1

Coordination-based Medium Access Control withSpace-reservation for Wireless Ad Hoc Networks

Kamal Rahimi Malekshan, Student member, IEEE, Weihua Zhuang, Fellow, IEEEand Yves Lostanlen, Senior member, IEEE

Abstract—Efficient radio spectrum utilization and low energyconsumption in mobile devices are essential in developing nextgeneration wireless networks. This paper presents a new mediumaccess control (MAC) mechanism to enhance spectrum efficiencyand reduce energy consumption in a wireless ad hoc network.A set of coordinator nodes, distributed in the network area,periodically schedule contention-free time slots for all datatransmissions/receptions in the network, based on transmissionrequests from source nodes. Adjacent coordinators exchangescheduling information to effectively increase spatial spectrumreuse and avoid transmission collisions. Moreover, the proposedMAC scheme allows a node to put its radio interface into a sleepmode when it is not transmitting/receiving a packet, in order toreduce energy consumption. Simulation results demonstrate thatthe proposed scheme achieves substantially higher throughputand has significantly lower energy consumption in comparisonwith existing schemes.

Index Terms—Medium access control, wireless ad hoc net-works, spectrum efficiency, energy efficiency.

I. INTRODUCTION

The number of mobile devices and the volume of mobiledata traffic have been constantly increasing. It is forecasted thatthere will be over 10 billion interconnected mobile devices,including machine-to-machine (M2M) modules, by 2018 [1].Overall mobile data traffic is expected to grow nearly 11-fold by 2018 from that in 2013 [1]. To meet the increasinggrowth of mobile data traffic, it is essential to efficiently utilizenetwork resources in the next generation wireless networks.A short communication range in small cells (or WiFi) forhot-spot mobile communications is a key to increase networkcapacity via spatial spectrum reuse. Such a dense network ofmobile nodes and access points and emerging M2M com-munications necessitate establishing self-organizing ad hocnetworks to opportunistically leverage spectrum. Yet, energyconsumption by radio interfaces should be minimized, becauseof limited battery storage of mobile devices. In [2], we presenta new energy efficient MAC protocol with high throughput andlow packet transmission delay for a fully connected network,in which only one node can transmit at each time instanceover the radio channel.

In a wireless ad hoc network, nodes that are not in thecommunication range of each other cannot hear each others’

This work was supported by a research grant from the Natural Sciencesand Engineering Research Council (NSERC) of Canada.

K. Rahimi Malekshan and W. Zhuang are with the Department of Elec-trical and Computer Engineering, University of Waterloo, Canada (e-mail:{krahimim, wzhuang}@uwaterloo.ca). Y. Lostanlen is with SIRADEL NorthAmerica, Toronto, Canada (e-mail: yves.lostanlen@ieee.org).

transmissions. However, their transmission may interfere eachother at the receiver nodes. On the other hand, nodes thatare sufficiently far apart in space can transmit simultaneouslywithout a collision (i.e., spatial spectrum reuse is possible).Thus, an effective medium access control (MAC) scheme fora wireless network should have the following features:

1) It should prevent simultaneous transmission of interfer-ing links. Otherwise, one or more of the transmissionswill fail because of transmission collision, which resultsin wastage of radio bandwidth and energy;

2) It should allow simultaneous transmissions of non-interfering links for spatial reuse of the radio channel,because preventing non-interfering links from simultane-ous transmission will unnecessarily degrade throughputof the network.

When a MAC scheme fails to accomplish the first feature, thehidden terminal problem arises. On the other hand, when aMAC scheme does not have the second feature, the exposedterminal problem occurs. A TDMA (Time Division MultipleAccess) MAC scheme can potentially solve both the hiddenterminal and exposed terminal problems in a wireless ad hocnetwork. However, finding an efficient time schedule requiresa central controller and the optimal solution is NP-hard [3],[4]. Moreover, in a wireless ad hoc network, the traffic loadand network topology change with time, which makes thestatic TDMA very inefficient. In addition, reassignment ofchannel time imposes a large overhead and requires globalchanges. The CSMA (Carrier Sense Multiple Access) MAC iscommonly used in wireless ad hoc (and wireless local area)networks because of its flexibility and simplicity. However, itsuffers from transmission collision and contention overhead,and cannot resolve the hidden and exposed terminal problemsin a wireless ad hoc network. The hidden terminal problemcan be avoided by increasing the carrier sensing range [5],which however aggravates the exposed terminal problem andresults in wastage of radio bandwidth. The RTS/CTS (Request-to-send/clear-to-send) mechanism is used in [6]–[10] to mit-igate the hidden terminal problem. However, this mechanismimposes a significant amount of overhead in bandwidth andenergy.

Radio interface is a main source of energy consumption inmobile devices, which can quickly drain the device’s limitedbattery [11]–[14]. For instance, the WiFi radio consumes morethan 70% of total energy in a smartphone when the screenis off [13], which is reduced to 44.5% in the power savingmode. A radio interface can be in one of the following

2

modes: transmit, receive, idle, and sleep. A radio interfaceconsumes a significant amount of energy in the idle mode,in which it is neither transmitting nor receiving a packet.For instance, a Cisco Aironet 350 series WLAN adapter [15]consumes 2.25W, 1.25W, 1.25W, and 0.075W in the transmit,receive, idle, and sleep modes respectively. In order to reduceenergy consumption, nodes should periodically put their radiointerfaces into the sleep mode. While a radio interface is in thesleep mode, the node cannot receive incoming packets. Thus, atransmitter node should be aware of the receiver node’s statusto successfully deliver a packet.

In a cellular network, network area is partitioned into cellsand nodes inside a cell only communicate with the cellbase station (BS) at the cell center. The BS schedules alltransmissions/receptions to and from nodes (downlink anduplink) inside its cell. Therefore, transmission collisions areprevented among nodes in the cell and idle listening energyconsumption of mobile nodes is minimized, because of de-terministic transmission/reception time which is assigned bythe BS. In the conventional cellular networks, each cell isassigned a fraction of total available radio spectrum to avoidinter-cell interference. For instance, in GSM a cell commonlyuses one-fourth of total available radio spectrum (frequencyreuse factor 4) to prevent inter-cell interference. Several inter-cell interference coordination techniques are proposed to im-prove network performance of cellular systems using fractionalfrequency reuse [16], [17]. In fractional frequency reuse, thetotal available radio spectrum is used for transmissions to andfrom the nodes close to the BS at the central region of a cell,but a fraction of spectrum is used for transmissions to and fromnodes that are outside the central region of the cell, in order toreduce inter-cell interference [16]–[19]. The dense deploymentof small cells in the next generation of wireless networks andthe direct device-to-device (D2D) and M2M communicationsform communication links in an ad hoc manner, which requirea new MAC mechanism to efficiently utilize the shared radiospectrum and minimize power consumption.

In this paper, we propose a novel medium access mechanismfor a wireless ad hoc network with arbitrary communicationlinks. Table I compares main characteristics of the proposedMAC mechanism and existing approaches. The proposedscheme combines the deterministic transmission/reception fea-ture of cellular networks and the opportunistic spectrum accessfeature of WiFi networks to efficiently utilize shared spectrumand minimize energy consumption. A set of coordinatorsdistributed in the network area are chosen to dynamicallycoordinate contention-free time slots for all data transmis-sions/receptions based on transmission requests from sourcenodes. Each coordinator periodically broadcasts a schedulingpacket to schedule all transmissions/receptions in its proximity.For each scheduled transmission/reception, the space aroundthe receiver node is reserved to avoid transmission collisionand enhance spatial radio spectrum reuse. A coordinator col-lects nodes’ transmission requests and overhears the schedul-ing packets of its neighboring coordinators. Accordingly, eachcoordinator schedules a transmission/reception only if thetransmission of the source node does not interfere with otherscheduled receptions and the other scheduled transmissions do

TABLE ICOMPARISON OF PROPOSED MAC MECHANISM WITH EXISTING

APPROACHES

Characteristic Proposed scheme CSMA/CA Cellular networks Data transmission time Deterministic Contention-based Deterministic Frequency reuse Space reservation Carrier sensing and RTS/CTS Frequency planning Communication links Arbitrary Arbitrary Between nodes and BS

not interfere with the reception at the destination. Dynamicassignment of the shared radio spectrum and adequate spatialspectrum reuse increase spectrum efficiency. Moreover, adeterministic transmission/reception time warrants nodes toput their radio interface into the sleep mode when they areneither transmitting nor receiving a packet, which reducesenergy consumption. Comparing with existing schemes, theproposed MAC provides significantly higher throughput andgreatly reduces node energy consumption.

