A Distributed Dynamic channel Allocation Scheme in cellular … · 2011. 1. 6. · in cellular communication networks (CCNs), such as the global system for mobile com-munication (GSM)

�2 Journal of Information Technology Research, 2(1), �2-�8, January-March 2009

Copyright © 2009, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Globalis prohibited.

AbSTrAcT

This article presents a description and performance evaluation of an efficient distributed dynamic channels allocation (DDCA) scheme, which can be used for channel allocation in cellular communication networks (CCNs), such as the global system for mobile com-munication (GSM). The scheme utilizes a well-known distributed artificial intelligence (DAI) algorithm, namely, the asynchronous weak-commitment (AWC) algorithm, in which a complete solution is established by extensive communication among a group of neighbouring collaborative cells forming a pattern, where each cell in the pattern uses a unique set of channels. To minimize communication overhead among cells, a token-based mechanism was introduced. The scheme achieved excellent average allocation efficiencies of over 85% for a number of simulations.

Keywords: asynchronous weak-commitment (AWC) algorithm; cellular communica-tion networks (CCNs); distributed artificial intelligence (DAI); dynamic channels allocation (DCA); GSM

INTrODUcTION

A number of techniques have been developed to combat impairments in rapidly varying radio channels and to obtain high spectral efficiencies in

CCNs. Some of those are channel coding and interleaving, adaptive modulation, transmitter-receiver antenna diversity, spectrum spreading, and dynamic chan-nel allocation (DCA) (Kostic 2002, Kostic 2001). In contrast to static chan-

A Distributed Dynamic channel Allocation Scheme in cellular

communication NetworksHussein Al-Bahadili, The Arab Academy for Banking & Financial Sciences,

JordanArafat Abu Mallouh, The Hashemite University, Jordan

Journal of Information Technology Research, 2(1), �2-�8, January-March 2009 ��

Copyright © 2009, IGI Global. Copying or distributing in print or electronic forms without written permission of IGI Global is prohibited.

nel allocation (SCA) schemes, DCA schemes provide better utilization of the channels at higher traffic loads albeit at the cost of higher acquisition time and some additional control messages. There are mainly two types of DCA schemes, these are: centralized DCA (CDCA) and distributed DCA (DDCA) (Modi 2001).

In this article, we develop and evaluate the performance of a DDCA scheme that convenes all requirements and constraints imposed by the user, service providers, and technology of CCNs. The scheme is based on the AWC algorithm, which is, in turn, based on a formalism that is widely used for various application problems in DAI, called distributed constraint satisfac-tion problem (DCSP) (Yokoo 2000, Yokoo 1998, Yokoo 1994). In order to minimize the data communication overheads of the AWC algorithm, a token-based mechanism is introduced, in which a token is circulating between a group of collaborative cells passing information about the channels that are allocated or in operation within each cell. In addition, the token controls the channels allocation process by only allowing the cell that hold the token to update its channels. The algorithm is characterized by minimizing: channel acquisition time, number of denied or failed calls, control message complex-ity, communication overheads, and network interruption.

The rest of this article is organized as follows. The following section presents a literature review that summarizes the

most recent and related work. An intro-duction to the DCA techniques and to the minimum interference constraints in CCN are given in the respective following sections. Description of the ACW algorithms is presented, followed by the description of the new proposed DDCA scheme. Definitions of param-eters that are used in evaluating the performance of the new scheme are also given. The simulations that were performed and the results obtained are described and discussed in a separate section. Finally, conclusions are drawn and recommendations for future work are pointed-out.

lITerATUre reVIeW

There are a number of techniques have been developed throughout the years to solve the channel allocation problem in CCNs, to make the system highly adaptive to traffic changes, to utilize the available spectrum efficiently, and to allocate channels optimally to the cells within the network. In this section, we review some of the most recent and related work.

