 07Expected channel OFF time 60msTotal number of flows 30Source-destination distance 800mSU CCC rate 1MbpsSU data channel rate 2MbpsPacket size 512 BytesSensing time 3msChannel switching time 100120583sRetransmission time 3

nodes and dividing it by the total number of receivedpackets A routing scheme with the lower end-to-enddelay is preferred for CRAHNs

(iii) The SU-PU interference ratio is defined as the ratioof the number of SUsrsquo packets interrupted by PUsrsquoactivities to the total number of packets delivered byan SU source node A routing scheme with lower SU-PU interference ratio is favorable for CRAHNs

(iv) The packet delivery ratio is defined as the ratio ofall successfully received data packets that are fullyreceived by the SU destination node to the total datapackets generated by the SU source node A routingscheme with higher packet delivery ratio is desirablefor CRAHNs

To ensure the statistical significance of the results we runeach experiment for 500 seconds and repeat it 1000 timeswithdifferent seeds to report the average value as final results

62 Simulation Results

621 Impact of Number of Flows In this part we evaluate theimpact of offered traffic load on the performance by changingthe number of flows We vary the number of flows from 15CBR flows to 35 flows

As shown in Figure 3(a) we observe that LBOR provideshigher throughput than SAAR and SAOR We first considerthe case that the number of flows is below the saturationlevel Recall that the throughput defined here is measuredby the aggregate throughput of all the flows In this casewhen the number of flows increases from 15 flows to 25flows the throughput of all three schemes starts increasing asmore SU receivers are involved andmore parallel concurrenttransmissions occur Then we consider the case that whenthe number of flows is larger than 25 this indicates that

the network traffic is becoming saturated In this case thethroughput gain of SAAR and SAOR starts to decline As thenetwork resources (available channels and links) are limitedthe increasing number of flows intensifies the congestion andinterference which explains the throughput degradation ofSAAR and SAOR However when more flows are involvedthe throughput of LBOR still increases with increasingnumber of flows This is because LBOR has a load balancingfeature and solves the congestion problem by routing thetraffic through alternate forwarders

Figure 3(b) shows that LBOR has superior end-to-enddelay performance as compared to SAAR and SAOR Whenthe number of flows is equal to 15 the performance gapbetween LBOR and SAAR (both with SAOR) is small In thiscase each relay node does not need to handle toomuch trafficfor multiple flows Thus the unfair distribution of networktraffic will not lead to extra end-to-end delay Neverthelessif the number of flows is above 25 the end-to-end delayof SAAR and SAOR increase more sharply than that ofLBOR This can be attributed to the ability of LBOR that itsends packets to the links with lower traffic load and delayMoreover it can be concluded from the result that LBORfacilitates a better distribution of traffic and the consequentreduction of queuing times contributes to a lower overalldelay

622 Impact of PUsrsquo Activities In this part we evaluate theperformance of LBOR under different PUsrsquo activity patternsThe expected channel OFF time 119864[119879119874119865119865119888ℎ ] varies from 20msto 120ms A small 119864[119879119874119865119865119888ℎ ] means that the PUsrsquo activitiesare high and thus the delivery of the secondary user ismore likely to be interrupted Different from the previousexperiment the number of CBR flows is fixed to be 30

Figure 4(a) shows the SU-PU interference ratio of thethree schemes under different values of 119864[119879119874119865119865119888ℎ ] In mostcases LBOR produces significantly lower SU-PU interferenceas compared to SAAR and SAOR This is because the trafficload of LBOR is distributed efficiently and fairly In LBORfewer flows will have a chance to share the same SU nodewhich is highly influenced by PUsrsquo behaviors Moreoverwe notice that the improvement of load balancing schemeis significant under high PUsrsquo activities (119864[119879119874119865119865119888ℎ ] is below60ms) but not under low PUsrsquo activities (119864[119879119874119865119865119888ℎ ]) is above60ms) The reason behind this phenomenon is that whenthe nodes are highly influenced by PUsrsquo behaviors andoverloaded LBOR will give priorities to other nodes with alower traffic load to release the burden of these nodes

Figure 4(b) displays the throughput of LBOR SAARand SAOR with different values of 119864[119879119874119865119865119888ℎ ] It shows thatthe throughput of these schemes increase as the value of119864[119879119874119865119865119888ℎ ] becomes larger The result also shows that LBORachieves higher throughput compared to SAAR and SAORWhen the PUsrsquo activities decrease (the value of 119864[119879119874119865119865119888ℎ ]increases) the performance difference between LBOR andSAAR (both with SAOR) becomes negligible Moreover thetraffic load of LBOR is evenly distributed across the SUsand different available channels Therefore it is expected that

Wireless Communications and Mobile Computing 11

15 20 25 30 35Number of flows

170

180

190

200

210

220

230

240

250

260

270

ro

ughp

ut (K

bps)

LBORSAARSAOR(a) Throughput versus varying number of flows

15 20 25 30 35Number of flows

25

30

35

40

45

50

55

60

65

70

End-

to-e

nd d

elay

(ms)

SAORSAARLBOR(b) End-to-end delay versus varying number of flows

Figure 3

more improvements can be obtained for LBOR when PUsappear frequently

The end-to-end delay and the packet delivery ratio per-formance with different types of PUsrsquo activities are shownin Figures 4(c) and 4(d) With the increase of 119864[119879119874119865119865119888ℎ ]the packet delivery ratio will increase and the end-to-enddelay will decrease When the value of 119864[119879119874119865119865119888ℎ ] is around20ms LBOR can reduce the end-to-end delay by up to36 and 40 compared to SAAR and SAOR HoweverLBOR can only increase the delivery ratio by up to 2and 3 respectively compared to these two schemes Thereare two main reasons for this case First the opportunisticrouting protocol of LBOR is based on multipath routingstrategy where the paths and relay nodes are determinedon-the-fly In addition SAAR and SAOR also utilize thesimilar opportunistic routing principle and they aim toincrease the packet delivery ratio for the dynamic changingspectrum environment Second SAAR and SAOR alwayschoose paths or forwarding nodes with the best performanceof packet delivery ratio Nevertheless LBOR jointly considersthe load balancing for selecting the relay nodes Thus theimprovement of packet delivery ratio from LBOR is limited

623 Impact ofNumber of Channels In this part we comparethe performance of the three schemes under different numberof channels The number of channels varies between 2 and 10and the expected channel OFF time is set to be 60ms

Figure 5(a) compares the SU-PU interference ratio of thethree schemes by varying the number of channels of thenetworkWe can observe that the SU-PU interference ratio ofLBOR is lower than that of SAAR and SAOR One significantobservationhere is thatwhen the number of channels is largerthan 6 the SU-PU interference ratio of SAAR and SAOR

tends to decrease as the number of channels is increased InCRAHNs a larger number of channels can service a largernumber of SUsThus in this case SUs can choosemore stablechannels to avoid PUsrsquo interruptions In addition if an SUis interrupted by PUsrsquo appearance it can easily find anotheravailable channel for data transmissions Another importantobservation made is that when the number of channelsincreases from 2 to 6 the SU-PU interference ratio of LBORwill increase but the other two schemes will decrease Anexplanation for this phenomenon is as follows Consideringthe load balancing strategy LBOR tries its best to distributetraffic evenly among multiple nodes When there are notenough available channels to avoid the mutual interferencesamong SUs the packets cannot be correctly decoded at thereceiver Due to the same reason we can also observe fromFigures 5(b) 5(c) and 5(d) that when the number of channelsis below 6 the improvements from load balancing are notsignificantly

Figure 5(b) shows the throughput of LBOR SAAR andSAOR with different number of channels in the network Weobserve that the throughput of these schemes increases as thenumber of channels becomes larger The result also confirmsthat nomatter howmany channels are available LBOR alwaysperforms better than SAAR and SAOR When the number ofchannels is above 6 more improvement can be obtained fromLBOR

As shown in Figures 5(c) and 5(d) they provide similarresults and demonstrate that LBOR can achieve lower end-to-end delay and higher packet delivery ratio in most casescompared to SAAR and SAOR The improvements comefrom the load balancing based metric and the load balanc-ing based relay selection algorithm On the one hand theproposed scheme adopts a load balancing based metric to

12 Wireless Communications and Mobile Computing

20 40 60 80 100 120E[Tch

OFF] (ms)

0

002

004

006

008

01

012

014

016

018

02

SU-P

U in

terf

eren

ce ra

tio

LBORSAARSAOR(a) SU-PU interference ratio versus PUsrsquo activities

E[TchOFF] (ms)

20 40 60 80 100 120100

150

200

250

300

350

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus PUsrsquo activities

20 40 60 80 100 12025

30

35

40

45

50

55

60

65

End-

to-e

nd d

elay

(ms)

E[TchOFF] (ms)

LBORSAARSAOR

(c) End-to-end delay versus PUsrsquo activities

20 40 60 80 100 120076

078

08

082

084

086

088

09

092

094

Pack

et d

eliv

ery

ratio

LBORSAARSAOR

E[TchOFF] (ms)

(d) Packet delivery ratio versus PUsrsquo activities

Figure 4

choose the relay nodes with higher throughput and lowerdelay On the other hand the load balancing based relayselection algorithm can predict the occupied capacity of eachlink and thus ensure that the traffic load is evenly balancedamong all of the relay nodes

624 Impact of Number of PUs In this experiment weevaluate the performance of three schemes under differentnumber of PUs The number of PUs varies between 10 and40 and the number of channels is set to be 6 Figure 6(a)

compares the SU-PU interference ratio of different schemesunder different number of PUs In most cases the SU-PU interference ratios of SAAR and SAOR rise sharplythan LBOR That is because the network traffic of nonloadbalancing methods (SAAR and SAOR) tends to concentrateon a specific number of SUs and links If these SUsrsquo channelsare occupied by PUs most packets of the network may beaffected On the contrary LBOR distributes traffic to a largernumber of SUs thus packets have less change to be affectedby the PUsrsquo appearance and it induces a load balancing

Wireless Communications and Mobile Computing 13

2 4 6 8 10Total number of channels

002

004

006

008

01

012

014

016

018SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of channels

2 4 6 8 10Total number of channels

160

180

200

220

240

260

280

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of channels

2 4 6 8 10Total number of channels

20

30

40

50

60

70

80

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of channels

2 4 6 8 10Total number of channels

07

075

08

085

09

095Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of channels

Figure 5

improvement compared with SAAR and SAOR When thenumber of PUs is below 30 the benefit from load balancingis not surprising since SUs are not heavily affected Whenthe number of PUs is 35 an interesting observation here isthat the SU-PU interference ratio of LBOR is lower than thecase of 30 The reason for this case is that the aggregatedbenefit from load balancing scheme overwhelms the negativeeffect of the increased number of PUs More specifically inLBOR SUs are not overloaded and the packets are properlydistributed Thus when data transmission failures are causedby the appearance of PUs it is much easier for LBOR toreroute the blocked trafficflows to the light-loaded SUswhose

channels are available compared with other two methodsWhen the number of PUs is 40 we observe that the SU-PUinterference ratio of LBOR is higher than the case of 35 In thissituation there are not enough available channels for SUs totransmit the traffic of thewhole networkThus it is difficult tofind light-loaded SUs to reroute packets for overloaded SUsand more packets would be interrupted by PUsrsquo activities

Figure 6(b) shows the throughput by varying the numberof PUs in the network for the three schemes In the figurewe observe that the proposed LBOR scheme achieves higherthroughput than SAAR and SAOR We can also find thatwhen the number of PUs increases the throughput decreases

14 Wireless Communications and Mobile Computing

10 15 20 25 30 35 40Total number of PUs

003

004

005

006

007

008

009

01

011

012

013SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

160

180

200

220

240

260

280

300

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

25

30

35

40

45

50

55

60

65

70

75

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

078

08

082

084

086

088

09

092Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of PUs

Figure 6

because of available resources are decaying Moreover whenthe number of PUs is small the improvement from LBOR islimited On the contrary when the number of PUs rises above25 the throughput of SAAR and SAOR drops drastically butthe throughput degradation of LBOR is limited This is dueto the fact that LBOR can predict the availability and theoccupied capacity of each link and thus relieve the networkrsquoscongestions

Figures 6(c) and 6(d) can also confirm that LBOR canprovide the lowest end-to-end delay and the highest packetdelivery ratio among the three schemes When the numberof PUs increases the unbalanced distribution of load amongnodes becomes a critical issue and has a negative impacton the end-to-end delay and the packet delivery ratio Thiscan be understood intuitively as follows On the one handthe previous link may become unavailable due to the PU

appearance and the packets which arrived earlier have towait for late packets in reordering buffers at the receivingdestination thus resulting in an increasing queuing delayOn the other hand if late packets do not arrive within theimposed deadline it will be treated as a lost one thus resultingin an increasing packet loss

7 Conclusion

In this paper we have proposed a load balancing oppor-tunistic routing scheme for cognitive radio networks Con-sidering the loading constraint of each node we analyzedthe throughput bound of one opportunistic module andformulated the problem of maximizing the total throughputof the network as a linear programming problem Moreoverwe designed a new metric to select and prioritize the

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

Wireless Communications and Mobile Computing 3

to take the problem of load balancing into account whichcauses the network prone to block

For the design of load balancing in wireless networksmany scholars have made some achievements The OELR[22] algorithm is used to solve the problem of networkservice life due to the excessive load of some nodes in mobilead hoc networks Although the algorithm can reduce theenergy consumption and ensure the reliability of the datatransmission it can only achieve load balancing for the localnetwork The ORPL-LB [23] algorithm is used to solve theinfluence of node active time on network load and energyconsumption in a wireless sensor network To achieve thebalance of energy consumption the algorithm firstly adjuststhe active time of the node according to the energy andthe flow Then it regulates the current working cycle of thenode based on the working period of the target The LBIA[24] algorithm is used to solve the problem of networkcongestion caused by the excessive load of some nodes Inthis algorithm in addition to load balancing the interdomaindata flow interference and domain data flow interferenceare considered So et al [25] proposed a load balancingopportunistic routing algorithm for wireless sensor networksThis opportunistic routing scheme designs a new forwardnode selectionmetric based on the residual energy of networknodes Xu et al [26] proposed a dynamic resource allocationmethod named DRAM which is designed through staticresource allocation and dynamic service migration to achievethe load balance for the fog computing systems Howeverthe existing network load balancing algorithms cannot bedirectly applied in CRAHNs

To the best of our knowledge jointly considering load bal-ancing and opportunistic routing is novel in CRAHNs Firstthe existing load balancing schemes [22ndash26] are designed fortraditional wireless networks and cannot be directly appliedin CRAHNS due to the varying availability of spectrumsSecond the forwarder set selection metric of existing oppor-tunistic routing schemes [12ndash18] only consider the routingparameters such as the end-to-end delay the throughput andthe number of transmissions Different from these worksour metric is load-dependent and takes care of the linkquality as well Most importantly our candidate forwarderselection and ordering algorithms can bypass blocked linksand distribute the excessive loads of high loaded nodes tonearby nodes

3 System Model

31 Network Model We consider a dense-scaling secondarynetwork (SN) coexisting with a primary network (PN)deployed in a specific area The PN consists of M Poissondistributed primary users We assume that the interferencerange of PU is 120588119875 If a PU is active and its transmissionfrequency overlaps an SUrsquos channel the SU which is insidethe interference range of the PU is not permitted to transmitat this time Meanwhile the transmission range of PU isassumed to be 119877119875 (without loss of generality 120588119875 gt 119877119875)The SN is composed of N SUs who know their locationsWe assume SUs are static or quasi-static Similar to PUsrsquonetwork the transmission and interference ranges of SUs

are assumed to be 119877119878 and 120588119878 respectively (without loss ofgenerality 120588119878 gt 119877119878) In our proposed scheme we need tocompute the distance between neighboring nodes and thedestination Thus it is essential to assume that all SUs canacquire their own location information by equipping an on-board localization device whose accuracy is around 5meters

In our system multiple PUs and SUs share a set oforthogonal channels denoted by 119862119867 = 119888ℎ1 119888ℎ2 119888ℎ119898SUs can exchange messages over a common control channel(CCC) Each SU is equipped with two radios one half-duplex cognitive radio that can switch among CH for datatransmissions and the other half-duplex normal radio inCCCfor exchanging control messages Note that PUs have therights of accessing the licensed bands for communicationswhile SUs can only opportunistically access the spectrum forad hoc device-to-device transmissions We model the occu-pation time of PUs in each data channel as an independentand identically distributed alternating ON (PU is active) andOFF (PU is inactive) process For ease of simulation andexplanation we assume that the mean time spent on bothON and OFF states is 119879119874119873119888ℎ and 119879119874119865119865119888ℎ respectively and theyfollow an exponential distribution with expectation 1120582119874119873119888ℎand 1120582119874119865119865119888ℎ Then we can obtain the probability of channelch to be busy (ON) and idle (OFF) respectively as follows

119875119888ℎ119887119906119904119910 = 120582119874119865119865119888ℎ120582119874119873119888ℎ

+ 120582119874119865119865119888ℎ

(1)

119875119888ℎ119894119889119897119890 = 120582119874119873119888ℎ120582119874119873119888ℎ

+ 120582119874119865119865119888ℎ

(2)

32 Traffic Model In this model each node 119899119894 can sense itsenvironment and find a set of available frequency bands (iethose bands that are not currently used by primary users)Weuse 119862119867119894119895 = 119862119867119894 cap 119862119867119895 to denote the set of channels thatare overlapped between node 119899119894 and node 119899119895 which can beused for communications Besides we use 119889119894119895 to represent theEuclidean distance between two nodes We assume that SUscan communicate with other nodes within their transmissionranges when they have common channels (ie there exists alinkedge 119897 between SUs) We use 119901119894119895 to represent the packetreception ratio (PRR) of the link 119897119894119895 between 119899119894 and 119899119895 We usean indicator 120595119894119895 to determine the state of the link 119897119894119895 between119899119894 and 119899119895 There exists a link between 119899119894 and 119899119895 if 119862119867119894119895 = 0and 119889119894119895 lt 119877119878 (ie 120595119894119895 = 1 otherwise 120595119894119895 = 0)

120595119894119895 = 1 119862119867119894119895 = 0 119886119899119889 119889119894119895 lt 1198771198780 otherwise (3)

Thus the SN can bemodeled as a graphG = (VE) whereV is the set of nodes (SUs) and E is the set of all the possiblelinks formed by nodes in V We divide the network time intoseveral time slotswith the length of120596Weuse C119862119897119906 and119862119897119886 torepresent the maximum capacity the used capacity and theavailable capacity of the link 119897 at the beginning of each timeslot119862119897119906 can be obtained through the exchange of informationbetween nodesMeanwhile119862119897119886 can be calculated according to

4 Wireless Communications and Mobile Computing

the used capacity (119862119897119906) and the maximum capacity (C) whichis shown as follows

119862119897119886 = C minus 119862119897119906 119862119897119906 lt C

0 119862119897119880 ge C (4)

where the first equation of (4) is adopted when the link is notoverloaded and the second is applied to the case that the linkis overloaded The available capacity of a whole link set canbe derived according to the available capacity of each link inthe set For a link set LS we denote the available capacity ofLS as 119862119871119878119886 Then we have

119862119871119878119886 = sum119897isin119871119878

119862119897119886 (5)

wheresum119897isin119871119878 119862119897119886 represents the sum of the available capacity ofeach link

33 Flow Model We assume that a data flow 119891 is composedof all the packets sent by a node at a time slot where 120578119898 isthe maximum bits of the data flow The average size of 119891 isV119891 (V119891120598(0 120578119898])We use ϝ to represent the data flow set whichincludes all the end-to-end data flows in the network For adata flow119891 (119891120598ϝ) we use s(119891) and d(119891) to represent its sourcenode and destination node respectively For a link 119897 (119897 isin 119864)we use V119897119891 to represent the average size of the data flow 119891transmitted among the link 119897 at a time slot Then accordingto the flow conservation for each node n (n isin V) we haveV119891 + sum

119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891 119894119891 119899 = 119904 (119891) 119886119899119889 119891 isin 119865sum

119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119894119891 119899 = 119904 (119891) 119889 (119891) 119886119899119889 119891 isin 119865

(6)

where 119871119868(119899) and 119871119874(119899) are the input link set and output linkset of node n respectively The first equation in (6) representsthe flow conservation formula of the source node whereV119891 is the data sent and sum119897isin119871119868(119899)

V119897119891 is the data successfullyreceived Similarly the second equation represents the flowconservation of the relay node Note that all the symbols ondata in this paper represent the data sentreceived within atime slot

According to (4) and (5) we can obtain the followingproperty

Property 1 (the data requirement) The total amount of datatransmitted link 119897 (119897 isin 119871119878) denoted by 119881119897119891 should meet therequirement 119881119897119891 lt 119862119897119886 The total amount of data transmittedon a set of links should meet the requirement sum119897isin119871119878119881119897119891 lt 119874119865

119886

A summary of major notations used in the paper is givenin Table 1 for easy reference

Table 1 Summary of notations

Symbol Definition120588P120588S The interference range of PUSURP RS The transmission range of PUSUCH=chj Channel set j=12 m119879119874119873119888ℎ 119879119874119865119865119888ℎ

Themean time that PU spent in ONOFFstates1120582119874119873119888ℎ The expectation of 119879119874119873119888ℎ1120582119874119865119865119888ℎ The expectation of 119879119874119865119865119888ℎ119875119888ℎ119887119906119904119910119875119888ℎ119894119889119897119890 The probability of channel ch to be busyidle

119862119867119894119895 = 119862119867119894 cap 119862119867119895

The set of common channels between node119899119894 and 119899119895119889119894119895 The Euclidean distance between node 119899119894 and119899119895119897119894119895 The link between node 119899119894 and 119899119895119901119894119895 The packet reception ratio of 119897119894119895120595119894119895 The state of 119897119894119895C119862119897119906119862119897119886 Themaximumusedavailable capacity of

link l119862119871119878119886 The available capacity of the link set LSV119891 The average size of the data flow 119891120578119898 Themaximum bits of the data flow 119891F The data flow set119881119897119891 The data flowing through link l119871119868(119899)119871119874(119899) The inputoutput link set of node n119889119903119890119897119886119910(119899119894 119899119895) The relay distance between node 119899119894 and 119899119895119873119894 The one-hop neighboring set of node 119899119894119865119894 The forwarding candidate set of node 119899119894119902119894 The data transfer rate of node 119899119894120588119898 The priority order of node 119899119898120591 The sum of 120582119874119865119865119888 and 120582119874119873119888120576 The state of channels120585 The effective forwarding rate

119875119903119890119897119886119910 (119899119895) The probability of the candidate node 119899119895successfully receiving the packet

119862119879119879119888ℎ119894 The cognitive transport throughput of theopportunistic module (119899119894 119865119894)119879119903119890119897119886119910 The time required for one-hop relayΘ The load score

4 Problem Formulation

In this section based on the system model we first describethe working process of one-hop opportunistic routing inCRAHNs and then analyze its throughput

41 Opportunistic Routing Primer We use one-hop oppor-tunistic routing to represent the process that sender node119899119894 (119899119894 isin V) sends data to its candidate node set For any node119899119895 (119899119895 isin 119865119894) in the candidate node set it should meet the

Wireless Communications and Mobile Computing 5

requirement that 119889119903119890119897119886119910(119899119894 119899119895) ge 0 where 119889119903119890119897119886119910(119899119894 119899119895) can becomputed as follows

119889119903119890119897119886119910 (119899119894 119899119895) = 119889119899119894119863 minus 119889119899119895119863 (7)

where D is the destination node 119889119899119894119863 is the Euclideandistance between 119899119894 and D and 119889119899119895119863 is the Euclidean distancebetween 119899119895 and D

Thenwe describe how to define anopportunisticmoduleFor any node 119899119894 (119899119894 isin V) we use119873119894 to represent its one-hopneighboring set Every node in 119873119894 must meet the followingconditions (1)The node operates on the same channel as thatfor node 119899119894 (2)The node is in the transmission range of node119899119894 Then a subset 119865119894 of 119873119894 can be selected as the forwardingcandidate set of node 119899119894 We use (119899119894 119873119894) and (119899119894 119865119894) to denotethe neighboring module and the opportunistic module ofnode 119899119894 respectively

Then we will use an example to illustrate the conceptof one opportunistic module The node S and D representsthe source node and the destination node respectively Thenode 119899 is the forward node and the neighbor set of 119899 isN = 1198991 1198992 1198993 1198994 1198995 In the neighboring set N nodes 11989911198992 and 1198993 meet the conditions of 119889119903119890119897119886119910(119899 119899119895) ge 0 (1 ≪j ≪ 3) Then the forwarding candidate node set of 119899 isF = 1198991 1198992 1198993 Therefore the neighboring set of 119899 is (119899119873)and the opportunistic module is (119899 119865) (see Figure 1)

In the opportunistic module every node has its relaypriority After the sender node broadcasts the packet to itsforwarding candidate set one of the candidate nodes contin-ues the forwarding process based on their relay priority Aforwarding candidate will send the message only when all thenodes with higher priorities fail to transmit Only when noneof the forwarding candidates has successfully received thepacket the senderwill retransmit the packetsThe forwardingprocess reiterates until all the packet is delivered to thedestination

In CRAHNs the process of one-hop opportunistic rout-ing is shown in Figure 2 The period of 1199051 to 1199052 is for channeldetection and the period of 1199052 to 1199053 is for data transmission

In the channel sensing step similarly to [27] the sendersearches for a temporarily unoccupied channel in collabora-tion with its neighbors using the energy detection techniqueBefore detecting the data channel the sender broadcasts ashort message (ie sensing invitation (SNSINV) in the CCC)to inform neighboring nodes of the selected data channelthe location information of the sender and the destinationand the used capacity of the output link The transmissionof SNSINV message in the CCC follows the CSMACAmechanism as specified in IEEE 80211 MAC After receivingthe SNSINV neighboring SUs set the selected data channel tobe the nonaccessible state so that no SU can transmit in thechosen data channel during the sensing period of the senderUsing the location information in SNSINV the neighboringSUs evaluate whether they are eligible relay candidates (iewhether the relay distance is nonnegative) Eligible relaycandidates will collaborate with the sender in channel sensingand data transmission In this situation other SUs cannottransmit in the selected data channel during the reservedperiod specified in SNSINV (the time for EnergyDetection in

S Dn

n5n1

n2

n3n4

Figure 1 An example of one opportunistic module (119899 119865)

Figure 2)When the channel is sensed idle (ie no PUactivityis detected) the sender and its candidate nodes will forwardthe data Otherwise the sender selects another channel andrepeats the channel sensing process

In the data transmission step the sender broadcasts thepacket to its candidate nodes The candidate nodes transmitACK messages in priority order after a shorter back-offwindow (SIFS) A candidate SU keeps listening to the datachannel until it overhears an ACK or it prepares to transmitthe ACK The waiting time required for each candidate nodeis 119906 If the sender receives no ACK message it will repeat thechannel sensing and data transmission steps

42 Throughput Bound of One Opportunistic Module In thissubsection we study the capacity region of one opportunisticmodule (119899119894 119865119894) This capacity region will serve as a bound ofthroughput corresponding to the links in the opportunisticmodule

Assume that the data transfer rate of the sender node 119899119894is 119902119894 and its candidate forwarding set is 119865119894 = 1198991 119899119898 inwhich the priority of nodes satisfies 1205881 gt gt 120588119898 We use119871119878119894 = 1198971198941 119897119894119898 to represent its forwarding candidate linkset For a candidate node 119899119895(1 le 119895 le 119898) its correspondingcandidate link is 119897119894119895 whose link state and PRR are 120595119894119895 and119901119894119895 respectively Assume that the probability of channel119888ℎ (119888ℎ120598119862) to be idle at 1199050 is 119875119888ℎ119894119889119897119890 Then we can estimate theidle probability of channel 119888ℎ at 1199051(1199051 gt 1199050) according to itsstate (busy or idle) at 1199050 which is shown as follows

119875119888ℎ119894119889119897119890 (1199051) = 119875119888ℎ119894119889119897119890 + (120576 minus 119875119888ℎ119894119889119897119890) 119890minus120591(1199051minus1199050) (8)

where 120591 = 120582119874119865119865119888 + 120582119874119873119888 and 120576 is a 0-1 variable The variable120576 indicates the state of channels where 120576 = 1 indicates thechannel 119888ℎ is idle at 1199051 and 120576 = 0 indicates the channel 119888ℎ isbusy at 1199051 Then the probability of channelrsquos idle state in theprocess of channel sensing denoted by119875119878119888ℎ119894119889119897119890 can be obtainedby

119875119878119888ℎ119894119889119897119890 = 119875119888ℎ119894119889119897119890 (1199051) ∙ 119875119888ℎ119894119889119897119890 (1199051 1199052) (9)

where119875119888ℎ119894119889119897119890(1199051 1199052) is the probability that channel stays idle overthe time (1199051 1199052) According to [16] 119875119888ℎ119894119889119897119890(1199051 1199052) can be derivedby

119875119888ℎ119894119889119897119890 (1199051 1199052) = intinfin1199052minus1199051

119865119888ℎ119874119865119865 (119909)119864 [119879119888ℎ119874119865119865]119889119909 (10)

6 Wireless Communications and Mobile Computing

One-hop relay

CCC

ch

ch

tt2t1t0

middot middot middot

PUActive

Energydetecti

on

data u ACK SNSIV SIFS

Figure 2 The process of one-hop opportunistic routing for CRAHNs

where 119865119888ℎ119874119865119865119864[119879119888ℎ119874119865119865] is the probability density function(PDF) of the residual time of the channel 119888ℎ remains idleMore specifically 119865119888ℎ119874119865119865(119909) is the cumulative distributionfunction (CDF) of the duration of the OFF state and 119864[119879119888ℎ119874119865119865]is the expected channel OFF time of the channel 119888ℎ

In the data transmission step due to the interferencesome candidate nodes cannot receive the packet successfullyWhen the SU node is disrupted by PUsrsquo appearance the linkbetween it and the sender becomes unusable Therefore thethroughput is directly related to the success rate of the linksin the opportunistic module According to the analysis resultabove we introduce the metric of effective forwarding rate(EFR) which indicates the success rate of data transmissionof each candidate link in the opportunistic module denotedby 120585 Thus for the candidate link 119897119894119895 of the candidate node119899119895(1 le 119895 le 119898) in the opportunistic module (119899119894 119865119894) theeffective forwarding rate can be denoted by 120585(119897119894119895) which isdefined in the following equation

120585 (119897119894119895) = 119902119894 ∙ 119901119894119895 ∙ 119895minus1prod119896=0

(1 minus 120595119894119896 ∙ 119901119894119896) (11)

where prod119895minus1

119896=0(1 minus 120595119894119896 ∙ 119901119894119896) is the probability that 119899119895 received

the packet successfully when all candidate nodes with higherpriority than node 119899119895 are failed to receive the packet

According to the (11) accumulating all EFR values oflinks in the candidate link set we can obtain the EFR of theopportunistic module as follows

120585 (119871119878119894) = 119898sum119895=1

120585 (119897119894119895) = 119902119894 ∙ (1 minus 119898prod119895=0

(1 minus 120593119894119895 ∙ 119901119894119895)) (12)

where prod119898119895=0(1 minus 120593119894119895 ∙ 119901119894119895) means that none of the candidate

nodes have received the packetAssume that the sender node 119899119894 chooses the channel119888ℎ as the data transmission channel Then according to

(9)-(11) we can deduce the equation for the probability of thecandidate node 119899119895 successfully receiving the packet denotedas 119875119903119890119897119886119910(119899119895) which is shown as follows

119875119903119890119897119886119910 (119899119895) = 120585 (119897119894119895) ∙ 119875119878119888ℎ119894119889119897119890 ∙ 119875119888ℎ119894119889119897119890 (1199052 1199053) (13)

According to the conclusions in [5] we can statethat when the link state is stable the channel remainsidle throughout the data forwarding process (ie 119875119878119888ℎ119894119889119897119890 ∙119875119888ℎ119894119889119897119890(1199052 1199053) = 1)

Based on (13) we can compute the cognitive transportthroughput (CTT) of one opportunistic module (119899119894 119865119894) inchannel 119888ℎ as follows

119862119879119879119888ℎ119894 = 119898sum119895=1

119875119888ℎ119903119890119897119886119910 (119899119895) ∙ ( 119871119879119903119890119897119886119910) (14)

where sum119898119895=1 119875119888ℎ119903119890119897119886119910(119899119895) is the data transmission success rate of

thewhole opportunisticmodule L is the data size transmittedby the opportunistic module and 119879119903119890119897119886119910 is the time requiredfor the whole process To facilitate the subsequent researchwe assume that the transmission time of all opportunisticmodules in the network is the same Therefore (14) can besimplified as follows

119862119879119879119888ℎ119894 = 119898sum119895=1

119875119888ℎ119903119890119897119886119910 (119899119895) ∙ 119892119871 (15)

Wireless Communications and Mobile Computing 7

43 Maximize the Total Throughput of the Network In thissubsection we formulate the optimization problem of net-work throughput as linear programming (LP) problem

P max sum119897isin119864119891isin119865

V119897119891 (16)

st V119891 + sum119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119899 = 119904 (119891) 119891 isin 119865 forall119899 isin 119881

(17)

sum119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119899 = 119904 (119891) 119899 = 119889 (119891) 119891 isin 119865 forall119899 isin 119881

(18)

sum119897isin119871119868(119899)

V119897119891 ge 0 forall119891 isin 119865 forall119899 isin 119881 (19)

sum119897isin1198710(119899)

V119897119891 = 0 119905 (119897) = 119889 (119891) forall119899 isin 119881 (20)

V119897119891 le 119862119897119886 119891 isin 119865 forall119897 isin 119864 (21)

sum119897isin119871119865

V119897119891 le 119862119871119878119886 119891 isin 119865 forall119897 isin 119871119878 (22)

sum119897isin119871119865

V119897119891 le 119862119879119879119888ℎ119894 forall119899 isin 119881 (23)

In the problem P we aim to maximize the throughput ofthe network as shown in (16) Equation (17) represents theflow conservation condition of the source and (18) representsthe flow conservation conditions of other nodes At eachnode except for the source and the destination the amountof incoming flow is equal to the amount of outgoing flowEquation (19) indicates that the amount of flow receivedby each node is nonnegative Equation (20) ensures thatthe outgoing flow from the destination of each data flowis 0 where 119905(119897) is the end node of link 119897 The physicalmeaning of (21) is that the actual flow delivered on each linkis constrained by the total amount of flow residing in thenetwork Equations (22) and (23) indicate that the amountof traffic passed through the opportunistic module is lessthan the available capacity and the throughput of the modulerespectively

In cognitive radio networks the state of the link willfrequently change due to the changing states (busy or idle)of the channel The dynamic link state may also lead tothe network topology change and make the candidate setbecome unavailable and nonoptimal Thus we cannot verifythe accuracy and effectiveness of the load balancing algo-rithm which is designed based on local network topologyinformation Moreover we cannot guarantee that the packetsare always transmitted through the links with lower trafficload Therefore a possible practical solution for this problemis distributing the traffic according to the load and thedata forwarding capacity of the link By this way we canimprove the network throughput while guaranteeing thecommunication quality

5 Opportunistic Routing

In this section we first propose a new metric to measurethe priority of candidate nodes and provide the sorting algo-rithm Then we design the load balancing based candidateforwarder selection algorithm under the stable condition ofthe link