The rest of this paper is organized as follows: Section II re-views related works. The system model is presented in SectionIII. In Section IV, we describe the proposed MAC mechanism.Simulation results are presented in Section V to evaluate theperformance of proposed MAC solution in comparison withexisting schemes. Finally, Section VI concludes this research.

II. RELATED WORKS

A dynamic TDMA MAC scheme is proposed in [20], [21].Time is partitioned into frames of F slots. Every node acquiresa transmission slot in each frame, in which it transmits apacket to inform the other nodes of the time slots that it willtransmit/receive data packets (frame information). A node canreserve additional transmission slots using ALOHA and/or viabroadcasting its frame information. A node can reserve a newtime slot only if none of the neighboring nodes has announceda transmission/reception in that time slot in the previous frame.This mechanism can mitigate the hidden terminal problem;however the imposed overhead of transmitting frame informa-tion by every node in each frame reduces network throughputand increases energy consumption. A hybrid TDMA-CSMAMAC scheme is proposed in [22] using CSMA as the baselineMAC scheme. A transmission time slot is assigned to eachnode such that none of the interfering nodes are assigned asame transmission slot. At each time slot, the owner has ahigher priority to transmit a packet. If a node experiencessuccessive collisions because of hidden nodes, it will trans-mit a request packet to prevent the interfering nodes fromtransmission in its assigned transmission slot for a requestedperiod of time.

The existing single-channel energy saving MAC proto-cols for a wireless ad hoc network can be classified intosynchronous and asynchronous energy saving protocols. Inthe synchronous energy saving schemes [6], [23], [24], all

3

nodes are synchronized in time and time is partitioned intobeacon intervals. All nodes wake up simultaneously at thebeginning of each beacon interval, in an ATIM (Ad hoctraffic indication message) window, to exchange ATIM packetswhich are transmitted by the sender nodes to inform theirintended receivers of buffered packets. During the rest ofbeacon interval (i.e., the communication period), the nodes thathave packets to transmit/receive stay awake to communicate.Other nodes switch their radio interfaces into the sleep modeto save energy. In the asynchronous energy saving schemes[25], [26], nodes are not synchronized in time and each nodehas its own clock. Each node evenly divides its time intobeacon intervals. There are two types of beacon intervals:active beacon intervals and energy saving beacon intervals.An active beacon interval starts with a beacon window, duringwhich the node should contend to transmit a beacon includingits clock and wake up pattern. Followed by the beacon windowis an MTIM (Multi-hop traffic indication message) window(similar to the ATIM window in the synchronous schemes),in which the nodes with buffered packets notify the intendedreceivers. After that, each node stays awake for the rest ofbeacon interval to receive the beacons transmitted by othernodes. An energy saving beacon interval starts by an MTIMwindow, and after the MTIM window, the node can power offif it has no packet to send or receive. The patterns of activeand energy saving beacon intervals should be chosen to ensurethat the beacon of each node will be heard by every single-hopnode at least once in a predefined period of time.

Existing energy saving schemes for wireless ad hoc net-works require all the nodes to stay awake during the ATIM (orMTIM) window in each beacon interval, and every node witha packet for transmission has to contend to send a notificationpacket to its intended receiver at each beacon interval. TheATIM (or MTIM) window overhead decreases the commu-nication period in each beacon interval and increases energyconsumption. Further, the ATIM (or MTIM) size significantlyaffects the network performance and should be adjusted basedon the networking condition. How to choose an optimal ATIM(or MTIM) size is an open issue.

The synchronous energy saving mode requires all nodes tobe synchronized in time. Time synchronization in a wirelessad hoc network is challenging because propagation delaysare long and the network may temporarily be partitioned.Although asynchronous power saving schemes do not needthe synchronization among nodes, no need for synchronizationcomes at expense of a requirement for periodically activebeacon intervals and more beacon transmissions than in syn-chronous power saving schemes. The more frequent beacontransmissions and periodically active beacon intervals causemore transmission overhead and more energy consumptionin asynchronous power saving schemes, in comparison withsynchronous power saving schemes.

In the existing power saving MAC schemes, the contentionand collision overhead during the communication period de-grades the network throughput and increases energy consump-tion. In a wireless network, the collision rate is further in-creased because of the hidden terminal problem, which furtherdecreases the performance of CSMA/CA MAC used in the

existing power saving protocols. Moreover, in the CSMA/CAMAC, the radio channel is not efficiently utilized becauseof the exposed terminal problem, which further reduces thenetwork performance. Even though the TMMAC [24] uses acontention-free MAC protocol in the communication periodto reduce overhead and energy consumption, it cannot fullyutilize the radio channel bandwidth because nodes reserve timeslots independently without coordination. Also, the TMMACrequires exchanging of three control packets (ATIM/ATIM-ACK/ATIM-RES) between the source and destination nodesin the ATIM window, which degrades channel utilization andincreases energy consumption.

In this work, we propose a new MAC mechanism to achievehigh throughput and low energy consumption in a wirelessad hoc network. Using a set of coordinators, the proposedMAC scheme dynamically reserves channel in both time andspace domains for data transmissions/receptions based onnodes’ transmission requests. Exchanging scheduling infor-mation among adjacent coordinators empowers the proposedMAC scheme to effectively increase spatial spectrum reuse andprevent transmission collision. Also, periodic assignment ofcontention-free data transmission/reception time slots enablesa node to put its radio interface into the sleep mode whenit is not transmitting/receiving a packet, which significantlyreduces energy consumption.

III. SYSTEM MODEL

Consider a wireless ad hoc network with N nodes whereall nodes are not in the communication range of each other.We focus on single-channel single-hop transmissions as, atthe MAC layer, each node communicates with one or moreof its one-hop neighboring nodes. Let lij denote single-hoplink from source node i ∈ {1, 2, ..., N} to destination nodej ∈ {1, 2, ..., N}, i 6= j. We denote the distance between thesource and destination nodes of lij by dij . The channel gainbetween source node i and destination node j is hij = cd−αij ,where c is a constant and α is the path loss exponent. Letp = (p1, p2, ..., pN ) denote the transmission power vector,where pi, i ∈ {1, 2, ..., N}, denotes the transmission powerlevel of source node i. Let u = (u1, u2, ..., uN ) denote thetransmission vector, where ui = 1 denotes that node i isscheduled for transmission and ui = 0 otherwise. Thus, thesignal to noise plus interference ratio (SINR) at the destinationof link lij is given by

γij =uipihij

N0 +∑k 6=i ukpkhkj

(1)

where N0 is background noise power and∑k 6=i ukjpkhkj ,

Iij is the amount of interference at the destination of link lij .All control/scheduling packets are transmitted at power level

Ps at rate Rs bps and all data packets are transmitted atpower level Pd at rate Rd bps. The corresponding minimumrequired SINR at a receiver node to successfully receivecontrol/scheduling and data packets are denoted by Γs andΓd respectively.