Abeysundara (2005) proposed a DCA technique using intelligent agents. Under his implementation, agents only interact with the environment and the network cells. An aspect of self-or-ganization is the reliance on multiple interactions, and the ability of agents to make use of the results of their own actions and the actions of others. The latter is not very apparent in his system;

�� Journal of Information Technology Research, 2(1), �2-�8, January-March 2009


agents make very little use of the ac-tions of others.

Yang et. al. (2005) proposed an ef-ficient fault-tolerant channel allocation algorithm which achieves high channel utilization. In the proposed algorithm, a cell may borrow a channel even based on some partial channel usage information it receives from some of its neighbours. Moreover, a cell can lend a channel to multiple borrowers (at most three) as long as any two of them are not neighbours.

Zhang et. al. (2005) formalized the distributed scheduling problems in distributed wireless sensor networks, in which computation and communi-cation resources are scarce, as DCSPs and distributed constraint optimization problems (DCOPs) and model them as distributed graph coloring. They found that to cope with limited resources and restricted real-time requirement, it is imperative to use distributed algorithms that have low overhead on resource consumption and high-quality anytime performance.

In order to meet these requirements, Zhang et. al. studied two existing DCSP algorithms, distributed stochastic search algorithm (DSA) and distributed breakout algorithm (DBA), for solving DCOPs and the distributed scheduling problems. Their results showed that DSA is superior to DBA when controlled properly, having better or competitive solution quality and significantly lower communication cost than DBA.

Bejar et. al. (2005) reported an experimental study of the average-case

computational complexity of two early algorithms, ABT and AWC search on an application in distributed sensor net-works. They also showed that random effects, both intentional such as random value selection and unintentional such as random delays, have a significant effect on the performance of the algorithms. Finally, they pointed-out that there are big performance differences between solvable and unsolvable instances.

Kostic et. al (2001) examined tech-niques for increasing spectral efficiency of CCNs by using slow frequency hopping (FH) with dynamic frequency-hop (DFH) pattern adaptation. Their analysis and simulations considered the effects of path loss, shadowing, Ray-leigh fading, co-channel interference, coherence bandwidth, voice activity, and occupancy. The results indicated that systems using DFH can support substantially more users than systems using RFH.

Salmenkaita et. al. (2001) presented a practical DCA scheme for cellular GSM networks that is called a dynamic frequency and channel assignment (DFCA) scheme. They showed that the behaviour of DFCA was satisfactory in very high load situations where the gain remains stable or is even increasing as happened with 50 km/hr mobile speed. In their model, a BS is capable of base band frequency hopping, and can only utilize part of the DFCA frequency band, therefore limiting the freedom of radio channel selection.

Also Salmenkaita et. al. proved that in a typical case the gain of DFCA was



reduced in a linear manner as the share of base band hopping BSs increases. Also DFCA can also be used to provide radio channel differentiation based on connection type, and this can be utilized to maximize the network performance when the frequency reuse does not have to be dimensioned from the worst case point of view.

Al-Agha (2000) proposed a multi-agent solution for intelligent BSs in wireless networks to verify its feasibility as a main target. He found agents are able to combine knowledge and experi-ence with neighbouring agents to make the best decisions, also he demonstrated that, the intelligent agent approach to introduce the self-adaptive resource allocation feature in mobile networks remains an attractive and formal way of integrating intelligence in BSs.

Yubin et. al. (2000) discussed the problem of DCA in CCNs. They de-veloped a simple but useful method to calculate the lower limit of call blocking probability of DCA, this method could be used to compare the performance of SCA with any kinds of DCA schemes easily and clearly. They found that the lower limit of blocking probability of DCA is related to the cluster size N, the lower limit of blocking probability of DCA will decrease if N increases. For example, they proved that in GSM system the application of DCA strategy would greatly improve the overall sys-tem capacity, while in CDMA systems in which N is less than 3 the improvement by DCA is not so obvious.

Prakash et. al. (1999) presented an efficient DDCA algorithm, which dis-tributes the responsibility for channel allocation among the mobile service stations of the network, and it keeps the involvement of mobile hosts in chan-nel selection to a minimum, thereby conserving the limited energy at their disposal. Also the algorithm keeps the number of handoffs to a minimum as it does not induce any intra cell hand-offs.