51 Load Balancing Candidate Forwarder Sorting AlgorithmFirst we take the metric of linkrsquos reliability as an example toanalyze the reason for the uneven distribution of link loadIn the algorithms which are based on the link reliabilitywith the rise of the delivery rate of the candidate link thepriority of the corresponding candidate node will ascendBased on opportunistic routing the nodewith higher prioritywill receive more data As a result for links in the candidatelink set with the rise of the priority the load on the linkwill elevate As the amount of transmitted data increase thehigher priority linkrsquos traffic load also increases This case maycause the link blocked In the meantime the amount of dataallocated to the link with lower priority is small which maylead to the waste of link resources Therefore with the rise ofthe delivery rate of the candidate link the available capacityof the link will decline

511 Metric Design According to the consideration abovewe redesign a metric called Load Score to measure thepriority of candidate nodes based on the delivery rate andavailable capacity of each candidate link of an opportunisticmodule denoted asΘ In the opportunisticmodule (n F) F =1198991 119899119898 is the candidate node set and LS = 1198971 119897119898 isthe candidate link set For any candidate node 119899119890(1 le 119890 le 119898)the Load Score can be derived as follows

Θ119890 = 119901119890 ∙ 119862119897119890119886 (24)

The physical meaning of the metric is the maximumamount of data transmitted in a time slot when the linkhas the highest priority This metric considers both thetransmission quality of the link and load of the link Weestimate the prioritization of candidate nodes based on theLoad Score Thus it can achieve the balanced distributionof traffic load and reduce the number of packet drops andqueuing delays Meanwhile it ensures that the link transfersthe maximum amount of data Next we will analyze thespecial situations that may occur in the process of sortingcandidate nodes

According to the condition that multiple candidate nodesmay have the same Load Score we propose the followingproposition

Proposition 2 When multiple candidate nodes have thesame Load Score a higher priority candidate node is alwaysassociated with a lower packet delivery rate

Proof We assume there are 119911 candidate nodes 1198991 119899119911 inthe opportunistic module (nF) and assume they have thesame value of Load Scores denoted asΘ The correspondinglinks of nodes are denoted by 1198971 119897119911 We use 119866 to represent

8 Wireless Communications and Mobile Computing

the maximum throughput of link set 1198971 119897119911 in a steadystate Then we have

G = q ∙ 1199011 ∙ 1198741198971119886 + q ∙ 1199012 ∙ 1198621198972119886 (1 minus 1199011) + + q ∙ 119901119911

∙ 119862119897119911119886 119911minus1sum119896=0

(1 minus 119901119896)= 119902 ∙ Θ ∙ (1 + (1 minus 1199011) + + 119911minus1sum

119896=0

(1 minus 119901119896))(25)

In (25) 119902 and Θ are constant values Then the value of 119866 isdecided by 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) For conveniencewe use the notation Υ to denote the formula 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) Then the maximum value of 119866 can beachieved when Υ has the maximum value Obviously Υ isthe sum of the polynomials As a result the necessary andsufficient condition forΥ to attain the maximum value is thateach term ofΥmust reach the maximum valueThat is to say(1minus1199011) sum119911minus1

119896=0(1minus119901119896)must take the maximum value at thesame time Intuitively these items are inversely proportionalto the delivery rate of the candidate link Therefore we canconclude that the maximum Υ can be achieved when 1199011 lt1199012 lt lt 119901119911 Accordingly the condition 1198621198971119886 gt 1198621198972119886 gt gt119862119897119911119886 is valid Then Proposition 2 holds

512 Load BalancingCandidate Forwarder Sorting AlgorithmIn this subsection we will explain how to sort the candidatenodes in LS In the opportunistic module (n F) F =1198991 119899119898 is the candidate node set FL = 1198971 119897119898 isthe candidate link set and PR = 1199011 119901119898 is a PRR setEach PRR corresponds to a candidate link Based on (24) wecan calculate the Load Score of each candidate node which isdenoted as H = Θ1 Θ119898 We develop a simple selectionsorting algorithm to sort H = Θ1 Θ119898 which is shownas in Algorithm 1

In Algorithm 1 the inputs are the candidate node set119865 the PRR set 119875119877 and the LS set 119867 The algorithm sortscandidate nodes based on their LS In the algorithm line(4) is used to mark the highest priority nodes for eachiteration and lines (6) to (12) indicate the priority orderingprocess According to the process of the algorithm the timecomplexity of this algorithm is O(m2) where m representsthe number of candidate nodes

For the ordered candidate node set 1198651015840 we can calcu-late the maximum throughput of the opportunistic module(n 1198651015840) according to (11) which is shown as follows

CTT = q ∙ Θ1 + q ∙ Θ2 (1 minus 1199011) + + q

∙ Θ119898

119898minus1sum119896=0

(1 minus 119901119897119896) (26)

52 Load Balancing Based Relay Selection Algorithm In thissection we further study the load balancing solution InCRAHNs the packet will be forwarded to the destinationthrough several opportunistic modules until the packet

Input F = 1198991 119899119898 PR = 1199011 119901119898 H = Θ1 Θ119898Output An ordered candidate node set 1198651015840(1) Initialize 1198651015840 = (2) Initialize a b k(3) for eachΘ119886120598H and 1 le a le m do(4) blarr997888 a(5) for eachΘ119896120598H and a + 1 le k le m do(6) if Θ119887 lt Θ119896 then(7) blarr997888 k(8) else if Θ119887 = Θ119896 then(9) if 119875119886 gt 119875119896 then(10) blarr997888 k(11) end if(12) end if(13) end for(14) 1198651015840119886 larr997888 119865119887(15) end for(16) output 1198651015840(17) end

Algorithm 1 Load Balancing Based Candidate Forwarder SortingAlgorithm

arrives to its destination In each opportunistic module theselection of the last hop is limited by the forwarding capabilityof the next hop To achieve the global optimal it is necessaryto predict the forwarding capability of the opportunisticmodule which is the final hop Then we determine thecandidate node set of the first hop opportunistic modulenamely the optimal candidate node set for the first hopopportunistic module However this method of selectingthe optimal opportunistic module is not appropriate Inthe cognitive radio network the link status is dynamic dueto the dynamic activities of PUs Hence we cannot verifythat the prediction of the forwarding ability of the last hopopportunistic module is convergent Thus we cannot provethe optimal opportunistic module is accurate In [16] theauthor chose an intermediate destinationnode to alleviate theinfluence of the dynamic change of link state Inspired by thismethod we adopt an opportunistic module iteration methodto determine the optimal candidate node set of the currentopportunistic module Since this method is based on theforwarding ability of the next hop opportunistic module it isnecessary to predict the occupied capacity of each candidatelink in this opportunistic module

In this paper we use Minimum Mean Square Error(MMSE) [28] based flow prediction model to estimate theoccupied capacity of the link in the next time slot Theformula of the model is as follows

119862119897119906 = MMSE (119883119905 | t = 0 1 2 ) (27)

where 119883119905|t = 0 1 2 is the current flow and historicalflow statistics In [28] the author describes the predictionscheme in detail and readers can refer to it for more details

For the opportunistic module (n F) the candidate nodeset is F = 1198991 119899119898 the candidate link set is LS =1198971 119897119898 and the corresponding LS set isH = Θ1 Θ119898

Wireless Communications and Mobile Computing 9

Input F = n1 nm H = Θ1 ΘmOutput The optimal node set Foptimal(1) Initialize arrays of H1015840 F1015840 F10158401015840 Foptimal and LS1015840(2) Initialize values of o1 o2 and ctt(3) for each ne120598F and 1 le e le m do(4) Search the opportunistic module(ne Fe) of ne(5) F1015840larr997888the candidate node set of (ne Fe)(6) LS1015840larr997888the candidate link set of (ne Fe)(7) for each n1015840k120598F and 1 le k le |Fe| do(8) o1 larr997888 the used capacity of l1015840k(9) o2 larr997888 the available capacity of l1015840k(10) H1015840

k larr997888 the LS of n1015840k(11) end for(12) F10158401015840 larr997888 sorting F1015840 according to Algorithm 1(13) cttlarr997888 the throughput of (ne F10158401015840)(14) if Θe gt ctt do(15) Θe larr997888 ctt(16) end if(17) end for(18) Foptimal larr997888 sorting F according to Algorithm 1(19) output Foptimal(20) end

Algorithm 2 Load Balancing Relay Selection Algorithm

The selection steps of the optimal opportunistic module aredescribed as follows

(1) For any candidate node 119899119890 isin 119865 we first determineits opportunistic module (119899119890 119865119890) according to relay distancesThen we perform the following steps (1) predicting theoccupied capacity of each link in candidate link set 119865119871119890 basedon (27) (2) calculating the available capacity of each link in119865119871119890 based on (14) (3) computing the LS of each node in thecandidate node set 119865119890 based on (24) (4) sorting the candidatenode set 119865119890 and obtaining an ordered candidate node set 1198651015840119890according to Algorithm 1 and (5) calculating the maximumthroughput CTT((119899119890 1198651015840119890)) of the opportunistic module basedon (26)

(2)We compare the LS valueΘ119890 of candidate node 119899119890 withCTT((119899119890 1198651015840119890)) and assign a smaller value to Θ119890

(3) We sort the candidate node set 119865 based on Algo-rithm 1 and the ordered set of candidate nodes is the optimalcandidate node set

The details of the Load Balancing Relay Selection Algo-rithm are shown in Algorithm 2 In Algorithm 2 lines (1)and (2) define the relevant parameters and their initializedvalues Lines (3) to (18) show the selection process for theoptimal candidate forwarding node According to the processof Algorithm 2 and the time complexity of Algorithm 1 thetime complexity of Algorithm 2 is O(m|1198651015840|2) where m isthe number of candidate nodes of the opportunistic module(nF) and |1198651015840| is the maximum number of candidate nodesin all the next hop opportunistic modules We can selectthe optimal opportunistic module for each hop according toAlgorithm 2 It balances the traffic load of each candidate linkwhile ensuring the quality of communication Meanwhile itimproves the throughput of the local network

6 Performance Evaluation

In this section we evaluate the performance of our proposedscheme with two well-known routing protocols for CRAHNsand present the simulation results on the SU-PU interferenceratio the packet delivery ratio the throughput and the end-to-end delay under different number of flows PUsrsquo activitiesnumber of channels and number of PUs

61 Simulation Settings The network parameter settings aresummarized in Table 2 In the simulations multiple SUsand PUs are randomly deployed in a 1000m times 1000marea and they are assumed to be stationary throughout thesimulation As stated in Section 31 the PU activities of eachchannel are modeled as an exponential ON-OFF processThe status of each channel is associated with two parametersone is the expected channel OFF time 119864[119879119874119865119865119888ℎ ] and theother is the probability of the idle state 119875119888ℎ119894119889119897119890 The channelstatus is estimated by utilizing periodic spectrum sensingand on-demand sensing before data transmissions In thesimulations 30 constant bit rate (CBR) flows are generatedbetween randomly chosen source-destination pairs of SUsand they are associated with the packet size of 512 Bytesand flow rate of 5Kbps Notice that the most representativeopportunistic routing scheme ExOR [9] adopted the methodin [29] to obtain the packet reception ratio between twonodes Accordingly in this simulation the packet receptionratios of the links are also based on the distance-to-deliveryratio relationshipmeasured in [29]Thus we require the posi-tion information of different nodes to compute the distancebetween neighboring nodes and obtain the packet receptionratios of the links We evaluate our scheme through NS-2simulation [30] In NS-2 our LBOR scheme is implementedas the routing agent In this work we only focus on thenetwork layer without considering MAC and physical layerissues Since the IEEE 80211 MAC protocol is customized forCR support we use the modified 80211cr as the CSMAMACagent in our simulation script We also apply CRN patch [31]to our simulations that can support multiple channels at thephysical layer

In the simulations we compare our LBOR scheme withSAAR scheme [15] and SAORprotocol [27]TheSAORproto-col is a well-knownmultipath opportunistic routing protocolcoupled with spectrum sensing and sharing in multichannelCRAHNs The SAAR scheme is a recently published any-path opportunistic routing scheme with consideration of theunreliable transmission feature and the uncertain spectrumavailability characteristic of CRAHNs

The performance metrics for our experiments are definedas follows

(i) The throughput is defined as the total amount oftraffic (in bits per second) an SU receiver receivesfrom the sender divided by the time it takes for thereceiver to obtain the last packet A routing schemewith higher throughput is desirable for CRAHNs

(ii) The end-to-end delay is defined as the average delaywhich is calculated by summing up the end-to-enddelays of all packets received by all SU destination

10 Wireless Communications and Mobile Computing

Table 2 Simulation parameters

Parameter valueNumber of SUs 100Number of PUs 25Transmission ranges of SUsand PUs 250m

Number of channels 6Channelsrsquo idle stateprobabilities1198751198881119894119889119897119890 1198751198886119894119889119897119890 03 03 05 05 07 07Expected channel OFF time 60msTotal number of flows 30Source-destination distance 800mSU CCC rate 1MbpsSU data channel rate 2MbpsPacket size 512 BytesSensing time 3msChannel switching time 100120583sRetransmission time 3

nodes and dividing it by the total number of receivedpackets A routing scheme with the lower end-to-enddelay is preferred for CRAHNs

(iii) The SU-PU interference ratio is defined as the ratioof the number of SUsrsquo packets interrupted by PUsrsquoactivities to the total number of packets delivered byan SU source node A routing scheme with lower SU-PU interference ratio is favorable for CRAHNs

(iv) The packet delivery ratio is defined as the ratio ofall successfully received data packets that are fullyreceived by the SU destination node to the total datapackets generated by the SU source node A routingscheme with higher packet delivery ratio is desirablefor CRAHNs

To ensure the statistical significance of the results we runeach experiment for 500 seconds and repeat it 1000 timeswithdifferent seeds to report the average value as final results

62 Simulation Results

621 Impact of Number of Flows In this part we evaluate theimpact of offered traffic load on the performance by changingthe number of flows We vary the number of flows from 15CBR flows to 35 flows

As shown in Figure 3(a) we observe that LBOR provideshigher throughput than SAAR and SAOR We first considerthe case that the number of flows is below the saturationlevel Recall that the throughput defined here is measuredby the aggregate throughput of all the flows In this casewhen the number of flows increases from 15 flows to 25flows the throughput of all three schemes starts increasing asmore SU receivers are involved andmore parallel concurrenttransmissions occur Then we consider the case that whenthe number of flows is larger than 25 this indicates that

the network traffic is becoming saturated In this case thethroughput gain of SAAR and SAOR starts to decline As thenetwork resources (available channels and links) are limitedthe increasing number of flows intensifies the congestion andinterference which explains the throughput degradation ofSAAR and SAOR However when more flows are involvedthe throughput of LBOR still increases with increasingnumber of flows This is because LBOR has a load balancingfeature and solves the congestion problem by routing thetraffic through alternate forwarders

Figure 3(b) shows that LBOR has superior end-to-enddelay performance as compared to SAAR and SAOR Whenthe number of flows is equal to 15 the performance gapbetween LBOR and SAAR (both with SAOR) is small In thiscase each relay node does not need to handle toomuch trafficfor multiple flows Thus the unfair distribution of networktraffic will not lead to extra end-to-end delay Neverthelessif the number of flows is above 25 the end-to-end delayof SAAR and SAOR increase more sharply than that ofLBOR This can be attributed to the ability of LBOR that itsends packets to the links with lower traffic load and delayMoreover it can be concluded from the result that LBORfacilitates a better distribution of traffic and the consequentreduction of queuing times contributes to a lower overalldelay

622 Impact of PUsrsquo Activities In this part we evaluate theperformance of LBOR under different PUsrsquo activity patternsThe expected channel OFF time 119864[119879119874119865119865119888ℎ ] varies from 20msto 120ms A small 119864[119879119874119865119865119888ℎ ] means that the PUsrsquo activitiesare high and thus the delivery of the secondary user ismore likely to be interrupted Different from the previousexperiment the number of CBR flows is fixed to be 30

Figure 4(a) shows the SU-PU interference ratio of thethree schemes under different values of 119864[119879119874119865119865119888ℎ ] In mostcases LBOR produces significantly lower SU-PU interferenceas compared to SAAR and SAOR This is because the trafficload of LBOR is distributed efficiently and fairly In LBORfewer flows will have a chance to share the same SU nodewhich is highly influenced by PUsrsquo behaviors Moreoverwe notice that the improvement of load balancing schemeis significant under high PUsrsquo activities (119864[119879119874119865119865119888ℎ ] is below60ms) but not under low PUsrsquo activities (119864[119879119874119865119865119888ℎ ]) is above60ms) The reason behind this phenomenon is that whenthe nodes are highly influenced by PUsrsquo behaviors andoverloaded LBOR will give priorities to other nodes with alower traffic load to release the burden of these nodes

Figure 4(b) displays the throughput of LBOR SAARand SAOR with different values of 119864[119879119874119865119865119888ℎ ] It shows thatthe throughput of these schemes increase as the value of119864[119879119874119865119865119888ℎ ] becomes larger The result also shows that LBORachieves higher throughput compared to SAAR and SAORWhen the PUsrsquo activities decrease (the value of 119864[119879119874119865119865119888ℎ ]increases) the performance difference between LBOR andSAAR (both with SAOR) becomes negligible Moreover thetraffic load of LBOR is evenly distributed across the SUsand different available channels Therefore it is expected that

Wireless Communications and Mobile Computing 11

15 20 25 30 35Number of flows

170

180

190

200

210

220

230

240

250

260

270

ro

ughp

ut (K

bps)

LBORSAARSAOR(a) Throughput versus varying number of flows

15 20 25 30 35Number of flows

25

30

35

40

45

50

55

60

65

70

End-

to-e

nd d

elay

(ms)

SAORSAARLBOR(b) End-to-end delay versus varying number of flows

Figure 3

more improvements can be obtained for LBOR when PUsappear frequently

The end-to-end delay and the packet delivery ratio per-formance with different types of PUsrsquo activities are shownin Figures 4(c) and 4(d) With the increase of 119864[119879119874119865119865119888ℎ ]the packet delivery ratio will increase and the end-to-enddelay will decrease When the value of 119864[119879119874119865119865119888ℎ ] is around20ms LBOR can reduce the end-to-end delay by up to36 and 40 compared to SAAR and SAOR HoweverLBOR can only increase the delivery ratio by up to 2and 3 respectively compared to these two schemes Thereare two main reasons for this case First the opportunisticrouting protocol of LBOR is based on multipath routingstrategy where the paths and relay nodes are determinedon-the-fly In addition SAAR and SAOR also utilize thesimilar opportunistic routing principle and they aim toincrease the packet delivery ratio for the dynamic changingspectrum environment Second SAAR and SAOR alwayschoose paths or forwarding nodes with the best performanceof packet delivery ratio Nevertheless LBOR jointly considersthe load balancing for selecting the relay nodes Thus theimprovement of packet delivery ratio from LBOR is limited

623 Impact ofNumber of Channels In this part we comparethe performance of the three schemes under different numberof channels The number of channels varies between 2 and 10and the expected channel OFF time is set to be 60ms

Figure 5(a) compares the SU-PU interference ratio of thethree schemes by varying the number of channels of thenetworkWe can observe that the SU-PU interference ratio ofLBOR is lower than that of SAAR and SAOR One significantobservationhere is thatwhen the number of channels is largerthan 6 the SU-PU interference ratio of SAAR and SAOR

tends to decrease as the number of channels is increased InCRAHNs a larger number of channels can service a largernumber of SUsThus in this case SUs can choosemore stablechannels to avoid PUsrsquo interruptions In addition if an SUis interrupted by PUsrsquo appearance it can easily find anotheravailable channel for data transmissions Another importantobservation made is that when the number of channelsincreases from 2 to 6 the SU-PU interference ratio of LBORwill increase but the other two schemes will decrease Anexplanation for this phenomenon is as follows Consideringthe load balancing strategy LBOR tries its best to distributetraffic evenly among multiple nodes When there are notenough available channels to avoid the mutual interferencesamong SUs the packets cannot be correctly decoded at thereceiver Due to the same reason we can also observe fromFigures 5(b) 5(c) and 5(d) that when the number of channelsis below 6 the improvements from load balancing are notsignificantly

Figure 5(b) shows the throughput of LBOR SAAR andSAOR with different number of channels in the network Weobserve that the throughput of these schemes increases as thenumber of channels becomes larger The result also confirmsthat nomatter howmany channels are available LBOR alwaysperforms better than SAAR and SAOR When the number ofchannels is above 6 more improvement can be obtained fromLBOR

As shown in Figures 5(c) and 5(d) they provide similarresults and demonstrate that LBOR can achieve lower end-to-end delay and higher packet delivery ratio in most casescompared to SAAR and SAOR The improvements comefrom the load balancing based metric and the load balanc-ing based relay selection algorithm On the one hand theproposed scheme adopts a load balancing based metric to

12 Wireless Communications and Mobile Computing

20 40 60 80 100 120E[Tch

OFF] (ms)

0

002

004

006

008

01

012

014

016

018

02

SU-P

U in

terf

eren

ce ra

tio

LBORSAARSAOR(a) SU-PU interference ratio versus PUsrsquo activities

E[TchOFF] (ms)

20 40 60 80 100 120100

150

200

250

300

350

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus PUsrsquo activities

20 40 60 80 100 12025

30

35

40

45

50

55

60

65

End-

to-e

nd d

elay

(ms)

E[TchOFF] (ms)

LBORSAARSAOR

(c) End-to-end delay versus PUsrsquo activities

20 40 60 80 100 120076

078

08

082

084

086

088

09

092

094

Pack

et d

eliv

ery

ratio

LBORSAARSAOR

E[TchOFF] (ms)

(d) Packet delivery ratio versus PUsrsquo activities

Figure 4

choose the relay nodes with higher throughput and lowerdelay On the other hand the load balancing based relayselection algorithm can predict the occupied capacity of eachlink and thus ensure that the traffic load is evenly balancedamong all of the relay nodes

624 Impact of Number of PUs In this experiment weevaluate the performance of three schemes under differentnumber of PUs The number of PUs varies between 10 and40 and the number of channels is set to be 6 Figure 6(a)

compares the SU-PU interference ratio of different schemesunder different number of PUs In most cases the SU-PU interference ratios of SAAR and SAOR rise sharplythan LBOR That is because the network traffic of nonloadbalancing methods (SAAR and SAOR) tends to concentrateon a specific number of SUs and links If these SUsrsquo channelsare occupied by PUs most packets of the network may beaffected On the contrary LBOR distributes traffic to a largernumber of SUs thus packets have less change to be affectedby the PUsrsquo appearance and it induces a load balancing

Wireless Communications and Mobile Computing 13

2 4 6 8 10Total number of channels

002

004

006

008

01

012

014

016

018SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of channels

2 4 6 8 10Total number of channels

160

180

200

220

240

260

280

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of channels

2 4 6 8 10Total number of channels

20

30

40

50

60

70

80

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of channels

2 4 6 8 10Total number of channels

07

075

08

085

09

095Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of channels

Figure 5

improvement compared with SAAR and SAOR When thenumber of PUs is below 30 the benefit from load balancingis not surprising since SUs are not heavily affected Whenthe number of PUs is 35 an interesting observation here isthat the SU-PU interference ratio of LBOR is lower than thecase of 30 The reason for this case is that the aggregatedbenefit from load balancing scheme overwhelms the negativeeffect of the increased number of PUs More specifically inLBOR SUs are not overloaded and the packets are properlydistributed Thus when data transmission failures are causedby the appearance of PUs it is much easier for LBOR toreroute the blocked trafficflows to the light-loaded SUswhose

channels are available compared with other two methodsWhen the number of PUs is 40 we observe that the SU-PUinterference ratio of LBOR is higher than the case of 35 In thissituation there are not enough available channels for SUs totransmit the traffic of thewhole networkThus it is difficult tofind light-loaded SUs to reroute packets for overloaded SUsand more packets would be interrupted by PUsrsquo activities

Figure 6(b) shows the throughput by varying the numberof PUs in the network for the three schemes In the figurewe observe that the proposed LBOR scheme achieves higherthroughput than SAAR and SAOR We can also find thatwhen the number of PUs increases the throughput decreases

14 Wireless Communications and Mobile Computing

10 15 20 25 30 35 40Total number of PUs

003

004

005

006

007

008

009

01

011

012

013SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

160

180

200

220

240

260

280

300

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

25

30

35

40

45

50

55

60

65

70

75

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

078

08

082

084

086

088

09

092Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of PUs

Figure 6

because of available resources are decaying Moreover whenthe number of PUs is small the improvement from LBOR islimited On the contrary when the number of PUs rises above25 the throughput of SAAR and SAOR drops drastically butthe throughput degradation of LBOR is limited This is dueto the fact that LBOR can predict the availability and theoccupied capacity of each link and thus relieve the networkrsquoscongestions

Figures 6(c) and 6(d) can also confirm that LBOR canprovide the lowest end-to-end delay and the highest packetdelivery ratio among the three schemes When the numberof PUs increases the unbalanced distribution of load amongnodes becomes a critical issue and has a negative impacton the end-to-end delay and the packet delivery ratio Thiscan be understood intuitively as follows On the one handthe previous link may become unavailable due to the PU

appearance and the packets which arrived earlier have towait for late packets in reordering buffers at the receivingdestination thus resulting in an increasing queuing delayOn the other hand if late packets do not arrive within theimposed deadline it will be treated as a lost one thus resultingin an increasing packet loss

7 Conclusion

In this paper we have proposed a load balancing oppor-tunistic routing scheme for cognitive radio networks Con-sidering the loading constraint of each node we analyzedthe throughput bound of one opportunistic module andformulated the problem of maximizing the total throughputof the network as a linear programming problem Moreoverwe designed a new metric to select and prioritize the

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

4 Wireless Communications and Mobile Computing

the used capacity (119862119897119906) and the maximum capacity (C) whichis shown as follows

119862119897119886 = C minus 119862119897119906 119862119897119906 lt C

0 119862119897119880 ge C (4)

where the first equation of (4) is adopted when the link is notoverloaded and the second is applied to the case that the linkis overloaded The available capacity of a whole link set canbe derived according to the available capacity of each link inthe set For a link set LS we denote the available capacity ofLS as 119862119871119878119886 Then we have

119862119871119878119886 = sum119897isin119871119878

119862119897119886 (5)

wheresum119897isin119871119878 119862119897119886 represents the sum of the available capacity ofeach link

33 Flow Model We assume that a data flow 119891 is composedof all the packets sent by a node at a time slot where 120578119898 isthe maximum bits of the data flow The average size of 119891 isV119891 (V119891120598(0 120578119898])We use ϝ to represent the data flow set whichincludes all the end-to-end data flows in the network For adata flow119891 (119891120598ϝ) we use s(119891) and d(119891) to represent its sourcenode and destination node respectively For a link 119897 (119897 isin 119864)we use V119897119891 to represent the average size of the data flow 119891transmitted among the link 119897 at a time slot Then accordingto the flow conservation for each node n (n isin V) we haveV119891 + sum

119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891 119894119891 119899 = 119904 (119891) 119886119899119889 119891 isin 119865sum

119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119894119891 119899 = 119904 (119891) 119889 (119891) 119886119899119889 119891 isin 119865

(6)

where 119871119868(119899) and 119871119874(119899) are the input link set and output linkset of node n respectively The first equation in (6) representsthe flow conservation formula of the source node whereV119891 is the data sent and sum119897isin119871119868(119899)

V119897119891 is the data successfullyreceived Similarly the second equation represents the flowconservation of the relay node Note that all the symbols ondata in this paper represent the data sentreceived within atime slot

According to (4) and (5) we can obtain the followingproperty

Property 1 (the data requirement) The total amount of datatransmitted link 119897 (119897 isin 119871119878) denoted by 119881119897119891 should meet therequirement 119881119897119891 lt 119862119897119886 The total amount of data transmittedon a set of links should meet the requirement sum119897isin119871119878119881119897119891 lt 119874119865

119886

A summary of major notations used in the paper is givenin Table 1 for easy reference

Table 1 Summary of notations

Symbol Definition120588P120588S The interference range of PUSURP RS The transmission range of PUSUCH=chj Channel set j=12 m119879119874119873119888ℎ 119879119874119865119865119888ℎ

Themean time that PU spent in ONOFFstates1120582119874119873119888ℎ The expectation of 119879119874119873119888ℎ1120582119874119865119865119888ℎ The expectation of 119879119874119865119865119888ℎ119875119888ℎ119887119906119904119910119875119888ℎ119894119889119897119890 The probability of channel ch to be busyidle

119862119867119894119895 = 119862119867119894 cap 119862119867119895

The set of common channels between node119899119894 and 119899119895119889119894119895 The Euclidean distance between node 119899119894 and119899119895119897119894119895 The link between node 119899119894 and 119899119895119901119894119895 The packet reception ratio of 119897119894119895120595119894119895 The state of 119897119894119895C119862119897119906119862119897119886 Themaximumusedavailable capacity of

link l119862119871119878119886 The available capacity of the link set LSV119891 The average size of the data flow 119891120578119898 Themaximum bits of the data flow 119891F The data flow set119881119897119891 The data flowing through link l119871119868(119899)119871119874(119899) The inputoutput link set of node n119889119903119890119897119886119910(119899119894 119899119895) The relay distance between node 119899119894 and 119899119895119873119894 The one-hop neighboring set of node 119899119894119865119894 The forwarding candidate set of node 119899119894119902119894 The data transfer rate of node 119899119894120588119898 The priority order of node 119899119898120591 The sum of 120582119874119865119865119888 and 120582119874119873119888120576 The state of channels120585 The effective forwarding rate

119875119903119890119897119886119910 (119899119895) The probability of the candidate node 119899119895successfully receiving the packet

119862119879119879119888ℎ119894 The cognitive transport throughput of theopportunistic module (119899119894 119865119894)119879119903119890119897119886119910 The time required for one-hop relayΘ The load score

4 Problem Formulation

In this section based on the system model we first describethe working process of one-hop opportunistic routing inCRAHNs and then analyze its throughput

41 Opportunistic Routing Primer We use one-hop oppor-tunistic routing to represent the process that sender node119899119894 (119899119894 isin V) sends data to its candidate node set For any node119899119895 (119899119895 isin 119865119894) in the candidate node set it should meet the

Wireless Communications and Mobile Computing 5

requirement that 119889119903119890119897119886119910(119899119894 119899119895) ge 0 where 119889119903119890119897119886119910(119899119894 119899119895) can becomputed as follows

119889119903119890119897119886119910 (119899119894 119899119895) = 119889119899119894119863 minus 119889119899119895119863 (7)

where D is the destination node 119889119899119894119863 is the Euclideandistance between 119899119894 and D and 119889119899119895119863 is the Euclidean distancebetween 119899119895 and D

Thenwe describe how to define anopportunisticmoduleFor any node 119899119894 (119899119894 isin V) we use119873119894 to represent its one-hopneighboring set Every node in 119873119894 must meet the followingconditions (1)The node operates on the same channel as thatfor node 119899119894 (2)The node is in the transmission range of node119899119894 Then a subset 119865119894 of 119873119894 can be selected as the forwardingcandidate set of node 119899119894 We use (119899119894 119873119894) and (119899119894 119865119894) to denotethe neighboring module and the opportunistic module ofnode 119899119894 respectively

Then we will use an example to illustrate the conceptof one opportunistic module The node S and D representsthe source node and the destination node respectively Thenode 119899 is the forward node and the neighbor set of 119899 isN = 1198991 1198992 1198993 1198994 1198995 In the neighboring set N nodes 11989911198992 and 1198993 meet the conditions of 119889119903119890119897119886119910(119899 119899119895) ge 0 (1 ≪j ≪ 3) Then the forwarding candidate node set of 119899 isF = 1198991 1198992 1198993 Therefore the neighboring set of 119899 is (119899119873)and the opportunistic module is (119899 119865) (see Figure 1)

In the opportunistic module every node has its relaypriority After the sender node broadcasts the packet to itsforwarding candidate set one of the candidate nodes contin-ues the forwarding process based on their relay priority Aforwarding candidate will send the message only when all thenodes with higher priorities fail to transmit Only when noneof the forwarding candidates has successfully received thepacket the senderwill retransmit the packetsThe forwardingprocess reiterates until all the packet is delivered to thedestination

In CRAHNs the process of one-hop opportunistic rout-ing is shown in Figure 2 The period of 1199051 to 1199052 is for channeldetection and the period of 1199052 to 1199053 is for data transmission

In the channel sensing step similarly to [27] the sendersearches for a temporarily unoccupied channel in collabora-tion with its neighbors using the energy detection techniqueBefore detecting the data channel the sender broadcasts ashort message (ie sensing invitation (SNSINV) in the CCC)to inform neighboring nodes of the selected data channelthe location information of the sender and the destinationand the used capacity of the output link The transmissionof SNSINV message in the CCC follows the CSMACAmechanism as specified in IEEE 80211 MAC After receivingthe SNSINV neighboring SUs set the selected data channel tobe the nonaccessible state so that no SU can transmit in thechosen data channel during the sensing period of the senderUsing the location information in SNSINV the neighboringSUs evaluate whether they are eligible relay candidates (iewhether the relay distance is nonnegative) Eligible relaycandidates will collaborate with the sender in channel sensingand data transmission In this situation other SUs cannottransmit in the selected data channel during the reservedperiod specified in SNSINV (the time for EnergyDetection in

S Dn

n5n1

n2

n3n4

Figure 1 An example of one opportunistic module (119899 119865)

Figure 2)When the channel is sensed idle (ie no PUactivityis detected) the sender and its candidate nodes will forwardthe data Otherwise the sender selects another channel andrepeats the channel sensing process

In the data transmission step the sender broadcasts thepacket to its candidate nodes The candidate nodes transmitACK messages in priority order after a shorter back-offwindow (SIFS) A candidate SU keeps listening to the datachannel until it overhears an ACK or it prepares to transmitthe ACK The waiting time required for each candidate nodeis 119906 If the sender receives no ACK message it will repeat thechannel sensing and data transmission steps

42 Throughput Bound of One Opportunistic Module In thissubsection we study the capacity region of one opportunisticmodule (119899119894 119865119894) This capacity region will serve as a bound ofthroughput corresponding to the links in the opportunisticmodule

Assume that the data transfer rate of the sender node 119899119894is 119902119894 and its candidate forwarding set is 119865119894 = 1198991 119899119898 inwhich the priority of nodes satisfies 1205881 gt gt 120588119898 We use119871119878119894 = 1198971198941 119897119894119898 to represent its forwarding candidate linkset For a candidate node 119899119895(1 le 119895 le 119898) its correspondingcandidate link is 119897119894119895 whose link state and PRR are 120595119894119895 and119901119894119895 respectively Assume that the probability of channel119888ℎ (119888ℎ120598119862) to be idle at 1199050 is 119875119888ℎ119894119889119897119890 Then we can estimate theidle probability of channel 119888ℎ at 1199051(1199051 gt 1199050) according to itsstate (busy or idle) at 1199050 which is shown as follows

119875119888ℎ119894119889119897119890 (1199051) = 119875119888ℎ119894119889119897119890 + (120576 minus 119875119888ℎ119894119889119897119890) 119890minus120591(1199051minus1199050) (8)

where 120591 = 120582119874119865119865119888 + 120582119874119873119888 and 120576 is a 0-1 variable The variable120576 indicates the state of channels where 120576 = 1 indicates thechannel 119888ℎ is idle at 1199051 and 120576 = 0 indicates the channel 119888ℎ isbusy at 1199051 Then the probability of channelrsquos idle state in theprocess of channel sensing denoted by119875119878119888ℎ119894119889119897119890 can be obtainedby

119875119878119888ℎ119894119889119897119890 = 119875119888ℎ119894119889119897119890 (1199051) ∙ 119875119888ℎ119894119889119897119890 (1199051 1199052) (9)

where119875119888ℎ119894119889119897119890(1199051 1199052) is the probability that channel stays idle overthe time (1199051 1199052) According to [16] 119875119888ℎ119894119889119897119890(1199051 1199052) can be derivedby