4

IV. MEDIUM ACCESS CONTROL

In order to efficiently utilize the radio channel and minimizeenergy consumption in a wireless ad hoc network, we use thefollowing main strategies:

1) Dynamic coordination of access to the shared mediumbased on instantaneous traffic load by a set of coordina-tors distributed in the network area;

2) Preventing transmission collisions and minimizingidle listening power consumption by periodic assign-ment of deterministic time slots for data transmis-sions/receptions;

3) Effective spatial channel reuse by space-reservation forscheduled transmissions/receptions and by exchangingscheduling information among adjacent coordinators.

The network coverage area is partitioned into hexagonal cells,as shown in Figure 1. The distance between the center anda vertex of a cell is denoted by rg , which is set such thatrg ≥ maxdij∈L dij , where L is the set of single-hop linksin the network. Therefore, the source and destination nodesof each single-hop link are either in one cell or adjacentcells. A node at the center of each cell coordinates all thetransmissions/receptions for nodes inside the cell. We assumethat coordinators have higher energy capacity and do not movefrequently (e.g., access points). Thus, the network planningdoes not need to be updated frequently.

All nodes are synchronized in time, and time is partitionedinto frames. Figure 2 shows the structure of a frame. Eachframe consists of three types of time slots, i.e., schedulingslots, contention-free slots, and contention slots. In schedulingtime slots, located at the beginning of each frame, coordi-nators transmit scheduling packets to coordinate transmis-sions/receptions of the current frame. The scheduling packetof a coordinator should be received by all nodes in the cell andadjacent coordinators. Data packet transmissions/receptionstake place in contention-free time slots, as scheduled bycoordinators. A source node scheduled for transmission incontention-free slots can notify the cell coordinator of itstransmission request for the next frame by including infor-mation in the header of one data packet. During contentionslots, source nodes that want to initiate a new transmissioncontend with each other to send a transmission request to thecell coordinator. In the following, we describe transmissionpolicy in each time slot, and then the detail operation of theMAC protocol.

A. Transmission policies in the different time slots

Scheduling slots: Scheduling time slots are assigned tocoordinators such that a scheduling time slot, assigned to acoordinator, is not assigned to any other two-hop neighboringcoordinator. Let Sj , j ∈ {1, 2, .., k}, denote the jth schedulingslot in a frame and Gi, i ∈ {0, 2, 3, ..., k − 1}, denote the setof coordinators that can be assigned same scheduling timeslot. Similar to frequency reuse in cellular networks, withk (= 7) scheduling time slots, as illustrated in Figure 3,every coordinator can acquire a scheduling time slot that isnot assigned to any other two-hop neighboring coordinatornode. To ensure fair channel access for nodes in different

C0

C1

C2

rg

C6

C3

C4

C5

ra

b cda

e

fr

Fig. 1. Partitioning the network area into hexagonal cells, where Ci, i ∈{0, 1, 2, ...,m}, denotes the coordinator of cell i, the dotted circle centred atCi shows the area that Ci broadcasts all scheduled transmissions/receptions,and the shaded area shows the space reserved for transmission from node fto node e.

D

3

C

1

D

2

D

1. . . . . . . . . . . . . . . . . . . . . . . . . . . .. . .

S

2

S

1

Frame n

S Scheduling slot Data transmission slotD C Contention slot

D

4

C

2. . .

Frame n+1Frame n-1

Fig. 2. Structure of one frame of the proposed scheme.

cells, we change transmission order of coordinators in eachframe as illustrated in Figure 3(b). In frame n, coordinatorsGi are assigned the j∗th scheduling time slot (Sj∗ ) wherej∗ = (n mod i + 1) + 1. Moreover, the size (rg) of cells,transmission power level (Ps) for scheduling packets, and datatransmission rate (Rs) of scheduling packets are selected suchthat a scheduling packet is received by all nodes inside thecell and all adjacent coordinators (with SINR ≥ Γs).

Contention-free slots: Data packet transmissions/receptionsare scheduled in contention-free time slots. For each scheduledtransmission/reception, no other node should be scheduledfor transmission in a reserved area around the receiver toguarantee required SINR at the destination. The shaded areain Figure 1 shows the reserved space for transmission fromnode f to node e, where no other node is scheduled fortransmission in the area to guarantee the required SINR atnode e. The reserved area for a scheduled link can be parts ofseveral adjacent cells (as in Figure 1), which is determined byexchanging real-time scheduling information among adjacentcoordinators. The proposed space-reservation mechanism is toprovide effective spatial spectrum reuse to improve spectrumefficiency while avoiding transmission collisions. In addition,for each scheduled source node in the current frame, thespace around the cell coordinator is reserved during onecontention-free slot to enure that the cell coordinator receives

5

G5

G2

G1

G3

G4

G6

G0

G5

G2

G1

G3

G4

G6

G0

G5

G2

G1

G3

G4

G6

G0

G5

G2

G1

G3

G4

G6

G0

G5

G2

G1

G3

G4

G6

G0

G5

G2

G1

G3

G4

G6

G0 G5

G2

G1

G3

G4

G6

G0

(a)

S

5

S

4

S

7

S

6

S

3

S

2

S

1

G4

S

5

S

4

S

7

S

6

S

3

S

2

S

1

Frame n Frame n+1

G5 G6 G1 G2 G3G0 G4 G5 G6 G1 G2G3 G0

(b)

Fig. 3. Assignment of scheduling time slots to coordinators, in which ascheduling time slot is assigned to all the coordinators of cells of a samegroup/color.

the transmission request of source node (for the next frame)that is included in the header of a data packet. When alink is scheduled for transmission, all other nodes in thereserved area around the receiver (and around coordinators)are denoted as interfering nodes and should not be scheduledfor transmission. Let r(d) denote the radius of the circularreserved area centered at the receiver node when the distancebetween the transmitter and receiver is d. The amount ofinterference imposed on the receiver due to transmissionsoutside the reserved area has an upper bound given by

I(d) ≤ I(d) , c′cPdr(d)

α , (2)

where c′ is a constant and depends on the node density andnetwork traffic load. Therefore, the received SINR at thedestination can be represented by

γ ≥cPddα

N0 + I(d). (3)

Using (2) and (3), the minimum radius of the reserved circulararea centred at the receiver to guarantee γ ≥ Γd can becalculated as

r(d) =( c′cPd

cPddαΓd

−N0

)1/α

. (4)

Under the assumption I(d)� N0,

r(d) ≈ (c′Γd)1/α

d. (5)

According to (4) and (5), as c′ increases, the reserved circulararea increases, which decreases the probability of packet

collisions. However, spectrum reuse is decreased as a resultof the larger reserved area per transmission/reception.

Contention slots: Each coordinator marks a few time slotsas contention slots, in which nodes inside the cell (that arenot currently scheduled for transmission) can send a requestto initiate a new transmission. In the contention slots, nodescontend with each other using a CSMA MAC scheme to senda transmission request to their cell coordinators. Adjacent co-ordinators mark the same idle time slot(s) as contention slots.Coordinators dynamically adjust the number of contentionslots and contention window size based on the traffic loadcondition. In Appendix, we present a mathematical modelto calculate the number of successful transmission requestsin the contention slots and the average delay to initiate anew transmission. Using the analytical model, we propose amechanism to dynamically adjust the contention window sizeand the number of contention slots based to the network loadand the required delay to initiate a new transmission.

B. Operation of the MAC protocol

A coordinator node stays awake during the following timeslots in a frame:

1) Scheduling slots – to transmit a scheduling packet andto receive the scheduling packets transmitted by adjacentcoordinators;

2) One of the contention-free slot(s) scheduled for thetransmission of each source inside the cell – to receivethe information of transmission request for the nextframe, included in the header of a packet transmittedby the source node scheduled for transmission;

3) Contention slots – to receive transmission requests fromnodes inside the cell that want to initiate a new trans-mission.

