an adaptive quality of service channel borrowing algorithm for cellular networks

INTERNATIONAL JOURNAL OF COMMUNICATION SYSTEMSInt. J. Commun. Syst. 2003; 16:759–777 (DOI: 10.1002/dac.616)

An adaptive quality of service channel borrowing algorithmfor cellular networks

Ibrahim Habib1,n,y, Mahmoud Sherif2, Mahmoud Naghshineh3 and Parviz Kermani3

1CUNY Graduate School and Department of Electrical Engineering, The City College of New York, U.S.A.2Lucent Technologies, NJ, U.S.A.

3 IBM Research Division, T.J. Watson Research Center, Yorktown, U.S.A.

SUMMARY

A multimedia call consists of three main sub-streams (i.e. video, audio and data) each with its own distinctquality of service (QoS) requirements (e.g. packet loss rate, delay tolerance, and jitter). These requirementsconstitute a specific fixed QoS level. In contrast to static approaches, we propose an adaptive QoS platformin which each sub-stream declares a preset range of acceptable QoS levels (e.g. maximum, acceptable,minimum) instead of just a single level. This range of QoS levels is pre-defined in a user-defined profile(UDP). In this paper, we propose a channel borrowing algorithm based on such adaptive QoS platform. Inour proposed algorithm, an acceptor cell that has used all its channels can borrow from any neighbouring(donor) cell as long as this donor cell has some channels available after satisfying a minimum QoS (minQ)level defined in the UDP. A donor cell that is assigning QoS levels (to calls under its coverage) higher thanthe minQ levels will declare those channels as available for borrowing by other acceptor cells. The criteriafor choosing the free channel include not only the number of free channels, but also the QoS levels in thedonor cell. The criteria are also extended to include the effect of channel locking on the number of freechannels, and the QoS levels on the locked cells. The algorithm is not triggered unless it does not cause anycall dropping in either the donor cell, or the cells affected by call blocking. In the meantime, the callblocking rate is significantly decreased while the offered load increases. Copyright # 2003 John Wiley &Sons, Ltd.

1. INTRODUCTION

In wireless networks, admission control is required to reserve resources in advance for callsrequiring guaranteed services. In the case of a multimedia call, each of its substreams (i.e. video,audio and data) has its own distinct quality of service (QoS) requirements (e.g. cell loss rate,delay, jitter). The network attempts to deliver the required QoS by allocating an appropriateamount of resources (e.g. bandwidth). The negotiated QoS requirements constitute a certainQoS level that remains fixed during the call (static allocation approach). Accordingly, thecorresponding allocated resources also remain unchanged. In a previous paper [1], we presented

Received 5 May 2003Revised 4 June 2003

Accepted 14 July 2003Copyright # 2003 John Wiley & Sons, Ltd.

yE-mail: [email protected]

nCorrespondence to: Ibrahim Habib, CUNY Graduate School and Department of Electrical Engineering, The CityCollege of New York, U.S.A.

and analysed an adaptive allocation of resources algorithm based on genetic algorithms. In suchan adaptive networking environment, calls can be admitted to the system even if the availablebandwidth is not sufficient to satisfy their highest QoS guarantees. To accomplish this, theproposed algorithm tries to degrade the QoS levels of the existing calls in order to free somebandwidth. Each call is considered to have a pre-defined minimum QoS level (minQ) and amaximum QoS level (maxQ) that are defined in a user-defined profile (UDP). The (minQ) levelcorresponds to the minimum set of QoS requirements that the application is willing to tolerate.On the other hand, the (maxQ) level corresponds to the best QoS requirements a call can obtain,whenever bandwidth is available. In the case of a multimedia call, each of its substreams (i.e.video, audio and data) will have its own QoS level ranging from high to medium to low. Forexample, one user might be willing to tolerate a low video quality, but requires high audioquality and high speed data service. Such user will be granted an amount of bandwidth thatcorresponds to his declared minQ level. Whenever there is available bandwidth, the user mightobtain an extra amount of bandwidth up to its declared maxQ level. It should be noted that theUDP approach also defines different ‘grades of service’ at different costs that the users cansubscribe to. For example, a network provider may choose to offer two grades of services withdifferent cost structures. A subscriber to a premium service will pay high dollar for the call butwill be allocated a UDP with the uppermost range of quality levels, whereas a subscriber to aneconomy service will have a UDP range of qualities that is significantly lower than the premiumone.

Figure 1 shows that when a call is admitted to the network it presents its traffic parametersand declared UDP to the admission controller/scheduler. The admission controller/schedulerwill, then, calculate the required bandwidth according to the traffic parameters presented to it.The calculated bandwidth and the declared UDP are saved in a database in order to be used bythe proposed optimization algorithm. Figure 2 shows an example of the UDP where a call ðiÞ ispresenting both the maxQ and the minQ levels to the admission controller/scheduler. All QoSlevels between the declared maxQ and minQ are allowed to be assigned to call i: In this example,the maxQ level corresponds to the highest video quality, the highest audio quality and thehighest speed for data service. The minQ level corresponds to medium video quality, low audioquality and low speed for data service. Any QoS level between these declared levels are allowedto be allocated to this call i:

In this paper, we suggest a channel borrowing algorithm based on the adaptive allocation ofresources algorithm described above. In a channel borrowing algorithm, an acceptor cell thathas used all its nominal channels can borrow free channels from its neighbouring cells(candidate donors) to accommodate new calls [2]. In our suggested algorithm, an acceptor cell

UDP

calculated B.W.

retrievedby the

proposedalgorithm

Traffic Parameters of

Call i Admission controller/ Scheduler

B.W.information +UDP database

UDP of Call i

Figure 1. Database used by the algorithm.