119875119888ℎ119894119889119897119890 (1199051 1199052) = intinfin1199052minus1199051

119865119888ℎ119874119865119865 (119909)119864 [119879119888ℎ119874119865119865]119889119909 (10)

6 Wireless Communications and Mobile Computing

One-hop relay

CCC

ch

ch

tt2t1t0

middot middot middot

PUActive

Energydetecti

on

data u ACK SNSIV SIFS

Figure 2 The process of one-hop opportunistic routing for CRAHNs

where 119865119888ℎ119874119865119865119864[119879119888ℎ119874119865119865] is the probability density function(PDF) of the residual time of the channel 119888ℎ remains idleMore specifically 119865119888ℎ119874119865119865(119909) is the cumulative distributionfunction (CDF) of the duration of the OFF state and 119864[119879119888ℎ119874119865119865]is the expected channel OFF time of the channel 119888ℎ

In the data transmission step due to the interferencesome candidate nodes cannot receive the packet successfullyWhen the SU node is disrupted by PUsrsquo appearance the linkbetween it and the sender becomes unusable Therefore thethroughput is directly related to the success rate of the linksin the opportunistic module According to the analysis resultabove we introduce the metric of effective forwarding rate(EFR) which indicates the success rate of data transmissionof each candidate link in the opportunistic module denotedby 120585 Thus for the candidate link 119897119894119895 of the candidate node119899119895(1 le 119895 le 119898) in the opportunistic module (119899119894 119865119894) theeffective forwarding rate can be denoted by 120585(119897119894119895) which isdefined in the following equation

120585 (119897119894119895) = 119902119894 ∙ 119901119894119895 ∙ 119895minus1prod119896=0

(1 minus 120595119894119896 ∙ 119901119894119896) (11)

where prod119895minus1

119896=0(1 minus 120595119894119896 ∙ 119901119894119896) is the probability that 119899119895 received

the packet successfully when all candidate nodes with higherpriority than node 119899119895 are failed to receive the packet

According to the (11) accumulating all EFR values oflinks in the candidate link set we can obtain the EFR of theopportunistic module as follows

120585 (119871119878119894) = 119898sum119895=1

120585 (119897119894119895) = 119902119894 ∙ (1 minus 119898prod119895=0

(1 minus 120593119894119895 ∙ 119901119894119895)) (12)

where prod119898119895=0(1 minus 120593119894119895 ∙ 119901119894119895) means that none of the candidate

nodes have received the packetAssume that the sender node 119899119894 chooses the channel119888ℎ as the data transmission channel Then according to

(9)-(11) we can deduce the equation for the probability of thecandidate node 119899119895 successfully receiving the packet denotedas 119875119903119890119897119886119910(119899119895) which is shown as follows

119875119903119890119897119886119910 (119899119895) = 120585 (119897119894119895) ∙ 119875119878119888ℎ119894119889119897119890 ∙ 119875119888ℎ119894119889119897119890 (1199052 1199053) (13)

According to the conclusions in [5] we can statethat when the link state is stable the channel remainsidle throughout the data forwarding process (ie 119875119878119888ℎ119894119889119897119890 ∙119875119888ℎ119894119889119897119890(1199052 1199053) = 1)

Based on (13) we can compute the cognitive transportthroughput (CTT) of one opportunistic module (119899119894 119865119894) inchannel 119888ℎ as follows

119862119879119879119888ℎ119894 = 119898sum119895=1

119875119888ℎ119903119890119897119886119910 (119899119895) ∙ ( 119871119879119903119890119897119886119910) (14)

where sum119898119895=1 119875119888ℎ119903119890119897119886119910(119899119895) is the data transmission success rate of

thewhole opportunisticmodule L is the data size transmittedby the opportunistic module and 119879119903119890119897119886119910 is the time requiredfor the whole process To facilitate the subsequent researchwe assume that the transmission time of all opportunisticmodules in the network is the same Therefore (14) can besimplified as follows

119862119879119879119888ℎ119894 = 119898sum119895=1

119875119888ℎ119903119890119897119886119910 (119899119895) ∙ 119892119871 (15)

Wireless Communications and Mobile Computing 7

43 Maximize the Total Throughput of the Network In thissubsection we formulate the optimization problem of net-work throughput as linear programming (LP) problem

P max sum119897isin119864119891isin119865

V119897119891 (16)

st V119891 + sum119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119899 = 119904 (119891) 119891 isin 119865 forall119899 isin 119881

(17)

sum119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119899 = 119904 (119891) 119899 = 119889 (119891) 119891 isin 119865 forall119899 isin 119881

(18)

sum119897isin119871119868(119899)

V119897119891 ge 0 forall119891 isin 119865 forall119899 isin 119881 (19)

sum119897isin1198710(119899)

V119897119891 = 0 119905 (119897) = 119889 (119891) forall119899 isin 119881 (20)

V119897119891 le 119862119897119886 119891 isin 119865 forall119897 isin 119864 (21)

sum119897isin119871119865

V119897119891 le 119862119871119878119886 119891 isin 119865 forall119897 isin 119871119878 (22)

sum119897isin119871119865

V119897119891 le 119862119879119879119888ℎ119894 forall119899 isin 119881 (23)

In the problem P we aim to maximize the throughput ofthe network as shown in (16) Equation (17) represents theflow conservation condition of the source and (18) representsthe flow conservation conditions of other nodes At eachnode except for the source and the destination the amountof incoming flow is equal to the amount of outgoing flowEquation (19) indicates that the amount of flow receivedby each node is nonnegative Equation (20) ensures thatthe outgoing flow from the destination of each data flowis 0 where 119905(119897) is the end node of link 119897 The physicalmeaning of (21) is that the actual flow delivered on each linkis constrained by the total amount of flow residing in thenetwork Equations (22) and (23) indicate that the amountof traffic passed through the opportunistic module is lessthan the available capacity and the throughput of the modulerespectively

In cognitive radio networks the state of the link willfrequently change due to the changing states (busy or idle)of the channel The dynamic link state may also lead tothe network topology change and make the candidate setbecome unavailable and nonoptimal Thus we cannot verifythe accuracy and effectiveness of the load balancing algo-rithm which is designed based on local network topologyinformation Moreover we cannot guarantee that the packetsare always transmitted through the links with lower trafficload Therefore a possible practical solution for this problemis distributing the traffic according to the load and thedata forwarding capacity of the link By this way we canimprove the network throughput while guaranteeing thecommunication quality

5 Opportunistic Routing

In this section we first propose a new metric to measurethe priority of candidate nodes and provide the sorting algo-rithm Then we design the load balancing based candidateforwarder selection algorithm under the stable condition ofthe link

51 Load Balancing Candidate Forwarder Sorting AlgorithmFirst we take the metric of linkrsquos reliability as an example toanalyze the reason for the uneven distribution of link loadIn the algorithms which are based on the link reliabilitywith the rise of the delivery rate of the candidate link thepriority of the corresponding candidate node will ascendBased on opportunistic routing the nodewith higher prioritywill receive more data As a result for links in the candidatelink set with the rise of the priority the load on the linkwill elevate As the amount of transmitted data increase thehigher priority linkrsquos traffic load also increases This case maycause the link blocked In the meantime the amount of dataallocated to the link with lower priority is small which maylead to the waste of link resources Therefore with the rise ofthe delivery rate of the candidate link the available capacityof the link will decline

511 Metric Design According to the consideration abovewe redesign a metric called Load Score to measure thepriority of candidate nodes based on the delivery rate andavailable capacity of each candidate link of an opportunisticmodule denoted asΘ In the opportunisticmodule (n F) F =1198991 119899119898 is the candidate node set and LS = 1198971 119897119898 isthe candidate link set For any candidate node 119899119890(1 le 119890 le 119898)the Load Score can be derived as follows

Θ119890 = 119901119890 ∙ 119862119897119890119886 (24)

The physical meaning of the metric is the maximumamount of data transmitted in a time slot when the linkhas the highest priority This metric considers both thetransmission quality of the link and load of the link Weestimate the prioritization of candidate nodes based on theLoad Score Thus it can achieve the balanced distributionof traffic load and reduce the number of packet drops andqueuing delays Meanwhile it ensures that the link transfersthe maximum amount of data Next we will analyze thespecial situations that may occur in the process of sortingcandidate nodes

According to the condition that multiple candidate nodesmay have the same Load Score we propose the followingproposition

Proposition 2 When multiple candidate nodes have thesame Load Score a higher priority candidate node is alwaysassociated with a lower packet delivery rate

Proof We assume there are 119911 candidate nodes 1198991 119899119911 inthe opportunistic module (nF) and assume they have thesame value of Load Scores denoted asΘ The correspondinglinks of nodes are denoted by 1198971 119897119911 We use 119866 to represent

8 Wireless Communications and Mobile Computing

the maximum throughput of link set 1198971 119897119911 in a steadystate Then we have

G = q ∙ 1199011 ∙ 1198741198971119886 + q ∙ 1199012 ∙ 1198621198972119886 (1 minus 1199011) + + q ∙ 119901119911

∙ 119862119897119911119886 119911minus1sum119896=0

(1 minus 119901119896)= 119902 ∙ Θ ∙ (1 + (1 minus 1199011) + + 119911minus1sum

119896=0

(1 minus 119901119896))(25)

In (25) 119902 and Θ are constant values Then the value of 119866 isdecided by 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) For conveniencewe use the notation Υ to denote the formula 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) Then the maximum value of 119866 can beachieved when Υ has the maximum value Obviously Υ isthe sum of the polynomials As a result the necessary andsufficient condition forΥ to attain the maximum value is thateach term ofΥmust reach the maximum valueThat is to say(1minus1199011) sum119911minus1

119896=0(1minus119901119896)must take the maximum value at thesame time Intuitively these items are inversely proportionalto the delivery rate of the candidate link Therefore we canconclude that the maximum Υ can be achieved when 1199011 lt1199012 lt lt 119901119911 Accordingly the condition 1198621198971119886 gt 1198621198972119886 gt gt119862119897119911119886 is valid Then Proposition 2 holds

512 Load BalancingCandidate Forwarder Sorting AlgorithmIn this subsection we will explain how to sort the candidatenodes in LS In the opportunistic module (n F) F =1198991 119899119898 is the candidate node set FL = 1198971 119897119898 isthe candidate link set and PR = 1199011 119901119898 is a PRR setEach PRR corresponds to a candidate link Based on (24) wecan calculate the Load Score of each candidate node which isdenoted as H = Θ1 Θ119898 We develop a simple selectionsorting algorithm to sort H = Θ1 Θ119898 which is shownas in Algorithm 1

In Algorithm 1 the inputs are the candidate node set119865 the PRR set 119875119877 and the LS set 119867 The algorithm sortscandidate nodes based on their LS In the algorithm line(4) is used to mark the highest priority nodes for eachiteration and lines (6) to (12) indicate the priority orderingprocess According to the process of the algorithm the timecomplexity of this algorithm is O(m2) where m representsthe number of candidate nodes

For the ordered candidate node set 1198651015840 we can calcu-late the maximum throughput of the opportunistic module(n 1198651015840) according to (11) which is shown as follows

CTT = q ∙ Θ1 + q ∙ Θ2 (1 minus 1199011) + + q

∙ Θ119898

119898minus1sum119896=0

(1 minus 119901119897119896) (26)

52 Load Balancing Based Relay Selection Algorithm In thissection we further study the load balancing solution InCRAHNs the packet will be forwarded to the destinationthrough several opportunistic modules until the packet

Input F = 1198991 119899119898 PR = 1199011 119901119898 H = Θ1 Θ119898Output An ordered candidate node set 1198651015840(1) Initialize 1198651015840 = (2) Initialize a b k(3) for eachΘ119886120598H and 1 le a le m do(4) blarr997888 a(5) for eachΘ119896120598H and a + 1 le k le m do(6) if Θ119887 lt Θ119896 then(7) blarr997888 k(8) else if Θ119887 = Θ119896 then(9) if 119875119886 gt 119875119896 then(10) blarr997888 k(11) end if(12) end if(13) end for(14) 1198651015840119886 larr997888 119865119887(15) end for(16) output 1198651015840(17) end

Algorithm 1 Load Balancing Based Candidate Forwarder SortingAlgorithm

arrives to its destination In each opportunistic module theselection of the last hop is limited by the forwarding capabilityof the next hop To achieve the global optimal it is necessaryto predict the forwarding capability of the opportunisticmodule which is the final hop Then we determine thecandidate node set of the first hop opportunistic modulenamely the optimal candidate node set for the first hopopportunistic module However this method of selectingthe optimal opportunistic module is not appropriate Inthe cognitive radio network the link status is dynamic dueto the dynamic activities of PUs Hence we cannot verifythat the prediction of the forwarding ability of the last hopopportunistic module is convergent Thus we cannot provethe optimal opportunistic module is accurate In [16] theauthor chose an intermediate destinationnode to alleviate theinfluence of the dynamic change of link state Inspired by thismethod we adopt an opportunistic module iteration methodto determine the optimal candidate node set of the currentopportunistic module Since this method is based on theforwarding ability of the next hop opportunistic module it isnecessary to predict the occupied capacity of each candidatelink in this opportunistic module

In this paper we use Minimum Mean Square Error(MMSE) [28] based flow prediction model to estimate theoccupied capacity of the link in the next time slot Theformula of the model is as follows

119862119897119906 = MMSE (119883119905 | t = 0 1 2 ) (27)

where 119883119905|t = 0 1 2 is the current flow and historicalflow statistics In [28] the author describes the predictionscheme in detail and readers can refer to it for more details

For the opportunistic module (n F) the candidate nodeset is F = 1198991 119899119898 the candidate link set is LS =1198971 119897119898 and the corresponding LS set isH = Θ1 Θ119898

Wireless Communications and Mobile Computing 9

Input F = n1 nm H = Θ1 ΘmOutput The optimal node set Foptimal(1) Initialize arrays of H1015840 F1015840 F10158401015840 Foptimal and LS1015840(2) Initialize values of o1 o2 and ctt(3) for each ne120598F and 1 le e le m do(4) Search the opportunistic module(ne Fe) of ne(5) F1015840larr997888the candidate node set of (ne Fe)(6) LS1015840larr997888the candidate link set of (ne Fe)(7) for each n1015840k120598F and 1 le k le |Fe| do(8) o1 larr997888 the used capacity of l1015840k(9) o2 larr997888 the available capacity of l1015840k(10) H1015840

k larr997888 the LS of n1015840k(11) end for(12) F10158401015840 larr997888 sorting F1015840 according to Algorithm 1(13) cttlarr997888 the throughput of (ne F10158401015840)(14) if Θe gt ctt do(15) Θe larr997888 ctt(16) end if(17) end for(18) Foptimal larr997888 sorting F according to Algorithm 1(19) output Foptimal(20) end

Algorithm 2 Load Balancing Relay Selection Algorithm

The selection steps of the optimal opportunistic module aredescribed as follows

(1) For any candidate node 119899119890 isin 119865 we first determineits opportunistic module (119899119890 119865119890) according to relay distancesThen we perform the following steps (1) predicting theoccupied capacity of each link in candidate link set 119865119871119890 basedon (27) (2) calculating the available capacity of each link in119865119871119890 based on (14) (3) computing the LS of each node in thecandidate node set 119865119890 based on (24) (4) sorting the candidatenode set 119865119890 and obtaining an ordered candidate node set 1198651015840119890according to Algorithm 1 and (5) calculating the maximumthroughput CTT((119899119890 1198651015840119890)) of the opportunistic module basedon (26)

(2)We compare the LS valueΘ119890 of candidate node 119899119890 withCTT((119899119890 1198651015840119890)) and assign a smaller value to Θ119890

(3) We sort the candidate node set 119865 based on Algo-rithm 1 and the ordered set of candidate nodes is the optimalcandidate node set

The details of the Load Balancing Relay Selection Algo-rithm are shown in Algorithm 2 In Algorithm 2 lines (1)and (2) define the relevant parameters and their initializedvalues Lines (3) to (18) show the selection process for theoptimal candidate forwarding node According to the processof Algorithm 2 and the time complexity of Algorithm 1 thetime complexity of Algorithm 2 is O(m|1198651015840|2) where m isthe number of candidate nodes of the opportunistic module(nF) and |1198651015840| is the maximum number of candidate nodesin all the next hop opportunistic modules We can selectthe optimal opportunistic module for each hop according toAlgorithm 2 It balances the traffic load of each candidate linkwhile ensuring the quality of communication Meanwhile itimproves the throughput of the local network

6 Performance Evaluation

In this section we evaluate the performance of our proposedscheme with two well-known routing protocols for CRAHNsand present the simulation results on the SU-PU interferenceratio the packet delivery ratio the throughput and the end-to-end delay under different number of flows PUsrsquo activitiesnumber of channels and number of PUs

61 Simulation Settings The network parameter settings aresummarized in Table 2 In the simulations multiple SUsand PUs are randomly deployed in a 1000m times 1000marea and they are assumed to be stationary throughout thesimulation As stated in Section 31 the PU activities of eachchannel are modeled as an exponential ON-OFF processThe status of each channel is associated with two parametersone is the expected channel OFF time 119864[119879119874119865119865119888ℎ ] and theother is the probability of the idle state 119875119888ℎ119894119889119897119890 The channelstatus is estimated by utilizing periodic spectrum sensingand on-demand sensing before data transmissions In thesimulations 30 constant bit rate (CBR) flows are generatedbetween randomly chosen source-destination pairs of SUsand they are associated with the packet size of 512 Bytesand flow rate of 5Kbps Notice that the most representativeopportunistic routing scheme ExOR [9] adopted the methodin [29] to obtain the packet reception ratio between twonodes Accordingly in this simulation the packet receptionratios of the links are also based on the distance-to-deliveryratio relationshipmeasured in [29]Thus we require the posi-tion information of different nodes to compute the distancebetween neighboring nodes and obtain the packet receptionratios of the links We evaluate our scheme through NS-2simulation [30] In NS-2 our LBOR scheme is implementedas the routing agent In this work we only focus on thenetwork layer without considering MAC and physical layerissues Since the IEEE 80211 MAC protocol is customized forCR support we use the modified 80211cr as the CSMAMACagent in our simulation script We also apply CRN patch [31]to our simulations that can support multiple channels at thephysical layer

In the simulations we compare our LBOR scheme withSAAR scheme [15] and SAORprotocol [27]TheSAORproto-col is a well-knownmultipath opportunistic routing protocolcoupled with spectrum sensing and sharing in multichannelCRAHNs The SAAR scheme is a recently published any-path opportunistic routing scheme with consideration of theunreliable transmission feature and the uncertain spectrumavailability characteristic of CRAHNs

The performance metrics for our experiments are definedas follows

(i) The throughput is defined as the total amount oftraffic (in bits per second) an SU receiver receivesfrom the sender divided by the time it takes for thereceiver to obtain the last packet A routing schemewith higher throughput is desirable for CRAHNs

(ii) The end-to-end delay is defined as the average delaywhich is calculated by summing up the end-to-enddelays of all packets received by all SU destination

10 Wireless Communications and Mobile Computing

Table 2 Simulation parameters

Parameter valueNumber of SUs 100Number of PUs 25Transmission ranges of SUsand PUs 250m

Number of channels 6Channelsrsquo idle stateprobabilities1198751198881119894119889119897119890 1198751198886119894119889119897119890 03 03 05 05 07 07Expected channel OFF time 60msTotal number of flows 30Source-destination distance 800mSU CCC rate 1MbpsSU data channel rate 2MbpsPacket size 512 BytesSensing time 3msChannel switching time 100120583sRetransmission time 3

nodes and dividing it by the total number of receivedpackets A routing scheme with the lower end-to-enddelay is preferred for CRAHNs

(iii) The SU-PU interference ratio is defined as the ratioof the number of SUsrsquo packets interrupted by PUsrsquoactivities to the total number of packets delivered byan SU source node A routing scheme with lower SU-PU interference ratio is favorable for CRAHNs

(iv) The packet delivery ratio is defined as the ratio ofall successfully received data packets that are fullyreceived by the SU destination node to the total datapackets generated by the SU source node A routingscheme with higher packet delivery ratio is desirablefor CRAHNs

To ensure the statistical significance of the results we runeach experiment for 500 seconds and repeat it 1000 timeswithdifferent seeds to report the average value as final results

62 Simulation Results

621 Impact of Number of Flows In this part we evaluate theimpact of offered traffic load on the performance by changingthe number of flows We vary the number of flows from 15CBR flows to 35 flows

As shown in Figure 3(a) we observe that LBOR provideshigher throughput than SAAR and SAOR We first considerthe case that the number of flows is below the saturationlevel Recall that the throughput defined here is measuredby the aggregate throughput of all the flows In this casewhen the number of flows increases from 15 flows to 25flows the throughput of all three schemes starts increasing asmore SU receivers are involved andmore parallel concurrenttransmissions occur Then we consider the case that whenthe number of flows is larger than 25 this indicates that

the network traffic is becoming saturated In this case thethroughput gain of SAAR and SAOR starts to decline As thenetwork resources (available channels and links) are limitedthe increasing number of flows intensifies the congestion andinterference which explains the throughput degradation ofSAAR and SAOR However when more flows are involvedthe throughput of LBOR still increases with increasingnumber of flows This is because LBOR has a load balancingfeature and solves the congestion problem by routing thetraffic through alternate forwarders

Figure 3(b) shows that LBOR has superior end-to-enddelay performance as compared to SAAR and SAOR Whenthe number of flows is equal to 15 the performance gapbetween LBOR and SAAR (both with SAOR) is small In thiscase each relay node does not need to handle toomuch trafficfor multiple flows Thus the unfair distribution of networktraffic will not lead to extra end-to-end delay Neverthelessif the number of flows is above 25 the end-to-end delayof SAAR and SAOR increase more sharply than that ofLBOR This can be attributed to the ability of LBOR that itsends packets to the links with lower traffic load and delayMoreover it can be concluded from the result that LBORfacilitates a better distribution of traffic and the consequentreduction of queuing times contributes to a lower overalldelay

622 Impact of PUsrsquo Activities In this part we evaluate theperformance of LBOR under different PUsrsquo activity patternsThe expected channel OFF time 119864[119879119874119865119865119888ℎ ] varies from 20msto 120ms A small 119864[119879119874119865119865119888ℎ ] means that the PUsrsquo activitiesare high and thus the delivery of the secondary user ismore likely to be interrupted Different from the previousexperiment the number of CBR flows is fixed to be 30

Figure 4(a) shows the SU-PU interference ratio of thethree schemes under different values of 119864[119879119874119865119865119888ℎ ] In mostcases LBOR produces significantly lower SU-PU interferenceas compared to SAAR and SAOR This is because the trafficload of LBOR is distributed efficiently and fairly In LBORfewer flows will have a chance to share the same SU nodewhich is highly influenced by PUsrsquo behaviors Moreoverwe notice that the improvement of load balancing schemeis significant under high PUsrsquo activities (119864[119879119874119865119865119888ℎ ] is below60ms) but not under low PUsrsquo activities (119864[119879119874119865119865119888ℎ ]) is above60ms) The reason behind this phenomenon is that whenthe nodes are highly influenced by PUsrsquo behaviors andoverloaded LBOR will give priorities to other nodes with alower traffic load to release the burden of these nodes

Figure 4(b) displays the throughput of LBOR SAARand SAOR with different values of 119864[119879119874119865119865119888ℎ ] It shows thatthe throughput of these schemes increase as the value of119864[119879119874119865119865119888ℎ ] becomes larger The result also shows that LBORachieves higher throughput compared to SAAR and SAORWhen the PUsrsquo activities decrease (the value of 119864[119879119874119865119865119888ℎ ]increases) the performance difference between LBOR andSAAR (both with SAOR) becomes negligible Moreover thetraffic load of LBOR is evenly distributed across the SUsand different available channels Therefore it is expected that

Wireless Communications and Mobile Computing 11

15 20 25 30 35Number of flows

170

180

190

200

210

220

230

240

250

260

270

ro

ughp

ut (K

bps)

LBORSAARSAOR(a) Throughput versus varying number of flows

15 20 25 30 35Number of flows

25

30

35

40

45

50

55

60

65

70

End-

to-e

nd d

elay

(ms)

SAORSAARLBOR(b) End-to-end delay versus varying number of flows

Figure 3

more improvements can be obtained for LBOR when PUsappear frequently

The end-to-end delay and the packet delivery ratio per-formance with different types of PUsrsquo activities are shownin Figures 4(c) and 4(d) With the increase of 119864[119879119874119865119865119888ℎ ]the packet delivery ratio will increase and the end-to-enddelay will decrease When the value of 119864[119879119874119865119865119888ℎ ] is around20ms LBOR can reduce the end-to-end delay by up to36 and 40 compared to SAAR and SAOR HoweverLBOR can only increase the delivery ratio by up to 2and 3 respectively compared to these two schemes Thereare two main reasons for this case First the opportunisticrouting protocol of LBOR is based on multipath routingstrategy where the paths and relay nodes are determinedon-the-fly In addition SAAR and SAOR also utilize thesimilar opportunistic routing principle and they aim toincrease the packet delivery ratio for the dynamic changingspectrum environment Second SAAR and SAOR alwayschoose paths or forwarding nodes with the best performanceof packet delivery ratio Nevertheless LBOR jointly considersthe load balancing for selecting the relay nodes Thus theimprovement of packet delivery ratio from LBOR is limited

623 Impact ofNumber of Channels In this part we comparethe performance of the three schemes under different numberof channels The number of channels varies between 2 and 10and the expected channel OFF time is set to be 60ms

Figure 5(a) compares the SU-PU interference ratio of thethree schemes by varying the number of channels of thenetworkWe can observe that the SU-PU interference ratio ofLBOR is lower than that of SAAR and SAOR One significantobservationhere is thatwhen the number of channels is largerthan 6 the SU-PU interference ratio of SAAR and SAOR

tends to decrease as the number of channels is increased InCRAHNs a larger number of channels can service a largernumber of SUsThus in this case SUs can choosemore stablechannels to avoid PUsrsquo interruptions In addition if an SUis interrupted by PUsrsquo appearance it can easily find anotheravailable channel for data transmissions Another importantobservation made is that when the number of channelsincreases from 2 to 6 the SU-PU interference ratio of LBORwill increase but the other two schemes will decrease Anexplanation for this phenomenon is as follows Consideringthe load balancing strategy LBOR tries its best to distributetraffic evenly among multiple nodes When there are notenough available channels to avoid the mutual interferencesamong SUs the packets cannot be correctly decoded at thereceiver Due to the same reason we can also observe fromFigures 5(b) 5(c) and 5(d) that when the number of channelsis below 6 the improvements from load balancing are notsignificantly

Figure 5(b) shows the throughput of LBOR SAAR andSAOR with different number of channels in the network Weobserve that the throughput of these schemes increases as thenumber of channels becomes larger The result also confirmsthat nomatter howmany channels are available LBOR alwaysperforms better than SAAR and SAOR When the number ofchannels is above 6 more improvement can be obtained fromLBOR

As shown in Figures 5(c) and 5(d) they provide similarresults and demonstrate that LBOR can achieve lower end-to-end delay and higher packet delivery ratio in most casescompared to SAAR and SAOR The improvements comefrom the load balancing based metric and the load balanc-ing based relay selection algorithm On the one hand theproposed scheme adopts a load balancing based metric to

12 Wireless Communications and Mobile Computing

20 40 60 80 100 120E[Tch

OFF] (ms)

0

002

004

006

008

01

012

014

016

018

02

SU-P

U in

terf

eren

ce ra

tio

LBORSAARSAOR(a) SU-PU interference ratio versus PUsrsquo activities

E[TchOFF] (ms)

20 40 60 80 100 120100

150

200

250

300

350

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus PUsrsquo activities

20 40 60 80 100 12025

30

35

40

45

50

55

60

65

End-

to-e

nd d

elay

(ms)

E[TchOFF] (ms)

LBORSAARSAOR

(c) End-to-end delay versus PUsrsquo activities

20 40 60 80 100 120076

078

08

082

084

086

088

09

092

094

Pack

et d

eliv

ery

ratio

LBORSAARSAOR

E[TchOFF] (ms)

(d) Packet delivery ratio versus PUsrsquo activities

Figure 4

choose the relay nodes with higher throughput and lowerdelay On the other hand the load balancing based relayselection algorithm can predict the occupied capacity of eachlink and thus ensure that the traffic load is evenly balancedamong all of the relay nodes

624 Impact of Number of PUs In this experiment weevaluate the performance of three schemes under differentnumber of PUs The number of PUs varies between 10 and40 and the number of channels is set to be 6 Figure 6(a)

compares the SU-PU interference ratio of different schemesunder different number of PUs In most cases the SU-PU interference ratios of SAAR and SAOR rise sharplythan LBOR That is because the network traffic of nonloadbalancing methods (SAAR and SAOR) tends to concentrateon a specific number of SUs and links If these SUsrsquo channelsare occupied by PUs most packets of the network may beaffected On the contrary LBOR distributes traffic to a largernumber of SUs thus packets have less change to be affectedby the PUsrsquo appearance and it induces a load balancing

Wireless Communications and Mobile Computing 13

2 4 6 8 10Total number of channels

002

004

006

008

01

012

014

016

018SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of channels

2 4 6 8 10Total number of channels

160

180

200

220

240

260

280

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of channels

2 4 6 8 10Total number of channels

20

30

40

50

60

70

80

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of channels

2 4 6 8 10Total number of channels

07

075

08

085

09

095Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of channels

Figure 5

improvement compared with SAAR and SAOR When thenumber of PUs is below 30 the benefit from load balancingis not surprising since SUs are not heavily affected Whenthe number of PUs is 35 an interesting observation here isthat the SU-PU interference ratio of LBOR is lower than thecase of 30 The reason for this case is that the aggregatedbenefit from load balancing scheme overwhelms the negativeeffect of the increased number of PUs More specifically inLBOR SUs are not overloaded and the packets are properlydistributed Thus when data transmission failures are causedby the appearance of PUs it is much easier for LBOR toreroute the blocked trafficflows to the light-loaded SUswhose

channels are available compared with other two methodsWhen the number of PUs is 40 we observe that the SU-PUinterference ratio of LBOR is higher than the case of 35 In thissituation there are not enough available channels for SUs totransmit the traffic of thewhole networkThus it is difficult tofind light-loaded SUs to reroute packets for overloaded SUsand more packets would be interrupted by PUsrsquo activities

Figure 6(b) shows the throughput by varying the numberof PUs in the network for the three schemes In the figurewe observe that the proposed LBOR scheme achieves higherthroughput than SAAR and SAOR We can also find thatwhen the number of PUs increases the throughput decreases

14 Wireless Communications and Mobile Computing

10 15 20 25 30 35 40Total number of PUs

003

004

005

006

007

008

009

01

011

012

013SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

160

180

200

220

240

260

280

300

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

25

30

35

40

45

50

55

60

65

70

75

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

078

08

082

084

086

088

09

092Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of PUs

Figure 6

because of available resources are decaying Moreover whenthe number of PUs is small the improvement from LBOR islimited On the contrary when the number of PUs rises above25 the throughput of SAAR and SAOR drops drastically butthe throughput degradation of LBOR is limited This is dueto the fact that LBOR can predict the availability and theoccupied capacity of each link and thus relieve the networkrsquoscongestions

Figures 6(c) and 6(d) can also confirm that LBOR canprovide the lowest end-to-end delay and the highest packetdelivery ratio among the three schemes When the numberof PUs increases the unbalanced distribution of load amongnodes becomes a critical issue and has a negative impacton the end-to-end delay and the packet delivery ratio Thiscan be understood intuitively as follows On the one handthe previous link may become unavailable due to the PU

appearance and the packets which arrived earlier have towait for late packets in reordering buffers at the receivingdestination thus resulting in an increasing queuing delayOn the other hand if late packets do not arrive within theimposed deadline it will be treated as a lost one thus resultingin an increasing packet loss

7 Conclusion

In this paper we have proposed a load balancing oppor-tunistic routing scheme for cognitive radio networks Con-sidering the loading constraint of each node we analyzedthe throughput bound of one opportunistic module andformulated the problem of maximizing the total throughputof the network as a linear programming problem Moreoverwe designed a new metric to select and prioritize the

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

Wireless Communications and Mobile Computing 5

requirement that 119889119903119890119897119886119910(119899119894 119899119895) ge 0 where 119889119903119890119897119886119910(119899119894 119899119895) can becomputed as follows

119889119903119890119897119886119910 (119899119894 119899119895) = 119889119899119894119863 minus 119889119899119895119863 (7)

where D is the destination node 119889119899119894119863 is the Euclideandistance between 119899119894 and D and 119889119899119895119863 is the Euclidean distancebetween 119899119895 and D

Thenwe describe how to define anopportunisticmoduleFor any node 119899119894 (119899119894 isin V) we use119873119894 to represent its one-hopneighboring set Every node in 119873119894 must meet the followingconditions (1)The node operates on the same channel as thatfor node 119899119894 (2)The node is in the transmission range of node119899119894 Then a subset 119865119894 of 119873119894 can be selected as the forwardingcandidate set of node 119899119894 We use (119899119894 119873119894) and (119899119894 119865119894) to denotethe neighboring module and the opportunistic module ofnode 119899119894 respectively

Then we will use an example to illustrate the conceptof one opportunistic module The node S and D representsthe source node and the destination node respectively Thenode 119899 is the forward node and the neighbor set of 119899 isN = 1198991 1198992 1198993 1198994 1198995 In the neighboring set N nodes 11989911198992 and 1198993 meet the conditions of 119889119903119890119897119886119910(119899 119899119895) ge 0 (1 ≪j ≪ 3) Then the forwarding candidate node set of 119899 isF = 1198991 1198992 1198993 Therefore the neighboring set of 119899 is (119899119873)and the opportunistic module is (119899 119865) (see Figure 1)

In the opportunistic module every node has its relaypriority After the sender node broadcasts the packet to itsforwarding candidate set one of the candidate nodes contin-ues the forwarding process based on their relay priority Aforwarding candidate will send the message only when all thenodes with higher priorities fail to transmit Only when noneof the forwarding candidates has successfully received thepacket the senderwill retransmit the packetsThe forwardingprocess reiterates until all the packet is delivered to thedestination

In CRAHNs the process of one-hop opportunistic rout-ing is shown in Figure 2 The period of 1199051 to 1199052 is for channeldetection and the period of 1199052 to 1199053 is for data transmission

In the channel sensing step similarly to [27] the sendersearches for a temporarily unoccupied channel in collabora-tion with its neighbors using the energy detection techniqueBefore detecting the data channel the sender broadcasts ashort message (ie sensing invitation (SNSINV) in the CCC)to inform neighboring nodes of the selected data channelthe location information of the sender and the destinationand the used capacity of the output link The transmissionof SNSINV message in the CCC follows the CSMACAmechanism as specified in IEEE 80211 MAC After receivingthe SNSINV neighboring SUs set the selected data channel tobe the nonaccessible state so that no SU can transmit in thechosen data channel during the sensing period of the senderUsing the location information in SNSINV the neighboringSUs evaluate whether they are eligible relay candidates (iewhether the relay distance is nonnegative) Eligible relaycandidates will collaborate with the sender in channel sensingand data transmission In this situation other SUs cannottransmit in the selected data channel during the reservedperiod specified in SNSINV (the time for EnergyDetection in

S Dn

n5n1

n2

n3n4

Figure 1 An example of one opportunistic module (119899 119865)

Figure 2)When the channel is sensed idle (ie no PUactivityis detected) the sender and its candidate nodes will forwardthe data Otherwise the sender selects another channel andrepeats the channel sensing process