Each coordinator has the location information of all nodesinside the cell and the nodes whose transmission/reception isadvertised by adjacent coordinators. A coordinator maintainstwo tables:

1) Demand table, which contains the transmission requestsof source nodes (i.e., source ID, destination ID, andthe number of packets ready for transmission), andis updated/generated based on the nodes’ transmissionrequests in previous frames and scheduling packets ofadjacent coordinators;

2) Scheduling table, which contains the information ofscheduled transmisssions/receptions (and correspond-ingly the reserved space for each scheduled transmis-sion/reception) for the current frame, and is updatedbased on scheduling packets of coordinator and schedul-ing packets broadcasted by adjacent coordinators.

Based on the demand table and scheduling table, each coordi-nator transmits a scheduling packet at its assigned schedulingtime slot in each frame. The scheduling packet contains thefollowing information:

1) the schedule of transmissions/receptions (scheduled bythe coordinator and/or adjacent coordinators) withindistance ra of the coordinator, where ra ∈ [rg, 2rg];

6

ra

C6

C5

C4

C3

C2

C1

rg ra

C0

(a) ra = rg

C5

C4

C3

C2

C0

C6

C1

rgra

(b) ra = 1.5rg

C5

C4

C3

C2

C1

C0

C6

rg

ra

(c) ra = 2rg

Fig. 4. The area centred at coordinator C0 in which the coordinator obtains the information of scheduled transmissions/receptions by overhearing schedulingpackets of adjacent coordinators, where a circular area centred at each coordinator denotes the area that the coordinator broadcasts the information of scheduledtransmissions/receptions.

2) cancelation of scheduled transmissions/receptions by ad-jacent coordinators within distance ra of the coordinatorthat interfere with transmissions/receptions scheduled byother adjacent coordinators;

3) announcement of the contention slots and contentionwindow size for the current frame.

Figure 4 shows the area centred at a coordinator inwhich the coordinator obtains the information of scheduledtransmissions/receptions by overhearing scheduling packets ofadjacent coordinators. A coordinator will schedule a trans-mission from a source to a destination in a contention-free time slot only when neither an interfering node to thesource is scheduled for reception nor an interfering nodeto the destination is scheduled for transmission. Also, eachcoordinator will cancel scheduled transmissions/receptions byadjacent coordinators within range ra that interfere with otherexisting scheduled transmissions/receptions. This mechanismensures that a scheduled link for transmission/reception by acoordinator does not interfere with transmissions/receptionsof nodes within range ra of the coordinator or adjacentcoordinators. In Figure 1, a scheduled link for transmissionby coordinator C0 does not interfere with any other scheduledtransmission/reception in area A0∪A1∪ ....A6, where Ai, i ∈{0, 1, ...6} denote the area within range ra from coordinatorCi. To illustrate, consider frame n where scheduling time slotsare assigned as in Figure 3(b) and Ci ∈ Gi, i ∈ {0, 1, 2, ..., 6}.A transmission/reception scheduled by coordinator C0 will notinterfere with any scheduled transmission/reception in areaA4, A5, A6, and A0, because coordinator C0 receives thescheduling packets of C4, C5, and C6 before transmittingits scheduling packet and it does not schedule an interferingtransmission/reception. Also, coordinators C1, C2, and C3,which overhear the scheduled transmission/reception from co-ordinator C0 before transmitting their own scheduling packets,will not schedule an interfering transmission/reception and willcancel any interring transmission/reception scheduled by theiradjacent coordinators in area A1, A2, and A3 respectively.

When both source and destination nodes are in one cell,the cell coordinator finds time slot(s) to schedule contention-free transmission/reception and broadcast the scheduled trans-

mission/reception in its scheduling time slot of current frame.However, when the source and destination nodes are locatedin adjacent cells, the coordinator of source schedules the trans-mission/reception in the current frame only if its schedulingtime slot is before the scheduling time slot of the coordinatorof destination node. Thus, the coordinator of destination nodecan inform the destination node of the scheduled transmis-sion/reception in its scheduling time slot of the current frame.Otherwise, the coordinator of source node finds time slotsto schedule contention-free transmission/reception in the nextframe and includes the scheduled transmission/reception in itsscheduling packet for the current frame. In the next frame,both the coordinators of source and destination again broadcastthe scheduled transmission/reception in their scheduling timeslots. Consider the network as illustrated in Figure 1, wherescheduling time slots are assigned to coordinators as in Figure3(b) and Ci ∈ Gi, i ∈ {0, 1, 2, ..., 6}. Coordinator C0 canschedule transmission/reception between nodes b and c (thatare inside the cell) in each frame and inform both source anddestination in its scheduling time slot. Also, it can scheduletransmission/reception from source node c to destination noded in frame n, in which coordinator C3 can inform destinationnode d of the scheduled transmission/reception in the sameframe. However, coordinator C0 will not schedule a transmis-sion from source nodes b to destination node a in frame n, inwhich the scheduling time slot of coordinator C5 comes beforeC0. In frame n, coordinator C0 finds time slots to schedulethe transmission/reception (from source node b to destinationnode a) for frame n + 1 and includes the information in itsscheduling packet of frame n. In frame n+1, both coordinatorsC0 and C5 broadcast the scheduled transmissions/receptionsin their scheduling time slots.

Figure 5 illustrates the operations of a coordinator nodeand a non-coordinator node in each time slot. Every non-coordinator node in the network stays awake during thescheduling time slot of its cell coordinator to receive theinformation of scheduled transmssions/receptions (in thecontention-free slots) and contention slots in the current frame.A node scheduled for transmission will also stay awake duringthe scheduling time slots of the adjacent coordinators within

7

distance ra from the node to receive cancelation information oftransmission/reception (from adjacent coordinators). In Figure3(b), nodes a and b stay awake during scheduling time slot ofcoordinator C0 in every time slot. Also, node b stays awakeduring scheduling time slot of C5 only if it is scheduled fortransmission in the current frame. The source and destinationnodes wake up at the assigned contention-free slots to performtransmissions/receptions as scheduled by cell coordinators.Source nodes will also include their transmission request fornext frame in the header of one packet (as determined by cellcoordinator). The cell coordinators will use this informationto update its demand table for next frame. The source nodesthat want to initiate a new transmission wake up at theassigned contention slots and contend with each other usinga CSMA MAC scheme to send transmission request to cellcoordinators. The coordinator will also record this informationto update/generate its demand table for the next frame.

V. NUMERICAL RESULTS

Consider single-hop transmissions in a wireless ad hocnetwork with dimensions 6dm × 6dm. Nodes are randomlydistributed over the network coverage area and the destinationof each source node is randomly selected from the rest nodesin its proximity at a distance less than dm.

We compare performance of the proposed scheme with theIEEE 802.11 DCF scheme without power saving (hereafter re-ferred to as DCF) and in power saving mode (hereafter referredto as PSM). Packets are generated according to a Poissonprocess at each source node. All control/scheduling packets(including RTS, ACK, ATIM, ATIM-Back, and schedulingpackets) are transmitted at the control/scheduling channelrate (Rs) and all data packets are transmitted at the datachannel rate (Rd). The required SINR at the destination forcontrol/sceduling and data packets are Γs = 6 dB and Γd = 9dB, 17 dB respectively1. The network load is defined as theaggregate packet generation rate in all the nodes. The follow-ing metrics are used as performance measures to compare theMAC schemes:

1) Throughput, which is defined as the summation of thenumbers of packets transmitted per second from allnodes in network, weighted by the packet transmissiondistance;

2) Energy consumption, which is the average energy con-sumption per data packet, and is calculated as the ratioof total energy consumption in all nodes (including co-ordinators in our proposed scheme) to the total numberof transmitted data packets in the network;

3) Collision rate, which is the ratio of collided data packetsto the total number of transmitted data packets in thenetwork.