The algorithm by Prakash et. al. is dynamic and can be easily adjusted to meet changes in the network load dis-tribution by transferring allocated chan-nels from lightly loaded cells to highly loaded cells. The distributed nature of the algorithm and the symmetry of the channel allocation procedure across the entire network make the system scal-able. The simplicity of the algorithm makes it easy to implement on a real-life network. The algorithm withstands fail-ures of mobile hosts and mobile service stations without significant degradation in performance. Even when the channel demand is high, a very small fraction of channel requests are dropped.

DcA TechNIQUeS

At it has been shown in the previous section that many channel allocation techniques have been proposed during the last two decades for infrastructure wireless networks to avoid channel interference, utilize efficiently the lim-ited frequencies (bandwidth) available,



provide an adequate QoS for the mo-bile users, reduced the rate of dropped calls, etc.

There other desirable characteristics that satisfactory channel allocation tech-niques should have, such as: minimum connection set-up time, adaptable to changing load distribution, fault toler-ance, scalability, low computation and communication overhead, minimum handoffs, maximum number of calls that can be accepted concurrently, etc.

One of the most widely used chan-nels allocation approach is the DCA. In which no set relationship exists between channels and cells. Instead, channels are part of a pool of resources. Whenever a channel is needed by a cell, the channel is allocated under the constraint that frequency reuse requirements can not be violated.

There are two problems that typi-cally occur with DCA based systems, these are (Zhang 1989):

• DCA methods typically have a degree of randomness associated with them, and this leads to the fact that frequency reuse is often not maximized, unlike the case for SCA systems in which cells using the same channel are separated by the minimum reuse distance.

• DCA methods often involve com-plex algorithms for deciding which available channel is most efficient. These algorithms can be very computationally intensive and may require large computing resources in order to be real-time. However,

this problem has been overcome with high increase of the speed of processing systems or using distrib-uted processing.

There are two main types of DCA schemes:

• Centralized DCA (CDCA). The CDCA scheme involves a single controller selecting a channel for each cell. CDCA schemes can theoretically provide the best performance. However, the enor-mous amount of computation and communication among BSs leads to excessive system latencies and renders CDCA schemes impracti-cal.

• Distributed DCA (DDCA). The DDCA scheme involves a number of controllers scattered across the network. In DDCA, a BS commu-nicates with each other without any central control to find a channel that does not interfere with neighbouring cells.

SIGNAl INTerFereNce IN ccNS

Radio signal interference depends on various parameters, such as cell shape, size, layout, applied modulation, etc. There are two main types of channels interferences in CCNs; these are (Stall-ings 2005):



Figure 1. First and second tier cells

E

A D

G

C

D

G C

B

F B

F

D E B

F

C

G E

Second-tier cell

First-tier cell

• Co-channel interference: occurs when two sufficiently close cells use the same channel simultane-ously. To reduce this type of inter-ference, channels should be reused simultaneously in cells which are sufficiently far apart so that an acceptable level of interference is maintained.

• Adjacent-channel interference: ap-pears when a signal is deteriorated by interference caused by signals in other radio channels used in the same cell or in one or more other cells. The term "adjacent" originates from interference caused by usage of two channels adjacent in radio-spectrum in FDMA.

For classification purposes and depending on the type of interference which may occur between two cells, we distinguish two types of radio neigh-bours as shown in Figure 1, these are (Tanenbaum 2003):

• The first-order neighbours: are cells in the first-tier around cell A.

• The second-order neighbours: are cells in the second-tier around cell A.

The group of cells that forms first and second tiers around a central cell, which are shown in Figure 1, is called frequency reuse pattern. Depending on the previous definitions we assume that the usage of channel k in cell A imposes the following constraints to a channel allocation algorithm (Stallings 2005):

• Channel k cannot be allocated to other wireless terminals in cell A or in the first and second tiers of cells.