Copyright # 2003 John Wiley & Sons, Ltd. Int. J. Commun. Syst. 2003; 16:759–777

I. HABIB ET AL.760

can borrow from any neighbouring (donor) cell as long as this donor cell has some channelsavailable after satisfying the minimum QoS (minQ) level defined in the (UDP). A donor cellassigning QoS levels (to calls under its coverage) higher than the minQ levels defined in the UDPwill declare those channels as available for borrowing by other acceptor cells. A channel can beborrowed by a cell if the borrowed channel does not interfere with existing calls. Furthermore,when a channel is borrowed, several other cells are prohibited from using it. This is calledchannel locking. The number of such cells depends on the cell layout and the type of initialallocation of channels to cells. In contrast to static borrowing, channel borrowing strategies dealwith short-term allocation of borrowed channels to cells. Once a call is completed, the borrowedchannels are returned to its nominal cell. The proposed channel borrowing algorithm differs inthe way a free channel is selected from a donor cell to be borrowed by an acceptor cell. In oursuggested algorithm, the criteria for choosing the free channel include not only the number offree channels but also the QoS levels in the donor cell. The criteria are also extended to includethe effect of channel locking on the number of free channels and the QoS levels on the lockedcells.

Let Cell j have N calls. The QoS levels of the N calls can expressed in the form of set½Q� ¼ ½Q1;Q2; . . . ;QN �; and the bandwidth requirement for these QoS levels can be expressed inthe form of set ½B� ¼ ½B1;B2; . . . ;BN �: The bandwidth requirements corresponding to the minQlevels can also be expressed using the set ½BM� ¼ ½BM1;BM2; . . . ;BMN �: Cell j can declare anumber of channels as available ðNavÞ for borrowing if

Nav ¼Cav �

PNi¼1 BMi

Chsizeð1Þ

where Cav is the capacity of cell j; and Chsize is the size of each channel.Throughout the description of the suggested channel borrowing algorithm, we are going to

consider the hexagonal planar layout of the cells. A cell cluster is a group of identical cells inwhich all of the available channels (frequencies) are evenly distributed. The most widely usedplan is the N ¼ 7 cell cluster [3–5] where the number of available channels are distributed evenlyamong seven cells, which then repeats itself over and over according to Figure 3. In hexagonalgeometry, this reuse plan is given by

DR¼

ffiffiffiffiffiffiffi3N

pð2Þ

where D is the reuse distance, R is the cell radii, and N is the modulus. We are going to considerN ¼ 7 throughout the discussion in this paper.

Video Audio Data maxQ High High High

High High MediumHigh High Low

…. …. ….…. …. ….

Medium Medium Low minQ Medium Low Low

Figure 2. Example of a user-defined profile (UDP).


QUALITY OF SERVICE CHANNEL BORROWING ALGORITHM 761

2. RELATED WORK

The channel borrowing algorithms can be divided into simple and hybrid [2]. In simple channelborrowing algorithms, any nominal channel in a cell can be borrowed by a neighbouring cell fortemporary use. In hybrid channel borrowing strategies, the set of channels assigned to each cellis divided into two subsets: (1) local and (2) borrowable. The local subset is used only in thenominally assigned cell, while the borrowable subset is allowed to be lent to neighbouring cells.The suggested algorithm presented in this chapter belongs to the simple channel borrowingalgorithms. A comprehensive survey of the channel borrowing algorithms is presented inReference [3]. A number of simple channel borrowing algorithms have been presented in theliterature. Borrow from the Richest [6] is a simple channel borrowing algorithm that requires theacceptor cell to borrow from the cell with the greatest number of channels available forborrowing. This algorithm does not take channel locking into account when choosing acandidate channel for borrowing. Basic Algorithm [7] is an improved version of the Borrow fromthe Richest strategy which takes channel locking into account when selecting a candidatechannel for borrowing. This algorithm tries to minimize the future call blocking probability inthe cell that is most affected by the channel borrowing. This affected cell might be the donor cellor any of the cells affected due to channel locking. The number of cells affected by channellocking depends on the cell layout and the cell reuse. Instead of trying to optimize whenborrowing, the Borrow First Available [6] algorithm selects the first candidate channel it finds.The objective is to try to minimize the complexity of the channel borrowing scheme. An EfficientBorrowing Channel Assignment (BCA) scheme is presented in Reference [8]. The BCA schemeconsists of two phases: (1) ordinary channel allocation phase, and (2) channel reallocation phaseto improve the efficiency of the scheme.

(a)

7

5

3

2

1

6

4 D

7

5

3

2

1

6

4

7

5

3

2

1

6

4 7

5

3

2

1

6

4

7

5

3

2

1

6

4

7

5

3

2

1

6

4

7

5

3

2

1

6

4

7

5

3

2

1

6

4

(b)

Figure 3. (a) Cell cluster ðN ¼ 7Þ; (b) a seven-cell reuse plan.