Similar metrics are used as performance measures in [2], [23],[24], and [28]–[31]. Each performance metric is calculatedas the average performance over 10 different random nodedistributions in the network area. In our proposed MAC

1The corresponding control/scheduling and data rates, according to datain [27] for IEEE 802.11g, are Rs = 6 Mbps and Rd = 18, 24 Mbpsrespectively.

Type of current

Time slot?

Contention

It is the

coordinator's

scheduling time

slot?

Scheduling

Data packet

header contains

request?

Data

Transmit

scheduling

packet

Yes No

Listen to scheduling

packets of adjacent

coordinators

Listen and record

transmission

requests

Stay at the

sleep mode

No

Listen and record

transmission

requests

Yes

(a) Coordinator

Type of current

Time slot?

Contention

The cell

coordinator's

scheduling time

slot?

Scheduling

Node scheduled

for transmission or

reception?

Data

Yes No

Listen to the scheduling

packet of coordinator

Transmit/receive

data packets

No

Send request packet

to cell coordinator

using CSMA/CA

Node scheduled for

data transmission?

Stay at the

sleep mode

No

Node wants to

initiate a new

transmission?

YesNo

Transmitting

coordinator is within

distance ra?

Yes

Yes No

Stay at the

sleep mode

Yes

(b) Non-coordinator

Fig. 5. The flowcharts of proposed MAC protocol in each time slot.

scheme, the network coverage area is partitioned into hexagoncells and a coordinator node is placed at the center of eachcell as in Figure 1. We set rg = hdm and ra = qrg , whereh ∈ {1, 1.5, 2} and q ∈ {1, 1.2, ..., 2}. The frame durationis 100 ms and the duration of each scheduling, contention-free, and contention slot is 1 ms. The carrier sensing rangeduring contention slots in the proposed scheme is set torc = 2rg . Since the performance of DCF in a wireless ad hocnetwork significantly depends on the carrier sensing range ofthe nodes, we vary carrier sensing range from rc = 1.8dm torc = 3.0dm. The beacon interval size of PSM is set to 100

8

TABLE IISIMULATION PARAMETERS

Parameter ValueMini-slot 20 µsSIFS 10 µsPHY preamble 192 µsRTS size 160 bitsCTS size 112 bitsACK size 112 bitsATIM size 224 bitsATIM-ACK size 112 bitsCWmin 15CWmax 1023Scheduling size for one transmission 200 bitsScheduling time slot 1msContention-free time slot 1msContention time slot 1msData packet+SIFS+ACK+DIFS duration 1msPd 100 mWPs 100− 180 mWc 0.0001c′ 3α 3.4dm 20Rd 18 Mbps, 24 MbpsRs 6 MbpsΓd 9 dB, 17 dBΓs 6 dBBeacon interval 100 msFrame duration 100 msPower consumption in sleep mode 0.075 WPower consumption in receive mode 1.15 WPower consumption in transmit mode 2.25− 3.15 W

ms [6]. The ATIM size varies from 2 ms to 10 ms, whichinclude the 4ms as specified in [6]. Simulations are performedusing MATLAB for 20 seconds of the channel time. Othersimulation parameters are given in Table II.

Figure 6 shows the throughput of IEEE 802.11 DCF MACscheme versus traffic load as the carrier sensing range changesfrom 1.8dm to 3.0dm. It is observed that the throughput ofDCF can be maximized by choosing rc = 2.0dm and rc =2.8dm when Γd = 9 dB and Γd = 17 dB respectively. Figure7 shows the performance of PSM as the ATIM size changesfrom 2 ms to 10 ms using carrier sensing range correspondingto the highest throughput of DCF in Figure 6. According toFigure 7, the optimal choice of ATIM size to maximize thethroughput depends on the network traffic load and requiredSINR at the receiver, Γd. We consider a DCF scheme and aPSM scheme whose carrier sensing range and ATIM size areadjusted for highest throughput, referred to as best-DCF andbest-PSM hereafter.

Figures 8-10 show the throughput, energy consumption andcollision rate of the proposed MAC (PMAC), best-DCF, andbest-PSM versus traffic load when Γs = 6 dB and Γd = 9dB. From Figure 8, the proposed MAC provides 20% higherthroughput than best-DCF and best-PSM. The proposed MACmechanism can achieve high throughput by opportunisticallyutilizing the spectrum in space and time domains and re-ducing signaling overhead. Reserving the required space foreach transmission/reception and sharing the information ofscheduled transmissions among adjacent coordinators facilitateefficient spatial channel reuse, while avoiding transmission

2000 4000 6000 8000 10000 120001

2

3

4

5

6x 10

4

Traffic load (p/s)

Thr

ough

put (

p·m

/s)

rc=1.8d

m

rc=2.0d

m

rc=2.2d

m

rc=2.4d

m

rc=2.6d

m

rc=2.8d

m

rc=3.0d

m

(a) Γd = 9 dB

2000 4000 6000 8000 10000 120000.5

1

1.5

2

2.5

3

3.5x 10

4

Traffic load (p/s)T

hrou

ghpu

t (p·

m/s

)

rc=1.8d

m

rc=2.0d

m

rc=2.2d

m

rc=2.4d

m

rc=2.6d

m

rc=2.8d

m

rc=3.0d

m

(b) Γd = 17 dB

Fig. 6. Throughput of the IEEE 802.11 DCF MAC vs traffic load for differentcarrier sensing ranges (N=100, Γs = 6 dB).

2000 4000 6000 8000 10000 120001

2

3

4

5

6x 10

4

Number of nodes

Thr

ough

put (

p·m

/s)

AT=2msAT=4msAT=6msAT=8msAT=10ms

(a) Γd = 9 dB

2000 4000 6000 8000 10000 120000.5

1

1.5

2

2.5

3

x 104

Number of nodes

Thr

ough

put (

p·m

/s)

AT=2msAT=4msAT=6msAT=8msAT=10ms

(b) Γd = 17 dB

Fig. 7. Throughput of the IEEE 802.11 DCF MAC in power saving mode(PSM) vs traffic load for different ATIM size when the carrier sensing rangeis set for highest throughput (N=100, Γs = 6 dB).

9

2000 4000 6000 8000 10000 120001

2

3

4

5

6

7x 10

4

Traffic load (p/s)

Thr

ough

put (

p·m

/s)

best−DCFbest−PSMPMAC, h=1, q=1PMAC, h=1, q=1.2

PMAC, h=1, q=1.4PMAC, h=1, q=1.6PMAC, h=1, q=1.8PMAC, h=1, q=2

(a)

2000 4000 6000 8000 10000 120001

2

3

4

5

6

7x 10

4

Traffic load (p/s)

Thr

ough

put (

p·m

/s)

best−DCFbest−PSMPMAC, h=1.5, q=1PMAC, h=1.5, q=1.2

PMAC, h=1.5, q=1.4PMAC, h=1.5, q=1.6PMAC, h=1.5, q=1.8PMAC, h=1.5, q=2

(b)

2000 4000 6000 8000 10000 120001

2

3

4

5

6

7x 10

4

Traffic load (p/s)

Thr

ough

put (

p·m

/s)

best−DCFbest−PSMPMAC, h=2, q=1PMAC, h=2, q=1.2

PMAC, h=2, q=1.4PMAC, h=2, q=1.6PMAC, h=2, q=1.8PMAC, h=2, q=2

(c)

Fig. 8. Throughput of the proposed MAC (PMAC), best-DCF, and best-PSM (N=100, Γs = 6 dB and Γd = 9 dB).