• The two adjacent channels of k can-not be allocated to other wireless terminals in cell A or in the first-tier of cells.

Considering the second constraint, that the adjacent channels of k are not allowed to be used in the first-tier of cells, reduces the total number of channels that can be allotted for each cell (f) to be calculated as: f = K/(N+2),



where K is the total number of channels allotted for the system, N is the reused factor (or number of cells within the pattern). Thus, for a GSM cellular net-work, since K = 124 and N = 7, then f is approximated to 13.

The AWc AlGOrIThM

The asynchronous weak-commitment (AWC) algorithm was proposed and de-veloped by Makoto Yokoo et. al. in 1995 for solving DCSP (Yokoo 1995). It is based upon the asynchronous backtrack-ing (ABT) algorithm and inspired by the weak-commitment (WC) algorithm (Yokoo 94). Two main keys represent the strength of this algorithm; it uses the min-conflict heuristic as a value ordering heuristic to reduce the risk of making bad decisions, and it abandons the partial solution and restarts the search process if there is no consistent value with the partial solution, which means AWC weakly commits to a par-tial solution that constructed from bad decisions.

In AWC agents asynchronously as-sign values to their variables and com-municate the values to neighbouring agents with shared binary constraints, the priority order is dynamically changed using the communicated pri-ority values. If the current value is not consistent because some constraint with variables of higher priority agents is not satisfied, the agent changes its value so that the value is consistent, and also the value minimizes the number

of constraint violations with variables of lower priority agents. But when it cannot find a consistent value, it sends nogood messages to other agents, and increments its priority value. If it has already sent an identical nogood, it will not change its priority value but will wait for the next message.

In the AWC algorithm, all variables have temporal initial values. A consis-tent partial solution is constructed for a subset of variables, and this partial solution is extended by adding variables one by one until a complete solution is found. When a variable is added to the partial solution, its tentative initial value is revised so that the new value satisfies all the constraints between the variables included in the partial solu-tion, and satisfies as many constraints as possible between variables that are not included in the partial solution, this value ordering heuristic is called the min-conflict heuristic which has a great effect on finding a solution with small number of cycles as possible (Faltings 2005).

When there exists no value for one variable that satisfies all the constraints between the variables included in the partial solution, this algorithm abandons the whole partial solution, and starts constructing a new partial solution from scratch, using the current value assign-ment as new tentative initial values. This algorithm records the abandoned partial solutions as new constraints, and avoids creating the same partial solution that has been created and abandoned before. Therefore, the completeness

Journal of Information Technology Research, 2(1), �2-�8, January-March 2009 �9


of the algorithm is guaranteed (Yokoo 1998).

The two main features of the AWC algorithm can be described as follows (Yokoo 1995):

• Min-conflict: when selecting a vari-able value, if there are multiple val-ues consistent with the agent_view, the agent prefers the value that minimizes the number of constraint violations with variables of lower priority agents.

• Weak-commitment: when an agent cannot find a value consistent with the current agent_view, it increases its priority value to be the maximum of neighbours. The mechanism of dynamically changing priority whenever a new nogood created; enables agents weakly commit to the partial solution, which means It abandons the partial solution and restarts the search process if there exists no consistent value with the partial solution. By increasing a priority value in this way, a wrong variable value of a high prior-ity agent can be revised without performing exhaustive search by lower priority agents, which is the main characteristic of the AWC algorithm.

Since agents act concurrently and asynchronously, and no agent has exact information about the partial solution, and also multiple agents may try to re-start the search process simultaneously, so that for establishing the priority order

and changing the priority values, the following rules will control the process (Yokoo 1998):

• For each agent, a nonnegative inte-ger value representing the priority order of the agent is defined; this value is called the priority value.

• The order is defined such that any agent with a larger priority value has higher priority.

• If the priority values of multiple agents are the same, the order is determined by the alphabetical order of the identifiers.