I. HABIB ET AL.762

3. QOS-BASED CHANNEL BORROWING ALGORITHM

As opposed to the Borrow from the richest [3,7] algorithm, the suggested algorithm takes intoaccount the effect of channel locking when choosing a candidate channel for borrowing.Furthermore, the suggested algorithm takes into account not only the number of availablechannels but also the average QoS level of each candidate cell. This allows the algorithm to tryto maximize the average QoS level of the calls existing in the system in addition to minimizingthe call blocking probability.

Figure 4 shows the block diagram of the adaptive allocation of resources algorithm when thechannel borrowing algorithm is added. As shown, the algorithm consists of three main modules:genetic algorithm modules I and II in addition to the channel borrowing module III. The reasonbehind using modules I and II is to divide the problem into smaller easier to handle sub-taskssince the search space is too large for a single GA. Module I assigns fair bandwidth allocationsto the existing calls; whereas module II maximizes the capacity utilization by assigning anyavailable bandwidth (left over from module I allocations) to the existing calls. Eventually, thesystem will not force a call to drop unless there is not enough bandwidth to satisfy its minQlevel. Module III is added as an important enhancement layer to allow for channel borrowing totake place in such an adaptive QoS environment.

Module I is triggered whenever a call arrival or departure takes place. The inputs to module Iinclude: (1) the capacity of the system (physical cell capacity), (2) the number of calls ðN Þ after

Genetic Algorithm Module I

Genetic Algorithm Module II

B.W.information

+ UDPdatabase

no. of calls (N)

Cell Capacity (B.W.)

(N) Fair allocated B.W and the

corresponding QoS

(N) Optimumallocated B.W and the

corresponding QoS levels

QoS levels fromthe previous step

Z-1

(N) Optimum allocated B.W and the

corresponding QoS levels (after channel

borrowing - if needed)

Module Chooser

Channel Borrowing Module III

Z-1

Figure 4. Block diagram of the proposed adaptive QoS platform.



the arrival/departure and (3) the bandwidth information for these calls and their preset UDPs(which are fetched from the database). Furthermore, QoS levels from the previous allocation arealso fed back to module I.

Figure 5 shows the flow chart for Module I. Module I searches for the QoS levels thatcorrespond to the bandwidth allocation closest to the ‘fair’ bandwidth. The evaluation of the

yes

yes

To Decision Module

no

no

Assign the minimum QoS levels for each call according to the user profile

If any assigned QoS < min QoS

*1 – Optimality criteria: Search for the QoS level that corresponds to a bandwidth that is the closest to the “fair” bandwidth (Cell Capacity/total number of calls), but must also be less than or equal to this “fair” bandwidth

END To Module II

Save the QoS levels found in the above step (this is set to be the minimum QoS level threshold for the next GA module)

Assign the required B.W. to this call

(maximum QoS index)

Run the GA to find the optimum B.W. allocated for this call *1

For each of the existing calls

Set all QoS indices for all current calls to the levels in the previous step

If calculatedB.W. > link

capacity

Calculate total B.W. needed by all existing calls inthis cell if all are granted the highest QoS level

Arrival/departure of a call triggers the B.W. allocation

scheme

Figure 5. Genetic algorithm module I.


I. HABIB ET AL.764

optimization function is based upon the following rules:

Bfair ¼Ctot

Nð3Þ

where Bfair is the fair bandwidth allocation, and Ctot is the total physical cell capacity.In Equation (3), the calculation of the fair bandwidth is applicable to calls of similar classes

(i.e. having the same type of multimedia substreams)For each of the existing call, the following equations are applied:

BmodIðiÞ 2 BiðQiÞ ð4Þ

BmodIðiÞ ¼ minimumðabsoluteðBfair � BiðQiÞÞÞ ð5Þ

BmodIðiÞ4Bfair ð6Þ

where i is the call number, BmodIðiÞ is the bandwidth requirement corresponding to the outputQoS level from module I, BiðQiÞ is the bandwidth requirement corresponding to the QoS levelQI; and Bfair is the fair bandwidth.

Once the fair allocations are determined by module I, if any of the assigned QoS levels inmodule I is less than the corresponding minimum QoS level minQ, then this minQ level isassigned to the corresponding call. Then this data is sent to the module chooser for furtherprocessing.

Figure 6 shows the module chooser. It starts by calculating the total bandwidth needed by allexisting calls if granted the QoS levels assigned by module I. If the total bandwidth exceeds thecell capacity, then module III is triggered to try to borrow some bandwidth from theneighbouring cell. Otherwise, module II is triggered to try to take advantage of any bandwidthleft over from module I.

Module II, on the other hand, is designed to take advantage of any available bandwidth leftover from module I. It tries to maximize the link capacity utilization and thus maximizing theQoS levels for the calls

In the following section, we are going to describe module III (channel borrowing algorithm)in more details.