In the data transmission step the sender broadcasts thepacket to its candidate nodes The candidate nodes transmitACK messages in priority order after a shorter back-offwindow (SIFS) A candidate SU keeps listening to the datachannel until it overhears an ACK or it prepares to transmitthe ACK The waiting time required for each candidate nodeis 119906 If the sender receives no ACK message it will repeat thechannel sensing and data transmission steps

42 Throughput Bound of One Opportunistic Module In thissubsection we study the capacity region of one opportunisticmodule (119899119894 119865119894) This capacity region will serve as a bound ofthroughput corresponding to the links in the opportunisticmodule

Assume that the data transfer rate of the sender node 119899119894is 119902119894 and its candidate forwarding set is 119865119894 = 1198991 119899119898 inwhich the priority of nodes satisfies 1205881 gt gt 120588119898 We use119871119878119894 = 1198971198941 119897119894119898 to represent its forwarding candidate linkset For a candidate node 119899119895(1 le 119895 le 119898) its correspondingcandidate link is 119897119894119895 whose link state and PRR are 120595119894119895 and119901119894119895 respectively Assume that the probability of channel119888ℎ (119888ℎ120598119862) to be idle at 1199050 is 119875119888ℎ119894119889119897119890 Then we can estimate theidle probability of channel 119888ℎ at 1199051(1199051 gt 1199050) according to itsstate (busy or idle) at 1199050 which is shown as follows

119875119888ℎ119894119889119897119890 (1199051) = 119875119888ℎ119894119889119897119890 + (120576 minus 119875119888ℎ119894119889119897119890) 119890minus120591(1199051minus1199050) (8)

where 120591 = 120582119874119865119865119888 + 120582119874119873119888 and 120576 is a 0-1 variable The variable120576 indicates the state of channels where 120576 = 1 indicates thechannel 119888ℎ is idle at 1199051 and 120576 = 0 indicates the channel 119888ℎ isbusy at 1199051 Then the probability of channelrsquos idle state in theprocess of channel sensing denoted by119875119878119888ℎ119894119889119897119890 can be obtainedby

119875119878119888ℎ119894119889119897119890 = 119875119888ℎ119894119889119897119890 (1199051) ∙ 119875119888ℎ119894119889119897119890 (1199051 1199052) (9)

where119875119888ℎ119894119889119897119890(1199051 1199052) is the probability that channel stays idle overthe time (1199051 1199052) According to [16] 119875119888ℎ119894119889119897119890(1199051 1199052) can be derivedby

119875119888ℎ119894119889119897119890 (1199051 1199052) = intinfin1199052minus1199051

119865119888ℎ119874119865119865 (119909)119864 [119879119888ℎ119874119865119865]119889119909 (10)

6 Wireless Communications and Mobile Computing

One-hop relay

CCC

ch

ch

tt2t1t0

middot middot middot

PUActive

Energydetecti

on

data u ACK SNSIV SIFS

Figure 2 The process of one-hop opportunistic routing for CRAHNs

where 119865119888ℎ119874119865119865119864[119879119888ℎ119874119865119865] is the probability density function(PDF) of the residual time of the channel 119888ℎ remains idleMore specifically 119865119888ℎ119874119865119865(119909) is the cumulative distributionfunction (CDF) of the duration of the OFF state and 119864[119879119888ℎ119874119865119865]is the expected channel OFF time of the channel 119888ℎ

In the data transmission step due to the interferencesome candidate nodes cannot receive the packet successfullyWhen the SU node is disrupted by PUsrsquo appearance the linkbetween it and the sender becomes unusable Therefore thethroughput is directly related to the success rate of the linksin the opportunistic module According to the analysis resultabove we introduce the metric of effective forwarding rate(EFR) which indicates the success rate of data transmissionof each candidate link in the opportunistic module denotedby 120585 Thus for the candidate link 119897119894119895 of the candidate node119899119895(1 le 119895 le 119898) in the opportunistic module (119899119894 119865119894) theeffective forwarding rate can be denoted by 120585(119897119894119895) which isdefined in the following equation

120585 (119897119894119895) = 119902119894 ∙ 119901119894119895 ∙ 119895minus1prod119896=0

(1 minus 120595119894119896 ∙ 119901119894119896) (11)

where prod119895minus1

119896=0(1 minus 120595119894119896 ∙ 119901119894119896) is the probability that 119899119895 received

the packet successfully when all candidate nodes with higherpriority than node 119899119895 are failed to receive the packet

According to the (11) accumulating all EFR values oflinks in the candidate link set we can obtain the EFR of theopportunistic module as follows

120585 (119871119878119894) = 119898sum119895=1

120585 (119897119894119895) = 119902119894 ∙ (1 minus 119898prod119895=0

(1 minus 120593119894119895 ∙ 119901119894119895)) (12)

where prod119898119895=0(1 minus 120593119894119895 ∙ 119901119894119895) means that none of the candidate

nodes have received the packetAssume that the sender node 119899119894 chooses the channel119888ℎ as the data transmission channel Then according to

(9)-(11) we can deduce the equation for the probability of thecandidate node 119899119895 successfully receiving the packet denotedas 119875119903119890119897119886119910(119899119895) which is shown as follows

119875119903119890119897119886119910 (119899119895) = 120585 (119897119894119895) ∙ 119875119878119888ℎ119894119889119897119890 ∙ 119875119888ℎ119894119889119897119890 (1199052 1199053) (13)

According to the conclusions in [5] we can statethat when the link state is stable the channel remainsidle throughout the data forwarding process (ie 119875119878119888ℎ119894119889119897119890 ∙119875119888ℎ119894119889119897119890(1199052 1199053) = 1)

Based on (13) we can compute the cognitive transportthroughput (CTT) of one opportunistic module (119899119894 119865119894) inchannel 119888ℎ as follows

119862119879119879119888ℎ119894 = 119898sum119895=1

119875119888ℎ119903119890119897119886119910 (119899119895) ∙ ( 119871119879119903119890119897119886119910) (14)

where sum119898119895=1 119875119888ℎ119903119890119897119886119910(119899119895) is the data transmission success rate of

thewhole opportunisticmodule L is the data size transmittedby the opportunistic module and 119879119903119890119897119886119910 is the time requiredfor the whole process To facilitate the subsequent researchwe assume that the transmission time of all opportunisticmodules in the network is the same Therefore (14) can besimplified as follows

119862119879119879119888ℎ119894 = 119898sum119895=1

119875119888ℎ119903119890119897119886119910 (119899119895) ∙ 119892119871 (15)

Wireless Communications and Mobile Computing 7

43 Maximize the Total Throughput of the Network In thissubsection we formulate the optimization problem of net-work throughput as linear programming (LP) problem

P max sum119897isin119864119891isin119865

V119897119891 (16)

st V119891 + sum119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119899 = 119904 (119891) 119891 isin 119865 forall119899 isin 119881

(17)

sum119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119899 = 119904 (119891) 119899 = 119889 (119891) 119891 isin 119865 forall119899 isin 119881

(18)

sum119897isin119871119868(119899)

V119897119891 ge 0 forall119891 isin 119865 forall119899 isin 119881 (19)

sum119897isin1198710(119899)

V119897119891 = 0 119905 (119897) = 119889 (119891) forall119899 isin 119881 (20)

V119897119891 le 119862119897119886 119891 isin 119865 forall119897 isin 119864 (21)

sum119897isin119871119865

V119897119891 le 119862119871119878119886 119891 isin 119865 forall119897 isin 119871119878 (22)

sum119897isin119871119865

V119897119891 le 119862119879119879119888ℎ119894 forall119899 isin 119881 (23)

In the problem P we aim to maximize the throughput ofthe network as shown in (16) Equation (17) represents theflow conservation condition of the source and (18) representsthe flow conservation conditions of other nodes At eachnode except for the source and the destination the amountof incoming flow is equal to the amount of outgoing flowEquation (19) indicates that the amount of flow receivedby each node is nonnegative Equation (20) ensures thatthe outgoing flow from the destination of each data flowis 0 where 119905(119897) is the end node of link 119897 The physicalmeaning of (21) is that the actual flow delivered on each linkis constrained by the total amount of flow residing in thenetwork Equations (22) and (23) indicate that the amountof traffic passed through the opportunistic module is lessthan the available capacity and the throughput of the modulerespectively

In cognitive radio networks the state of the link willfrequently change due to the changing states (busy or idle)of the channel The dynamic link state may also lead tothe network topology change and make the candidate setbecome unavailable and nonoptimal Thus we cannot verifythe accuracy and effectiveness of the load balancing algo-rithm which is designed based on local network topologyinformation Moreover we cannot guarantee that the packetsare always transmitted through the links with lower trafficload Therefore a possible practical solution for this problemis distributing the traffic according to the load and thedata forwarding capacity of the link By this way we canimprove the network throughput while guaranteeing thecommunication quality

5 Opportunistic Routing

In this section we first propose a new metric to measurethe priority of candidate nodes and provide the sorting algo-rithm Then we design the load balancing based candidateforwarder selection algorithm under the stable condition ofthe link

51 Load Balancing Candidate Forwarder Sorting AlgorithmFirst we take the metric of linkrsquos reliability as an example toanalyze the reason for the uneven distribution of link loadIn the algorithms which are based on the link reliabilitywith the rise of the delivery rate of the candidate link thepriority of the corresponding candidate node will ascendBased on opportunistic routing the nodewith higher prioritywill receive more data As a result for links in the candidatelink set with the rise of the priority the load on the linkwill elevate As the amount of transmitted data increase thehigher priority linkrsquos traffic load also increases This case maycause the link blocked In the meantime the amount of dataallocated to the link with lower priority is small which maylead to the waste of link resources Therefore with the rise ofthe delivery rate of the candidate link the available capacityof the link will decline

511 Metric Design According to the consideration abovewe redesign a metric called Load Score to measure thepriority of candidate nodes based on the delivery rate andavailable capacity of each candidate link of an opportunisticmodule denoted asΘ In the opportunisticmodule (n F) F =1198991 119899119898 is the candidate node set and LS = 1198971 119897119898 isthe candidate link set For any candidate node 119899119890(1 le 119890 le 119898)the Load Score can be derived as follows

Θ119890 = 119901119890 ∙ 119862119897119890119886 (24)

The physical meaning of the metric is the maximumamount of data transmitted in a time slot when the linkhas the highest priority This metric considers both thetransmission quality of the link and load of the link Weestimate the prioritization of candidate nodes based on theLoad Score Thus it can achieve the balanced distributionof traffic load and reduce the number of packet drops andqueuing delays Meanwhile it ensures that the link transfersthe maximum amount of data Next we will analyze thespecial situations that may occur in the process of sortingcandidate nodes

According to the condition that multiple candidate nodesmay have the same Load Score we propose the followingproposition

Proposition 2 When multiple candidate nodes have thesame Load Score a higher priority candidate node is alwaysassociated with a lower packet delivery rate

Proof We assume there are 119911 candidate nodes 1198991 119899119911 inthe opportunistic module (nF) and assume they have thesame value of Load Scores denoted asΘ The correspondinglinks of nodes are denoted by 1198971 119897119911 We use 119866 to represent

8 Wireless Communications and Mobile Computing

the maximum throughput of link set 1198971 119897119911 in a steadystate Then we have

G = q ∙ 1199011 ∙ 1198741198971119886 + q ∙ 1199012 ∙ 1198621198972119886 (1 minus 1199011) + + q ∙ 119901119911

∙ 119862119897119911119886 119911minus1sum119896=0

(1 minus 119901119896)= 119902 ∙ Θ ∙ (1 + (1 minus 1199011) + + 119911minus1sum

119896=0

(1 minus 119901119896))(25)

In (25) 119902 and Θ are constant values Then the value of 119866 isdecided by 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) For conveniencewe use the notation Υ to denote the formula 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) Then the maximum value of 119866 can beachieved when Υ has the maximum value Obviously Υ isthe sum of the polynomials As a result the necessary andsufficient condition forΥ to attain the maximum value is thateach term ofΥmust reach the maximum valueThat is to say(1minus1199011) sum119911minus1

119896=0(1minus119901119896)must take the maximum value at thesame time Intuitively these items are inversely proportionalto the delivery rate of the candidate link Therefore we canconclude that the maximum Υ can be achieved when 1199011 lt1199012 lt lt 119901119911 Accordingly the condition 1198621198971119886 gt 1198621198972119886 gt gt119862119897119911119886 is valid Then Proposition 2 holds

512 Load BalancingCandidate Forwarder Sorting AlgorithmIn this subsection we will explain how to sort the candidatenodes in LS In the opportunistic module (n F) F =1198991 119899119898 is the candidate node set FL = 1198971 119897119898 isthe candidate link set and PR = 1199011 119901119898 is a PRR setEach PRR corresponds to a candidate link Based on (24) wecan calculate the Load Score of each candidate node which isdenoted as H = Θ1 Θ119898 We develop a simple selectionsorting algorithm to sort H = Θ1 Θ119898 which is shownas in Algorithm 1

In Algorithm 1 the inputs are the candidate node set119865 the PRR set 119875119877 and the LS set 119867 The algorithm sortscandidate nodes based on their LS In the algorithm line(4) is used to mark the highest priority nodes for eachiteration and lines (6) to (12) indicate the priority orderingprocess According to the process of the algorithm the timecomplexity of this algorithm is O(m2) where m representsthe number of candidate nodes

For the ordered candidate node set 1198651015840 we can calcu-late the maximum throughput of the opportunistic module(n 1198651015840) according to (11) which is shown as follows

CTT = q ∙ Θ1 + q ∙ Θ2 (1 minus 1199011) + + q

∙ Θ119898

119898minus1sum119896=0

(1 minus 119901119897119896) (26)

52 Load Balancing Based Relay Selection Algorithm In thissection we further study the load balancing solution InCRAHNs the packet will be forwarded to the destinationthrough several opportunistic modules until the packet

Input F = 1198991 119899119898 PR = 1199011 119901119898 H = Θ1 Θ119898Output An ordered candidate node set 1198651015840(1) Initialize 1198651015840 = (2) Initialize a b k(3) for eachΘ119886120598H and 1 le a le m do(4) blarr997888 a(5) for eachΘ119896120598H and a + 1 le k le m do(6) if Θ119887 lt Θ119896 then(7) blarr997888 k(8) else if Θ119887 = Θ119896 then(9) if 119875119886 gt 119875119896 then(10) blarr997888 k(11) end if(12) end if(13) end for(14) 1198651015840119886 larr997888 119865119887(15) end for(16) output 1198651015840(17) end

Algorithm 1 Load Balancing Based Candidate Forwarder SortingAlgorithm

arrives to its destination In each opportunistic module theselection of the last hop is limited by the forwarding capabilityof the next hop To achieve the global optimal it is necessaryto predict the forwarding capability of the opportunisticmodule which is the final hop Then we determine thecandidate node set of the first hop opportunistic modulenamely the optimal candidate node set for the first hopopportunistic module However this method of selectingthe optimal opportunistic module is not appropriate Inthe cognitive radio network the link status is dynamic dueto the dynamic activities of PUs Hence we cannot verifythat the prediction of the forwarding ability of the last hopopportunistic module is convergent Thus we cannot provethe optimal opportunistic module is accurate In [16] theauthor chose an intermediate destinationnode to alleviate theinfluence of the dynamic change of link state Inspired by thismethod we adopt an opportunistic module iteration methodto determine the optimal candidate node set of the currentopportunistic module Since this method is based on theforwarding ability of the next hop opportunistic module it isnecessary to predict the occupied capacity of each candidatelink in this opportunistic module

In this paper we use Minimum Mean Square Error(MMSE) [28] based flow prediction model to estimate theoccupied capacity of the link in the next time slot Theformula of the model is as follows

119862119897119906 = MMSE (119883119905 | t = 0 1 2 ) (27)

where 119883119905|t = 0 1 2 is the current flow and historicalflow statistics In [28] the author describes the predictionscheme in detail and readers can refer to it for more details

For the opportunistic module (n F) the candidate nodeset is F = 1198991 119899119898 the candidate link set is LS =1198971 119897119898 and the corresponding LS set isH = Θ1 Θ119898

Wireless Communications and Mobile Computing 9

Input F = n1 nm H = Θ1 ΘmOutput The optimal node set Foptimal(1) Initialize arrays of H1015840 F1015840 F10158401015840 Foptimal and LS1015840(2) Initialize values of o1 o2 and ctt(3) for each ne120598F and 1 le e le m do(4) Search the opportunistic module(ne Fe) of ne(5) F1015840larr997888the candidate node set of (ne Fe)(6) LS1015840larr997888the candidate link set of (ne Fe)(7) for each n1015840k120598F and 1 le k le |Fe| do(8) o1 larr997888 the used capacity of l1015840k(9) o2 larr997888 the available capacity of l1015840k(10) H1015840

k larr997888 the LS of n1015840k(11) end for(12) F10158401015840 larr997888 sorting F1015840 according to Algorithm 1(13) cttlarr997888 the throughput of (ne F10158401015840)(14) if Θe gt ctt do(15) Θe larr997888 ctt(16) end if(17) end for(18) Foptimal larr997888 sorting F according to Algorithm 1(19) output Foptimal(20) end

Algorithm 2 Load Balancing Relay Selection Algorithm

The selection steps of the optimal opportunistic module aredescribed as follows

(1) For any candidate node 119899119890 isin 119865 we first determineits opportunistic module (119899119890 119865119890) according to relay distancesThen we perform the following steps (1) predicting theoccupied capacity of each link in candidate link set 119865119871119890 basedon (27) (2) calculating the available capacity of each link in119865119871119890 based on (14) (3) computing the LS of each node in thecandidate node set 119865119890 based on (24) (4) sorting the candidatenode set 119865119890 and obtaining an ordered candidate node set 1198651015840119890according to Algorithm 1 and (5) calculating the maximumthroughput CTT((119899119890 1198651015840119890)) of the opportunistic module basedon (26)

(2)We compare the LS valueΘ119890 of candidate node 119899119890 withCTT((119899119890 1198651015840119890)) and assign a smaller value to Θ119890

(3) We sort the candidate node set 119865 based on Algo-rithm 1 and the ordered set of candidate nodes is the optimalcandidate node set

The details of the Load Balancing Relay Selection Algo-rithm are shown in Algorithm 2 In Algorithm 2 lines (1)and (2) define the relevant parameters and their initializedvalues Lines (3) to (18) show the selection process for theoptimal candidate forwarding node According to the processof Algorithm 2 and the time complexity of Algorithm 1 thetime complexity of Algorithm 2 is O(m|1198651015840|2) where m isthe number of candidate nodes of the opportunistic module(nF) and |1198651015840| is the maximum number of candidate nodesin all the next hop opportunistic modules We can selectthe optimal opportunistic module for each hop according toAlgorithm 2 It balances the traffic load of each candidate linkwhile ensuring the quality of communication Meanwhile itimproves the throughput of the local network

6 Performance Evaluation

In this section we evaluate the performance of our proposedscheme with two well-known routing protocols for CRAHNsand present the simulation results on the SU-PU interferenceratio the packet delivery ratio the throughput and the end-to-end delay under different number of flows PUsrsquo activitiesnumber of channels and number of PUs

61 Simulation Settings The network parameter settings aresummarized in Table 2 In the simulations multiple SUsand PUs are randomly deployed in a 1000m times 1000marea and they are assumed to be stationary throughout thesimulation As stated in Section 31 the PU activities of eachchannel are modeled as an exponential ON-OFF processThe status of each channel is associated with two parametersone is the expected channel OFF time 119864[119879119874119865119865119888ℎ ] and theother is the probability of the idle state 119875119888ℎ119894119889119897119890 The channelstatus is estimated by utilizing periodic spectrum sensingand on-demand sensing before data transmissions In thesimulations 30 constant bit rate (CBR) flows are generatedbetween randomly chosen source-destination pairs of SUsand they are associated with the packet size of 512 Bytesand flow rate of 5Kbps Notice that the most representativeopportunistic routing scheme ExOR [9] adopted the methodin [29] to obtain the packet reception ratio between twonodes Accordingly in this simulation the packet receptionratios of the links are also based on the distance-to-deliveryratio relationshipmeasured in [29]Thus we require the posi-tion information of different nodes to compute the distancebetween neighboring nodes and obtain the packet receptionratios of the links We evaluate our scheme through NS-2simulation [30] In NS-2 our LBOR scheme is implementedas the routing agent In this work we only focus on thenetwork layer without considering MAC and physical layerissues Since the IEEE 80211 MAC protocol is customized forCR support we use the modified 80211cr as the CSMAMACagent in our simulation script We also apply CRN patch [31]to our simulations that can support multiple channels at thephysical layer

In the simulations we compare our LBOR scheme withSAAR scheme [15] and SAORprotocol [27]TheSAORproto-col is a well-knownmultipath opportunistic routing protocolcoupled with spectrum sensing and sharing in multichannelCRAHNs The SAAR scheme is a recently published any-path opportunistic routing scheme with consideration of theunreliable transmission feature and the uncertain spectrumavailability characteristic of CRAHNs

The performance metrics for our experiments are definedas follows

(i) The throughput is defined as the total amount oftraffic (in bits per second) an SU receiver receivesfrom the sender divided by the time it takes for thereceiver to obtain the last packet A routing schemewith higher throughput is desirable for CRAHNs

(ii) The end-to-end delay is defined as the average delaywhich is calculated by summing up the end-to-enddelays of all packets received by all SU destination

10 Wireless Communications and Mobile Computing

Table 2 Simulation parameters

Parameter valueNumber of SUs 100Number of PUs 25Transmission ranges of SUsand PUs 250m

Number of channels 6Channelsrsquo idle stateprobabilities1198751198881119894119889119897119890 1198751198886119894119889119897119890 03 03 05 05 07 07Expected channel OFF time 60msTotal number of flows 30Source-destination distance 800mSU CCC rate 1MbpsSU data channel rate 2MbpsPacket size 512 BytesSensing time 3msChannel switching time 100120583sRetransmission time 3

nodes and dividing it by the total number of receivedpackets A routing scheme with the lower end-to-enddelay is preferred for CRAHNs

(iii) The SU-PU interference ratio is defined as the ratioof the number of SUsrsquo packets interrupted by PUsrsquoactivities to the total number of packets delivered byan SU source node A routing scheme with lower SU-PU interference ratio is favorable for CRAHNs

(iv) The packet delivery ratio is defined as the ratio ofall successfully received data packets that are fullyreceived by the SU destination node to the total datapackets generated by the SU source node A routingscheme with higher packet delivery ratio is desirablefor CRAHNs

To ensure the statistical significance of the results we runeach experiment for 500 seconds and repeat it 1000 timeswithdifferent seeds to report the average value as final results

62 Simulation Results

621 Impact of Number of Flows In this part we evaluate theimpact of offered traffic load on the performance by changingthe number of flows We vary the number of flows from 15CBR flows to 35 flows

As shown in Figure 3(a) we observe that LBOR provideshigher throughput than SAAR and SAOR We first considerthe case that the number of flows is below the saturationlevel Recall that the throughput defined here is measuredby the aggregate throughput of all the flows In this casewhen the number of flows increases from 15 flows to 25flows the throughput of all three schemes starts increasing asmore SU receivers are involved andmore parallel concurrenttransmissions occur Then we consider the case that whenthe number of flows is larger than 25 this indicates that

the network traffic is becoming saturated In this case thethroughput gain of SAAR and SAOR starts to decline As thenetwork resources (available channels and links) are limitedthe increasing number of flows intensifies the congestion andinterference which explains the throughput degradation ofSAAR and SAOR However when more flows are involvedthe throughput of LBOR still increases with increasingnumber of flows This is because LBOR has a load balancingfeature and solves the congestion problem by routing thetraffic through alternate forwarders

Figure 3(b) shows that LBOR has superior end-to-enddelay performance as compared to SAAR and SAOR Whenthe number of flows is equal to 15 the performance gapbetween LBOR and SAAR (both with SAOR) is small In thiscase each relay node does not need to handle toomuch trafficfor multiple flows Thus the unfair distribution of networktraffic will not lead to extra end-to-end delay Neverthelessif the number of flows is above 25 the end-to-end delayof SAAR and SAOR increase more sharply than that ofLBOR This can be attributed to the ability of LBOR that itsends packets to the links with lower traffic load and delayMoreover it can be concluded from the result that LBORfacilitates a better distribution of traffic and the consequentreduction of queuing times contributes to a lower overalldelay

622 Impact of PUsrsquo Activities In this part we evaluate theperformance of LBOR under different PUsrsquo activity patternsThe expected channel OFF time 119864[119879119874119865119865119888ℎ ] varies from 20msto 120ms A small 119864[119879119874119865119865119888ℎ ] means that the PUsrsquo activitiesare high and thus the delivery of the secondary user ismore likely to be interrupted Different from the previousexperiment the number of CBR flows is fixed to be 30

Figure 4(a) shows the SU-PU interference ratio of thethree schemes under different values of 119864[119879119874119865119865119888ℎ ] In mostcases LBOR produces significantly lower SU-PU interferenceas compared to SAAR and SAOR This is because the trafficload of LBOR is distributed efficiently and fairly In LBORfewer flows will have a chance to share the same SU nodewhich is highly influenced by PUsrsquo behaviors Moreoverwe notice that the improvement of load balancing schemeis significant under high PUsrsquo activities (119864[119879119874119865119865119888ℎ ] is below60ms) but not under low PUsrsquo activities (119864[119879119874119865119865119888ℎ ]) is above60ms) The reason behind this phenomenon is that whenthe nodes are highly influenced by PUsrsquo behaviors andoverloaded LBOR will give priorities to other nodes with alower traffic load to release the burden of these nodes

Figure 4(b) displays the throughput of LBOR SAARand SAOR with different values of 119864[119879119874119865119865119888ℎ ] It shows thatthe throughput of these schemes increase as the value of119864[119879119874119865119865119888ℎ ] becomes larger The result also shows that LBORachieves higher throughput compared to SAAR and SAORWhen the PUsrsquo activities decrease (the value of 119864[119879119874119865119865119888ℎ ]increases) the performance difference between LBOR andSAAR (both with SAOR) becomes negligible Moreover thetraffic load of LBOR is evenly distributed across the SUsand different available channels Therefore it is expected that

Wireless Communications and Mobile Computing 11

15 20 25 30 35Number of flows

170

180

190

200

210

220

230

240

250

260

270

ro

ughp

ut (K

bps)

LBORSAARSAOR(a) Throughput versus varying number of flows

15 20 25 30 35Number of flows

25

30

35

40

45

50

55

60

65

70

End-

to-e

nd d

elay

(ms)

SAORSAARLBOR(b) End-to-end delay versus varying number of flows

Figure 3

more improvements can be obtained for LBOR when PUsappear frequently

The end-to-end delay and the packet delivery ratio per-formance with different types of PUsrsquo activities are shownin Figures 4(c) and 4(d) With the increase of 119864[119879119874119865119865119888ℎ ]the packet delivery ratio will increase and the end-to-enddelay will decrease When the value of 119864[119879119874119865119865119888ℎ ] is around20ms LBOR can reduce the end-to-end delay by up to36 and 40 compared to SAAR and SAOR HoweverLBOR can only increase the delivery ratio by up to 2and 3 respectively compared to these two schemes Thereare two main reasons for this case First the opportunisticrouting protocol of LBOR is based on multipath routingstrategy where the paths and relay nodes are determinedon-the-fly In addition SAAR and SAOR also utilize thesimilar opportunistic routing principle and they aim toincrease the packet delivery ratio for the dynamic changingspectrum environment Second SAAR and SAOR alwayschoose paths or forwarding nodes with the best performanceof packet delivery ratio Nevertheless LBOR jointly considersthe load balancing for selecting the relay nodes Thus theimprovement of packet delivery ratio from LBOR is limited

623 Impact ofNumber of Channels In this part we comparethe performance of the three schemes under different numberof channels The number of channels varies between 2 and 10and the expected channel OFF time is set to be 60ms

Figure 5(a) compares the SU-PU interference ratio of thethree schemes by varying the number of channels of thenetworkWe can observe that the SU-PU interference ratio ofLBOR is lower than that of SAAR and SAOR One significantobservationhere is thatwhen the number of channels is largerthan 6 the SU-PU interference ratio of SAAR and SAOR

tends to decrease as the number of channels is increased InCRAHNs a larger number of channels can service a largernumber of SUsThus in this case SUs can choosemore stablechannels to avoid PUsrsquo interruptions In addition if an SUis interrupted by PUsrsquo appearance it can easily find anotheravailable channel for data transmissions Another importantobservation made is that when the number of channelsincreases from 2 to 6 the SU-PU interference ratio of LBORwill increase but the other two schemes will decrease Anexplanation for this phenomenon is as follows Consideringthe load balancing strategy LBOR tries its best to distributetraffic evenly among multiple nodes When there are notenough available channels to avoid the mutual interferencesamong SUs the packets cannot be correctly decoded at thereceiver Due to the same reason we can also observe fromFigures 5(b) 5(c) and 5(d) that when the number of channelsis below 6 the improvements from load balancing are notsignificantly

Figure 5(b) shows the throughput of LBOR SAAR andSAOR with different number of channels in the network Weobserve that the throughput of these schemes increases as thenumber of channels becomes larger The result also confirmsthat nomatter howmany channels are available LBOR alwaysperforms better than SAAR and SAOR When the number ofchannels is above 6 more improvement can be obtained fromLBOR

As shown in Figures 5(c) and 5(d) they provide similarresults and demonstrate that LBOR can achieve lower end-to-end delay and higher packet delivery ratio in most casescompared to SAAR and SAOR The improvements comefrom the load balancing based metric and the load balanc-ing based relay selection algorithm On the one hand theproposed scheme adopts a load balancing based metric to

12 Wireless Communications and Mobile Computing

20 40 60 80 100 120E[Tch

OFF] (ms)

0

002

004

006

008

01

012

014

016

018

02

SU-P

U in

terf

eren

ce ra

tio

LBORSAARSAOR(a) SU-PU interference ratio versus PUsrsquo activities

E[TchOFF] (ms)

20 40 60 80 100 120100

150

200

250

300

350

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus PUsrsquo activities

20 40 60 80 100 12025

30

35

40

45

50

55

60

65

End-

to-e

nd d

elay

(ms)

E[TchOFF] (ms)

LBORSAARSAOR

(c) End-to-end delay versus PUsrsquo activities

20 40 60 80 100 120076

078

08

082

084

086

088

09

092

094

Pack

et d

eliv

ery

ratio

LBORSAARSAOR

E[TchOFF] (ms)

(d) Packet delivery ratio versus PUsrsquo activities

Figure 4

choose the relay nodes with higher throughput and lowerdelay On the other hand the load balancing based relayselection algorithm can predict the occupied capacity of eachlink and thus ensure that the traffic load is evenly balancedamong all of the relay nodes

624 Impact of Number of PUs In this experiment weevaluate the performance of three schemes under differentnumber of PUs The number of PUs varies between 10 and40 and the number of channels is set to be 6 Figure 6(a)

compares the SU-PU interference ratio of different schemesunder different number of PUs In most cases the SU-PU interference ratios of SAAR and SAOR rise sharplythan LBOR That is because the network traffic of nonloadbalancing methods (SAAR and SAOR) tends to concentrateon a specific number of SUs and links If these SUsrsquo channelsare occupied by PUs most packets of the network may beaffected On the contrary LBOR distributes traffic to a largernumber of SUs thus packets have less change to be affectedby the PUsrsquo appearance and it induces a load balancing

Wireless Communications and Mobile Computing 13

2 4 6 8 10Total number of channels

002

004

006

008

01

012

014

016

018SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of channels

2 4 6 8 10Total number of channels

160

180

200

220

240

260

280

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of channels

2 4 6 8 10Total number of channels

20

30

40

50

60

70

80

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of channels

2 4 6 8 10Total number of channels

07

075

08

085

09

095Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of channels

Figure 5

improvement compared with SAAR and SAOR When thenumber of PUs is below 30 the benefit from load balancingis not surprising since SUs are not heavily affected Whenthe number of PUs is 35 an interesting observation here isthat the SU-PU interference ratio of LBOR is lower than thecase of 30 The reason for this case is that the aggregatedbenefit from load balancing scheme overwhelms the negativeeffect of the increased number of PUs More specifically inLBOR SUs are not overloaded and the packets are properlydistributed Thus when data transmission failures are causedby the appearance of PUs it is much easier for LBOR toreroute the blocked trafficflows to the light-loaded SUswhose

channels are available compared with other two methodsWhen the number of PUs is 40 we observe that the SU-PUinterference ratio of LBOR is higher than the case of 35 In thissituation there are not enough available channels for SUs totransmit the traffic of thewhole networkThus it is difficult tofind light-loaded SUs to reroute packets for overloaded SUsand more packets would be interrupted by PUsrsquo activities

Figure 6(b) shows the throughput by varying the numberof PUs in the network for the three schemes In the figurewe observe that the proposed LBOR scheme achieves higherthroughput than SAAR and SAOR We can also find thatwhen the number of PUs increases the throughput decreases

14 Wireless Communications and Mobile Computing

10 15 20 25 30 35 40Total number of PUs

003

004

005

006

007

008

009

01

011

012

013SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

160

180

200

220

240

260

280

300

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

25

30

35

40

45

50

55

60

65

70

75

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

078

08

082

084

086

088

09

092Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of PUs

Figure 6

because of available resources are decaying Moreover whenthe number of PUs is small the improvement from LBOR islimited On the contrary when the number of PUs rises above25 the throughput of SAAR and SAOR drops drastically butthe throughput degradation of LBOR is limited This is dueto the fact that LBOR can predict the availability and theoccupied capacity of each link and thus relieve the networkrsquoscongestions

Figures 6(c) and 6(d) can also confirm that LBOR canprovide the lowest end-to-end delay and the highest packetdelivery ratio among the three schemes When the numberof PUs increases the unbalanced distribution of load amongnodes becomes a critical issue and has a negative impacton the end-to-end delay and the packet delivery ratio Thiscan be understood intuitively as follows On the one handthe previous link may become unavailable due to the PU

appearance and the packets which arrived earlier have towait for late packets in reordering buffers at the receivingdestination thus resulting in an increasing queuing delayOn the other hand if late packets do not arrive within theimposed deadline it will be treated as a lost one thus resultingin an increasing packet loss

7 Conclusion

In this paper we have proposed a load balancing oppor-tunistic routing scheme for cognitive radio networks Con-sidering the loading constraint of each node we analyzedthe throughput bound of one opportunistic module andformulated the problem of maximizing the total throughputof the network as a linear programming problem Moreoverwe designed a new metric to select and prioritize the

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

6 Wireless Communications and Mobile Computing

One-hop relay

CCC

ch

ch

tt2t1t0

middot middot middot

PUActive

Energydetecti

on

data u ACK SNSIV SIFS

Figure 2 The process of one-hop opportunistic routing for CRAHNs

where 119865119888ℎ119874119865119865119864[119879119888ℎ119874119865119865] is the probability density function(PDF) of the residual time of the channel 119888ℎ remains idleMore specifically 119865119888ℎ119874119865119865(119909) is the cumulative distributionfunction (CDF) of the duration of the OFF state and 119864[119879119888ℎ119874119865119865]is the expected channel OFF time of the channel 119888ℎ

In the data transmission step due to the interferencesome candidate nodes cannot receive the packet successfullyWhen the SU node is disrupted by PUsrsquo appearance the linkbetween it and the sender becomes unusable Therefore thethroughput is directly related to the success rate of the linksin the opportunistic module According to the analysis resultabove we introduce the metric of effective forwarding rate(EFR) which indicates the success rate of data transmissionof each candidate link in the opportunistic module denotedby 120585 Thus for the candidate link 119897119894119895 of the candidate node119899119895(1 le 119895 le 119898) in the opportunistic module (119899119894 119865119894) theeffective forwarding rate can be denoted by 120585(119897119894119895) which isdefined in the following equation