2000 4000 6000 8000 10000 120000

0.02

0.04

0.06

0.08

0.1

0.12

0.14

Traffic load (p/s)

Ene

rgy

cons

umpt

ion

(J/p

)

best−DCFbest−PSMPMAC, h=1, q=1PMAC, h=1, q=1.2

PMAC, h=1, q=1.4PMAC, h=1, q=1.6PMAC, h=1, q=1.8PMAC, h=1, q=2

(a)

2000 4000 6000 8000 10000 120000

0.02

0.04

0.06

0.08

0.1

0.12

0.14

Traffic load (p/s)

Ene

rgy

cons

umpt

ion

(J/p

)

best−DCFbest−PSMPMAC, h=1.5, q=1PMAC, h=1.5, q=1.2

PMAC, h=1.5, q=1.4PMAC, h=1.5, q=1.6PMAC, h=1.5, q=1.8PMAC, h=1.5, q=2

(b)

2000 4000 6000 8000 10000 120000

0.02

0.04

0.06

0.08

0.1

0.12

0.14

Traffic load (p/s)

Ene

rgy

cons

umpt

ion

(J/p

)

best−DCFbest−PSMPMAC, h=2, q=1PMAC, h=2, q=1.2

PMAC, h=2, q=1.4PMAC, h=2, q=1.6PMAC, h=2, q=1.8PMAC, h=2, q=2

(c)

Fig. 9. Energy consumption of the proposed MAC (PMAC), best-DCF, and best-PSM (N=100, Γs = 6 dB and Γd = 9 dB).

2000 4000 6000 8000 10000 1200010

−3

10−2

10−1

100

Traffic load (p/s)

Col

lisio

n ra

te

best−DCFbest−PSMPMAC, h=1, q=1PMAC, h=1, q=1.2

PMAC, h=1, q=1.4PMAC, h=1, q=1.6PMAC, h=1, q=1.8PMAC, h=1, q=2

(a)

2000 4000 6000 8000 10000 1200010

−3

10−2

10−1

100

Traffic load (p/s)

Col

lisio

n ra

te

best−DCFbest−PSMPMAC, h=1.5, q=1PMAC, h=1.5, q=1.2

PMAC, h=1.5, q=1.4PMAC, h=1.5, q=1.6PMAC, h=1.5, q=1.8PMAC, h=1.5, q=2

(b)

2000 4000 6000 8000 10000 1200010

−3

10−2

10−1

100

Traffic load (p/s)

Col

lisio

n ra

te

best−DCFbest−PSMPMAC, h=2, q=1PMAC, h=2, q=1.2

PMAC, h=2, q=1.4PMAC, h=2, q=1.6PMAC, h=2, q=1.8PMAC, h=2, q=2

(c)

Fig. 10. Collision rate of the proposed MAC (PMAC), best-DCF, and best-PSM (N=100, Γs = 6 dB and Γd = 9 dB).

collisions, which significantly improve the network through-put. In addition, a cell coordinator schedules all data trans-missions/receptions for nodes inside the cell by transmittingonly a scheduling packet in each frame. The small schedulingoverhead allows more data packet transmissions/receptions toincrease throughput.

Energy consumption per transmitted data packet is shownin Figure 9. Although the total energy consumption in eachscheme increases as the network load increases, the highestenergy consumption per packet occurs at the lowest networktraffic load. The results indicate that the proposed MAChas significantly lower energy consumption per transmitteddata packet, which is 25%-50% of the best-PSM energyconsumption. The high energy efficiency of the proposed MACscheme is the result of minimizing energy wastage becauseof node idle listening and transmission collisions, which isachieved by periodic assignment of deterministic time slotsfor transmissions/receptions. In the proposed scheme, a nodestays awake only during the scheduling time slot of cell

coordinator, in its data time slot(s) either for transmissionor reception, and when initiating a new transmission in thecontention slots. Also, energy wastage caused by transmissioncollisions is minimized by reserving space exclusively foreach scheduled data transmission/reception and sharing thescheduling information among adjacent cell coordinators.

The packet collision rate for the different protocols isdemonstrated in Figure 10. The high transmission collisionrate in the DCF and PSM MAC schemes is due to thehidden terminal problem of CSMA MAC in a wireless adhoc network. In the proposed MAC, the packet collision rateis reduced as ra and/or rg increases, which increases thearea range around a coordinator that it is aware of scheduledtransmissions/receptions. As the results indicate, the proposedMAC has a much lower packet collision rate in compassionwith the best-DCF and best-PSM. The proposed MAC schemecan effectively minimize transmission collisions by assigningcontention-free time slots for data transmissions/receptionsand reserving space around a scheduled link to prevent colli-

10

100 200 300 400 500 600 700

1

2

3

4

5

6

7

x 104

Number of nodes (N)

Thr

ough

put (

p·m

/s)

Γd=9dB

Γd=17dB

best−DCFbest−PSMPMAC, h=1.5, q=1.5

(a) Throughput

100 200 300 400 500 600 7000

0.02

0.04

0.06

0.08

0.1

0.12

0.14

Number of nodes (N)

Ene

rgy

cons

umpt

ion

(J/p

)

Γd=9dB

Γd=17dB

best−DCFbest−PSMPMAC, h=1.5, q=1.5

(b) Energy consumption

100 200 300 400 500 600 70010

−3

10−2

10−1

100

Number of nodes (N)

Col

lisio

n ra

te

Γd=9dB

Γd=17dB

best−DCFbest−PSMPMAC, h=1.5, q=1.5

(c) Collision rate

Fig. 11. Performance of the proposed MAC (PMAC), best-DCF, and best-PSM versus node density (Traffic load=8000 p/s, Γs = 6 dB and Γd = 9, 17 dB).

sions.Figure 11 shows the performance of the proposed MAC

(PMAC), best-DCF, and best-PSM in high traffic load (8000packets/s) as the node density changes and for Γd = 9 dB and17 dB respectively. The number of transmitted packets persecond decreases in each scheme as Γd increases, because alarger channel space required for each transmission/receptionin each MAC scheme to meet the higher SINR requirementat the receiver node. According to Figure 11(a), the pro-posed MAC scheme provides 30%-40% higher throughputthan best-DCF and best-PSM. Figure 11(b) shows that theenergy consumption per packet increases in each scheme asthe node density and/or Γd increases. It is observed thatthe energy consumption of the proposed MAC mechanism isabout 35%-45% of the best-PSM. Figure 11(c) shows that thetransmission collision rate in the proposed MAC scheme isalways lower than 0.02, which is about 10 times smaller thanthe transmission collision rate in the best-DCF and best-PSM.

VI. CONCLUSION

In this paper, we present a novel coordination-based MACprotocol for a wireless ad hoc network. In the proposedMAC scheme, the network area is partitioned into cellsand a coordinator node periodically schedules all transmis-sions/receptions for nodes inside its cell. For each scheduledtransmission/reception, the channel in both time and space do-mains are reserved to avoid transmission collisions. Adjacentcoordinators exchange scheduling information to maximizespatial spectrum reuse while avoiding transmission collision.A source node contends only once to transmit a batch ofpackets. After that it can request for transmission by includingthe information in the header of one data packet. Moreover,periodic scheduling of transmission time slots for data packetsallows a node to put its radio interface into the sleep modewhen not transmitting/receiving a packet in order to reduceenergy consumption. We compare the performance of theproposed scheme with the IEEE 802.11 DCF scheme withoutpower saving and in power saving mode, whose carrier sensingrange and ATIM size are dynamically adjusted to providehighest throughput. The performance measures include ag-gregate throughput, average energy consumption per packetand packet collision rate. Simulation results show that theproposed scheme achievers substantially higher throughput,significantly reduces energy consumption, and has a much

smaller packet collision rate in comparison with the exist-ing protocols. Distributing coordinators in the network areaon the basis of network environment analysis and adjustingtransmission power level of network links are further researchdirections to enhance network capacity and reduce energyconsumption.