4. MODULE III CHANNEL BORROWING ALGORITHM

Throughout the description of module III, we are going to consider the cell layout shown inFigure 7. The cell layout shown is a seven-cell cluster surrounded by a second tier of 11 cells.The total number of cells in this two-tier layout is 19 cells. The cell reuse plan is N ¼ 7: Eachcolour in Figure 7 represents a distinctive set of channel frequencies. The reason for choosingthis specific layout is that module III is going to consider the first two tiers of the neighbouringcells only. Cell A is the acceptor cell. It is surrounded by two tiers of cells. Cell A can borrowfrom any of its first tier neighbours (cell 1, cell 2; . . . ; cell 6). Furthermore, it can borrowchannels from one and only one neighbour (no multiple donor cells are allowed). Borrowing anynumber of channels from any of these six cells will cause these channels to be locked in anothertwo cells. To illustrate the concept of channel locking, let us assume that cell A is going tochoose to borrow a number of free channels from cell 2. This will cause these borrowed channelsto be locked in both cells 14 and 17. For this cell layout ðN ¼ 7Þ; the number of cells affected by



channel locking is always two cells. We are going to denote these two cells as affected cell1 andaffected cell2. In the suggested channel borrowing algorithm, each cell is required to keep trackof the following parameters: (1) The number of available free channels if the existing calls areassigned their minQ levels, (2) the number of available free channels in affected cell1 if theexisting calls are assigned their minQ levels and (3) the number of available free channels inaffected cell2 if the existing calls are assigned their minQ levels. The information regarding theavailable number of channels in the affected cell can be collected either periodically or on a needbasis (a message sent from the cell to the affected cell requesting immediate informationregarding the number of available free channels).

Assuming that Cell A (the acceptor cell) needs Nb number of channels to borrow. The numberof channels needed ðNbÞ should be calculated from the following equation:

Nb ¼ ðBtot � CtotÞ=Chsize ð7Þ

where Btot is the total bandwidth needed by existing calls if assigned their minQ level, Ctot thetotal physical cell capacity, Chsize the size of each channel.

yes

End Decision

no Go to Module II (still some bandwidth

available

Go to Module III (channel borrowing)

If calculatedB.W. > link

capacity

Calculate total B.W. needed by all existing calls inthis cell if all are granted the QoS levels assigned in

Module I

Start Decision Module

Figure 6. Module chooser.


I. HABIB ET AL.766

Cell A will then send a borrow request message to each of its first tier neighbour cells. Each ofthe neighbouring cells receiving the borrow request message will calculate the number ofavailable channels ðNavÞ if its calls are granted their minQ level and compare it to the requestednumber of channels ðNbÞ: If Nav5Nb; then the cell will send back a negative acknowledge(NACK) message informing the acceptor of the denial of its request. This is shown in Figure8(a). Otherwise, it will send a message to each of the affected cells (due to channel locking)requesting information regarding the number of available channels (Nav(affected cell)). Theinformation regarding the available channels does not have to be gathered when theborrow request message is received. As mentioned before, this information can be gatheredperiodically and kept ready for any borrow request message. If the returned valuesNav(affected cell)> Nb then this means that the requested number of channels is available forborrowing and that the locked channels (in the affected cells) are also available. An

Cell 13

Cell 8

Cell 7

Cell 15

Cell 14

Cell 9

Cell 18

Cell 16

Cell 17

Cell 12

Cell 10

Cell 11

Cell 1

Cell 6

Cell 5

Cell 4Cell 4

Cell 3Cell 3

Cell 2Cell 2

Cell A

Figure 7. Two tier ðN ¼ 7Þ cell layout.



acknowledgement (ACK) message is then sent back to the acceptor cell. Otherwise, a NACKmessage is sent back to the acceptor cell denoting that the locked channels will cause some callsto be dropped in the affected cell.

Figure 8(b) shows the flow of the messages if there are enough available channels in theneighbouring (candidate) cell ðNav > NbÞ: If the affected cells return an ACK message, thisimplies that the locked channels due to the borrowing process will not cause any calls to bedropped in these cells. This is due to the fact that when the candidate cell sends acheck no channel message to any of the affected cells, it will calculate the number of availablechannels after assigning the minQ levels to the calls existing in this cell. Therefore, immediatelyafterwards, the candidate cell send an ACK message to the acceptor cell informing it that it canfulfill the borrow request message.

The candidate cell will then wait for each of the affected cells to send a message carryinginformation regarding the average QoS (average qos) levels of the existing calls and the numberof channels that would be available if the borrowing process is executed ðNavÞ: The average qosis calculated assuming the borrowing process has been executed. This will give an indication ofhow the borrowing process affects the QoS levels of the existing calls. In the meantime, thecandidate cell will calculate the average QoS levels of its existing calls based on the same criteria.Notice that the candidate cell can calculate the average QoS level while the other affected cellsare doing the same thing in parallel. Once all the information is available for the candidate cell,it will send it in a message to the acceptor cell.