120585 (119897119894119895) = 119902119894 ∙ 119901119894119895 ∙ 119895minus1prod119896=0

(1 minus 120595119894119896 ∙ 119901119894119896) (11)

where prod119895minus1

119896=0(1 minus 120595119894119896 ∙ 119901119894119896) is the probability that 119899119895 received

the packet successfully when all candidate nodes with higherpriority than node 119899119895 are failed to receive the packet

According to the (11) accumulating all EFR values oflinks in the candidate link set we can obtain the EFR of theopportunistic module as follows

120585 (119871119878119894) = 119898sum119895=1

120585 (119897119894119895) = 119902119894 ∙ (1 minus 119898prod119895=0

(1 minus 120593119894119895 ∙ 119901119894119895)) (12)

where prod119898119895=0(1 minus 120593119894119895 ∙ 119901119894119895) means that none of the candidate

nodes have received the packetAssume that the sender node 119899119894 chooses the channel119888ℎ as the data transmission channel Then according to

(9)-(11) we can deduce the equation for the probability of thecandidate node 119899119895 successfully receiving the packet denotedas 119875119903119890119897119886119910(119899119895) which is shown as follows

119875119903119890119897119886119910 (119899119895) = 120585 (119897119894119895) ∙ 119875119878119888ℎ119894119889119897119890 ∙ 119875119888ℎ119894119889119897119890 (1199052 1199053) (13)

According to the conclusions in [5] we can statethat when the link state is stable the channel remainsidle throughout the data forwarding process (ie 119875119878119888ℎ119894119889119897119890 ∙119875119888ℎ119894119889119897119890(1199052 1199053) = 1)

Based on (13) we can compute the cognitive transportthroughput (CTT) of one opportunistic module (119899119894 119865119894) inchannel 119888ℎ as follows

119862119879119879119888ℎ119894 = 119898sum119895=1

119875119888ℎ119903119890119897119886119910 (119899119895) ∙ ( 119871119879119903119890119897119886119910) (14)

where sum119898119895=1 119875119888ℎ119903119890119897119886119910(119899119895) is the data transmission success rate of

thewhole opportunisticmodule L is the data size transmittedby the opportunistic module and 119879119903119890119897119886119910 is the time requiredfor the whole process To facilitate the subsequent researchwe assume that the transmission time of all opportunisticmodules in the network is the same Therefore (14) can besimplified as follows

119862119879119879119888ℎ119894 = 119898sum119895=1

119875119888ℎ119903119890119897119886119910 (119899119895) ∙ 119892119871 (15)

Wireless Communications and Mobile Computing 7

43 Maximize the Total Throughput of the Network In thissubsection we formulate the optimization problem of net-work throughput as linear programming (LP) problem

P max sum119897isin119864119891isin119865

V119897119891 (16)

st V119891 + sum119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119899 = 119904 (119891) 119891 isin 119865 forall119899 isin 119881

(17)

sum119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119899 = 119904 (119891) 119899 = 119889 (119891) 119891 isin 119865 forall119899 isin 119881

(18)

sum119897isin119871119868(119899)

V119897119891 ge 0 forall119891 isin 119865 forall119899 isin 119881 (19)

sum119897isin1198710(119899)

V119897119891 = 0 119905 (119897) = 119889 (119891) forall119899 isin 119881 (20)

V119897119891 le 119862119897119886 119891 isin 119865 forall119897 isin 119864 (21)

sum119897isin119871119865

V119897119891 le 119862119871119878119886 119891 isin 119865 forall119897 isin 119871119878 (22)

sum119897isin119871119865

V119897119891 le 119862119879119879119888ℎ119894 forall119899 isin 119881 (23)

In the problem P we aim to maximize the throughput ofthe network as shown in (16) Equation (17) represents theflow conservation condition of the source and (18) representsthe flow conservation conditions of other nodes At eachnode except for the source and the destination the amountof incoming flow is equal to the amount of outgoing flowEquation (19) indicates that the amount of flow receivedby each node is nonnegative Equation (20) ensures thatthe outgoing flow from the destination of each data flowis 0 where 119905(119897) is the end node of link 119897 The physicalmeaning of (21) is that the actual flow delivered on each linkis constrained by the total amount of flow residing in thenetwork Equations (22) and (23) indicate that the amountof traffic passed through the opportunistic module is lessthan the available capacity and the throughput of the modulerespectively

In cognitive radio networks the state of the link willfrequently change due to the changing states (busy or idle)of the channel The dynamic link state may also lead tothe network topology change and make the candidate setbecome unavailable and nonoptimal Thus we cannot verifythe accuracy and effectiveness of the load balancing algo-rithm which is designed based on local network topologyinformation Moreover we cannot guarantee that the packetsare always transmitted through the links with lower trafficload Therefore a possible practical solution for this problemis distributing the traffic according to the load and thedata forwarding capacity of the link By this way we canimprove the network throughput while guaranteeing thecommunication quality

5 Opportunistic Routing

In this section we first propose a new metric to measurethe priority of candidate nodes and provide the sorting algo-rithm Then we design the load balancing based candidateforwarder selection algorithm under the stable condition ofthe link

51 Load Balancing Candidate Forwarder Sorting AlgorithmFirst we take the metric of linkrsquos reliability as an example toanalyze the reason for the uneven distribution of link loadIn the algorithms which are based on the link reliabilitywith the rise of the delivery rate of the candidate link thepriority of the corresponding candidate node will ascendBased on opportunistic routing the nodewith higher prioritywill receive more data As a result for links in the candidatelink set with the rise of the priority the load on the linkwill elevate As the amount of transmitted data increase thehigher priority linkrsquos traffic load also increases This case maycause the link blocked In the meantime the amount of dataallocated to the link with lower priority is small which maylead to the waste of link resources Therefore with the rise ofthe delivery rate of the candidate link the available capacityof the link will decline

511 Metric Design According to the consideration abovewe redesign a metric called Load Score to measure thepriority of candidate nodes based on the delivery rate andavailable capacity of each candidate link of an opportunisticmodule denoted asΘ In the opportunisticmodule (n F) F =1198991 119899119898 is the candidate node set and LS = 1198971 119897119898 isthe candidate link set For any candidate node 119899119890(1 le 119890 le 119898)the Load Score can be derived as follows

Θ119890 = 119901119890 ∙ 119862119897119890119886 (24)

The physical meaning of the metric is the maximumamount of data transmitted in a time slot when the linkhas the highest priority This metric considers both thetransmission quality of the link and load of the link Weestimate the prioritization of candidate nodes based on theLoad Score Thus it can achieve the balanced distributionof traffic load and reduce the number of packet drops andqueuing delays Meanwhile it ensures that the link transfersthe maximum amount of data Next we will analyze thespecial situations that may occur in the process of sortingcandidate nodes

According to the condition that multiple candidate nodesmay have the same Load Score we propose the followingproposition

Proposition 2 When multiple candidate nodes have thesame Load Score a higher priority candidate node is alwaysassociated with a lower packet delivery rate

Proof We assume there are 119911 candidate nodes 1198991 119899119911 inthe opportunistic module (nF) and assume they have thesame value of Load Scores denoted asΘ The correspondinglinks of nodes are denoted by 1198971 119897119911 We use 119866 to represent

8 Wireless Communications and Mobile Computing

the maximum throughput of link set 1198971 119897119911 in a steadystate Then we have

G = q ∙ 1199011 ∙ 1198741198971119886 + q ∙ 1199012 ∙ 1198621198972119886 (1 minus 1199011) + + q ∙ 119901119911

∙ 119862119897119911119886 119911minus1sum119896=0

(1 minus 119901119896)= 119902 ∙ Θ ∙ (1 + (1 minus 1199011) + + 119911minus1sum

119896=0

(1 minus 119901119896))(25)

In (25) 119902 and Θ are constant values Then the value of 119866 isdecided by 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) For conveniencewe use the notation Υ to denote the formula 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) Then the maximum value of 119866 can beachieved when Υ has the maximum value Obviously Υ isthe sum of the polynomials As a result the necessary andsufficient condition forΥ to attain the maximum value is thateach term ofΥmust reach the maximum valueThat is to say(1minus1199011) sum119911minus1

119896=0(1minus119901119896)must take the maximum value at thesame time Intuitively these items are inversely proportionalto the delivery rate of the candidate link Therefore we canconclude that the maximum Υ can be achieved when 1199011 lt1199012 lt lt 119901119911 Accordingly the condition 1198621198971119886 gt 1198621198972119886 gt gt119862119897119911119886 is valid Then Proposition 2 holds

512 Load BalancingCandidate Forwarder Sorting AlgorithmIn this subsection we will explain how to sort the candidatenodes in LS In the opportunistic module (n F) F =1198991 119899119898 is the candidate node set FL = 1198971 119897119898 isthe candidate link set and PR = 1199011 119901119898 is a PRR setEach PRR corresponds to a candidate link Based on (24) wecan calculate the Load Score of each candidate node which isdenoted as H = Θ1 Θ119898 We develop a simple selectionsorting algorithm to sort H = Θ1 Θ119898 which is shownas in Algorithm 1

In Algorithm 1 the inputs are the candidate node set119865 the PRR set 119875119877 and the LS set 119867 The algorithm sortscandidate nodes based on their LS In the algorithm line(4) is used to mark the highest priority nodes for eachiteration and lines (6) to (12) indicate the priority orderingprocess According to the process of the algorithm the timecomplexity of this algorithm is O(m2) where m representsthe number of candidate nodes

For the ordered candidate node set 1198651015840 we can calcu-late the maximum throughput of the opportunistic module(n 1198651015840) according to (11) which is shown as follows

CTT = q ∙ Θ1 + q ∙ Θ2 (1 minus 1199011) + + q

∙ Θ119898

119898minus1sum119896=0

(1 minus 119901119897119896) (26)

52 Load Balancing Based Relay Selection Algorithm In thissection we further study the load balancing solution InCRAHNs the packet will be forwarded to the destinationthrough several opportunistic modules until the packet

Input F = 1198991 119899119898 PR = 1199011 119901119898 H = Θ1 Θ119898Output An ordered candidate node set 1198651015840(1) Initialize 1198651015840 = (2) Initialize a b k(3) for eachΘ119886120598H and 1 le a le m do(4) blarr997888 a(5) for eachΘ119896120598H and a + 1 le k le m do(6) if Θ119887 lt Θ119896 then(7) blarr997888 k(8) else if Θ119887 = Θ119896 then(9) if 119875119886 gt 119875119896 then(10) blarr997888 k(11) end if(12) end if(13) end for(14) 1198651015840119886 larr997888 119865119887(15) end for(16) output 1198651015840(17) end

Algorithm 1 Load Balancing Based Candidate Forwarder SortingAlgorithm

arrives to its destination In each opportunistic module theselection of the last hop is limited by the forwarding capabilityof the next hop To achieve the global optimal it is necessaryto predict the forwarding capability of the opportunisticmodule which is the final hop Then we determine thecandidate node set of the first hop opportunistic modulenamely the optimal candidate node set for the first hopopportunistic module However this method of selectingthe optimal opportunistic module is not appropriate Inthe cognitive radio network the link status is dynamic dueto the dynamic activities of PUs Hence we cannot verifythat the prediction of the forwarding ability of the last hopopportunistic module is convergent Thus we cannot provethe optimal opportunistic module is accurate In [16] theauthor chose an intermediate destinationnode to alleviate theinfluence of the dynamic change of link state Inspired by thismethod we adopt an opportunistic module iteration methodto determine the optimal candidate node set of the currentopportunistic module Since this method is based on theforwarding ability of the next hop opportunistic module it isnecessary to predict the occupied capacity of each candidatelink in this opportunistic module

In this paper we use Minimum Mean Square Error(MMSE) [28] based flow prediction model to estimate theoccupied capacity of the link in the next time slot Theformula of the model is as follows

119862119897119906 = MMSE (119883119905 | t = 0 1 2 ) (27)

where 119883119905|t = 0 1 2 is the current flow and historicalflow statistics In [28] the author describes the predictionscheme in detail and readers can refer to it for more details

For the opportunistic module (n F) the candidate nodeset is F = 1198991 119899119898 the candidate link set is LS =1198971 119897119898 and the corresponding LS set isH = Θ1 Θ119898

Wireless Communications and Mobile Computing 9

Input F = n1 nm H = Θ1 ΘmOutput The optimal node set Foptimal(1) Initialize arrays of H1015840 F1015840 F10158401015840 Foptimal and LS1015840(2) Initialize values of o1 o2 and ctt(3) for each ne120598F and 1 le e le m do(4) Search the opportunistic module(ne Fe) of ne(5) F1015840larr997888the candidate node set of (ne Fe)(6) LS1015840larr997888the candidate link set of (ne Fe)(7) for each n1015840k120598F and 1 le k le |Fe| do(8) o1 larr997888 the used capacity of l1015840k(9) o2 larr997888 the available capacity of l1015840k(10) H1015840

k larr997888 the LS of n1015840k(11) end for(12) F10158401015840 larr997888 sorting F1015840 according to Algorithm 1(13) cttlarr997888 the throughput of (ne F10158401015840)(14) if Θe gt ctt do(15) Θe larr997888 ctt(16) end if(17) end for(18) Foptimal larr997888 sorting F according to Algorithm 1(19) output Foptimal(20) end

Algorithm 2 Load Balancing Relay Selection Algorithm

The selection steps of the optimal opportunistic module aredescribed as follows

(1) For any candidate node 119899119890 isin 119865 we first determineits opportunistic module (119899119890 119865119890) according to relay distancesThen we perform the following steps (1) predicting theoccupied capacity of each link in candidate link set 119865119871119890 basedon (27) (2) calculating the available capacity of each link in119865119871119890 based on (14) (3) computing the LS of each node in thecandidate node set 119865119890 based on (24) (4) sorting the candidatenode set 119865119890 and obtaining an ordered candidate node set 1198651015840119890according to Algorithm 1 and (5) calculating the maximumthroughput CTT((119899119890 1198651015840119890)) of the opportunistic module basedon (26)

(2)We compare the LS valueΘ119890 of candidate node 119899119890 withCTT((119899119890 1198651015840119890)) and assign a smaller value to Θ119890

(3) We sort the candidate node set 119865 based on Algo-rithm 1 and the ordered set of candidate nodes is the optimalcandidate node set

The details of the Load Balancing Relay Selection Algo-rithm are shown in Algorithm 2 In Algorithm 2 lines (1)and (2) define the relevant parameters and their initializedvalues Lines (3) to (18) show the selection process for theoptimal candidate forwarding node According to the processof Algorithm 2 and the time complexity of Algorithm 1 thetime complexity of Algorithm 2 is O(m|1198651015840|2) where m isthe number of candidate nodes of the opportunistic module(nF) and |1198651015840| is the maximum number of candidate nodesin all the next hop opportunistic modules We can selectthe optimal opportunistic module for each hop according toAlgorithm 2 It balances the traffic load of each candidate linkwhile ensuring the quality of communication Meanwhile itimproves the throughput of the local network

6 Performance Evaluation

In this section we evaluate the performance of our proposedscheme with two well-known routing protocols for CRAHNsand present the simulation results on the SU-PU interferenceratio the packet delivery ratio the throughput and the end-to-end delay under different number of flows PUsrsquo activitiesnumber of channels and number of PUs

61 Simulation Settings The network parameter settings aresummarized in Table 2 In the simulations multiple SUsand PUs are randomly deployed in a 1000m times 1000marea and they are assumed to be stationary throughout thesimulation As stated in Section 31 the PU activities of eachchannel are modeled as an exponential ON-OFF processThe status of each channel is associated with two parametersone is the expected channel OFF time 119864[119879119874119865119865119888ℎ ] and theother is the probability of the idle state 119875119888ℎ119894119889119897119890 The channelstatus is estimated by utilizing periodic spectrum sensingand on-demand sensing before data transmissions In thesimulations 30 constant bit rate (CBR) flows are generatedbetween randomly chosen source-destination pairs of SUsand they are associated with the packet size of 512 Bytesand flow rate of 5Kbps Notice that the most representativeopportunistic routing scheme ExOR [9] adopted the methodin [29] to obtain the packet reception ratio between twonodes Accordingly in this simulation the packet receptionratios of the links are also based on the distance-to-deliveryratio relationshipmeasured in [29]Thus we require the posi-tion information of different nodes to compute the distancebetween neighboring nodes and obtain the packet receptionratios of the links We evaluate our scheme through NS-2simulation [30] In NS-2 our LBOR scheme is implementedas the routing agent In this work we only focus on thenetwork layer without considering MAC and physical layerissues Since the IEEE 80211 MAC protocol is customized forCR support we use the modified 80211cr as the CSMAMACagent in our simulation script We also apply CRN patch [31]to our simulations that can support multiple channels at thephysical layer

In the simulations we compare our LBOR scheme withSAAR scheme [15] and SAORprotocol [27]TheSAORproto-col is a well-knownmultipath opportunistic routing protocolcoupled with spectrum sensing and sharing in multichannelCRAHNs The SAAR scheme is a recently published any-path opportunistic routing scheme with consideration of theunreliable transmission feature and the uncertain spectrumavailability characteristic of CRAHNs

The performance metrics for our experiments are definedas follows

(i) The throughput is defined as the total amount oftraffic (in bits per second) an SU receiver receivesfrom the sender divided by the time it takes for thereceiver to obtain the last packet A routing schemewith higher throughput is desirable for CRAHNs

(ii) The end-to-end delay is defined as the average delaywhich is calculated by summing up the end-to-enddelays of all packets received by all SU destination

10 Wireless Communications and Mobile Computing

Table 2 Simulation parameters

Parameter valueNumber of SUs 100Number of PUs 25Transmission ranges of SUsand PUs 250m

Number of channels 6Channelsrsquo idle stateprobabilities1198751198881119894119889119897119890 1198751198886119894119889119897119890 03 03 05 05 07 07Expected channel OFF time 60msTotal number of flows 30Source-destination distance 800mSU CCC rate 1MbpsSU data channel rate 2MbpsPacket size 512 BytesSensing time 3msChannel switching time 100120583sRetransmission time 3

nodes and dividing it by the total number of receivedpackets A routing scheme with the lower end-to-enddelay is preferred for CRAHNs

(iii) The SU-PU interference ratio is defined as the ratioof the number of SUsrsquo packets interrupted by PUsrsquoactivities to the total number of packets delivered byan SU source node A routing scheme with lower SU-PU interference ratio is favorable for CRAHNs

(iv) The packet delivery ratio is defined as the ratio ofall successfully received data packets that are fullyreceived by the SU destination node to the total datapackets generated by the SU source node A routingscheme with higher packet delivery ratio is desirablefor CRAHNs

To ensure the statistical significance of the results we runeach experiment for 500 seconds and repeat it 1000 timeswithdifferent seeds to report the average value as final results

62 Simulation Results

621 Impact of Number of Flows In this part we evaluate theimpact of offered traffic load on the performance by changingthe number of flows We vary the number of flows from 15CBR flows to 35 flows

As shown in Figure 3(a) we observe that LBOR provideshigher throughput than SAAR and SAOR We first considerthe case that the number of flows is below the saturationlevel Recall that the throughput defined here is measuredby the aggregate throughput of all the flows In this casewhen the number of flows increases from 15 flows to 25flows the throughput of all three schemes starts increasing asmore SU receivers are involved andmore parallel concurrenttransmissions occur Then we consider the case that whenthe number of flows is larger than 25 this indicates that

the network traffic is becoming saturated In this case thethroughput gain of SAAR and SAOR starts to decline As thenetwork resources (available channels and links) are limitedthe increasing number of flows intensifies the congestion andinterference which explains the throughput degradation ofSAAR and SAOR However when more flows are involvedthe throughput of LBOR still increases with increasingnumber of flows This is because LBOR has a load balancingfeature and solves the congestion problem by routing thetraffic through alternate forwarders

Figure 3(b) shows that LBOR has superior end-to-enddelay performance as compared to SAAR and SAOR Whenthe number of flows is equal to 15 the performance gapbetween LBOR and SAAR (both with SAOR) is small In thiscase each relay node does not need to handle toomuch trafficfor multiple flows Thus the unfair distribution of networktraffic will not lead to extra end-to-end delay Neverthelessif the number of flows is above 25 the end-to-end delayof SAAR and SAOR increase more sharply than that ofLBOR This can be attributed to the ability of LBOR that itsends packets to the links with lower traffic load and delayMoreover it can be concluded from the result that LBORfacilitates a better distribution of traffic and the consequentreduction of queuing times contributes to a lower overalldelay

622 Impact of PUsrsquo Activities In this part we evaluate theperformance of LBOR under different PUsrsquo activity patternsThe expected channel OFF time 119864[119879119874119865119865119888ℎ ] varies from 20msto 120ms A small 119864[119879119874119865119865119888ℎ ] means that the PUsrsquo activitiesare high and thus the delivery of the secondary user ismore likely to be interrupted Different from the previousexperiment the number of CBR flows is fixed to be 30

Figure 4(a) shows the SU-PU interference ratio of thethree schemes under different values of 119864[119879119874119865119865119888ℎ ] In mostcases LBOR produces significantly lower SU-PU interferenceas compared to SAAR and SAOR This is because the trafficload of LBOR is distributed efficiently and fairly In LBORfewer flows will have a chance to share the same SU nodewhich is highly influenced by PUsrsquo behaviors Moreoverwe notice that the improvement of load balancing schemeis significant under high PUsrsquo activities (119864[119879119874119865119865119888ℎ ] is below60ms) but not under low PUsrsquo activities (119864[119879119874119865119865119888ℎ ]) is above60ms) The reason behind this phenomenon is that whenthe nodes are highly influenced by PUsrsquo behaviors andoverloaded LBOR will give priorities to other nodes with alower traffic load to release the burden of these nodes

Figure 4(b) displays the throughput of LBOR SAARand SAOR with different values of 119864[119879119874119865119865119888ℎ ] It shows thatthe throughput of these schemes increase as the value of119864[119879119874119865119865119888ℎ ] becomes larger The result also shows that LBORachieves higher throughput compared to SAAR and SAORWhen the PUsrsquo activities decrease (the value of 119864[119879119874119865119865119888ℎ ]increases) the performance difference between LBOR andSAAR (both with SAOR) becomes negligible Moreover thetraffic load of LBOR is evenly distributed across the SUsand different available channels Therefore it is expected that

Wireless Communications and Mobile Computing 11

15 20 25 30 35Number of flows

170

180

190

200

210

220

230

240

250

260

270

ro

ughp

ut (K

bps)

LBORSAARSAOR(a) Throughput versus varying number of flows

15 20 25 30 35Number of flows

25

30

35

40

45

50

55

60

65

70

End-

to-e

nd d

elay

(ms)

SAORSAARLBOR(b) End-to-end delay versus varying number of flows

Figure 3

more improvements can be obtained for LBOR when PUsappear frequently

The end-to-end delay and the packet delivery ratio per-formance with different types of PUsrsquo activities are shownin Figures 4(c) and 4(d) With the increase of 119864[119879119874119865119865119888ℎ ]the packet delivery ratio will increase and the end-to-enddelay will decrease When the value of 119864[119879119874119865119865119888ℎ ] is around20ms LBOR can reduce the end-to-end delay by up to36 and 40 compared to SAAR and SAOR HoweverLBOR can only increase the delivery ratio by up to 2and 3 respectively compared to these two schemes Thereare two main reasons for this case First the opportunisticrouting protocol of LBOR is based on multipath routingstrategy where the paths and relay nodes are determinedon-the-fly In addition SAAR and SAOR also utilize thesimilar opportunistic routing principle and they aim toincrease the packet delivery ratio for the dynamic changingspectrum environment Second SAAR and SAOR alwayschoose paths or forwarding nodes with the best performanceof packet delivery ratio Nevertheless LBOR jointly considersthe load balancing for selecting the relay nodes Thus theimprovement of packet delivery ratio from LBOR is limited

623 Impact ofNumber of Channels In this part we comparethe performance of the three schemes under different numberof channels The number of channels varies between 2 and 10and the expected channel OFF time is set to be 60ms

Figure 5(a) compares the SU-PU interference ratio of thethree schemes by varying the number of channels of thenetworkWe can observe that the SU-PU interference ratio ofLBOR is lower than that of SAAR and SAOR One significantobservationhere is thatwhen the number of channels is largerthan 6 the SU-PU interference ratio of SAAR and SAOR

tends to decrease as the number of channels is increased InCRAHNs a larger number of channels can service a largernumber of SUsThus in this case SUs can choosemore stablechannels to avoid PUsrsquo interruptions In addition if an SUis interrupted by PUsrsquo appearance it can easily find anotheravailable channel for data transmissions Another importantobservation made is that when the number of channelsincreases from 2 to 6 the SU-PU interference ratio of LBORwill increase but the other two schemes will decrease Anexplanation for this phenomenon is as follows Consideringthe load balancing strategy LBOR tries its best to distributetraffic evenly among multiple nodes When there are notenough available channels to avoid the mutual interferencesamong SUs the packets cannot be correctly decoded at thereceiver Due to the same reason we can also observe fromFigures 5(b) 5(c) and 5(d) that when the number of channelsis below 6 the improvements from load balancing are notsignificantly

Figure 5(b) shows the throughput of LBOR SAAR andSAOR with different number of channels in the network Weobserve that the throughput of these schemes increases as thenumber of channels becomes larger The result also confirmsthat nomatter howmany channels are available LBOR alwaysperforms better than SAAR and SAOR When the number ofchannels is above 6 more improvement can be obtained fromLBOR

As shown in Figures 5(c) and 5(d) they provide similarresults and demonstrate that LBOR can achieve lower end-to-end delay and higher packet delivery ratio in most casescompared to SAAR and SAOR The improvements comefrom the load balancing based metric and the load balanc-ing based relay selection algorithm On the one hand theproposed scheme adopts a load balancing based metric to

12 Wireless Communications and Mobile Computing

20 40 60 80 100 120E[Tch

OFF] (ms)

0

002

004

006

008

01

012

014

016

018

02

SU-P

U in

terf

eren

ce ra

tio

LBORSAARSAOR(a) SU-PU interference ratio versus PUsrsquo activities

E[TchOFF] (ms)

20 40 60 80 100 120100

150

200

250

300

350

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus PUsrsquo activities

20 40 60 80 100 12025

30

35

40

45

50

55

60

65

End-

to-e

nd d

elay

(ms)

E[TchOFF] (ms)

LBORSAARSAOR

(c) End-to-end delay versus PUsrsquo activities

20 40 60 80 100 120076

078

08

082

084

086

088

09

092

094

Pack

et d

eliv

ery

ratio

LBORSAARSAOR

E[TchOFF] (ms)

(d) Packet delivery ratio versus PUsrsquo activities

Figure 4

choose the relay nodes with higher throughput and lowerdelay On the other hand the load balancing based relayselection algorithm can predict the occupied capacity of eachlink and thus ensure that the traffic load is evenly balancedamong all of the relay nodes

624 Impact of Number of PUs In this experiment weevaluate the performance of three schemes under differentnumber of PUs The number of PUs varies between 10 and40 and the number of channels is set to be 6 Figure 6(a)

compares the SU-PU interference ratio of different schemesunder different number of PUs In most cases the SU-PU interference ratios of SAAR and SAOR rise sharplythan LBOR That is because the network traffic of nonloadbalancing methods (SAAR and SAOR) tends to concentrateon a specific number of SUs and links If these SUsrsquo channelsare occupied by PUs most packets of the network may beaffected On the contrary LBOR distributes traffic to a largernumber of SUs thus packets have less change to be affectedby the PUsrsquo appearance and it induces a load balancing

Wireless Communications and Mobile Computing 13

2 4 6 8 10Total number of channels

002

004

006

008

01

012

014

016

018SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of channels

2 4 6 8 10Total number of channels

160

180

200

220

240

260

280

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of channels

2 4 6 8 10Total number of channels

20

30

40

50

60

70

80

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of channels

2 4 6 8 10Total number of channels

07

075

08

085

09

095Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of channels

Figure 5

improvement compared with SAAR and SAOR When thenumber of PUs is below 30 the benefit from load balancingis not surprising since SUs are not heavily affected Whenthe number of PUs is 35 an interesting observation here isthat the SU-PU interference ratio of LBOR is lower than thecase of 30 The reason for this case is that the aggregatedbenefit from load balancing scheme overwhelms the negativeeffect of the increased number of PUs More specifically inLBOR SUs are not overloaded and the packets are properlydistributed Thus when data transmission failures are causedby the appearance of PUs it is much easier for LBOR toreroute the blocked trafficflows to the light-loaded SUswhose

channels are available compared with other two methodsWhen the number of PUs is 40 we observe that the SU-PUinterference ratio of LBOR is higher than the case of 35 In thissituation there are not enough available channels for SUs totransmit the traffic of thewhole networkThus it is difficult tofind light-loaded SUs to reroute packets for overloaded SUsand more packets would be interrupted by PUsrsquo activities

Figure 6(b) shows the throughput by varying the numberof PUs in the network for the three schemes In the figurewe observe that the proposed LBOR scheme achieves higherthroughput than SAAR and SAOR We can also find thatwhen the number of PUs increases the throughput decreases

14 Wireless Communications and Mobile Computing

10 15 20 25 30 35 40Total number of PUs

003

004

005

006

007

008

009

01

011

012

013SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

160

180

200

220

240

260

280

300

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

25

30

35

40

45

50

55

60

65

70

75

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

078

08

082

084

086

088

09

092Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of PUs

Figure 6

because of available resources are decaying Moreover whenthe number of PUs is small the improvement from LBOR islimited On the contrary when the number of PUs rises above25 the throughput of SAAR and SAOR drops drastically butthe throughput degradation of LBOR is limited This is dueto the fact that LBOR can predict the availability and theoccupied capacity of each link and thus relieve the networkrsquoscongestions

Figures 6(c) and 6(d) can also confirm that LBOR canprovide the lowest end-to-end delay and the highest packetdelivery ratio among the three schemes When the numberof PUs increases the unbalanced distribution of load amongnodes becomes a critical issue and has a negative impacton the end-to-end delay and the packet delivery ratio Thiscan be understood intuitively as follows On the one handthe previous link may become unavailable due to the PU

appearance and the packets which arrived earlier have towait for late packets in reordering buffers at the receivingdestination thus resulting in an increasing queuing delayOn the other hand if late packets do not arrive within theimposed deadline it will be treated as a lost one thus resultingin an increasing packet loss

7 Conclusion

In this paper we have proposed a load balancing oppor-tunistic routing scheme for cognitive radio networks Con-sidering the loading constraint of each node we analyzedthe throughput bound of one opportunistic module andformulated the problem of maximizing the total throughputof the network as a linear programming problem Moreoverwe designed a new metric to select and prioritize the

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

Wireless Communications and Mobile Computing 7

43 Maximize the Total Throughput of the Network In thissubsection we formulate the optimization problem of net-work throughput as linear programming (LP) problem

P max sum119897isin119864119891isin119865

V119897119891 (16)

st V119891 + sum119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119899 = 119904 (119891) 119891 isin 119865 forall119899 isin 119881

(17)

sum119897isin119871119868(119899)

V119897119891 = sum119897isin119871119874(119899)

V119897119891119899 = 119904 (119891) 119899 = 119889 (119891) 119891 isin 119865 forall119899 isin 119881

(18)

sum119897isin119871119868(119899)

V119897119891 ge 0 forall119891 isin 119865 forall119899 isin 119881 (19)

sum119897isin1198710(119899)

V119897119891 = 0 119905 (119897) = 119889 (119891) forall119899 isin 119881 (20)

V119897119891 le 119862119897119886 119891 isin 119865 forall119897 isin 119864 (21)

sum119897isin119871119865

V119897119891 le 119862119871119878119886 119891 isin 119865 forall119897 isin 119871119878 (22)

sum119897isin119871119865

V119897119891 le 119862119879119879119888ℎ119894 forall119899 isin 119881 (23)

In the problem P we aim to maximize the throughput ofthe network as shown in (16) Equation (17) represents theflow conservation condition of the source and (18) representsthe flow conservation conditions of other nodes At eachnode except for the source and the destination the amountof incoming flow is equal to the amount of outgoing flowEquation (19) indicates that the amount of flow receivedby each node is nonnegative Equation (20) ensures thatthe outgoing flow from the destination of each data flowis 0 where 119905(119897) is the end node of link 119897 The physicalmeaning of (21) is that the actual flow delivered on each linkis constrained by the total amount of flow residing in thenetwork Equations (22) and (23) indicate that the amountof traffic passed through the opportunistic module is lessthan the available capacity and the throughput of the modulerespectively

In cognitive radio networks the state of the link willfrequently change due to the changing states (busy or idle)of the channel The dynamic link state may also lead tothe network topology change and make the candidate setbecome unavailable and nonoptimal Thus we cannot verifythe accuracy and effectiveness of the load balancing algo-rithm which is designed based on local network topologyinformation Moreover we cannot guarantee that the packetsare always transmitted through the links with lower trafficload Therefore a possible practical solution for this problemis distributing the traffic according to the load and thedata forwarding capacity of the link By this way we canimprove the network throughput while guaranteeing thecommunication quality

5 Opportunistic Routing

In this section we first propose a new metric to measurethe priority of candidate nodes and provide the sorting algo-rithm Then we design the load balancing based candidateforwarder selection algorithm under the stable condition ofthe link

51 Load Balancing Candidate Forwarder Sorting AlgorithmFirst we take the metric of linkrsquos reliability as an example toanalyze the reason for the uneven distribution of link loadIn the algorithms which are based on the link reliabilitywith the rise of the delivery rate of the candidate link thepriority of the corresponding candidate node will ascendBased on opportunistic routing the nodewith higher prioritywill receive more data As a result for links in the candidatelink set with the rise of the priority the load on the linkwill elevate As the amount of transmitted data increase thehigher priority linkrsquos traffic load also increases This case maycause the link blocked In the meantime the amount of dataallocated to the link with lower priority is small which maylead to the waste of link resources Therefore with the rise ofthe delivery rate of the candidate link the available capacityof the link will decline

511 Metric Design According to the consideration abovewe redesign a metric called Load Score to measure thepriority of candidate nodes based on the delivery rate andavailable capacity of each candidate link of an opportunisticmodule denoted asΘ In the opportunisticmodule (n F) F =1198991 119899119898 is the candidate node set and LS = 1198971 119897119898 isthe candidate link set For any candidate node 119899119890(1 le 119890 le 119898)the Load Score can be derived as follows

Θ119890 = 119901119890 ∙ 119862119897119890119886 (24)

The physical meaning of the metric is the maximumamount of data transmitted in a time slot when the linkhas the highest priority This metric considers both thetransmission quality of the link and load of the link Weestimate the prioritization of candidate nodes based on theLoad Score Thus it can achieve the balanced distributionof traffic load and reduce the number of packet drops andqueuing delays Meanwhile it ensures that the link transfersthe maximum amount of data Next we will analyze thespecial situations that may occur in the process of sortingcandidate nodes

According to the condition that multiple candidate nodesmay have the same Load Score we propose the followingproposition

Proposition 2 When multiple candidate nodes have thesame Load Score a higher priority candidate node is alwaysassociated with a lower packet delivery rate

Proof We assume there are 119911 candidate nodes 1198991 119899119911 inthe opportunistic module (nF) and assume they have thesame value of Load Scores denoted asΘ The correspondinglinks of nodes are denoted by 1198971 119897119911 We use 119866 to represent

8 Wireless Communications and Mobile Computing

the maximum throughput of link set 1198971 119897119911 in a steadystate Then we have

G = q ∙ 1199011 ∙ 1198741198971119886 + q ∙ 1199012 ∙ 1198621198972119886 (1 minus 1199011) + + q ∙ 119901119911