APPENDIX

In this section, we present a mathematical model to an-alyze the number of successful transmission requests in thecontention slots and the average delay to initiate a newtransmission. Based on the analytical model, we propose amechanism to dynamically adjust the contention window sizeand the number of contention slots according to the traffic loadand required delay to initiate a new transmission.

In the contention slots, the nodes that want to initiate anew transmission contend with each other using CSMA MACto send a request packet to their cell coordinators. Eachcontending node chooses a random back-off time uniformlydistributed in the range [0,W −1], where W is the contentionwindow size that is dynamically set by coordinators. Aftereach idle mini-slot, a contending node decreases its back-off window by one and transmits its request packet whenits back-off window reaches zero. Nodes freeze their back-off window while the channel is busy and restart reducingthe back-off window when the channel is idle again. Weset the carrier sensing range, rc, large enough (comparing tothe maximum transmission range of requests, rg) such thatthe hidden node problem is avoided, in order to reduce theprobability of transmission request collisions. We also assumethat contending nodes are uniformly distributed in the networkarea. Let N ′ denote the number of contending nodes withina circular area with radius rc. Thus, when a node starts totransmit a request packet, N ′ − 1 other nodes (which arein the transmitting node’s carrier sensing range) have to staysilent until the nodes finishes the transmission of its requestpacket. Let Tcp denote the total duration of contention slots ina frame, ts denote the duration of a mini-slot, and Tr denotethe duration of a transmission request packet. Since contendingnodes choose their back-off time uniformly distributed in therange [0,W − 1], when the channel is not busy a contendingnode starts to initiate transmission request packet in a mini-slot with probability 1/W . Therefore, the probability thatX ∈ [0, N ′] nodes within a circular area with radius rc start

11

4 8 12 16 20 24 28 32 36 40 48 480

0.5

1

1.5

2

2.5

3

W

Q

N’=5N’=10N’=15N’=20

(a)

1 2 3 4 5 62

4

6

8

10

12

14

Tcp

(ms)

Q

N’=5 (W=12)N’=10 (W=24)N’=15 (W=36)N’=20 (W=48)

(b)

1 2 3 4 5 61

2

3

4

5

Tcp

(ms)

D

N’=5 (W=12)N’=10 (W=24)N’=15 (W=36)N’=20 (W=48)

(c)

Fig. 12. a) The number of successful transmission requests in 1 ms; b) The expected number of successful transmission requests in one frame (with theoptimal contention window size); c) The average delay to initiate a new transmission normalized to frame duration (with the optimal contention window size).

transmission in a mini-slot (when the channel is not busy) canbe written as

p(X = i) =

(N ′

i

)(

1

W)i(1− 1

W)N′−i. (6)

Using (6), the probability that a mini-slot is idle is

δI = p(X = 0) = (1− 1

W)N′, (7)

the probability of starting a successful transmission request ina mini-slot is

δS = p(X = 1) =N ′

W(1− 1

W)N′−1, (8)

and the probability of a transmission collision in a mini-slotis

δC = p(X ≥ 2) = 1− pi − ps. (9)

Consider a cycle as the time between two consecutive idledetection of mini-slots. The probability of initiating a trans-mission (successful or collision) after M idle mini-slots is

P (M = m) = δm−1I (1− δI). (10)

Thus, the average number of idle mini-slots in a cycle is

m =∑m≥1

mP (m) =1

1− δI=

1

1− (1− 1W )N ′

(11)

and the average duration of a cycle is

Tcy = mts + Tr =ts

1− (1− 1W )N ′

+ Tr. (12)

Since the contention window size is W and on average m idlemini-slots exits in a cycle, the expected number of cycles inthe contention slots of a frame is

u = min (W

m,TcpTcy

) = min (W1

1−(1− 1W )N′

,Tcp

ts1−(1− 1

W )N′+ Tr

).

(13)Therefore, in a circular area with radius rc, the expectednumber of successful transmission requests in the contentionslots of one frame can be written as

Q = u× δsδs + δc

= min (W1

1−(1− 1W )N′

,Tcp

ts1−(1− 1

W )N′+ Tr

)

×N ′

W (1− 1W )N

′−1

1− (1− 1W )N ′

. (14)

Figure 12(a) shows the the expected number of successfultransmission requests (during 1 ms contention time) usingdifferent contention window sizes as the number of contendingnodes varies. Figure 12(b) shows the expected number ofsuccessful transmission requests in one frame as the totalduration of contention slots increases (in each case, the con-tention window is optimized). Using (14), the probability thata contending node successfully sends a transmission requestto the coordinator in a frame is

λs =Q

N ′. (15)

Therefore, the probability that a node successfully sends itsrequest to the coordinator after contending in Y frames is

P (Y = y) = λs(1− λs)y−1 (16)

and the expected delay to initiate a new transmission is

D =∑y

yP (Y = y)Tf =Tfλs

=TfN

′

Q, (17)

where Tf is the duration of a frame. Figure 12(c) showsthe average delay to initiate a new transmission as the totalcontention slot duration increases for a different number ofcontending nodes in the carrier sensing range (the contentionwindow is optimized).

The coordinators measure δI , δS and δC by monitoringcontention slots of the most recent frame. Based on (7), (8) and(9) and the value of contention window size in the previousframe, they estimate the number of contending nodes withinthe carrier sensing range. The optimal value of contentionwindow size can be calculated using (14) for the estimatednumber of contending nodes. Also, the number of contentionslots can be adjusted for the required transmission requestdelay using (17). Accordingly, the number of contention slotsand the contention window size are dynamically updated andannounced by the coordinators.

REFERENCES

[1] “Cisco visual networking index: Global mobile data traffic forecastupdate, 2013-2018,” White Paper, Cisco, Feb. 2014.

[2] K. Rahimi Malekshan, W. Zhuang, and Y. Lostanlen, “An energyefficient MAC protocol for fully connected wireless ad hoc networks,”IEEE Trans. Wireless Communications, vol. 13, no. 10, pp. 5729–5740,Oct. 2014.

12

[3] S. Ramanathan, “A unified framework and algorithm for (T/F/C)DMAchannel assignment in wireless networks,” in Proc. IEEE INFOCOM’97,Apr. 1997.

[4] A. Ephremides and T. Truong, “Scheduling broadcasts in multihop radionetworks,” IEEE Trans. Communications, vol. 38, no. 4, pp. 456–460,Apr. 1990.

[5] J. Zhu, X. Guo, L. Yang, and W. Conner, “Leveraging spatial reuse in802.11 mesh networks with enhanced physical carrier sensing,” in Proc.IEEE ICC’04, June 2004.

[6] IEEE 802.11 WG, Part 11: Wireless LAN Medium Access Control (MAC)and Physical Layer (PHY) Specification, Standard, Aug. 1999.

[7] Z. Abichar and J. Chang, “Group-based medium access control for IEEE802.11n wireless LANs,” IEEE Tran. Mobile Computing,, vol. 12, no. 2,pp. 304–317, July 2013.

[8] W. Ye, J. Heidemann, and D. Estrin, “Medium access control withcoordinated adaptive sleeping for wireless sensor networks,” IEEE/ACMTran. Networking, vol. 12, no. 3, pp. 493–506, June 2004.

[9] T. van Dam and K. Langendoen, “An adaptive energy-efficient MACprotocol for wireless sensor networks,” in Proc. ACM SenSys’03, Nov.2003.

[10] L. B. Jiang and S.-C. Liew, “Improving throughput and fairness byreducing exposed and hidden nodes in 802.11 networks,” IEEE Trans.Mobile Computing, vol. 7, no. 1, pp. 34–49, Jan. 2008.

[11] S.-L. Tsao and C.-H. Huang, “Review: a survey of energy efficientMAC protocols for IEEE 802.11 WLAN,” ACM Trans. ComputerCommunications, vol. 34, no. 1, pp. 54–67, Jan. 2011.