Once all messages from all the neighbouring (candidate) cells are received, the acceptor cellwill start processing the information searching for the best candidate cell to become the donor.First, all cells sending back NACK messages are removed from the candidate list. Theinformation of each of the remaining cells in the candidate list include: (1) a vector of average

acceptor candidate acceptor candidate Affected_cell

Average_qos[],Nav[]

Average_qos,Nav

ACK/NACK

ACK/NACK

Check_no_channel

Borrow_request

NACK

Borrow_request

(a) (b)

Figure 8. Flow of messages between acceptor, candidate and affected cells. (a) Nav5Nb; (b) Nav > Nb:


I. HABIB ET AL.768

QoS levels of the candidate cell, affected cell 1 and affected cell 2 (average qos[]), and (2) a vectorof available number of channels ðNavÞ:

The acceptor cell will use the following equation to calculate the cost of borrowing therequested number of channels from each cell. The cost value of each cell in the current candidatelist is denoted by borrow cost.

borrow cost ¼minðNavÞ

max channels�

minðaverage qosÞ64

� �ð8Þ

where (max channels) is the maximum number of channels in each of the cells. Note that thevalue of average qos ranges from 1 to 64 (1 being the best QoS level, and 64 being the least).Therefore, the value of borrow cost ranges from �1 to 1. The higher the borrow cost value, thebetter is the candidate cell. Therefore, the acceptor cell will choose the one with the highestborrow cost value.

5. SIMULATION RESULTS

The system under study is shown in Figure 9. The system consists of an acceptor cell (cell A)surrounded by two tiers of cells. The cell layout is a seven-cell cluster system ðN ¼ 7Þ: Each cellhas a physical capacity of 25 Mbps. For a 30 Kbps channels, each cell is assigned 848 channels

Cell 13

Cell 7

Cell A

Cell 2Cell 2

Cell 3Cell 3

Cell 4Cell 4

Cell 5

Cell 6

Cell 1

Cell 11

Cell 10

Cell 12

Cell 17

Cell 16

Cell 18

Cell 9

Cell 14

Cell 15

Cell 8

Figure 9. System under study.



ðmax no channels ¼ 848Þ: The dotted line represents a street with high call traffic load. Alongthe street are cells 15, 5, A, 2 and 9. In our simulations, the traffic load in these cells ischaracterized by a high traffic load. As shown in the cell layout, each of the following six setsrepresent a group of cells assigned the same set of channel frequencies: (1) cells f1; 12; 15g; (2)cells f2; 14; 17g; (3) cells f3; 7; 16g; (4) cells f4; 9; 18g; (5) cells f5; 8; 11g and (6) cells f6; 10; 13g:Therefore, the no of affected cells ¼ 2 (i.e. when one cell is a donor cell, 2 other cells areaffected due to channel locking).

The traffic load is multimedia traffic. In our simulation, the video component of themultimedia call was obtained using the MPEG-1 coded movie Star Wars as an example of highactivity video traffic. Voice and Data substreams were generated according to an exponentialdistribution. The MPEG-1 coded movie generates a frame sequence of I B B P B B P B B P B B.There are 24 frames per second [9]. This set of data is described in Reference [10]. Each call hasan average duration of 5 min of movie time. Therefore the movie was actually divided into 24different calls. To generate three QoS levels for the video substream, we adopted the algorithmdescribed in Section 2 where the B frames are dropped to obtain a lesser QoS level, then both theB and P frames are dropped to obtain the least video quality. Whereas, the high, medium andlow QoS levels for audio were generated by varying the average bit rate of the audio source from32 to 16 to 8 Kbps, respectively. Similarly, QoS levels for data were generated by varying thedata rate from 64 to 32 to 16 Kbps.

The calls statistics were as follows: (a) 33% of the calls had a minQ level equals to 1, and (b)the rest of the calls had a minQ level equals to 18. The initial distribution of the calls is shown inFigure 10 (the number between brackets represent the number of calls).

Assuming that a 9th call needs to be admitted to cell A and that there is not enough numberof free channels to satisfy the minQ level of such a call. This will trigger module III to try toborrow a number of channels from any of the six neighbouring cells (cells 1; 2; . . . ; 6). Table Ishows all the information gathered by the simulator after running the algorithm includingmodule III. The information gathered include: (1) the average QoS level of each candidate cell,affected cell 1, and affected cell 2 (average qos(1), average qos(2), average qos(3)), (2) thenumber of available channels of each candidate cell, affected cell 1 and affected cell 2 (Navð1Þ;Navð2Þ; Navð3Þ) and (3) the borrow cost value of each candidate cell.

As shown in Table I, two of the candidate cells returned a NACK message. Cell 2 returned aNACK because there are no channels available ðNavð1Þ ¼ 0Þ in the candidate cell itself. On theother hand, cell 4 returned a NACK because one of the affected cell (cell 9) does not anychannels available ðNavð2Þ ¼ 0Þ and channel locking would cause one or more calls to bedropped. Therefore, the borrow cost value is calculated for only four cells. Cell 6 has the highercost value and therefore was chosen to be the donor cell.

Table II shows the same information for a higher traffic call load. As shown, the cell with thehighest cost value was cell 6. The main advantage of any channel borrowing algorithm is toreduce the call blocking rate. To study the effect of the suggested channel borrowing algorithmon the system under study, we had to calculate the call blocking rate at different values of trafficload. Furthermore, our suggested algorithm not only tries to reduce the call blocking rate, butalso enhances the QoS offered to the existing calls. This is mainly due to the existence of a presetrange of acceptable QoS levels instead of one. Therefore, the algorithm will always try to takeadvantage of any free channels available to increase the offered QoS level. Figure 11 shows thepercentage call blocking rate of the system under study versus the percentage offered load. Usingthe suggested channel borrowing algorithm reduced the call blocking rate of the system. This


I. HABIB ET AL.770

takes place under an offered load between 40% and 80%. Below 40%, there is no need for usingthe channel borrowing algorithm as there are enough channel in the system to satisfy at least theminQ levels of the existing calls.