∙ 119862119897119911119886 119911minus1sum119896=0

(1 minus 119901119896)= 119902 ∙ Θ ∙ (1 + (1 minus 1199011) + + 119911minus1sum

119896=0

(1 minus 119901119896))(25)

In (25) 119902 and Θ are constant values Then the value of 119866 isdecided by 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) For conveniencewe use the notation Υ to denote the formula 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) Then the maximum value of 119866 can beachieved when Υ has the maximum value Obviously Υ isthe sum of the polynomials As a result the necessary andsufficient condition forΥ to attain the maximum value is thateach term ofΥmust reach the maximum valueThat is to say(1minus1199011) sum119911minus1

119896=0(1minus119901119896)must take the maximum value at thesame time Intuitively these items are inversely proportionalto the delivery rate of the candidate link Therefore we canconclude that the maximum Υ can be achieved when 1199011 lt1199012 lt lt 119901119911 Accordingly the condition 1198621198971119886 gt 1198621198972119886 gt gt119862119897119911119886 is valid Then Proposition 2 holds

512 Load BalancingCandidate Forwarder Sorting AlgorithmIn this subsection we will explain how to sort the candidatenodes in LS In the opportunistic module (n F) F =1198991 119899119898 is the candidate node set FL = 1198971 119897119898 isthe candidate link set and PR = 1199011 119901119898 is a PRR setEach PRR corresponds to a candidate link Based on (24) wecan calculate the Load Score of each candidate node which isdenoted as H = Θ1 Θ119898 We develop a simple selectionsorting algorithm to sort H = Θ1 Θ119898 which is shownas in Algorithm 1

In Algorithm 1 the inputs are the candidate node set119865 the PRR set 119875119877 and the LS set 119867 The algorithm sortscandidate nodes based on their LS In the algorithm line(4) is used to mark the highest priority nodes for eachiteration and lines (6) to (12) indicate the priority orderingprocess According to the process of the algorithm the timecomplexity of this algorithm is O(m2) where m representsthe number of candidate nodes

For the ordered candidate node set 1198651015840 we can calcu-late the maximum throughput of the opportunistic module(n 1198651015840) according to (11) which is shown as follows

CTT = q ∙ Θ1 + q ∙ Θ2 (1 minus 1199011) + + q

∙ Θ119898

119898minus1sum119896=0

(1 minus 119901119897119896) (26)

52 Load Balancing Based Relay Selection Algorithm In thissection we further study the load balancing solution InCRAHNs the packet will be forwarded to the destinationthrough several opportunistic modules until the packet

Input F = 1198991 119899119898 PR = 1199011 119901119898 H = Θ1 Θ119898Output An ordered candidate node set 1198651015840(1) Initialize 1198651015840 = (2) Initialize a b k(3) for eachΘ119886120598H and 1 le a le m do(4) blarr997888 a(5) for eachΘ119896120598H and a + 1 le k le m do(6) if Θ119887 lt Θ119896 then(7) blarr997888 k(8) else if Θ119887 = Θ119896 then(9) if 119875119886 gt 119875119896 then(10) blarr997888 k(11) end if(12) end if(13) end for(14) 1198651015840119886 larr997888 119865119887(15) end for(16) output 1198651015840(17) end

Algorithm 1 Load Balancing Based Candidate Forwarder SortingAlgorithm

arrives to its destination In each opportunistic module theselection of the last hop is limited by the forwarding capabilityof the next hop To achieve the global optimal it is necessaryto predict the forwarding capability of the opportunisticmodule which is the final hop Then we determine thecandidate node set of the first hop opportunistic modulenamely the optimal candidate node set for the first hopopportunistic module However this method of selectingthe optimal opportunistic module is not appropriate Inthe cognitive radio network the link status is dynamic dueto the dynamic activities of PUs Hence we cannot verifythat the prediction of the forwarding ability of the last hopopportunistic module is convergent Thus we cannot provethe optimal opportunistic module is accurate In [16] theauthor chose an intermediate destinationnode to alleviate theinfluence of the dynamic change of link state Inspired by thismethod we adopt an opportunistic module iteration methodto determine the optimal candidate node set of the currentopportunistic module Since this method is based on theforwarding ability of the next hop opportunistic module it isnecessary to predict the occupied capacity of each candidatelink in this opportunistic module

In this paper we use Minimum Mean Square Error(MMSE) [28] based flow prediction model to estimate theoccupied capacity of the link in the next time slot Theformula of the model is as follows

119862119897119906 = MMSE (119883119905 | t = 0 1 2 ) (27)

where 119883119905|t = 0 1 2 is the current flow and historicalflow statistics In [28] the author describes the predictionscheme in detail and readers can refer to it for more details

For the opportunistic module (n F) the candidate nodeset is F = 1198991 119899119898 the candidate link set is LS =1198971 119897119898 and the corresponding LS set isH = Θ1 Θ119898

Wireless Communications and Mobile Computing 9

Input F = n1 nm H = Θ1 ΘmOutput The optimal node set Foptimal(1) Initialize arrays of H1015840 F1015840 F10158401015840 Foptimal and LS1015840(2) Initialize values of o1 o2 and ctt(3) for each ne120598F and 1 le e le m do(4) Search the opportunistic module(ne Fe) of ne(5) F1015840larr997888the candidate node set of (ne Fe)(6) LS1015840larr997888the candidate link set of (ne Fe)(7) for each n1015840k120598F and 1 le k le |Fe| do(8) o1 larr997888 the used capacity of l1015840k(9) o2 larr997888 the available capacity of l1015840k(10) H1015840

k larr997888 the LS of n1015840k(11) end for(12) F10158401015840 larr997888 sorting F1015840 according to Algorithm 1(13) cttlarr997888 the throughput of (ne F10158401015840)(14) if Θe gt ctt do(15) Θe larr997888 ctt(16) end if(17) end for(18) Foptimal larr997888 sorting F according to Algorithm 1(19) output Foptimal(20) end

Algorithm 2 Load Balancing Relay Selection Algorithm

The selection steps of the optimal opportunistic module aredescribed as follows

(1) For any candidate node 119899119890 isin 119865 we first determineits opportunistic module (119899119890 119865119890) according to relay distancesThen we perform the following steps (1) predicting theoccupied capacity of each link in candidate link set 119865119871119890 basedon (27) (2) calculating the available capacity of each link in119865119871119890 based on (14) (3) computing the LS of each node in thecandidate node set 119865119890 based on (24) (4) sorting the candidatenode set 119865119890 and obtaining an ordered candidate node set 1198651015840119890according to Algorithm 1 and (5) calculating the maximumthroughput CTT((119899119890 1198651015840119890)) of the opportunistic module basedon (26)

(2)We compare the LS valueΘ119890 of candidate node 119899119890 withCTT((119899119890 1198651015840119890)) and assign a smaller value to Θ119890

(3) We sort the candidate node set 119865 based on Algo-rithm 1 and the ordered set of candidate nodes is the optimalcandidate node set

The details of the Load Balancing Relay Selection Algo-rithm are shown in Algorithm 2 In Algorithm 2 lines (1)and (2) define the relevant parameters and their initializedvalues Lines (3) to (18) show the selection process for theoptimal candidate forwarding node According to the processof Algorithm 2 and the time complexity of Algorithm 1 thetime complexity of Algorithm 2 is O(m|1198651015840|2) where m isthe number of candidate nodes of the opportunistic module(nF) and |1198651015840| is the maximum number of candidate nodesin all the next hop opportunistic modules We can selectthe optimal opportunistic module for each hop according toAlgorithm 2 It balances the traffic load of each candidate linkwhile ensuring the quality of communication Meanwhile itimproves the throughput of the local network

6 Performance Evaluation

In this section we evaluate the performance of our proposedscheme with two well-known routing protocols for CRAHNsand present the simulation results on the SU-PU interferenceratio the packet delivery ratio the throughput and the end-to-end delay under different number of flows PUsrsquo activitiesnumber of channels and number of PUs

61 Simulation Settings The network parameter settings aresummarized in Table 2 In the simulations multiple SUsand PUs are randomly deployed in a 1000m times 1000marea and they are assumed to be stationary throughout thesimulation As stated in Section 31 the PU activities of eachchannel are modeled as an exponential ON-OFF processThe status of each channel is associated with two parametersone is the expected channel OFF time 119864[119879119874119865119865119888ℎ ] and theother is the probability of the idle state 119875119888ℎ119894119889119897119890 The channelstatus is estimated by utilizing periodic spectrum sensingand on-demand sensing before data transmissions In thesimulations 30 constant bit rate (CBR) flows are generatedbetween randomly chosen source-destination pairs of SUsand they are associated with the packet size of 512 Bytesand flow rate of 5Kbps Notice that the most representativeopportunistic routing scheme ExOR [9] adopted the methodin [29] to obtain the packet reception ratio between twonodes Accordingly in this simulation the packet receptionratios of the links are also based on the distance-to-deliveryratio relationshipmeasured in [29]Thus we require the posi-tion information of different nodes to compute the distancebetween neighboring nodes and obtain the packet receptionratios of the links We evaluate our scheme through NS-2simulation [30] In NS-2 our LBOR scheme is implementedas the routing agent In this work we only focus on thenetwork layer without considering MAC and physical layerissues Since the IEEE 80211 MAC protocol is customized forCR support we use the modified 80211cr as the CSMAMACagent in our simulation script We also apply CRN patch [31]to our simulations that can support multiple channels at thephysical layer

In the simulations we compare our LBOR scheme withSAAR scheme [15] and SAORprotocol [27]TheSAORproto-col is a well-knownmultipath opportunistic routing protocolcoupled with spectrum sensing and sharing in multichannelCRAHNs The SAAR scheme is a recently published any-path opportunistic routing scheme with consideration of theunreliable transmission feature and the uncertain spectrumavailability characteristic of CRAHNs

The performance metrics for our experiments are definedas follows

(i) The throughput is defined as the total amount oftraffic (in bits per second) an SU receiver receivesfrom the sender divided by the time it takes for thereceiver to obtain the last packet A routing schemewith higher throughput is desirable for CRAHNs

(ii) The end-to-end delay is defined as the average delaywhich is calculated by summing up the end-to-enddelays of all packets received by all SU destination

10 Wireless Communications and Mobile Computing

Table 2 Simulation parameters

Parameter valueNumber of SUs 100Number of PUs 25Transmission ranges of SUsand PUs 250m

Number of channels 6Channelsrsquo idle stateprobabilities1198751198881119894119889119897119890 1198751198886119894119889119897119890 03 03 05 05 07 07Expected channel OFF time 60msTotal number of flows 30Source-destination distance 800mSU CCC rate 1MbpsSU data channel rate 2MbpsPacket size 512 BytesSensing time 3msChannel switching time 100120583sRetransmission time 3

nodes and dividing it by the total number of receivedpackets A routing scheme with the lower end-to-enddelay is preferred for CRAHNs

(iii) The SU-PU interference ratio is defined as the ratioof the number of SUsrsquo packets interrupted by PUsrsquoactivities to the total number of packets delivered byan SU source node A routing scheme with lower SU-PU interference ratio is favorable for CRAHNs

(iv) The packet delivery ratio is defined as the ratio ofall successfully received data packets that are fullyreceived by the SU destination node to the total datapackets generated by the SU source node A routingscheme with higher packet delivery ratio is desirablefor CRAHNs

To ensure the statistical significance of the results we runeach experiment for 500 seconds and repeat it 1000 timeswithdifferent seeds to report the average value as final results

62 Simulation Results

621 Impact of Number of Flows In this part we evaluate theimpact of offered traffic load on the performance by changingthe number of flows We vary the number of flows from 15CBR flows to 35 flows

As shown in Figure 3(a) we observe that LBOR provideshigher throughput than SAAR and SAOR We first considerthe case that the number of flows is below the saturationlevel Recall that the throughput defined here is measuredby the aggregate throughput of all the flows In this casewhen the number of flows increases from 15 flows to 25flows the throughput of all three schemes starts increasing asmore SU receivers are involved andmore parallel concurrenttransmissions occur Then we consider the case that whenthe number of flows is larger than 25 this indicates that

the network traffic is becoming saturated In this case thethroughput gain of SAAR and SAOR starts to decline As thenetwork resources (available channels and links) are limitedthe increasing number of flows intensifies the congestion andinterference which explains the throughput degradation ofSAAR and SAOR However when more flows are involvedthe throughput of LBOR still increases with increasingnumber of flows This is because LBOR has a load balancingfeature and solves the congestion problem by routing thetraffic through alternate forwarders

Figure 3(b) shows that LBOR has superior end-to-enddelay performance as compared to SAAR and SAOR Whenthe number of flows is equal to 15 the performance gapbetween LBOR and SAAR (both with SAOR) is small In thiscase each relay node does not need to handle toomuch trafficfor multiple flows Thus the unfair distribution of networktraffic will not lead to extra end-to-end delay Neverthelessif the number of flows is above 25 the end-to-end delayof SAAR and SAOR increase more sharply than that ofLBOR This can be attributed to the ability of LBOR that itsends packets to the links with lower traffic load and delayMoreover it can be concluded from the result that LBORfacilitates a better distribution of traffic and the consequentreduction of queuing times contributes to a lower overalldelay

622 Impact of PUsrsquo Activities In this part we evaluate theperformance of LBOR under different PUsrsquo activity patternsThe expected channel OFF time 119864[119879119874119865119865119888ℎ ] varies from 20msto 120ms A small 119864[119879119874119865119865119888ℎ ] means that the PUsrsquo activitiesare high and thus the delivery of the secondary user ismore likely to be interrupted Different from the previousexperiment the number of CBR flows is fixed to be 30

Figure 4(a) shows the SU-PU interference ratio of thethree schemes under different values of 119864[119879119874119865119865119888ℎ ] In mostcases LBOR produces significantly lower SU-PU interferenceas compared to SAAR and SAOR This is because the trafficload of LBOR is distributed efficiently and fairly In LBORfewer flows will have a chance to share the same SU nodewhich is highly influenced by PUsrsquo behaviors Moreoverwe notice that the improvement of load balancing schemeis significant under high PUsrsquo activities (119864[119879119874119865119865119888ℎ ] is below60ms) but not under low PUsrsquo activities (119864[119879119874119865119865119888ℎ ]) is above60ms) The reason behind this phenomenon is that whenthe nodes are highly influenced by PUsrsquo behaviors andoverloaded LBOR will give priorities to other nodes with alower traffic load to release the burden of these nodes

Figure 4(b) displays the throughput of LBOR SAARand SAOR with different values of 119864[119879119874119865119865119888ℎ ] It shows thatthe throughput of these schemes increase as the value of119864[119879119874119865119865119888ℎ ] becomes larger The result also shows that LBORachieves higher throughput compared to SAAR and SAORWhen the PUsrsquo activities decrease (the value of 119864[119879119874119865119865119888ℎ ]increases) the performance difference between LBOR andSAAR (both with SAOR) becomes negligible Moreover thetraffic load of LBOR is evenly distributed across the SUsand different available channels Therefore it is expected that

Wireless Communications and Mobile Computing 11

15 20 25 30 35Number of flows

170

180

190

200

210

220

230

240

250

260

270

ro

ughp

ut (K

bps)

LBORSAARSAOR(a) Throughput versus varying number of flows

15 20 25 30 35Number of flows

25

30

35

40

45

50

55

60

65

70

End-

to-e

nd d

elay

(ms)

SAORSAARLBOR(b) End-to-end delay versus varying number of flows

Figure 3

more improvements can be obtained for LBOR when PUsappear frequently

The end-to-end delay and the packet delivery ratio per-formance with different types of PUsrsquo activities are shownin Figures 4(c) and 4(d) With the increase of 119864[119879119874119865119865119888ℎ ]the packet delivery ratio will increase and the end-to-enddelay will decrease When the value of 119864[119879119874119865119865119888ℎ ] is around20ms LBOR can reduce the end-to-end delay by up to36 and 40 compared to SAAR and SAOR HoweverLBOR can only increase the delivery ratio by up to 2and 3 respectively compared to these two schemes Thereare two main reasons for this case First the opportunisticrouting protocol of LBOR is based on multipath routingstrategy where the paths and relay nodes are determinedon-the-fly In addition SAAR and SAOR also utilize thesimilar opportunistic routing principle and they aim toincrease the packet delivery ratio for the dynamic changingspectrum environment Second SAAR and SAOR alwayschoose paths or forwarding nodes with the best performanceof packet delivery ratio Nevertheless LBOR jointly considersthe load balancing for selecting the relay nodes Thus theimprovement of packet delivery ratio from LBOR is limited

623 Impact ofNumber of Channels In this part we comparethe performance of the three schemes under different numberof channels The number of channels varies between 2 and 10and the expected channel OFF time is set to be 60ms

Figure 5(a) compares the SU-PU interference ratio of thethree schemes by varying the number of channels of thenetworkWe can observe that the SU-PU interference ratio ofLBOR is lower than that of SAAR and SAOR One significantobservationhere is thatwhen the number of channels is largerthan 6 the SU-PU interference ratio of SAAR and SAOR

tends to decrease as the number of channels is increased InCRAHNs a larger number of channels can service a largernumber of SUsThus in this case SUs can choosemore stablechannels to avoid PUsrsquo interruptions In addition if an SUis interrupted by PUsrsquo appearance it can easily find anotheravailable channel for data transmissions Another importantobservation made is that when the number of channelsincreases from 2 to 6 the SU-PU interference ratio of LBORwill increase but the other two schemes will decrease Anexplanation for this phenomenon is as follows Consideringthe load balancing strategy LBOR tries its best to distributetraffic evenly among multiple nodes When there are notenough available channels to avoid the mutual interferencesamong SUs the packets cannot be correctly decoded at thereceiver Due to the same reason we can also observe fromFigures 5(b) 5(c) and 5(d) that when the number of channelsis below 6 the improvements from load balancing are notsignificantly

Figure 5(b) shows the throughput of LBOR SAAR andSAOR with different number of channels in the network Weobserve that the throughput of these schemes increases as thenumber of channels becomes larger The result also confirmsthat nomatter howmany channels are available LBOR alwaysperforms better than SAAR and SAOR When the number ofchannels is above 6 more improvement can be obtained fromLBOR

As shown in Figures 5(c) and 5(d) they provide similarresults and demonstrate that LBOR can achieve lower end-to-end delay and higher packet delivery ratio in most casescompared to SAAR and SAOR The improvements comefrom the load balancing based metric and the load balanc-ing based relay selection algorithm On the one hand theproposed scheme adopts a load balancing based metric to

12 Wireless Communications and Mobile Computing

20 40 60 80 100 120E[Tch

OFF] (ms)

0

002

004

006

008

01

012

014

016

018

02

SU-P

U in

terf

eren

ce ra

tio

LBORSAARSAOR(a) SU-PU interference ratio versus PUsrsquo activities

E[TchOFF] (ms)

20 40 60 80 100 120100

150

200

250

300

350

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus PUsrsquo activities

20 40 60 80 100 12025

30

35

40

45

50

55

60

65

End-

to-e

nd d

elay

(ms)

E[TchOFF] (ms)

LBORSAARSAOR

(c) End-to-end delay versus PUsrsquo activities

20 40 60 80 100 120076

078

08

082

084

086

088

09

092

094

Pack

et d

eliv

ery

ratio

LBORSAARSAOR

E[TchOFF] (ms)

(d) Packet delivery ratio versus PUsrsquo activities

Figure 4

choose the relay nodes with higher throughput and lowerdelay On the other hand the load balancing based relayselection algorithm can predict the occupied capacity of eachlink and thus ensure that the traffic load is evenly balancedamong all of the relay nodes

624 Impact of Number of PUs In this experiment weevaluate the performance of three schemes under differentnumber of PUs The number of PUs varies between 10 and40 and the number of channels is set to be 6 Figure 6(a)

compares the SU-PU interference ratio of different schemesunder different number of PUs In most cases the SU-PU interference ratios of SAAR and SAOR rise sharplythan LBOR That is because the network traffic of nonloadbalancing methods (SAAR and SAOR) tends to concentrateon a specific number of SUs and links If these SUsrsquo channelsare occupied by PUs most packets of the network may beaffected On the contrary LBOR distributes traffic to a largernumber of SUs thus packets have less change to be affectedby the PUsrsquo appearance and it induces a load balancing

Wireless Communications and Mobile Computing 13

2 4 6 8 10Total number of channels

002

004

006

008

01

012

014

016

018SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of channels

2 4 6 8 10Total number of channels

160

180

200

220

240

260

280

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of channels

2 4 6 8 10Total number of channels

20

30

40

50

60

70

80

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of channels

2 4 6 8 10Total number of channels

07

075

08

085

09

095Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of channels

Figure 5

improvement compared with SAAR and SAOR When thenumber of PUs is below 30 the benefit from load balancingis not surprising since SUs are not heavily affected Whenthe number of PUs is 35 an interesting observation here isthat the SU-PU interference ratio of LBOR is lower than thecase of 30 The reason for this case is that the aggregatedbenefit from load balancing scheme overwhelms the negativeeffect of the increased number of PUs More specifically inLBOR SUs are not overloaded and the packets are properlydistributed Thus when data transmission failures are causedby the appearance of PUs it is much easier for LBOR toreroute the blocked trafficflows to the light-loaded SUswhose

channels are available compared with other two methodsWhen the number of PUs is 40 we observe that the SU-PUinterference ratio of LBOR is higher than the case of 35 In thissituation there are not enough available channels for SUs totransmit the traffic of thewhole networkThus it is difficult tofind light-loaded SUs to reroute packets for overloaded SUsand more packets would be interrupted by PUsrsquo activities

Figure 6(b) shows the throughput by varying the numberof PUs in the network for the three schemes In the figurewe observe that the proposed LBOR scheme achieves higherthroughput than SAAR and SAOR We can also find thatwhen the number of PUs increases the throughput decreases

14 Wireless Communications and Mobile Computing

10 15 20 25 30 35 40Total number of PUs

003

004

005

006

007

008

009

01

011

012

013SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

160

180

200

220

240

260

280

300

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

25

30

35

40

45

50

55

60

65

70

75

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

078

08

082

084

086

088

09

092Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of PUs

Figure 6

because of available resources are decaying Moreover whenthe number of PUs is small the improvement from LBOR islimited On the contrary when the number of PUs rises above25 the throughput of SAAR and SAOR drops drastically butthe throughput degradation of LBOR is limited This is dueto the fact that LBOR can predict the availability and theoccupied capacity of each link and thus relieve the networkrsquoscongestions

Figures 6(c) and 6(d) can also confirm that LBOR canprovide the lowest end-to-end delay and the highest packetdelivery ratio among the three schemes When the numberof PUs increases the unbalanced distribution of load amongnodes becomes a critical issue and has a negative impacton the end-to-end delay and the packet delivery ratio Thiscan be understood intuitively as follows On the one handthe previous link may become unavailable due to the PU

appearance and the packets which arrived earlier have towait for late packets in reordering buffers at the receivingdestination thus resulting in an increasing queuing delayOn the other hand if late packets do not arrive within theimposed deadline it will be treated as a lost one thus resultingin an increasing packet loss

7 Conclusion

In this paper we have proposed a load balancing oppor-tunistic routing scheme for cognitive radio networks Con-sidering the loading constraint of each node we analyzedthe throughput bound of one opportunistic module andformulated the problem of maximizing the total throughputof the network as a linear programming problem Moreoverwe designed a new metric to select and prioritize the

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

8 Wireless Communications and Mobile Computing

the maximum throughput of link set 1198971 119897119911 in a steadystate Then we have

G = q ∙ 1199011 ∙ 1198741198971119886 + q ∙ 1199012 ∙ 1198621198972119886 (1 minus 1199011) + + q ∙ 119901119911

∙ 119862119897119911119886 119911minus1sum119896=0

(1 minus 119901119896)= 119902 ∙ Θ ∙ (1 + (1 minus 1199011) + + 119911minus1sum

119896=0

(1 minus 119901119896))(25)

In (25) 119902 and Θ are constant values Then the value of 119866 isdecided by 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) For conveniencewe use the notation Υ to denote the formula 1 + (1 minus 119901119886) + + sum119911minus1

119896=0(1 minus 119901119896) Then the maximum value of 119866 can beachieved when Υ has the maximum value Obviously Υ isthe sum of the polynomials As a result the necessary andsufficient condition forΥ to attain the maximum value is thateach term ofΥmust reach the maximum valueThat is to say(1minus1199011) sum119911minus1

119896=0(1minus119901119896)must take the maximum value at thesame time Intuitively these items are inversely proportionalto the delivery rate of the candidate link Therefore we canconclude that the maximum Υ can be achieved when 1199011 lt1199012 lt lt 119901119911 Accordingly the condition 1198621198971119886 gt 1198621198972119886 gt gt119862119897119911119886 is valid Then Proposition 2 holds

512 Load BalancingCandidate Forwarder Sorting AlgorithmIn this subsection we will explain how to sort the candidatenodes in LS In the opportunistic module (n F) F =1198991 119899119898 is the candidate node set FL = 1198971 119897119898 isthe candidate link set and PR = 1199011 119901119898 is a PRR setEach PRR corresponds to a candidate link Based on (24) wecan calculate the Load Score of each candidate node which isdenoted as H = Θ1 Θ119898 We develop a simple selectionsorting algorithm to sort H = Θ1 Θ119898 which is shownas in Algorithm 1

In Algorithm 1 the inputs are the candidate node set119865 the PRR set 119875119877 and the LS set 119867 The algorithm sortscandidate nodes based on their LS In the algorithm line(4) is used to mark the highest priority nodes for eachiteration and lines (6) to (12) indicate the priority orderingprocess According to the process of the algorithm the timecomplexity of this algorithm is O(m2) where m representsthe number of candidate nodes

For the ordered candidate node set 1198651015840 we can calcu-late the maximum throughput of the opportunistic module(n 1198651015840) according to (11) which is shown as follows

CTT = q ∙ Θ1 + q ∙ Θ2 (1 minus 1199011) + + q

∙ Θ119898

119898minus1sum119896=0

(1 minus 119901119897119896) (26)

52 Load Balancing Based Relay Selection Algorithm In thissection we further study the load balancing solution InCRAHNs the packet will be forwarded to the destinationthrough several opportunistic modules until the packet

Input F = 1198991 119899119898 PR = 1199011 119901119898 H = Θ1 Θ119898Output An ordered candidate node set 1198651015840(1) Initialize 1198651015840 = (2) Initialize a b k(3) for eachΘ119886120598H and 1 le a le m do(4) blarr997888 a(5) for eachΘ119896120598H and a + 1 le k le m do(6) if Θ119887 lt Θ119896 then(7) blarr997888 k(8) else if Θ119887 = Θ119896 then(9) if 119875119886 gt 119875119896 then(10) blarr997888 k(11) end if(12) end if(13) end for(14) 1198651015840119886 larr997888 119865119887(15) end for(16) output 1198651015840(17) end

Algorithm 1 Load Balancing Based Candidate Forwarder SortingAlgorithm

arrives to its destination In each opportunistic module theselection of the last hop is limited by the forwarding capabilityof the next hop To achieve the global optimal it is necessaryto predict the forwarding capability of the opportunisticmodule which is the final hop Then we determine thecandidate node set of the first hop opportunistic modulenamely the optimal candidate node set for the first hopopportunistic module However this method of selectingthe optimal opportunistic module is not appropriate Inthe cognitive radio network the link status is dynamic dueto the dynamic activities of PUs Hence we cannot verifythat the prediction of the forwarding ability of the last hopopportunistic module is convergent Thus we cannot provethe optimal opportunistic module is accurate In [16] theauthor chose an intermediate destinationnode to alleviate theinfluence of the dynamic change of link state Inspired by thismethod we adopt an opportunistic module iteration methodto determine the optimal candidate node set of the currentopportunistic module Since this method is based on theforwarding ability of the next hop opportunistic module it isnecessary to predict the occupied capacity of each candidatelink in this opportunistic module

In this paper we use Minimum Mean Square Error(MMSE) [28] based flow prediction model to estimate theoccupied capacity of the link in the next time slot Theformula of the model is as follows

119862119897119906 = MMSE (119883119905 | t = 0 1 2 ) (27)

where 119883119905|t = 0 1 2 is the current flow and historicalflow statistics In [28] the author describes the predictionscheme in detail and readers can refer to it for more details

For the opportunistic module (n F) the candidate nodeset is F = 1198991 119899119898 the candidate link set is LS =1198971 119897119898 and the corresponding LS set isH = Θ1 Θ119898

Wireless Communications and Mobile Computing 9

Input F = n1 nm H = Θ1 ΘmOutput The optimal node set Foptimal(1) Initialize arrays of H1015840 F1015840 F10158401015840 Foptimal and LS1015840(2) Initialize values of o1 o2 and ctt(3) for each ne120598F and 1 le e le m do(4) Search the opportunistic module(ne Fe) of ne(5) F1015840larr997888the candidate node set of (ne Fe)(6) LS1015840larr997888the candidate link set of (ne Fe)(7) for each n1015840k120598F and 1 le k le |Fe| do(8) o1 larr997888 the used capacity of l1015840k(9) o2 larr997888 the available capacity of l1015840k(10) H1015840

k larr997888 the LS of n1015840k(11) end for(12) F10158401015840 larr997888 sorting F1015840 according to Algorithm 1(13) cttlarr997888 the throughput of (ne F10158401015840)(14) if Θe gt ctt do(15) Θe larr997888 ctt(16) end if(17) end for(18) Foptimal larr997888 sorting F according to Algorithm 1(19) output Foptimal(20) end

Algorithm 2 Load Balancing Relay Selection Algorithm

The selection steps of the optimal opportunistic module aredescribed as follows

(1) For any candidate node 119899119890 isin 119865 we first determineits opportunistic module (119899119890 119865119890) according to relay distancesThen we perform the following steps (1) predicting theoccupied capacity of each link in candidate link set 119865119871119890 basedon (27) (2) calculating the available capacity of each link in119865119871119890 based on (14) (3) computing the LS of each node in thecandidate node set 119865119890 based on (24) (4) sorting the candidatenode set 119865119890 and obtaining an ordered candidate node set 1198651015840119890according to Algorithm 1 and (5) calculating the maximumthroughput CTT((119899119890 1198651015840119890)) of the opportunistic module basedon (26)

(2)We compare the LS valueΘ119890 of candidate node 119899119890 withCTT((119899119890 1198651015840119890)) and assign a smaller value to Θ119890

(3) We sort the candidate node set 119865 based on Algo-rithm 1 and the ordered set of candidate nodes is the optimalcandidate node set

The details of the Load Balancing Relay Selection Algo-rithm are shown in Algorithm 2 In Algorithm 2 lines (1)and (2) define the relevant parameters and their initializedvalues Lines (3) to (18) show the selection process for theoptimal candidate forwarding node According to the processof Algorithm 2 and the time complexity of Algorithm 1 thetime complexity of Algorithm 2 is O(m|1198651015840|2) where m isthe number of candidate nodes of the opportunistic module(nF) and |1198651015840| is the maximum number of candidate nodesin all the next hop opportunistic modules We can selectthe optimal opportunistic module for each hop according toAlgorithm 2 It balances the traffic load of each candidate linkwhile ensuring the quality of communication Meanwhile itimproves the throughput of the local network

6 Performance Evaluation

In this section we evaluate the performance of our proposedscheme with two well-known routing protocols for CRAHNsand present the simulation results on the SU-PU interferenceratio the packet delivery ratio the throughput and the end-to-end delay under different number of flows PUsrsquo activitiesnumber of channels and number of PUs

61 Simulation Settings The network parameter settings aresummarized in Table 2 In the simulations multiple SUsand PUs are randomly deployed in a 1000m times 1000marea and they are assumed to be stationary throughout thesimulation As stated in Section 31 the PU activities of eachchannel are modeled as an exponential ON-OFF processThe status of each channel is associated with two parametersone is the expected channel OFF time 119864[119879119874119865119865119888ℎ ] and theother is the probability of the idle state 119875119888ℎ119894119889119897119890 The channelstatus is estimated by utilizing periodic spectrum sensingand on-demand sensing before data transmissions In thesimulations 30 constant bit rate (CBR) flows are generatedbetween randomly chosen source-destination pairs of SUsand they are associated with the packet size of 512 Bytesand flow rate of 5Kbps Notice that the most representativeopportunistic routing scheme ExOR [9] adopted the methodin [29] to obtain the packet reception ratio between twonodes Accordingly in this simulation the packet receptionratios of the links are also based on the distance-to-deliveryratio relationshipmeasured in [29]Thus we require the posi-tion information of different nodes to compute the distancebetween neighboring nodes and obtain the packet receptionratios of the links We evaluate our scheme through NS-2simulation [30] In NS-2 our LBOR scheme is implementedas the routing agent In this work we only focus on thenetwork layer without considering MAC and physical layerissues Since the IEEE 80211 MAC protocol is customized forCR support we use the modified 80211cr as the CSMAMACagent in our simulation script We also apply CRN patch [31]to our simulations that can support multiple channels at thephysical layer

In the simulations we compare our LBOR scheme withSAAR scheme [15] and SAORprotocol [27]TheSAORproto-col is a well-knownmultipath opportunistic routing protocolcoupled with spectrum sensing and sharing in multichannelCRAHNs The SAAR scheme is a recently published any-path opportunistic routing scheme with consideration of theunreliable transmission feature and the uncertain spectrumavailability characteristic of CRAHNs

The performance metrics for our experiments are definedas follows

(i) The throughput is defined as the total amount oftraffic (in bits per second) an SU receiver receivesfrom the sender divided by the time it takes for thereceiver to obtain the last packet A routing schemewith higher throughput is desirable for CRAHNs

(ii) The end-to-end delay is defined as the average delaywhich is calculated by summing up the end-to-enddelays of all packets received by all SU destination

10 Wireless Communications and Mobile Computing

Table 2 Simulation parameters

Parameter valueNumber of SUs 100Number of PUs 25Transmission ranges of SUsand PUs 250m

Number of channels 6Channelsrsquo idle stateprobabilities1198751198881119894119889119897119890 1198751198886119894119889119897119890 03 03 05 05 07 07Expected channel OFF time 60msTotal number of flows 30Source-destination distance 800mSU CCC rate 1MbpsSU data channel rate 2MbpsPacket size 512 BytesSensing time 3msChannel switching time 100120583sRetransmission time 3

nodes and dividing it by the total number of receivedpackets A routing scheme with the lower end-to-enddelay is preferred for CRAHNs

(iii) The SU-PU interference ratio is defined as the ratioof the number of SUsrsquo packets interrupted by PUsrsquoactivities to the total number of packets delivered byan SU source node A routing scheme with lower SU-PU interference ratio is favorable for CRAHNs

(iv) The packet delivery ratio is defined as the ratio ofall successfully received data packets that are fullyreceived by the SU destination node to the total datapackets generated by the SU source node A routingscheme with higher packet delivery ratio is desirablefor CRAHNs

To ensure the statistical significance of the results we runeach experiment for 500 seconds and repeat it 1000 timeswithdifferent seeds to report the average value as final results

62 Simulation Results

621 Impact of Number of Flows In this part we evaluate theimpact of offered traffic load on the performance by changingthe number of flows We vary the number of flows from 15CBR flows to 35 flows

As shown in Figure 3(a) we observe that LBOR provideshigher throughput than SAAR and SAOR We first considerthe case that the number of flows is below the saturationlevel Recall that the throughput defined here is measuredby the aggregate throughput of all the flows In this casewhen the number of flows increases from 15 flows to 25flows the throughput of all three schemes starts increasing asmore SU receivers are involved andmore parallel concurrenttransmissions occur Then we consider the case that whenthe number of flows is larger than 25 this indicates that

the network traffic is becoming saturated In this case thethroughput gain of SAAR and SAOR starts to decline As thenetwork resources (available channels and links) are limitedthe increasing number of flows intensifies the congestion andinterference which explains the throughput degradation ofSAAR and SAOR However when more flows are involvedthe throughput of LBOR still increases with increasingnumber of flows This is because LBOR has a load balancingfeature and solves the congestion problem by routing thetraffic through alternate forwarders