• For each agent, the initial priority value is 0.

• If there exists no consistent value for agent Xi, the priority value of Xi is changed to k +1, where k is the largest priority value of the related agents.

The AWC algorithm works as fol-lows:

• There exist N agents.• Each agent has to know only the

identifiers of an agent with which it must establish a constraint in order to direct the constraint.

• The priority value is initially zero for all agents, but it is dynamically changed during the execution, if two agents have the same priority value the alphabetical order of identifiers will determine who is greater (i.e., X1 priority is higher than X2).



• Each agent must establish a link from itself to lower priority agents if it has a constraint with.

• Each agent has a set of values from the agents that are connected by incoming links. These values constitute the agent’s agent_view which has the same definition as in the ABT algorithm.

• Agents communicate their location values and priority values with constrained agents, by sending ok messages for both lower and higher priority agents, the lower and higher priority agents are called neigh-bours.

• After that, the agents wait for and respond to messages.

• If an ok message is received by some agent by an incoming link, the evaluating agent adds the message sending agent index, priority, and value to his neighbours, and if the agent who sent the message has a higher priority, then it will be added to the evaluating agent agent_view, and then checks whether its own value assignment is consistent with its agent_view.

• The agent’s own assignment is consistent with the agent_view, if all constraints the agent evaluates are true under the value assignments described in the agent_view.

• If an agent’s own assignment is not consistent with the agent_view, the agent will try to change the current value so that it will be consistent with its agent_view and also the value minimizes the constraint vio-

lations with lower priority agents, and then send ok message to his neighbours.

• If an agent’s own assignment is not consistent with the agent_view and the agent is unable to find any consistent value, then a subset of the agent_view, which caused the problem is defined and is called a nogood, after that the agent will check if it has already sent an identi-cal nogood message to this subset, if yes it will not change its priority nor send nogood message but will wait for another messages.

• If the agent has not sent an identical nogood message to the same subset, then the agent will increase it prior-ity value to be the maximum prior-ity value of all agents connected to it by constraints plus one, then it will change the current value to another value that minimizes the constraint violations with lower priority agents, and then send ok message to his neighbours.

• If a nogood message is received by some agent by an outgoing link, the agent will check if its current value is consistent with its agent_view, depending on the result the agent will respond as described before.

• The algorithm will terminates when a solution is found or when there is an empty nogood element set is found, so there is no solution.

Journal of Information Technology Research, 2(1), �2-�8, January-March 2009 �1


The PrOPOSeD DDcA ScheMe

In order to formulate the channel alloca-tion problem in CCNs as DCSP so that can be solved using the AWC algorithm, we assume the following:

• An agent stands for a BS (cell), which is responsible for controlling all phones within the cell.

• Variables stand for the channels allocated for each cell

• Constraints stand for the legal us-age of the available channels and the legal channel reuse to eliminate any radio interferences.

There are two mechanisms that can used in implementing the ACW algorithm for DCA, these are:

• On-demand mechanism. In this mechanism, each cell collabora-tively updates its resources (allo-cate new channels as needed and as available). Cells instantly and simultaneously exchange informa-tion on the allocated frequencies as depicted by the algorithm. It is clear from the description of the ACW algorithm; it involves a huge data communications between cells and may cause a lot of interruption during the network operation. This mechanism is very much similar to the implementation suggested by M. Yokoo et. al (Yokoo 2000, Yokoo 1998, Yokoo 1995).

• Token-base mechanism. In order to practically implement the ACW algorithm efficiently, for DCA in CCNs with minimum communica-tion overheads, minimum system interruption, and not affecting the system operation, a token-based mechanism is introduced. In a to-ken-based mechanism, each pattern (a group of collaborative cells) will have a token that circulates between the cells carrying information about the channels that are currently al-located for each cell. Also, only allows a cell that holds the token to update its resources. So that the data communication is minimized and the process of resource update does not interrupt the network operation. Since, the token is continuously circulating between cells; if pos-sible, a cell updates its resources if it is equal to or less than a certain threshold value. Thus, each cell can update its resources independently and as long as the resource is avail-able. However, in this mechanism, a cell needs to have a mechanism to initiate a resource update, such as when the cell resources below a certain threshold value, otherwise it just bypasses the token to minimize delay. In this work the token-based mechanism is used.