Above 78%, the acceptor cells cannot borrow any cells due to the fact that the algorithm doesnot allow the borrowing process to take place if the locked channels will cause any current callto be dropped. Therefore, the performance of the channel borrowing algorithm for an offeredload above 78% is exactly the same as without using the algorithm. It is also shown that whenusing the algorithm and up to an offered load of 53%, the call blocking is kept at 0%.

Figure 12 shows the reduction in call blocking versus the offered load. This figure shows thatthe advantage of using the suggested algorithm exists when the offered load is in the range from40% to 78%. The reduction in call blocking reaches its maximum at 53% offered load and is ofa value of 3.7% reduction. This reduction is kept flat in the offered load range from 53% to72%.

Figure 13 shows the increase in offered load for the same call blocking rates. This curve is ofsignificant importance as it shows how much enhancement can be achieved if the system isdesigned to operate at a certain call blocking rate. It is shown that for a call blocking of 2.7%,there is a 10% enhancement in the offered load. The enhancement keeps increasing till it reachesa peak value of 14.5% at 6.4% blocking rate. Above 6.4% blocking rate, the enhancement startsto decrease till it reaches 0% at 10% blocking rate. Again this is due to the channel blockingaffect as the traffic load increases.

Cell A (8)

Cell 4

Cell 5 (7)

Cell 6 (4)

Cell 1 (5)

Cell 7 (2)

Cell 11 (2)

Cell 10 (4)

Cell 12 (1)

Cell 17 (2)

Cell 16 (5)

Cell 18 (2)

Cell 9 (8)

Cell 14 (4)

Cell 15 (7)

Cell 8 (4)

Cell 13 (1)

Cell 4 (3)

Cell 3Cell 3 (3)

Cell 2Cell 2 (7)

Figure 10. Initial call distribution.



TableI.

Module

IIIsimulatorinform

ation(initialcalldistribution).

CellA

Cell1

Cell2

Cell3

Cell4

Cell5

Cell6

No.ofcalls

95

73

37

4CallsIds

1,2,3,4,5,6,7,8,9

8,9,10,11,12

13,14,15,16,17,18,19

19,20,21

22,23,24

1,2,3,4,5,6,7

8,9,10,11

MinQ

1,1,1,18,18,18,18,18,18

1,18,18,18,18

1,1,18,18,18,18,18

1,18,18

1,18,18

1,1,18,18,18,18,18

1,18,18,18

Nav(1)

997

471

0502

426

178

515

ACK

Needed

bw¼

10478=149

YN

YN

YY

Averageqos(1)

12.4

13.4

13.2

12.4

12.4

13.2

13.8

AffectedCell1

12(1

call)

14(4

calls)

7(2

calls)

9(8

calls)

8(4

calls)

10(4

calls)

Nav(2)

645

405

0515

515

Averageqos(2)

1.0

1.0

13.8

13.8

AffectedCell2

15(7

calls)

17(2

calls)

16(5

calls)

18(2

calls)

11(2

calls)

13(1

call)

Nav(3)

178

471

405

645

Averageqos(3)

13.2

13.4

1.0

1.0

Borrow

cost

0:531*10�3

NACK

0.2682

NACK

�5:719*10�3

0.3917

Cap¼

60000cell=s

¼848channelsðchannel¼

30KbpsÞ:Cell6ischosenasthedonorcell(highestborrow

cost

value).

TableII.Module

IIIsimulatorinform

ation(higher

trafficload).

CellA

Cell1

Cell2

Cell3

Cell4

Cell5

Cell6

No.ofcalls

96

84

48

5CallsIds

1,2,3,4,5,6,7,8,9

8,9,10,11,12,13

1,2,3,4,5,6,7,8

8,9,10,11

8,9,10,11

1,2,3,4,5,6,7

8,9,10,11,12

MinQ

1,1,1,18,18,18,18,18,18

1,1,18,18,18,18

1,1,18,18,18,18,18,18

1,18,18,18

1,18,18,18

1,1,18,18,18,18,18

1,18,18,18,18

Nav(1)

70478/997

0/0

0/0

36503/515

36503/515

0/0

33357/471

ACK

Needed

bw¼

10478=149

NN

YN

NY

Averageqos(1)

12.4

13.8

13.8

13.4

AffectedCell1

12(1

call)

14(4

calls)

7(2

calls)

9(8

calls)

8(4

calls)

10(4

calls)

Nav(2)

45707/645

28698/405

0/0

36503/515

36503/515

Averageqos(2)

1.0

1.0

13.8

13.8

AffectedCell2

15(7

calls)

17(2

calls)

16(5

calls)

18(2

calls)

11(2

calls)

13(1

call)

Nav(3)

12623/178

33357/471

28698/405

45707/645

Averageqos(3)

13.2

13.4

1.0

1.0

Borrow

cost

NACK

NACK

0.2620

NACK

NACK

0.3398

Cap¼

60000cell=s

¼848channelsðchannel¼

30Kbps).Cell6ischosenasthedonorcell(highestborrow

cost

value).