Figure 3(b) shows that LBOR has superior end-to-enddelay performance as compared to SAAR and SAOR Whenthe number of flows is equal to 15 the performance gapbetween LBOR and SAAR (both with SAOR) is small In thiscase each relay node does not need to handle toomuch trafficfor multiple flows Thus the unfair distribution of networktraffic will not lead to extra end-to-end delay Neverthelessif the number of flows is above 25 the end-to-end delayof SAAR and SAOR increase more sharply than that ofLBOR This can be attributed to the ability of LBOR that itsends packets to the links with lower traffic load and delayMoreover it can be concluded from the result that LBORfacilitates a better distribution of traffic and the consequentreduction of queuing times contributes to a lower overalldelay

622 Impact of PUsrsquo Activities In this part we evaluate theperformance of LBOR under different PUsrsquo activity patternsThe expected channel OFF time 119864[119879119874119865119865119888ℎ ] varies from 20msto 120ms A small 119864[119879119874119865119865119888ℎ ] means that the PUsrsquo activitiesare high and thus the delivery of the secondary user ismore likely to be interrupted Different from the previousexperiment the number of CBR flows is fixed to be 30

Figure 4(a) shows the SU-PU interference ratio of thethree schemes under different values of 119864[119879119874119865119865119888ℎ ] In mostcases LBOR produces significantly lower SU-PU interferenceas compared to SAAR and SAOR This is because the trafficload of LBOR is distributed efficiently and fairly In LBORfewer flows will have a chance to share the same SU nodewhich is highly influenced by PUsrsquo behaviors Moreoverwe notice that the improvement of load balancing schemeis significant under high PUsrsquo activities (119864[119879119874119865119865119888ℎ ] is below60ms) but not under low PUsrsquo activities (119864[119879119874119865119865119888ℎ ]) is above60ms) The reason behind this phenomenon is that whenthe nodes are highly influenced by PUsrsquo behaviors andoverloaded LBOR will give priorities to other nodes with alower traffic load to release the burden of these nodes

Figure 4(b) displays the throughput of LBOR SAARand SAOR with different values of 119864[119879119874119865119865119888ℎ ] It shows thatthe throughput of these schemes increase as the value of119864[119879119874119865119865119888ℎ ] becomes larger The result also shows that LBORachieves higher throughput compared to SAAR and SAORWhen the PUsrsquo activities decrease (the value of 119864[119879119874119865119865119888ℎ ]increases) the performance difference between LBOR andSAAR (both with SAOR) becomes negligible Moreover thetraffic load of LBOR is evenly distributed across the SUsand different available channels Therefore it is expected that

Wireless Communications and Mobile Computing 11

15 20 25 30 35Number of flows

170

180

190

200

210

220

230

240

250

260

270

ro

ughp

ut (K

bps)

LBORSAARSAOR(a) Throughput versus varying number of flows

15 20 25 30 35Number of flows

25

30

35

40

45

50

55

60

65

70

End-

to-e

nd d

elay

(ms)

SAORSAARLBOR(b) End-to-end delay versus varying number of flows

Figure 3

more improvements can be obtained for LBOR when PUsappear frequently

The end-to-end delay and the packet delivery ratio per-formance with different types of PUsrsquo activities are shownin Figures 4(c) and 4(d) With the increase of 119864[119879119874119865119865119888ℎ ]the packet delivery ratio will increase and the end-to-enddelay will decrease When the value of 119864[119879119874119865119865119888ℎ ] is around20ms LBOR can reduce the end-to-end delay by up to36 and 40 compared to SAAR and SAOR HoweverLBOR can only increase the delivery ratio by up to 2and 3 respectively compared to these two schemes Thereare two main reasons for this case First the opportunisticrouting protocol of LBOR is based on multipath routingstrategy where the paths and relay nodes are determinedon-the-fly In addition SAAR and SAOR also utilize thesimilar opportunistic routing principle and they aim toincrease the packet delivery ratio for the dynamic changingspectrum environment Second SAAR and SAOR alwayschoose paths or forwarding nodes with the best performanceof packet delivery ratio Nevertheless LBOR jointly considersthe load balancing for selecting the relay nodes Thus theimprovement of packet delivery ratio from LBOR is limited

623 Impact ofNumber of Channels In this part we comparethe performance of the three schemes under different numberof channels The number of channels varies between 2 and 10and the expected channel OFF time is set to be 60ms

Figure 5(a) compares the SU-PU interference ratio of thethree schemes by varying the number of channels of thenetworkWe can observe that the SU-PU interference ratio ofLBOR is lower than that of SAAR and SAOR One significantobservationhere is thatwhen the number of channels is largerthan 6 the SU-PU interference ratio of SAAR and SAOR

tends to decrease as the number of channels is increased InCRAHNs a larger number of channels can service a largernumber of SUsThus in this case SUs can choosemore stablechannels to avoid PUsrsquo interruptions In addition if an SUis interrupted by PUsrsquo appearance it can easily find anotheravailable channel for data transmissions Another importantobservation made is that when the number of channelsincreases from 2 to 6 the SU-PU interference ratio of LBORwill increase but the other two schemes will decrease Anexplanation for this phenomenon is as follows Consideringthe load balancing strategy LBOR tries its best to distributetraffic evenly among multiple nodes When there are notenough available channels to avoid the mutual interferencesamong SUs the packets cannot be correctly decoded at thereceiver Due to the same reason we can also observe fromFigures 5(b) 5(c) and 5(d) that when the number of channelsis below 6 the improvements from load balancing are notsignificantly

Figure 5(b) shows the throughput of LBOR SAAR andSAOR with different number of channels in the network Weobserve that the throughput of these schemes increases as thenumber of channels becomes larger The result also confirmsthat nomatter howmany channels are available LBOR alwaysperforms better than SAAR and SAOR When the number ofchannels is above 6 more improvement can be obtained fromLBOR

As shown in Figures 5(c) and 5(d) they provide similarresults and demonstrate that LBOR can achieve lower end-to-end delay and higher packet delivery ratio in most casescompared to SAAR and SAOR The improvements comefrom the load balancing based metric and the load balanc-ing based relay selection algorithm On the one hand theproposed scheme adopts a load balancing based metric to

12 Wireless Communications and Mobile Computing

20 40 60 80 100 120E[Tch

OFF] (ms)

0

002

004

006

008

01

012

014

016

018

02

SU-P

U in

terf

eren

ce ra

tio

LBORSAARSAOR(a) SU-PU interference ratio versus PUsrsquo activities

E[TchOFF] (ms)

20 40 60 80 100 120100

150

200

250

300

350

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus PUsrsquo activities

20 40 60 80 100 12025

30

35

40

45

50

55

60

65

End-

to-e

nd d

elay

(ms)

E[TchOFF] (ms)

LBORSAARSAOR

(c) End-to-end delay versus PUsrsquo activities

20 40 60 80 100 120076

078

08

082

084

086

088

09

092

094

Pack

et d

eliv

ery

ratio

LBORSAARSAOR

E[TchOFF] (ms)

(d) Packet delivery ratio versus PUsrsquo activities

Figure 4

choose the relay nodes with higher throughput and lowerdelay On the other hand the load balancing based relayselection algorithm can predict the occupied capacity of eachlink and thus ensure that the traffic load is evenly balancedamong all of the relay nodes

624 Impact of Number of PUs In this experiment weevaluate the performance of three schemes under differentnumber of PUs The number of PUs varies between 10 and40 and the number of channels is set to be 6 Figure 6(a)

compares the SU-PU interference ratio of different schemesunder different number of PUs In most cases the SU-PU interference ratios of SAAR and SAOR rise sharplythan LBOR That is because the network traffic of nonloadbalancing methods (SAAR and SAOR) tends to concentrateon a specific number of SUs and links If these SUsrsquo channelsare occupied by PUs most packets of the network may beaffected On the contrary LBOR distributes traffic to a largernumber of SUs thus packets have less change to be affectedby the PUsrsquo appearance and it induces a load balancing

Wireless Communications and Mobile Computing 13

2 4 6 8 10Total number of channels

002

004

006

008

01

012

014

016

018SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of channels

2 4 6 8 10Total number of channels

160

180

200

220

240

260

280

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of channels

2 4 6 8 10Total number of channels

20

30

40

50

60

70

80

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of channels

2 4 6 8 10Total number of channels

07

075

08

085

09

095Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of channels

Figure 5

improvement compared with SAAR and SAOR When thenumber of PUs is below 30 the benefit from load balancingis not surprising since SUs are not heavily affected Whenthe number of PUs is 35 an interesting observation here isthat the SU-PU interference ratio of LBOR is lower than thecase of 30 The reason for this case is that the aggregatedbenefit from load balancing scheme overwhelms the negativeeffect of the increased number of PUs More specifically inLBOR SUs are not overloaded and the packets are properlydistributed Thus when data transmission failures are causedby the appearance of PUs it is much easier for LBOR toreroute the blocked trafficflows to the light-loaded SUswhose

channels are available compared with other two methodsWhen the number of PUs is 40 we observe that the SU-PUinterference ratio of LBOR is higher than the case of 35 In thissituation there are not enough available channels for SUs totransmit the traffic of thewhole networkThus it is difficult tofind light-loaded SUs to reroute packets for overloaded SUsand more packets would be interrupted by PUsrsquo activities

Figure 6(b) shows the throughput by varying the numberof PUs in the network for the three schemes In the figurewe observe that the proposed LBOR scheme achieves higherthroughput than SAAR and SAOR We can also find thatwhen the number of PUs increases the throughput decreases

14 Wireless Communications and Mobile Computing

10 15 20 25 30 35 40Total number of PUs

003

004

005

006

007

008

009

01

011

012

013SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

160

180

200

220

240

260

280

300

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

25

30

35

40

45

50

55

60

65

70

75

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

078

08

082

084

086

088

09

092Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of PUs

Figure 6

because of available resources are decaying Moreover whenthe number of PUs is small the improvement from LBOR islimited On the contrary when the number of PUs rises above25 the throughput of SAAR and SAOR drops drastically butthe throughput degradation of LBOR is limited This is dueto the fact that LBOR can predict the availability and theoccupied capacity of each link and thus relieve the networkrsquoscongestions

Figures 6(c) and 6(d) can also confirm that LBOR canprovide the lowest end-to-end delay and the highest packetdelivery ratio among the three schemes When the numberof PUs increases the unbalanced distribution of load amongnodes becomes a critical issue and has a negative impacton the end-to-end delay and the packet delivery ratio Thiscan be understood intuitively as follows On the one handthe previous link may become unavailable due to the PU

appearance and the packets which arrived earlier have towait for late packets in reordering buffers at the receivingdestination thus resulting in an increasing queuing delayOn the other hand if late packets do not arrive within theimposed deadline it will be treated as a lost one thus resultingin an increasing packet loss

7 Conclusion

In this paper we have proposed a load balancing oppor-tunistic routing scheme for cognitive radio networks Con-sidering the loading constraint of each node we analyzedthe throughput bound of one opportunistic module andformulated the problem of maximizing the total throughputof the network as a linear programming problem Moreoverwe designed a new metric to select and prioritize the

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

Wireless Communications and Mobile Computing 9

Input F = n1 nm H = Θ1 ΘmOutput The optimal node set Foptimal(1) Initialize arrays of H1015840 F1015840 F10158401015840 Foptimal and LS1015840(2) Initialize values of o1 o2 and ctt(3) for each ne120598F and 1 le e le m do(4) Search the opportunistic module(ne Fe) of ne(5) F1015840larr997888the candidate node set of (ne Fe)(6) LS1015840larr997888the candidate link set of (ne Fe)(7) for each n1015840k120598F and 1 le k le |Fe| do(8) o1 larr997888 the used capacity of l1015840k(9) o2 larr997888 the available capacity of l1015840k(10) H1015840

k larr997888 the LS of n1015840k(11) end for(12) F10158401015840 larr997888 sorting F1015840 according to Algorithm 1(13) cttlarr997888 the throughput of (ne F10158401015840)(14) if Θe gt ctt do(15) Θe larr997888 ctt(16) end if(17) end for(18) Foptimal larr997888 sorting F according to Algorithm 1(19) output Foptimal(20) end

Algorithm 2 Load Balancing Relay Selection Algorithm

The selection steps of the optimal opportunistic module aredescribed as follows

(1) For any candidate node 119899119890 isin 119865 we first determineits opportunistic module (119899119890 119865119890) according to relay distancesThen we perform the following steps (1) predicting theoccupied capacity of each link in candidate link set 119865119871119890 basedon (27) (2) calculating the available capacity of each link in119865119871119890 based on (14) (3) computing the LS of each node in thecandidate node set 119865119890 based on (24) (4) sorting the candidatenode set 119865119890 and obtaining an ordered candidate node set 1198651015840119890according to Algorithm 1 and (5) calculating the maximumthroughput CTT((119899119890 1198651015840119890)) of the opportunistic module basedon (26)

(2)We compare the LS valueΘ119890 of candidate node 119899119890 withCTT((119899119890 1198651015840119890)) and assign a smaller value to Θ119890

(3) We sort the candidate node set 119865 based on Algo-rithm 1 and the ordered set of candidate nodes is the optimalcandidate node set

The details of the Load Balancing Relay Selection Algo-rithm are shown in Algorithm 2 In Algorithm 2 lines (1)and (2) define the relevant parameters and their initializedvalues Lines (3) to (18) show the selection process for theoptimal candidate forwarding node According to the processof Algorithm 2 and the time complexity of Algorithm 1 thetime complexity of Algorithm 2 is O(m|1198651015840|2) where m isthe number of candidate nodes of the opportunistic module(nF) and |1198651015840| is the maximum number of candidate nodesin all the next hop opportunistic modules We can selectthe optimal opportunistic module for each hop according toAlgorithm 2 It balances the traffic load of each candidate linkwhile ensuring the quality of communication Meanwhile itimproves the throughput of the local network

6 Performance Evaluation

In this section we evaluate the performance of our proposedscheme with two well-known routing protocols for CRAHNsand present the simulation results on the SU-PU interferenceratio the packet delivery ratio the throughput and the end-to-end delay under different number of flows PUsrsquo activitiesnumber of channels and number of PUs

61 Simulation Settings The network parameter settings aresummarized in Table 2 In the simulations multiple SUsand PUs are randomly deployed in a 1000m times 1000marea and they are assumed to be stationary throughout thesimulation As stated in Section 31 the PU activities of eachchannel are modeled as an exponential ON-OFF processThe status of each channel is associated with two parametersone is the expected channel OFF time 119864[119879119874119865119865119888ℎ ] and theother is the probability of the idle state 119875119888ℎ119894119889119897119890 The channelstatus is estimated by utilizing periodic spectrum sensingand on-demand sensing before data transmissions In thesimulations 30 constant bit rate (CBR) flows are generatedbetween randomly chosen source-destination pairs of SUsand they are associated with the packet size of 512 Bytesand flow rate of 5Kbps Notice that the most representativeopportunistic routing scheme ExOR [9] adopted the methodin [29] to obtain the packet reception ratio between twonodes Accordingly in this simulation the packet receptionratios of the links are also based on the distance-to-deliveryratio relationshipmeasured in [29]Thus we require the posi-tion information of different nodes to compute the distancebetween neighboring nodes and obtain the packet receptionratios of the links We evaluate our scheme through NS-2simulation [30] In NS-2 our LBOR scheme is implementedas the routing agent In this work we only focus on thenetwork layer without considering MAC and physical layerissues Since the IEEE 80211 MAC protocol is customized forCR support we use the modified 80211cr as the CSMAMACagent in our simulation script We also apply CRN patch [31]to our simulations that can support multiple channels at thephysical layer

In the simulations we compare our LBOR scheme withSAAR scheme [15] and SAORprotocol [27]TheSAORproto-col is a well-knownmultipath opportunistic routing protocolcoupled with spectrum sensing and sharing in multichannelCRAHNs The SAAR scheme is a recently published any-path opportunistic routing scheme with consideration of theunreliable transmission feature and the uncertain spectrumavailability characteristic of CRAHNs

The performance metrics for our experiments are definedas follows

(i) The throughput is defined as the total amount oftraffic (in bits per second) an SU receiver receivesfrom the sender divided by the time it takes for thereceiver to obtain the last packet A routing schemewith higher throughput is desirable for CRAHNs

(ii) The end-to-end delay is defined as the average delaywhich is calculated by summing up the end-to-enddelays of all packets received by all SU destination

10 Wireless Communications and Mobile Computing

Table 2 Simulation parameters

Parameter valueNumber of SUs 100Number of PUs 25Transmission ranges of SUsand PUs 250m

Number of channels 6Channelsrsquo idle stateprobabilities1198751198881119894119889119897119890 1198751198886119894119889119897119890 03 03 05 05 07 07Expected channel OFF time 60msTotal number of flows 30Source-destination distance 800mSU CCC rate 1MbpsSU data channel rate 2MbpsPacket size 512 BytesSensing time 3msChannel switching time 100120583sRetransmission time 3

nodes and dividing it by the total number of receivedpackets A routing scheme with the lower end-to-enddelay is preferred for CRAHNs

(iii) The SU-PU interference ratio is defined as the ratioof the number of SUsrsquo packets interrupted by PUsrsquoactivities to the total number of packets delivered byan SU source node A routing scheme with lower SU-PU interference ratio is favorable for CRAHNs

(iv) The packet delivery ratio is defined as the ratio ofall successfully received data packets that are fullyreceived by the SU destination node to the total datapackets generated by the SU source node A routingscheme with higher packet delivery ratio is desirablefor CRAHNs

To ensure the statistical significance of the results we runeach experiment for 500 seconds and repeat it 1000 timeswithdifferent seeds to report the average value as final results

62 Simulation Results

621 Impact of Number of Flows In this part we evaluate theimpact of offered traffic load on the performance by changingthe number of flows We vary the number of flows from 15CBR flows to 35 flows

As shown in Figure 3(a) we observe that LBOR provideshigher throughput than SAAR and SAOR We first considerthe case that the number of flows is below the saturationlevel Recall that the throughput defined here is measuredby the aggregate throughput of all the flows In this casewhen the number of flows increases from 15 flows to 25flows the throughput of all three schemes starts increasing asmore SU receivers are involved andmore parallel concurrenttransmissions occur Then we consider the case that whenthe number of flows is larger than 25 this indicates that

the network traffic is becoming saturated In this case thethroughput gain of SAAR and SAOR starts to decline As thenetwork resources (available channels and links) are limitedthe increasing number of flows intensifies the congestion andinterference which explains the throughput degradation ofSAAR and SAOR However when more flows are involvedthe throughput of LBOR still increases with increasingnumber of flows This is because LBOR has a load balancingfeature and solves the congestion problem by routing thetraffic through alternate forwarders

Figure 3(b) shows that LBOR has superior end-to-enddelay performance as compared to SAAR and SAOR Whenthe number of flows is equal to 15 the performance gapbetween LBOR and SAAR (both with SAOR) is small In thiscase each relay node does not need to handle toomuch trafficfor multiple flows Thus the unfair distribution of networktraffic will not lead to extra end-to-end delay Neverthelessif the number of flows is above 25 the end-to-end delayof SAAR and SAOR increase more sharply than that ofLBOR This can be attributed to the ability of LBOR that itsends packets to the links with lower traffic load and delayMoreover it can be concluded from the result that LBORfacilitates a better distribution of traffic and the consequentreduction of queuing times contributes to a lower overalldelay

622 Impact of PUsrsquo Activities In this part we evaluate theperformance of LBOR under different PUsrsquo activity patternsThe expected channel OFF time 119864[119879119874119865119865119888ℎ ] varies from 20msto 120ms A small 119864[119879119874119865119865119888ℎ ] means that the PUsrsquo activitiesare high and thus the delivery of the secondary user ismore likely to be interrupted Different from the previousexperiment the number of CBR flows is fixed to be 30

Figure 4(a) shows the SU-PU interference ratio of thethree schemes under different values of 119864[119879119874119865119865119888ℎ ] In mostcases LBOR produces significantly lower SU-PU interferenceas compared to SAAR and SAOR This is because the trafficload of LBOR is distributed efficiently and fairly In LBORfewer flows will have a chance to share the same SU nodewhich is highly influenced by PUsrsquo behaviors Moreoverwe notice that the improvement of load balancing schemeis significant under high PUsrsquo activities (119864[119879119874119865119865119888ℎ ] is below60ms) but not under low PUsrsquo activities (119864[119879119874119865119865119888ℎ ]) is above60ms) The reason behind this phenomenon is that whenthe nodes are highly influenced by PUsrsquo behaviors andoverloaded LBOR will give priorities to other nodes with alower traffic load to release the burden of these nodes

Figure 4(b) displays the throughput of LBOR SAARand SAOR with different values of 119864[119879119874119865119865119888ℎ ] It shows thatthe throughput of these schemes increase as the value of119864[119879119874119865119865119888ℎ ] becomes larger The result also shows that LBORachieves higher throughput compared to SAAR and SAORWhen the PUsrsquo activities decrease (the value of 119864[119879119874119865119865119888ℎ ]increases) the performance difference between LBOR andSAAR (both with SAOR) becomes negligible Moreover thetraffic load of LBOR is evenly distributed across the SUsand different available channels Therefore it is expected that

Wireless Communications and Mobile Computing 11

15 20 25 30 35Number of flows

170

180

190

200

210

220

230

240

250

260

270

ro

ughp

ut (K

bps)

LBORSAARSAOR(a) Throughput versus varying number of flows

15 20 25 30 35Number of flows

25

30

35

40

45

50

55

60

65

70

End-

to-e

nd d

elay

(ms)

SAORSAARLBOR(b) End-to-end delay versus varying number of flows

Figure 3

more improvements can be obtained for LBOR when PUsappear frequently

The end-to-end delay and the packet delivery ratio per-formance with different types of PUsrsquo activities are shownin Figures 4(c) and 4(d) With the increase of 119864[119879119874119865119865119888ℎ ]the packet delivery ratio will increase and the end-to-enddelay will decrease When the value of 119864[119879119874119865119865119888ℎ ] is around20ms LBOR can reduce the end-to-end delay by up to36 and 40 compared to SAAR and SAOR HoweverLBOR can only increase the delivery ratio by up to 2and 3 respectively compared to these two schemes Thereare two main reasons for this case First the opportunisticrouting protocol of LBOR is based on multipath routingstrategy where the paths and relay nodes are determinedon-the-fly In addition SAAR and SAOR also utilize thesimilar opportunistic routing principle and they aim toincrease the packet delivery ratio for the dynamic changingspectrum environment Second SAAR and SAOR alwayschoose paths or forwarding nodes with the best performanceof packet delivery ratio Nevertheless LBOR jointly considersthe load balancing for selecting the relay nodes Thus theimprovement of packet delivery ratio from LBOR is limited

623 Impact ofNumber of Channels In this part we comparethe performance of the three schemes under different numberof channels The number of channels varies between 2 and 10and the expected channel OFF time is set to be 60ms

Figure 5(a) compares the SU-PU interference ratio of thethree schemes by varying the number of channels of thenetworkWe can observe that the SU-PU interference ratio ofLBOR is lower than that of SAAR and SAOR One significantobservationhere is thatwhen the number of channels is largerthan 6 the SU-PU interference ratio of SAAR and SAOR

tends to decrease as the number of channels is increased InCRAHNs a larger number of channels can service a largernumber of SUsThus in this case SUs can choosemore stablechannels to avoid PUsrsquo interruptions In addition if an SUis interrupted by PUsrsquo appearance it can easily find anotheravailable channel for data transmissions Another importantobservation made is that when the number of channelsincreases from 2 to 6 the SU-PU interference ratio of LBORwill increase but the other two schemes will decrease Anexplanation for this phenomenon is as follows Consideringthe load balancing strategy LBOR tries its best to distributetraffic evenly among multiple nodes When there are notenough available channels to avoid the mutual interferencesamong SUs the packets cannot be correctly decoded at thereceiver Due to the same reason we can also observe fromFigures 5(b) 5(c) and 5(d) that when the number of channelsis below 6 the improvements from load balancing are notsignificantly

Figure 5(b) shows the throughput of LBOR SAAR andSAOR with different number of channels in the network Weobserve that the throughput of these schemes increases as thenumber of channels becomes larger The result also confirmsthat nomatter howmany channels are available LBOR alwaysperforms better than SAAR and SAOR When the number ofchannels is above 6 more improvement can be obtained fromLBOR

As shown in Figures 5(c) and 5(d) they provide similarresults and demonstrate that LBOR can achieve lower end-to-end delay and higher packet delivery ratio in most casescompared to SAAR and SAOR The improvements comefrom the load balancing based metric and the load balanc-ing based relay selection algorithm On the one hand theproposed scheme adopts a load balancing based metric to

12 Wireless Communications and Mobile Computing

20 40 60 80 100 120E[Tch

OFF] (ms)

0

002

004

006

008

01

012

014

016

018

02

SU-P

U in

terf

eren

ce ra

tio

LBORSAARSAOR(a) SU-PU interference ratio versus PUsrsquo activities

E[TchOFF] (ms)

20 40 60 80 100 120100

150

200

250

300

350

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus PUsrsquo activities

20 40 60 80 100 12025

30

35

40

45

50

55

60

65

End-

to-e

nd d

elay

(ms)

E[TchOFF] (ms)

LBORSAARSAOR

(c) End-to-end delay versus PUsrsquo activities

20 40 60 80 100 120076

078

08

082

084

086

088

09

092

094

Pack

et d

eliv

ery

ratio

LBORSAARSAOR

E[TchOFF] (ms)

(d) Packet delivery ratio versus PUsrsquo activities

Figure 4

choose the relay nodes with higher throughput and lowerdelay On the other hand the load balancing based relayselection algorithm can predict the occupied capacity of eachlink and thus ensure that the traffic load is evenly balancedamong all of the relay nodes

624 Impact of Number of PUs In this experiment weevaluate the performance of three schemes under differentnumber of PUs The number of PUs varies between 10 and40 and the number of channels is set to be 6 Figure 6(a)

compares the SU-PU interference ratio of different schemesunder different number of PUs In most cases the SU-PU interference ratios of SAAR and SAOR rise sharplythan LBOR That is because the network traffic of nonloadbalancing methods (SAAR and SAOR) tends to concentrateon a specific number of SUs and links If these SUsrsquo channelsare occupied by PUs most packets of the network may beaffected On the contrary LBOR distributes traffic to a largernumber of SUs thus packets have less change to be affectedby the PUsrsquo appearance and it induces a load balancing

Wireless Communications and Mobile Computing 13

2 4 6 8 10Total number of channels

002

004

006

008

01

012

014

016

018SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of channels

2 4 6 8 10Total number of channels

160

180

200

220

240

260

280

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of channels

2 4 6 8 10Total number of channels

20

30

40

50

60

70

80

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of channels

2 4 6 8 10Total number of channels

07

075

08

085

09

095Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of channels

Figure 5

improvement compared with SAAR and SAOR When thenumber of PUs is below 30 the benefit from load balancingis not surprising since SUs are not heavily affected Whenthe number of PUs is 35 an interesting observation here isthat the SU-PU interference ratio of LBOR is lower than thecase of 30 The reason for this case is that the aggregatedbenefit from load balancing scheme overwhelms the negativeeffect of the increased number of PUs More specifically inLBOR SUs are not overloaded and the packets are properlydistributed Thus when data transmission failures are causedby the appearance of PUs it is much easier for LBOR toreroute the blocked trafficflows to the light-loaded SUswhose

channels are available compared with other two methodsWhen the number of PUs is 40 we observe that the SU-PUinterference ratio of LBOR is higher than the case of 35 In thissituation there are not enough available channels for SUs totransmit the traffic of thewhole networkThus it is difficult tofind light-loaded SUs to reroute packets for overloaded SUsand more packets would be interrupted by PUsrsquo activities

Figure 6(b) shows the throughput by varying the numberof PUs in the network for the three schemes In the figurewe observe that the proposed LBOR scheme achieves higherthroughput than SAAR and SAOR We can also find thatwhen the number of PUs increases the throughput decreases

14 Wireless Communications and Mobile Computing

10 15 20 25 30 35 40Total number of PUs

003

004

005

006

007

008

009

01

011

012

013SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

160

180

200

220

240

260

280

300

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

25

30

35

40

45

50

55

60

65

70

75

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

078

08

082

084

086

088

09

092Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of PUs

Figure 6

because of available resources are decaying Moreover whenthe number of PUs is small the improvement from LBOR islimited On the contrary when the number of PUs rises above25 the throughput of SAAR and SAOR drops drastically butthe throughput degradation of LBOR is limited This is dueto the fact that LBOR can predict the availability and theoccupied capacity of each link and thus relieve the networkrsquoscongestions

Figures 6(c) and 6(d) can also confirm that LBOR canprovide the lowest end-to-end delay and the highest packetdelivery ratio among the three schemes When the numberof PUs increases the unbalanced distribution of load amongnodes becomes a critical issue and has a negative impacton the end-to-end delay and the packet delivery ratio Thiscan be understood intuitively as follows On the one handthe previous link may become unavailable due to the PU

appearance and the packets which arrived earlier have towait for late packets in reordering buffers at the receivingdestination thus resulting in an increasing queuing delayOn the other hand if late packets do not arrive within theimposed deadline it will be treated as a lost one thus resultingin an increasing packet loss

7 Conclusion

In this paper we have proposed a load balancing oppor-tunistic routing scheme for cognitive radio networks Con-sidering the loading constraint of each node we analyzedthe throughput bound of one opportunistic module andformulated the problem of maximizing the total throughputof the network as a linear programming problem Moreoverwe designed a new metric to select and prioritize the

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

10 Wireless Communications and Mobile Computing

Table 2 Simulation parameters

Parameter valueNumber of SUs 100Number of PUs 25Transmission ranges of SUsand PUs 250m

Number of channels 6Channelsrsquo idle stateprobabilities1198751198881119894119889119897119890 1198751198886119894119889119897119890 03 03 05 05 07 07Expected channel OFF time 60msTotal number of flows 30Source-destination distance 800mSU CCC rate 1MbpsSU data channel rate 2MbpsPacket size 512 BytesSensing time 3msChannel switching time 100120583sRetransmission time 3

nodes and dividing it by the total number of receivedpackets A routing scheme with the lower end-to-enddelay is preferred for CRAHNs

(iii) The SU-PU interference ratio is defined as the ratioof the number of SUsrsquo packets interrupted by PUsrsquoactivities to the total number of packets delivered byan SU source node A routing scheme with lower SU-PU interference ratio is favorable for CRAHNs

(iv) The packet delivery ratio is defined as the ratio ofall successfully received data packets that are fullyreceived by the SU destination node to the total datapackets generated by the SU source node A routingscheme with higher packet delivery ratio is desirablefor CRAHNs

To ensure the statistical significance of the results we runeach experiment for 500 seconds and repeat it 1000 timeswithdifferent seeds to report the average value as final results

62 Simulation Results

621 Impact of Number of Flows In this part we evaluate theimpact of offered traffic load on the performance by changingthe number of flows We vary the number of flows from 15CBR flows to 35 flows

As shown in Figure 3(a) we observe that LBOR provideshigher throughput than SAAR and SAOR We first considerthe case that the number of flows is below the saturationlevel Recall that the throughput defined here is measuredby the aggregate throughput of all the flows In this casewhen the number of flows increases from 15 flows to 25flows the throughput of all three schemes starts increasing asmore SU receivers are involved andmore parallel concurrenttransmissions occur Then we consider the case that whenthe number of flows is larger than 25 this indicates that

the network traffic is becoming saturated In this case thethroughput gain of SAAR and SAOR starts to decline As thenetwork resources (available channels and links) are limitedthe increasing number of flows intensifies the congestion andinterference which explains the throughput degradation ofSAAR and SAOR However when more flows are involvedthe throughput of LBOR still increases with increasingnumber of flows This is because LBOR has a load balancingfeature and solves the congestion problem by routing thetraffic through alternate forwarders

Figure 3(b) shows that LBOR has superior end-to-enddelay performance as compared to SAAR and SAOR Whenthe number of flows is equal to 15 the performance gapbetween LBOR and SAAR (both with SAOR) is small In thiscase each relay node does not need to handle toomuch trafficfor multiple flows Thus the unfair distribution of networktraffic will not lead to extra end-to-end delay Neverthelessif the number of flows is above 25 the end-to-end delayof SAAR and SAOR increase more sharply than that ofLBOR This can be attributed to the ability of LBOR that itsends packets to the links with lower traffic load and delayMoreover it can be concluded from the result that LBORfacilitates a better distribution of traffic and the consequentreduction of queuing times contributes to a lower overalldelay

622 Impact of PUsrsquo Activities In this part we evaluate theperformance of LBOR under different PUsrsquo activity patternsThe expected channel OFF time 119864[119879119874119865119865119888ℎ ] varies from 20msto 120ms A small 119864[119879119874119865119865119888ℎ ] means that the PUsrsquo activitiesare high and thus the delivery of the secondary user ismore likely to be interrupted Different from the previousexperiment the number of CBR flows is fixed to be 30

Figure 4(a) shows the SU-PU interference ratio of thethree schemes under different values of 119864[119879119874119865119865119888ℎ ] In mostcases LBOR produces significantly lower SU-PU interferenceas compared to SAAR and SAOR This is because the trafficload of LBOR is distributed efficiently and fairly In LBORfewer flows will have a chance to share the same SU nodewhich is highly influenced by PUsrsquo behaviors Moreoverwe notice that the improvement of load balancing schemeis significant under high PUsrsquo activities (119864[119879119874119865119865119888ℎ ] is below60ms) but not under low PUsrsquo activities (119864[119879119874119865119865119888ℎ ]) is above60ms) The reason behind this phenomenon is that whenthe nodes are highly influenced by PUsrsquo behaviors andoverloaded LBOR will give priorities to other nodes with alower traffic load to release the burden of these nodes

Figure 4(b) displays the throughput of LBOR SAARand SAOR with different values of 119864[119879119874119865119865119888ℎ ] It shows thatthe throughput of these schemes increase as the value of119864[119879119874119865119865119888ℎ ] becomes larger The result also shows that LBORachieves higher throughput compared to SAAR and SAORWhen the PUsrsquo activities decrease (the value of 119864[119879119874119865119865119888ℎ ]increases) the performance difference between LBOR andSAAR (both with SAOR) becomes negligible Moreover thetraffic load of LBOR is evenly distributed across the SUsand different available channels Therefore it is expected that

Wireless Communications and Mobile Computing 11

15 20 25 30 35Number of flows

170

180

190

200

210

220

230

240

250

260

270

ro

ughp

ut (K

bps)

LBORSAARSAOR(a) Throughput versus varying number of flows

15 20 25 30 35Number of flows

25

30

35

40

45

50

55

60

65

70

End-

to-e

nd d

elay

(ms)

SAORSAARLBOR(b) End-to-end delay versus varying number of flows

Figure 3

more improvements can be obtained for LBOR when PUsappear frequently

The end-to-end delay and the packet delivery ratio per-formance with different types of PUsrsquo activities are shownin Figures 4(c) and 4(d) With the increase of 119864[119879119874119865119865119888ℎ ]the packet delivery ratio will increase and the end-to-enddelay will decrease When the value of 119864[119879119874119865119865119888ℎ ] is around20ms LBOR can reduce the end-to-end delay by up to36 and 40 compared to SAAR and SAOR HoweverLBOR can only increase the delivery ratio by up to 2and 3 respectively compared to these two schemes Thereare two main reasons for this case First the opportunisticrouting protocol of LBOR is based on multipath routingstrategy where the paths and relay nodes are determinedon-the-fly In addition SAAR and SAOR also utilize thesimilar opportunistic routing principle and they aim toincrease the packet delivery ratio for the dynamic changingspectrum environment Second SAAR and SAOR alwayschoose paths or forwarding nodes with the best performanceof packet delivery ratio Nevertheless LBOR jointly considersthe load balancing for selecting the relay nodes Thus theimprovement of packet delivery ratio from LBOR is limited