[12] P. Serrano, A. d. l. Oliva, P. Patras, V. Mancuso, and A. Banchs,“Greening wireless communications: status and future directions,” ACMTrans. Computer Communications, vol. 35, no. 14, pp. 1651 – 1661,Aug. 2012.

[13] T. Pering, Y. Agarwal, R. Gupta, and R. Want, “CoolSpots: reducingthe power consumption of wireless mobile devices with multiple radiointerfaces,” in Proc. MobiSys’06, June 2006.

[14] T. Jin, G. Noubir, and B. Sheng, “WiZi-Cloud: application-transparentdual ZigBee-WiFi radios for low power Internet access,” in Proc. IEEEINFOCOM’11, Apr. 2011.

[15] Cisco, “Aironet wireless LAN adapters 350 series technical specifi-cations,” http://www.cisco.com/en/US/docs/wireless/wlan adapter/350cb20a/user/linux/configuration/guide/icglappA.html, Sept. 2013.

[16] A. Hamza, S. Khalifa, H. Hamza, and K. Elsayed, “A survey on inter-Cell interference coordination techniques in OFDMA-Based cellularnetworks,” IEEE Communications Surveys Tutorials, vol. 15, no. 4, pp.1642–1670, Nov. 2013.

[17] A. Damnjanovic, J. Montojo, Y. Wei, T. Ji, T. Luo, M. Vajapeyam,T. Yoo, O. Song, and D. Malladi, “A survey on 3GPP heterogeneousnetworks,” IEEE Tran. Wireless Communications, vol. 18, no. 3, pp.10–21, June 2011.

[18] M. Ismail, A. Gamage, W. Zhuang, X. Shen, E. Serpedin, and K. Qaraqe,“Uplink decentralized joint bandwidth and power allocation for energy-efficient operation in a heterogeneous wireless medium,” IEEE Trans.Communications, vol. 63, no. 4, pp. 1483–1495, April 2015.

[19] R. Zhang, M. Wang, L. Cai, Z. Zheng, and X. Shen, “LTE-unlicensed:the future of spectrum aggregation for cellular networks,” IEEE WirelessCommunications, vol. 22, no. 3, pp. 150–159, June 2015.

[20] F. Borgonovo, A. Capone, M. Cesana, and L. Fratta, “ADHOC MAC:New MAC architecture for ad hoc networks providing efficient andreliable point-to-point and broadcast services,” Wirel. Netw., vol. 10,no. 4, pp. 359–366, July 2004.

[21] H. A. Omar, W. Zhuang, and L. Li, “VeMAC: A TDMA-Based MACprotocol for reliable broadcast in VANETs,” Mobile Computing, IEEETransactions on, vol. 12, no. 9, pp. 1724–1736, Sept. 2013.

[22] I. Rhee, A. Warrier, M. Aia, J. Min, and M. Sichitiu, “Z-MAC: A hybridMAC for wireless sensor networks,” IEEE/ACM Trans. Networking,vol. 16, no. 3, pp. 511–524, June 2008.

[23] E.-S. Jung and N. H. Vaidya, “An energy efficient MAC protocol forwireless LANs,” in Proc. IEEE INFOCOM’02, Jan. 2002.

[24] J. Zhang, G. Zhou, C. Huang, S. Son, and J. A. Stankovic, “TMMAC:An energy efficient multi-channel MAC protocol for ad hoc networks,”in Proc. IEEE ICC’07, June 2007.

[25] Y.-C. Tseng, C.-S. Hsu, and T.-Y. Hsieh, “Power-saving protocolsfor IEEE 802.11-based multi-hop ad hoc networks,” in Proc. IEEEINFOCOM’02, June 2002.

[26] J.-R. Jiang, Y.-C. Tseng, C.-S. Hsu, and T.-H. Lai, “Quorum-basedasynchronous power-saving protocols for IEEE 802.11 ad hoc networks,”Mob. Netw. Appl., vol. 10, no. 1-2, pp. 169–181, Feb. 2005.

[27] Cisco, “The minimum SINR required to support the corresponding datarate in IEEE 802.11a/g,” http://www.cisco.com/en/US/docs/wireless/controller/7.0MR1/configuration/guide/cg mesh.html, Dec. 2013.

[28] R. Zheng and R. Kravets, “On-demand power management for ad hocnetworks,” in Proc. IEEE INFOCOM’03, March 2003.

[29] K. Rahimi Malekshan and W. Zhunag, “An energy efficient MACprotocol for fully-connected wireless networks,” in Proc. IEEE ICC’13,June 2013.

[30] E. Rozner, V. Navda, R. Ramjee, and S. Rayanchu, “NAPman: network-assisted power management for WiFi devices,” in Proc. MobiSys’10,June 2010.

[31] X. Zhang and K. G. Shin, “E-mili: energy-minimizing idle listening inwireless networks,” in Proc. MobiCom’11, Sept. 2011.

Kamal Rahimi Malekshan (S’11) is currently aPhD student at the Department of Electrical andComputer Engineering, University of Waterloo, ON,Canada. He has received M.Sc. degree in electricalengineering from the University of Tehran, Iran in2011 and B.Sc. degree in electrical engineering fromthe University of Isfahan, Iran in 2008.

His current research interest include medium ac-cess control (MAC), power management and trans-mission power control (TPC) in wireless ad hocnetworks.

Weihua Zhuang (M’93-SM’01-F’08) has been withthe Department of Electrical and Computer Engi-neering, University of Waterloo, Canada, since 1993,where she is a Professor and a Tier I Canada Re-search Chair in Wireless Communication Networks.Her current research focuses on resource allocationand QoS provisioning in wireless networks, and onsmart grid. She is a co-recipient of several best paperawards from IEEE conferences. She received theOutstanding Performance Award 4 times since 2005from the University of Waterloo and the Premiers

Research Excellence Award in 2001 from the Ontario Government. Dr. Zhuangwas the Editor-in-Chief of IEEE Transactions on Vehicular Technology (2007-2013), and the Technical Program Symposia Chair of the IEEE Globecom2011. She is a Fellow of the IEEE, a Fellow of the Canadian Academy ofEngineering (CAE), a Fellow of the Engineering Institute of Canada (EIC),and an elected member in the Board of Governors and VP Mobile Radio ofthe IEEE Vehicular Technology Society. She was an IEEE CommunicationsSociety Distinguished Lecturer (2008-2011).

Yves Lostanlen (S’M98-M’01-SM’09) received theDipl.-Ing (MSc EE) magna cum laude in 1996 fromNational Institute for Applied Sciences (INSA) inFrance. After three years of research at UniversityCollege London and INSA he accomplished a Eu-ropean PhD summa cum laude in 2000. In 2009,he obtained a ScD (Habilitation) from Universityof Rennes, France and in 2013 he graduated fromMIT Sloan School of Management, USA (ExecutiveMBA.) Dr. Yves Lostanlen is currently CEO ofSIRADEL North America and is based in Toronto,

Canada, serving many top-tier companies in the ICT, Energy, Healthcare,Broadcast & Media Industries: Government, policy makers, regulators, in-frastructure operators and equipment manufacturers. His current scientificinterests are innovative technologies (hardware, software, data analytics) andservices enabling energy-efficient infrastructure networks (telecom, energy,water) in under-developed regions in order to catalyze competitive advantage,productivity and growth. Yves Lostanlen is also an Adjunct Professor in theFaculty of Applied Science and Engineering at the University of Toronto,Canada. A senior member of IEEE, Yves Lostanlen has written more than100 papers for international conferences, periodicals, book chapters and hasbeen technical committee chairman, and session chairman at several inter-national conferences. A frequent keynote speaker at international scientificand industrial conferences, Prof Lostanlen enjoys combining academic andindustrial insights and technology and business constraints. He received a”Young Scientist” Award” for two papers at the EuroEM 2000 conference.