I. HABIB ET AL.772

Figure 14 shows the enhancement in the QoS offered versus the offered load. This curve showsanother contribution of the suggested algorithm. Due to the adaptive QoS platform of thesuggested system, the suggested channel borrowing algorithm is also able to enhance the QoS

call blocking versus offered load

0

2

4

6

8

10

12

0 10 20 30 40 50 60 70 80 90

offered load

call

blo

ckin

g

call blocking (without)

call blocking (channel borrowing)

Figure 11. Call blocking versus offered load.

% reduction in call blocking

0

0.5

1

1.5

2

2.5

3

3.5

4

0 10 20 30 40 50 60 70 80 9

offered load

% r

edu

ctio

n in

cal

l blo

ckin

g

0

Figure 12. Reduction in call blocking versus offered load.



level of existing calls. The enhancement starts when the offered load is at 40%. It reaches a peakvalue of 7.8% at 45% offered load. It then starts to decrease reaching 0% at 78% offered load.

Figure 15 shows a sample of the simulator output for 1:5 h simulation time. The figure showsthe enhancement in the percentage QoS level versus time. It is clear that most of the time, the

% increase in offered load

0

2

4

6

8

10

12

14

16

0 2 4 6 8 10

call blocking

incr

ease

in o

ffer

ed lo

ad

12

Figure 13. Increase in offered load versus call blocking.

Enhancement in QoS

0

1

2

3

4

5

6

7

8

9

0 10 20 30 40 50 60 70 80 9

offered load

% Q

oS

en

han

cem

ent

0

Figure 14. QoS enhancement versus offered load.


I. HABIB ET AL.774

channel borrowing algorithm makes a difference and enhances the QoS level offered to the calls.This is due to the fact that statistically, most of the time the system is operated in the rangebetween 50 and 80% of the offered load and this is when the suggested algorithm is of great useand advantage. The percentage enhancement varies between 7% and 1%. The averagepercentage of QoS enhancement is 3.82%.

6. CONCLUSIONS

In this paper we presented an adaptive algorithm for channel borrowing in wireless cellularnetworks. The algorithm utilizes Genetic Algorithms to optimize the network resources whilemaintaining the quality of service requirements of the users. This is achieved by (1) adopting adynamic range of QoS levels that are acceptable to the user, (2) increase the efficiency of thenetwork by allocating to the users an amount of bandwidth that is just enough to satisfy theusers’ minimum QoS requirements, (3) adaptively borrow the channels available fromneighbouring cells where capacity is available to cells where capacity is needed to meet trafficdemands. The algorithm is adaptive in the sense that it re-allocates the channels allocated todifferent cells based upon the dynamic traffic pattern demands from the users. Hence, in hotcells, channels are borrowed to meet increasing demands for capacity and decrease callsblocking probabilities. The channel borrowing algorithm is essentially a multi-objectiveoptimization function that tries to re-allocate channels to cells in order to minimize the callblocking probability, while in the mean time maintaining the QoS requirements of each callabove a minimum contracted level. The simulation results show several advantages of theproposed channel borrowing algorithm: (1) the network efficiency is increased as more calls canbe accepted for the same amount of resources at a certain call blocking probability, (2) the

% QoS Enahncement

0

1

2

3

4

5

6

7

8

0.00

7.00

14.0

021

.00

28.0

035

.00

42.0

049

.00

56.0

063

.00

70.0

077

.00

84.0

0

time (min)

% Q

oS

Figure 15. QoS enhancement versus time.



quality of service provided by the network is enhanced as the call blocking probability isdecreased for the same offered traffic loads, and finally (3) the QoS levels provided to theaccepted calls are also enhanced.

REFERENCES

1. Sherif MR, Habib IW, Naghshineh M, Kermani P. Adaptive allocation of resources and call admission control forwireless ATM using genetic algorithms. IEEE Journal on Selected Areas in Communications (JSAC) special issue onneurocomputing application on wireless communications, February, 2000; 18(2).

2. Katzela I, Naghshineh M. Channel assignment schemes for cellular mobile telecommunication systems: acomprehensive survey. IEEE Personal Communications, June 1996.

3. Farouque S. Cellular Mobile Systems Engineering. Artech House: Boston, 1996.4. Holtzman J, Goodman D. Wireless Communications, Future Directions. Kluwer Academic Publishers: Dordrecht,

1993.5. Rappaport F. Wireless Personal Communications. Kluwer Academic Publishers: Dordrecht, 1993.6. Anderson L. A simulation study of sonle dynamic channel assignment algorithms in high capacity mobile

telecommunications system. IEEE Transactions on Vehicular Technology 1973; VT-22:210.7. Engel J, Peritsky M. Statistically optimum dynamic server assignment in systems with interfering servers. IEEE

Transactions on Vehicular Technology 1973; VT-22:203.8. Chang K, Kim J, Yim C, Kim S. An efficient borrowing channel assignment scheme for cellular mobile systems.