623 Impact ofNumber of Channels In this part we comparethe performance of the three schemes under different numberof channels The number of channels varies between 2 and 10and the expected channel OFF time is set to be 60ms

Figure 5(a) compares the SU-PU interference ratio of thethree schemes by varying the number of channels of thenetworkWe can observe that the SU-PU interference ratio ofLBOR is lower than that of SAAR and SAOR One significantobservationhere is thatwhen the number of channels is largerthan 6 the SU-PU interference ratio of SAAR and SAOR

tends to decrease as the number of channels is increased InCRAHNs a larger number of channels can service a largernumber of SUsThus in this case SUs can choosemore stablechannels to avoid PUsrsquo interruptions In addition if an SUis interrupted by PUsrsquo appearance it can easily find anotheravailable channel for data transmissions Another importantobservation made is that when the number of channelsincreases from 2 to 6 the SU-PU interference ratio of LBORwill increase but the other two schemes will decrease Anexplanation for this phenomenon is as follows Consideringthe load balancing strategy LBOR tries its best to distributetraffic evenly among multiple nodes When there are notenough available channels to avoid the mutual interferencesamong SUs the packets cannot be correctly decoded at thereceiver Due to the same reason we can also observe fromFigures 5(b) 5(c) and 5(d) that when the number of channelsis below 6 the improvements from load balancing are notsignificantly

Figure 5(b) shows the throughput of LBOR SAAR andSAOR with different number of channels in the network Weobserve that the throughput of these schemes increases as thenumber of channels becomes larger The result also confirmsthat nomatter howmany channels are available LBOR alwaysperforms better than SAAR and SAOR When the number ofchannels is above 6 more improvement can be obtained fromLBOR

As shown in Figures 5(c) and 5(d) they provide similarresults and demonstrate that LBOR can achieve lower end-to-end delay and higher packet delivery ratio in most casescompared to SAAR and SAOR The improvements comefrom the load balancing based metric and the load balanc-ing based relay selection algorithm On the one hand theproposed scheme adopts a load balancing based metric to

12 Wireless Communications and Mobile Computing

20 40 60 80 100 120E[Tch

OFF] (ms)

0

002

004

006

008

01

012

014

016

018

02

SU-P

U in

terf

eren

ce ra

tio

LBORSAARSAOR(a) SU-PU interference ratio versus PUsrsquo activities

E[TchOFF] (ms)

20 40 60 80 100 120100

150

200

250

300

350

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus PUsrsquo activities

20 40 60 80 100 12025

30

35

40

45

50

55

60

65

End-

to-e

nd d

elay

(ms)

E[TchOFF] (ms)

LBORSAARSAOR

(c) End-to-end delay versus PUsrsquo activities

20 40 60 80 100 120076

078

08

082

084

086

088

09

092

094

Pack

et d

eliv

ery

ratio

LBORSAARSAOR

E[TchOFF] (ms)

(d) Packet delivery ratio versus PUsrsquo activities

Figure 4

choose the relay nodes with higher throughput and lowerdelay On the other hand the load balancing based relayselection algorithm can predict the occupied capacity of eachlink and thus ensure that the traffic load is evenly balancedamong all of the relay nodes

624 Impact of Number of PUs In this experiment weevaluate the performance of three schemes under differentnumber of PUs The number of PUs varies between 10 and40 and the number of channels is set to be 6 Figure 6(a)

compares the SU-PU interference ratio of different schemesunder different number of PUs In most cases the SU-PU interference ratios of SAAR and SAOR rise sharplythan LBOR That is because the network traffic of nonloadbalancing methods (SAAR and SAOR) tends to concentrateon a specific number of SUs and links If these SUsrsquo channelsare occupied by PUs most packets of the network may beaffected On the contrary LBOR distributes traffic to a largernumber of SUs thus packets have less change to be affectedby the PUsrsquo appearance and it induces a load balancing

Wireless Communications and Mobile Computing 13

2 4 6 8 10Total number of channels

002

004

006

008

01

012

014

016

018SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of channels

2 4 6 8 10Total number of channels

160

180

200

220

240

260

280

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of channels

2 4 6 8 10Total number of channels

20

30

40

50

60

70

80

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of channels

2 4 6 8 10Total number of channels

07

075

08

085

09

095Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of channels

Figure 5

improvement compared with SAAR and SAOR When thenumber of PUs is below 30 the benefit from load balancingis not surprising since SUs are not heavily affected Whenthe number of PUs is 35 an interesting observation here isthat the SU-PU interference ratio of LBOR is lower than thecase of 30 The reason for this case is that the aggregatedbenefit from load balancing scheme overwhelms the negativeeffect of the increased number of PUs More specifically inLBOR SUs are not overloaded and the packets are properlydistributed Thus when data transmission failures are causedby the appearance of PUs it is much easier for LBOR toreroute the blocked trafficflows to the light-loaded SUswhose

channels are available compared with other two methodsWhen the number of PUs is 40 we observe that the SU-PUinterference ratio of LBOR is higher than the case of 35 In thissituation there are not enough available channels for SUs totransmit the traffic of thewhole networkThus it is difficult tofind light-loaded SUs to reroute packets for overloaded SUsand more packets would be interrupted by PUsrsquo activities

Figure 6(b) shows the throughput by varying the numberof PUs in the network for the three schemes In the figurewe observe that the proposed LBOR scheme achieves higherthroughput than SAAR and SAOR We can also find thatwhen the number of PUs increases the throughput decreases

14 Wireless Communications and Mobile Computing

10 15 20 25 30 35 40Total number of PUs

003

004

005

006

007

008

009

01

011

012

013SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

160

180

200

220

240

260

280

300

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

25

30

35

40

45

50

55

60

65

70

75

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

078

08

082

084

086

088

09

092Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of PUs

Figure 6

because of available resources are decaying Moreover whenthe number of PUs is small the improvement from LBOR islimited On the contrary when the number of PUs rises above25 the throughput of SAAR and SAOR drops drastically butthe throughput degradation of LBOR is limited This is dueto the fact that LBOR can predict the availability and theoccupied capacity of each link and thus relieve the networkrsquoscongestions

Figures 6(c) and 6(d) can also confirm that LBOR canprovide the lowest end-to-end delay and the highest packetdelivery ratio among the three schemes When the numberof PUs increases the unbalanced distribution of load amongnodes becomes a critical issue and has a negative impacton the end-to-end delay and the packet delivery ratio Thiscan be understood intuitively as follows On the one handthe previous link may become unavailable due to the PU

appearance and the packets which arrived earlier have towait for late packets in reordering buffers at the receivingdestination thus resulting in an increasing queuing delayOn the other hand if late packets do not arrive within theimposed deadline it will be treated as a lost one thus resultingin an increasing packet loss

7 Conclusion

In this paper we have proposed a load balancing oppor-tunistic routing scheme for cognitive radio networks Con-sidering the loading constraint of each node we analyzedthe throughput bound of one opportunistic module andformulated the problem of maximizing the total throughputof the network as a linear programming problem Moreoverwe designed a new metric to select and prioritize the

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

Wireless Communications and Mobile Computing 11

15 20 25 30 35Number of flows

170

180

190

200

210

220

230

240

250

260

270

ro

ughp

ut (K

bps)

LBORSAARSAOR(a) Throughput versus varying number of flows

15 20 25 30 35Number of flows

25

30

35

40

45

50

55

60

65

70

End-

to-e

nd d

elay

(ms)

SAORSAARLBOR(b) End-to-end delay versus varying number of flows

Figure 3

more improvements can be obtained for LBOR when PUsappear frequently

The end-to-end delay and the packet delivery ratio per-formance with different types of PUsrsquo activities are shownin Figures 4(c) and 4(d) With the increase of 119864[119879119874119865119865119888ℎ ]the packet delivery ratio will increase and the end-to-enddelay will decrease When the value of 119864[119879119874119865119865119888ℎ ] is around20ms LBOR can reduce the end-to-end delay by up to36 and 40 compared to SAAR and SAOR HoweverLBOR can only increase the delivery ratio by up to 2and 3 respectively compared to these two schemes Thereare two main reasons for this case First the opportunisticrouting protocol of LBOR is based on multipath routingstrategy where the paths and relay nodes are determinedon-the-fly In addition SAAR and SAOR also utilize thesimilar opportunistic routing principle and they aim toincrease the packet delivery ratio for the dynamic changingspectrum environment Second SAAR and SAOR alwayschoose paths or forwarding nodes with the best performanceof packet delivery ratio Nevertheless LBOR jointly considersthe load balancing for selecting the relay nodes Thus theimprovement of packet delivery ratio from LBOR is limited

623 Impact ofNumber of Channels In this part we comparethe performance of the three schemes under different numberof channels The number of channels varies between 2 and 10and the expected channel OFF time is set to be 60ms

Figure 5(a) compares the SU-PU interference ratio of thethree schemes by varying the number of channels of thenetworkWe can observe that the SU-PU interference ratio ofLBOR is lower than that of SAAR and SAOR One significantobservationhere is thatwhen the number of channels is largerthan 6 the SU-PU interference ratio of SAAR and SAOR

tends to decrease as the number of channels is increased InCRAHNs a larger number of channels can service a largernumber of SUsThus in this case SUs can choosemore stablechannels to avoid PUsrsquo interruptions In addition if an SUis interrupted by PUsrsquo appearance it can easily find anotheravailable channel for data transmissions Another importantobservation made is that when the number of channelsincreases from 2 to 6 the SU-PU interference ratio of LBORwill increase but the other two schemes will decrease Anexplanation for this phenomenon is as follows Consideringthe load balancing strategy LBOR tries its best to distributetraffic evenly among multiple nodes When there are notenough available channels to avoid the mutual interferencesamong SUs the packets cannot be correctly decoded at thereceiver Due to the same reason we can also observe fromFigures 5(b) 5(c) and 5(d) that when the number of channelsis below 6 the improvements from load balancing are notsignificantly

Figure 5(b) shows the throughput of LBOR SAAR andSAOR with different number of channels in the network Weobserve that the throughput of these schemes increases as thenumber of channels becomes larger The result also confirmsthat nomatter howmany channels are available LBOR alwaysperforms better than SAAR and SAOR When the number ofchannels is above 6 more improvement can be obtained fromLBOR

As shown in Figures 5(c) and 5(d) they provide similarresults and demonstrate that LBOR can achieve lower end-to-end delay and higher packet delivery ratio in most casescompared to SAAR and SAOR The improvements comefrom the load balancing based metric and the load balanc-ing based relay selection algorithm On the one hand theproposed scheme adopts a load balancing based metric to

12 Wireless Communications and Mobile Computing

20 40 60 80 100 120E[Tch

OFF] (ms)

0

002

004

006

008

01

012

014

016

018

02

SU-P

U in

terf

eren

ce ra

tio

LBORSAARSAOR(a) SU-PU interference ratio versus PUsrsquo activities

E[TchOFF] (ms)

20 40 60 80 100 120100

150

200

250

300

350

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus PUsrsquo activities

20 40 60 80 100 12025

30

35

40

45

50

55

60

65

End-

to-e

nd d

elay

(ms)

E[TchOFF] (ms)

LBORSAARSAOR

(c) End-to-end delay versus PUsrsquo activities

20 40 60 80 100 120076

078

08

082

084

086

088

09

092

094

Pack

et d

eliv

ery

ratio

LBORSAARSAOR

E[TchOFF] (ms)

(d) Packet delivery ratio versus PUsrsquo activities

Figure 4

choose the relay nodes with higher throughput and lowerdelay On the other hand the load balancing based relayselection algorithm can predict the occupied capacity of eachlink and thus ensure that the traffic load is evenly balancedamong all of the relay nodes

624 Impact of Number of PUs In this experiment weevaluate the performance of three schemes under differentnumber of PUs The number of PUs varies between 10 and40 and the number of channels is set to be 6 Figure 6(a)

compares the SU-PU interference ratio of different schemesunder different number of PUs In most cases the SU-PU interference ratios of SAAR and SAOR rise sharplythan LBOR That is because the network traffic of nonloadbalancing methods (SAAR and SAOR) tends to concentrateon a specific number of SUs and links If these SUsrsquo channelsare occupied by PUs most packets of the network may beaffected On the contrary LBOR distributes traffic to a largernumber of SUs thus packets have less change to be affectedby the PUsrsquo appearance and it induces a load balancing

Wireless Communications and Mobile Computing 13

2 4 6 8 10Total number of channels

002

004

006

008

01

012

014

016

018SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of channels

2 4 6 8 10Total number of channels

160

180

200

220

240

260

280

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of channels

2 4 6 8 10Total number of channels

20

30

40

50

60

70

80

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of channels

2 4 6 8 10Total number of channels

07

075

08

085

09

095Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of channels

Figure 5

improvement compared with SAAR and SAOR When thenumber of PUs is below 30 the benefit from load balancingis not surprising since SUs are not heavily affected Whenthe number of PUs is 35 an interesting observation here isthat the SU-PU interference ratio of LBOR is lower than thecase of 30 The reason for this case is that the aggregatedbenefit from load balancing scheme overwhelms the negativeeffect of the increased number of PUs More specifically inLBOR SUs are not overloaded and the packets are properlydistributed Thus when data transmission failures are causedby the appearance of PUs it is much easier for LBOR toreroute the blocked trafficflows to the light-loaded SUswhose

channels are available compared with other two methodsWhen the number of PUs is 40 we observe that the SU-PUinterference ratio of LBOR is higher than the case of 35 In thissituation there are not enough available channels for SUs totransmit the traffic of thewhole networkThus it is difficult tofind light-loaded SUs to reroute packets for overloaded SUsand more packets would be interrupted by PUsrsquo activities

Figure 6(b) shows the throughput by varying the numberof PUs in the network for the three schemes In the figurewe observe that the proposed LBOR scheme achieves higherthroughput than SAAR and SAOR We can also find thatwhen the number of PUs increases the throughput decreases

14 Wireless Communications and Mobile Computing

10 15 20 25 30 35 40Total number of PUs

003

004

005

006

007

008

009

01

011

012

013SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

160

180

200

220

240

260

280

300

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

25

30

35

40

45

50

55

60

65

70

75

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

078

08

082

084

086

088

09

092Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of PUs

Figure 6

because of available resources are decaying Moreover whenthe number of PUs is small the improvement from LBOR islimited On the contrary when the number of PUs rises above25 the throughput of SAAR and SAOR drops drastically butthe throughput degradation of LBOR is limited This is dueto the fact that LBOR can predict the availability and theoccupied capacity of each link and thus relieve the networkrsquoscongestions

Figures 6(c) and 6(d) can also confirm that LBOR canprovide the lowest end-to-end delay and the highest packetdelivery ratio among the three schemes When the numberof PUs increases the unbalanced distribution of load amongnodes becomes a critical issue and has a negative impacton the end-to-end delay and the packet delivery ratio Thiscan be understood intuitively as follows On the one handthe previous link may become unavailable due to the PU

appearance and the packets which arrived earlier have towait for late packets in reordering buffers at the receivingdestination thus resulting in an increasing queuing delayOn the other hand if late packets do not arrive within theimposed deadline it will be treated as a lost one thus resultingin an increasing packet loss

7 Conclusion

In this paper we have proposed a load balancing oppor-tunistic routing scheme for cognitive radio networks Con-sidering the loading constraint of each node we analyzedthe throughput bound of one opportunistic module andformulated the problem of maximizing the total throughputof the network as a linear programming problem Moreoverwe designed a new metric to select and prioritize the

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

12 Wireless Communications and Mobile Computing

20 40 60 80 100 120E[Tch

OFF] (ms)

0

002

004

006

008

01

012

014

016

018

02

SU-P

U in

terf

eren

ce ra

tio

LBORSAARSAOR(a) SU-PU interference ratio versus PUsrsquo activities

E[TchOFF] (ms)

20 40 60 80 100 120100

150

200

250

300

350

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus PUsrsquo activities

20 40 60 80 100 12025

30

35

40

45

50

55

60

65

End-

to-e

nd d

elay

(ms)

E[TchOFF] (ms)

LBORSAARSAOR

(c) End-to-end delay versus PUsrsquo activities

20 40 60 80 100 120076

078

08

082

084

086

088

09

092

094

Pack

et d

eliv

ery

ratio

LBORSAARSAOR

E[TchOFF] (ms)

(d) Packet delivery ratio versus PUsrsquo activities

Figure 4

choose the relay nodes with higher throughput and lowerdelay On the other hand the load balancing based relayselection algorithm can predict the occupied capacity of eachlink and thus ensure that the traffic load is evenly balancedamong all of the relay nodes

624 Impact of Number of PUs In this experiment weevaluate the performance of three schemes under differentnumber of PUs The number of PUs varies between 10 and40 and the number of channels is set to be 6 Figure 6(a)

compares the SU-PU interference ratio of different schemesunder different number of PUs In most cases the SU-PU interference ratios of SAAR and SAOR rise sharplythan LBOR That is because the network traffic of nonloadbalancing methods (SAAR and SAOR) tends to concentrateon a specific number of SUs and links If these SUsrsquo channelsare occupied by PUs most packets of the network may beaffected On the contrary LBOR distributes traffic to a largernumber of SUs thus packets have less change to be affectedby the PUsrsquo appearance and it induces a load balancing

Wireless Communications and Mobile Computing 13

2 4 6 8 10Total number of channels

002

004

006

008

01

012

014

016

018SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of channels

2 4 6 8 10Total number of channels

160

180

200

220

240

260

280

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of channels

2 4 6 8 10Total number of channels

20

30

40

50

60

70

80

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of channels

2 4 6 8 10Total number of channels

07

075

08

085

09

095Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of channels

Figure 5

improvement compared with SAAR and SAOR When thenumber of PUs is below 30 the benefit from load balancingis not surprising since SUs are not heavily affected Whenthe number of PUs is 35 an interesting observation here isthat the SU-PU interference ratio of LBOR is lower than thecase of 30 The reason for this case is that the aggregatedbenefit from load balancing scheme overwhelms the negativeeffect of the increased number of PUs More specifically inLBOR SUs are not overloaded and the packets are properlydistributed Thus when data transmission failures are causedby the appearance of PUs it is much easier for LBOR toreroute the blocked trafficflows to the light-loaded SUswhose

channels are available compared with other two methodsWhen the number of PUs is 40 we observe that the SU-PUinterference ratio of LBOR is higher than the case of 35 In thissituation there are not enough available channels for SUs totransmit the traffic of thewhole networkThus it is difficult tofind light-loaded SUs to reroute packets for overloaded SUsand more packets would be interrupted by PUsrsquo activities

Figure 6(b) shows the throughput by varying the numberof PUs in the network for the three schemes In the figurewe observe that the proposed LBOR scheme achieves higherthroughput than SAAR and SAOR We can also find thatwhen the number of PUs increases the throughput decreases

14 Wireless Communications and Mobile Computing

10 15 20 25 30 35 40Total number of PUs

003

004

005

006

007

008

009

01

011

012

013SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

160

180

200

220

240

260

280

300

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

25

30

35

40

45

50

55

60

65

70

75

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

078

08

082

084

086

088

09

092Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of PUs

Figure 6

because of available resources are decaying Moreover whenthe number of PUs is small the improvement from LBOR islimited On the contrary when the number of PUs rises above25 the throughput of SAAR and SAOR drops drastically butthe throughput degradation of LBOR is limited This is dueto the fact that LBOR can predict the availability and theoccupied capacity of each link and thus relieve the networkrsquoscongestions

Figures 6(c) and 6(d) can also confirm that LBOR canprovide the lowest end-to-end delay and the highest packetdelivery ratio among the three schemes When the numberof PUs increases the unbalanced distribution of load amongnodes becomes a critical issue and has a negative impacton the end-to-end delay and the packet delivery ratio Thiscan be understood intuitively as follows On the one handthe previous link may become unavailable due to the PU

appearance and the packets which arrived earlier have towait for late packets in reordering buffers at the receivingdestination thus resulting in an increasing queuing delayOn the other hand if late packets do not arrive within theimposed deadline it will be treated as a lost one thus resultingin an increasing packet loss

7 Conclusion

In this paper we have proposed a load balancing oppor-tunistic routing scheme for cognitive radio networks Con-sidering the loading constraint of each node we analyzedthe throughput bound of one opportunistic module andformulated the problem of maximizing the total throughputof the network as a linear programming problem Moreoverwe designed a new metric to select and prioritize the

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

Wireless Communications and Mobile Computing 13

2 4 6 8 10Total number of channels

002

004

006

008

01

012

014

016

018SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of channels

2 4 6 8 10Total number of channels

160

180

200

220

240

260

280

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of channels

2 4 6 8 10Total number of channels

20

30

40

50

60

70

80

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of channels

2 4 6 8 10Total number of channels

07

075

08

085

09

095Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of channels

Figure 5

improvement compared with SAAR and SAOR When thenumber of PUs is below 30 the benefit from load balancingis not surprising since SUs are not heavily affected Whenthe number of PUs is 35 an interesting observation here isthat the SU-PU interference ratio of LBOR is lower than thecase of 30 The reason for this case is that the aggregatedbenefit from load balancing scheme overwhelms the negativeeffect of the increased number of PUs More specifically inLBOR SUs are not overloaded and the packets are properlydistributed Thus when data transmission failures are causedby the appearance of PUs it is much easier for LBOR toreroute the blocked trafficflows to the light-loaded SUswhose

channels are available compared with other two methodsWhen the number of PUs is 40 we observe that the SU-PUinterference ratio of LBOR is higher than the case of 35 In thissituation there are not enough available channels for SUs totransmit the traffic of thewhole networkThus it is difficult tofind light-loaded SUs to reroute packets for overloaded SUsand more packets would be interrupted by PUsrsquo activities

Figure 6(b) shows the throughput by varying the numberof PUs in the network for the three schemes In the figurewe observe that the proposed LBOR scheme achieves higherthroughput than SAAR and SAOR We can also find thatwhen the number of PUs increases the throughput decreases

14 Wireless Communications and Mobile Computing

10 15 20 25 30 35 40Total number of PUs

003

004

005

006

007

008

009

01

011

012

013SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

160

180

200

220

240

260

280

300

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

25

30

35

40

45

50

55

60

65

70

75

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

078

08

082

084

086

088

09

092Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of PUs

Figure 6

because of available resources are decaying Moreover whenthe number of PUs is small the improvement from LBOR islimited On the contrary when the number of PUs rises above25 the throughput of SAAR and SAOR drops drastically butthe throughput degradation of LBOR is limited This is dueto the fact that LBOR can predict the availability and theoccupied capacity of each link and thus relieve the networkrsquoscongestions

Figures 6(c) and 6(d) can also confirm that LBOR canprovide the lowest end-to-end delay and the highest packetdelivery ratio among the three schemes When the numberof PUs increases the unbalanced distribution of load amongnodes becomes a critical issue and has a negative impacton the end-to-end delay and the packet delivery ratio Thiscan be understood intuitively as follows On the one handthe previous link may become unavailable due to the PU

appearance and the packets which arrived earlier have towait for late packets in reordering buffers at the receivingdestination thus resulting in an increasing queuing delayOn the other hand if late packets do not arrive within theimposed deadline it will be treated as a lost one thus resultingin an increasing packet loss

7 Conclusion

In this paper we have proposed a load balancing oppor-tunistic routing scheme for cognitive radio networks Con-sidering the loading constraint of each node we analyzedthe throughput bound of one opportunistic module andformulated the problem of maximizing the total throughputof the network as a linear programming problem Moreoverwe designed a new metric to select and prioritize the

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

14 Wireless Communications and Mobile Computing

10 15 20 25 30 35 40Total number of PUs

003

004

005

006

007

008

009

01

011

012

013SU

-PU

inte

rfer

ence

ratio

LBORSAARSAOR

(a) SU-PU interference ratio versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

160

180

200

220

240

260

280

300

ro

ughp

ut (K

bps)

LBORSAARSAOR

(b) Throughput versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

25

30

35

40

45

50

55

60

65

70

75

End-

to-e

nd d

elay

(ms)

LBORSAARSAOR

(c) End-to-end delay versus varying number of PUs

Total number of PUs10 15 20 25 30 35 40

078

08

082

084

086

088

09

092Pa

cket

del

iver

y ra

tio

LBORSAARSAOR

(d) Packet delivery ratio versus varying number of PUs

Figure 6

because of available resources are decaying Moreover whenthe number of PUs is small the improvement from LBOR islimited On the contrary when the number of PUs rises above25 the throughput of SAAR and SAOR drops drastically butthe throughput degradation of LBOR is limited This is dueto the fact that LBOR can predict the availability and theoccupied capacity of each link and thus relieve the networkrsquoscongestions

Figures 6(c) and 6(d) can also confirm that LBOR canprovide the lowest end-to-end delay and the highest packetdelivery ratio among the three schemes When the numberof PUs increases the unbalanced distribution of load amongnodes becomes a critical issue and has a negative impacton the end-to-end delay and the packet delivery ratio Thiscan be understood intuitively as follows On the one handthe previous link may become unavailable due to the PU

appearance and the packets which arrived earlier have towait for late packets in reordering buffers at the receivingdestination thus resulting in an increasing queuing delayOn the other hand if late packets do not arrive within theimposed deadline it will be treated as a lost one thus resultingin an increasing packet loss

7 Conclusion

In this paper we have proposed a load balancing oppor-tunistic routing scheme for cognitive radio networks Con-sidering the loading constraint of each node we analyzedthe throughput bound of one opportunistic module andformulated the problem of maximizing the total throughputof the network as a linear programming problem Moreoverwe designed a new metric to select and prioritize the

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

Wireless Communications and Mobile Computing 15

candidate nodes which considers the traffic load We alsoproposed load balancing based candidate forwarder sortingand relay selection algorithms Simulation results have shownthe advantages of LBOR over SAAR and SAOR concerningthe SU-PU interference ratio the packet delivery ratio thethroughput and the end-to-end delay of CRAHNs

Data Availability

The simulation data used to support the findings of thisstudy were supplied by Wenxuan Duan under license andso cannot be made freely available Requests for access tothese data should be made to [Wenxuan Duan duanwenx-uanwhuteducn]

Conflicts of Interest

The authors declare that they have no conflicts of interest

Acknowledgments

This work was funded by the National Natural ScienceFoundation of China (Grant No 61601337 and 61771354)the Hubei Provincial Natural Science Foundation of China(Grant No 2017CFB593) and the Fundamental ResearchFunds for the Central Universities (WUT2017II14XZ)

References

[1] J Mitola III and G Q Maguire Jr ldquoCognitive radio makingsoftware radios more personalrdquo IEEE Personal Communica-tions vol 6 no 4 pp 13ndash18 1999

[2] Y Saleem K-L A Yau H Mohamad N Ramli and M HRehmani ldquoSMARTA SpectruM-AwareClusteR-based rouTingscheme for distributed cognitive radio networksrdquo ComputerNetworks vol 91 pp 196ndash224 2015

[3] Y Saleem K-L A Yau H Mohamad N Ramli M HRehmani and Q Ni ldquoClustering and Reinforcement-Learning-Based Routing for Cognitive Radio Networksrdquo IEEE WirelessCommunications Magazine vol 24 no 4 pp 146ndash151 2017

[4] Y Saleem K-L A Yau HMohamad et al ldquoJoint channel selec-tion and cluster-based routing scheme based on reinforcementlearning for cognitive radio networksrdquo in International Confer-ence on Computer Communications and Control Technologypp 21ndash25 2015

[5] M Zareei E M Mohamed M H Anisi C V Rosales KTsukamoto and M K Khan ldquoOn-Demand Hybrid Routingfor Cognitive Radio Ad-Hoc Networkrdquo IEEE Access vol 4 pp8294ndash8302 2016

[6] K J Prasanna Venkatesan and V Vijayarangan ldquoSecure andreliable routing in cognitive radio networksrdquoWireless Networksvol 23 no 6 pp 1689ndash1696 2017

[7] A Guirguis M Karmoose K Habak M El-Nainay and MYoussef ldquoCooperation-based multi-hop routing protocol forcognitive radio networksrdquo Journal of Network and ComputerApplications vol 110 pp 27ndash42 2018

[8] J Lu Z Cai and X Wang ldquoUser social activity-based routingfor cognitive radio networksrdquo Personal Ubiquitous Computingvol 13 pp 1ndash17 2018

[9] S Biswas and R Morris ldquoExOR opportunistic multi-hoprouting for wireless networksrdquo in Proceedings of the Confer-ence on Applications Technologies Architectures and Protocolsfor Computer Communications (SIGCOMM rsquo05) pp 133ndash144ACM 2005

[10] R W Coutinho A Boukerche L F Vieira and A A LoureiroldquoGeographic and opportunistic routing for underwater sen-sor networksrdquo Institute of Electrical and Electronics EngineersTransactions on Computers vol 65 no 2 pp 548ndash561 2016

[11] M A Rahman Y Lee and I Koo ldquoEECOR An Energy-Efficient Cooperative Opportunistic Routing Protocol forUnderwater Acoustic Sensor Networksrdquo IEEE Access vol 5 pp14119ndash14132 2017

[12] X Tang Y Chang and K Zhou ldquoGeographical opportunis-tic routing in dynamic multi-hop cognitive radio networksrdquoin Proceedings of the 2012 Computing Communications andApplications Conference ComComAp 2012 pp 256ndash261 ChinaJanuary 2012

[13] X Tang and Q Liu ldquoNetwork coding based geographicalopportunistic routing for ad hoc cognitive radio networksrdquo inProceedings of the 2012 IEEE Globecom Workshops GC Wkshps2012 pp 503ndash507 USA December 2012

[14] X Tang J Zhou S Xiong J Wang and K Zhou ldquoGeographicSegmented Opportunistic Routing in Cognitive Radio Ad HocNetworks Using Network Codingrdquo IEEE Access 2018

[15] JWang H Yue L Hai and Y Fang ldquoSpectrum-AwareAnypathRouting in Multi-Hop Cognitive Radio Networksrdquo IEEE Trans-actions on Mobile Computing vol 16 no 4 pp 1176ndash1187 2017

[16] R Sanchez-Iborra and M-D Cano ldquoJOKER A Novel Oppor-tunistic Routing Protocolrdquo IEEE Journal on Selected Areas inCommunications vol 34 no 5 pp 1690ndash1703 2016

[17] C Cui H Man Y Wang and S Liu ldquoOptimal CooperativeSpectrumAware Opportunistic Routing in Cognitive Radio AdHoc Networksrdquo Wireless Personal Communications vol 91 no1 pp 101ndash118 2016

[18] X Zhong R Lu L Li and S Zhang ldquoETOR Energy andTrust Aware Opportunistic Routing in Cognitive Radio SocialInternet ofThingsrdquo in Proceedings of the 2017 IEEEGlobal Com-munications Conference (GLOBECOM2017) pp 1ndash6 SingaporeDecember 2017

[19] X Tang H Zhang K Zhou and J Wang ldquoExtending accesspoint service coverage area through opportunistic forwardingin multi-hop collaborative relay WLANsrdquo in Proceedings of the20th IEEE International Conference on Parallel and DistributedSystems ICPADS 2014 pp 787ndash792 Taiwan December 2014

[20] H Ben Fradj R Anane M Bouallegue and R Bouallegue ldquoArange-based opportunistic routing protocol forWireless Sensornetworksrdquo in Proceedings of the 13th IEEE International WirelessCommunications and Mobile Computing Conference IWCMC2017 pp 770ndash774 Spain June 2017

[21] Y B Zikria S Nosheen J-G Choi and S W Kim ldquoHeuris-tic Approach to Select Opportunistic Routing Forwarders(HASORF) to Enhance Throughput for Wireless Sensor Net-worksrdquo Journal of Sensors vol 2015 Article ID 634759 10 pages2015

[22] H Abdulwahid B Dai B Huang and Z Chen ldquoOptimalenergy and Load Balance Routing for MANETsrdquo in Proceedingsof the 2015 4th International Conference on Computer ScienceandNetwork Technology (ICCSNT) pp 981ndash986Harbin ChinaDecember 2015

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

16 Wireless Communications and Mobile Computing

[23] M Michel S Duquennoy B Quoitin and T Voigt ldquoLoad-balanced data collection through opportunistic routingrdquo inPro-ceedings of the 11th IEEE International Conference onDistributedComputing in Sensor Systems DCOSS 2015 pp 62ndash70 BrazilJune 2015

[24] W Shi E Wang Z Li et al ldquoA load-balance and interference-aware routing algorithm for multicast in the wireless meshnetworkrdquo in IEEE International Conference on Information andCommunication Technology Convergence pp 206ndash210 2016

[25] J So and H Byun ldquoLoad-Balanced Opportunistic Routing forDuty-Cycled Wireless Sensor Networksrdquo IEEE Transactions onMobile Computing vol 16 no 7 pp 1940ndash1955 2017

[26] X Xu S Fu Q Cai et al ldquoDynamic Resource Allocation forLoad Balancing in Fog EnvironmentrdquoWireless Communicationsand Mobile Computing vol 2018 Article ID 6421607 15 pages2018

[27] Y Liu L X Cai and X S Shen ldquoSpectrum-aware opportunisticrouting inmulti-hop cognitive radio networksrdquo IEEE Journal onSelectedAreas in Communications vol 30 no 10 pp 1958ndash19682012

[28] J Li X Wang R Yu and J Jia ldquoThe adaptive routing algo-rithm depending on the traffic prediction model in cognitivenetworksrdquo in Proceedings of the 1st International Conference onPervasive Computing Signal Processing andApplications PCSPA2010 pp 319ndash322 China September 2010

[29] DGanesan B Krishnamachari AWoo et al ldquoComplex Behav-ior at Scale An Experimental Study of Low-Power WirelessSensor Networksrdquo Tech Rep 02-0013 Univ of California LosAngeles 2002

[30] ldquoThe Network Simulator-ns-2rdquo httpwwwisiedunsnamns[31] Cognitive Radio Network patch for ns2 httpsdocsgoogle

comfiled0B7S255p3kFXNNXEtS3ozTHpmRHMedit 2009

International Journal of

AerospaceEngineeringHindawiwwwhindawicom Volume 2018

RoboticsJournal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Active and Passive Electronic Components

VLSI Design

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Shock and Vibration

Hindawiwwwhindawicom Volume 2018

Civil EngineeringAdvances in

Acoustics and VibrationAdvances in

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Electrical and Computer Engineering

Journal of

Advances inOptoElectronics

Hindawiwwwhindawicom

Volume 2018

Hindawi Publishing Corporation httpwwwhindawicom Volume 2013Hindawiwwwhindawicom

The Scientific World Journal

Volume 2018

Control Scienceand Engineering

Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom

Journal ofEngineeringVolume 2018

SensorsJournal of

Hindawiwwwhindawicom Volume 2018

International Journal of

RotatingMachinery

Hindawiwwwhindawicom Volume 2018

Modelling ampSimulationin EngineeringHindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Chemical EngineeringInternational Journal of Antennas and

Propagation

International Journal of

Hindawiwwwhindawicom Volume 2018

Hindawiwwwhindawicom Volume 2018

Navigation and Observation

International Journal of

Hindawi

wwwhindawicom Volume 2018

Advances in

Multimedia

Submit your manuscripts atwwwhindawicom