IEEE Transactions on Vehicular Technology 1998; 47:602–608.9. Garret M, Fernandez A. Variable bit rate video bandwidth trace using MPEG code. Bellcore, ftp://ftp.bellcore.com/

pub/vbr.video.trace/, 1992.10. Garret M. Contributions towards real-time services on packet networks. Ph.D. Dissertation, Columbia University,

May 1993.

AUTHORS’ BIOGRAPHIES

Dr Mahmoud Naghshineh is Director of emerging markets at IBM TechnologyGroup. He is responsible for technology roadmap in the embedded space anddefining software, services and solutions in support of Technology Group’s offering.Prior to his current position, he was a Senior Manager at the IBM Thomas J.Watson Research Center, Yorktown Heights, New York, where he managed thePervasive Computing Infrastructure Department and had world wide responsibilityfor IBM research’s projects in Pervasive Computing and directs the work of morethan 50 researchers at the TJ Watson Research Center. He manages projects insoftware and services infrastructure pervasive computing, mobile and wirelessInternet, embedded software, secure platforms, network processor technologies,telecommunications services, and web-based payment infrastructure. He joined IBMin 1988. Prior to his current position, he has worked on communication and

networking protocols, fast packet-switched/broadband IP networks, wireless and mobile networkingoptical networking, QoS provisioning, call admission, routing, and resource allocation, and networksecurity. He has had several main technical contributions to IBM products in areas of networkingtechnologies and software. He has contributed to IETF mobility protocols, IrDA, Bluetooth and IEEE802.11 and .15 standards.He received his doctoral degree form Columbia University, New York, in 1994. He is a Fellow member

of the IEEE. Aside from IBM, he has been very active in the external research. He is currently the Editor-in-Chief of IEEE Wireless Communications Magazine a leading magazine in mobile and wireless Internet.He has served as Program co-chair of MobiCom 2001 and as a technical editorial board member of manywireless and mobile networking/computing journals, as a member of technical program committee, sessionorganizer and chairperson for many IEEE/ACM, NSF and Government conferences and workshops.Currently, he is an adjunct faculty member of the department of electrical engineering at ColumbiaUniversity teaching a graduate course on wireless and mobile networking. He has published over 100technical papers and holds a number of IBM awards and patents.


I. HABIB ET AL.776

Parviz Kermani holds a PhD degree in Computer Science from UCLA. Since 1978,he has been with IBM at T.J. Watson Research Center in Yorktown Heights, NY.While at UCLA he participated in the ARPA network project, which later led tothe creation of the Internet. At UCLA he did research in the design and evaluationof switching and flow control techniques in computer communication. Hispioneering work on a new switching techniques, Cut-Though switching, later wasused in many innovative switching and networking architectures. Since joiningIBM, he has been involved in number of diverse research projects spanning fromtheoretical works to product developments, mostly in computer communicationareas. He has made many contributions to number of IBM products for whichIBM holds patents. Dr Kermani is a senior member of IEEE and has manypublications in diverse fields of computers and communications. He has chaired

technical sessions in number of international conferences. He was the General Chair of Infocom 2002, amajor conference in computer communications. He is also an adjunct professor at the graduate center ofPolytechnic University in Westchester, New York, where he has been teaching graduate courses incomputer communication networks for the last 17 years. His current interests are in the area of digitalmedia distribution with emphasis on education and distance learning applications, mobile computing,personal system communications, Internet security, distributed and pervasive computing systems. He canbe reached at [email protected]

Ibrahim Habib received the PhD degree in 1991 from the City University of NewYork, USA, the MSc degree in 1984 from Polytechnic University of New York,USA, and the BSc degree in 1981 from Ain Shams University, Cairo, Egypt all inElectrical Engineering. From 1981 till 1983 and from 1984 till 1988, he was acomputer networks engineer involved in the planning, system engineering andinstallation of several IBM SNA networking projects in both Egypt and SaudiArabia. In 1991 he joined the Faculty of the City University of New York where isnow a tenured Associate Professor. From 1997 till 2001 he was also a consultant atthe industry; first with AT&T, and then with Telcordia Technologies. His researchinterests span different areas of traffic engineering in IP, wireless, and opticalnetworking. He has published more than 70 technical papers and reports in thoseareas. He was a Guest Editor of the IEEE Journal on Selected Areas in

Communications (JSAC) twice in 1997, and 2000; and a Guest Editor of the IEEE CommunicationsMagazine twice in 1995 and 1997. He also served as an Editor of the same Magazine from 1994 till 1997.He was Guest Editor of the John Wiley Journal on Wireless Communications and Mobile Computing in2002; he is currently an Editor of the same Journal. He is currently a Guest Editor of JSAC special issue onMetro Optical Networks to appear in 2003. He was the co-chairman of the Wireless Networking andOptical Networking Symposia at the IEEE GLOBECOM 2002; and the High Speed NetworkingSymposium at ICC 2002. He is the chairman of the IEEE Optical Network Symposium at GLOBECOM’2003; and the Wireless Networking Symposium at ICC’ 2004.He is listed in the Marquis’s who is who in the World 2001, and who is who in America 2002, and 2003

editions.



an adaptive quality of service channel borrowing algorithm for cellular networks

Documents