PerFOrMANce MeASUreS

There are a number of parameters that can be used in evaluating the perfor-



mance of the new DDCA scheme, such as:

• Number of successfully allocated channels (Ca,i). It is defined as the number of channels that are suc-cessfully allocated for the ith cell during network operation, to satisfy specific values and distributions of initial and traffic loads.

• Number of failed channels (Cf,i). It is defined as the number of channels that are failed to be allocated (i.e., calls are dropped), for the ith cell during network operation, to satisfy specific values and distributions of initial and traffic loads.

• Allocation efficiency (Ei). It is defined as the number of channels that are successfully allocated (Ca,i) divided by the total number of channels that are requested during the simulation period (Ct,i), where Ct,i=Ca,i+Cf,i. Thus, Ei can be calcu-lated by:

, ,

, , ,

a i a ii

t i a i f i

C CE

C C C= =

+ (1)

Since, the scheme is usually used to compute these parameters for a number of neighbouring cells; it has been found that it is more indicative to evaluate the performance in terms of the average values of the above parameters over a network of k cells, for a certain period of operation and network environment. The average values of the channels allocated, failed, and the allocation ef-

ficiency are referred to as aC , fC , and E, respectively.

Evaluating the performance of the DDCA scheme in terms of the computed average values is very useful. This is because the algorithm may perform well and allocate all channels required by a certain cell, but it may fail to allocate enough channels for another demand-ing cell. In this article, the effects of a number of networks parameters are investigated, these include:

• Cell traffic load (α) which represents the number of calls arrives to the cell (calls/sec).

• Cell initial load (β) which represents the number of channels that are initially allocated to a cell (chan-nel/cell)

SIMUlATIONS AND reSUlTS

A number of simulations are carried-out to evaluate and analyze the performance of the DDCA scheme and consequently the AWC algorithm in providing a DCA solution in GSM CCNs, which satisfies all constraints within a rea-sonable processing time and minimum data exchange. These simulations are grouped into two different scenarios as follows:

• Scenario #1. Uniform initial and traffic loads: It simulates a GSM network in suburban or rural areas, where the traffic load is low and uniformally distributed.



• Scenario #2. Uniform initial load and nonuniform traffic load: It simu-lates a GSM network in an urban downtown area, where the traffic load is centralized in shopping or business districts.

The main difference between these two scenarios is in the traffic require-ments, while the infrastructure topology of the network is unchanged, as shown in Figure 2, and the initial load is al-ways uniform regarding its initial value (number of channels per cell). The total number of cells that are considered is 37 cells, with one central cell encircled by three tiers.

In both scenarios, the duration of calls is modelled as a random vari-able that varies between 20 and 360 sec with an average value of 180 sec approximately. The arrivals of calls in each cell are modelled as independent and distributed processes with various arrival rates assigned for the neighbour-

ing cells. For the simplicity of analysis, the hand-offs is not considered. The total simulation time is taken to be 1000 cycles, and each cycle is equivalent to 1 sec.

In all simulations, the DDCA scheme begins with a certain initial load (β), where each cell will be allo-cated a specified number of channels considered as busy channels. Thus, after initialization, if the number of incoming (arrived) calls is more than the normally concluded calls, a new channels need to be allocated. In other words, each cell requests channels for the new coming calls depending on the traffic load. It is important to realize that each channel can accommodate 8 calls, because GSM divides its channels into 8 time slots (TDMA).

Scenario #1. Uniform loads

The first scenario simulates a network with a uniform traffic load (α=0.2 calls/sec) in each cell, and uniform initial loads of β =8 and β =10 chan-nels/cell. This scenario may represent a GSM in a suburban or a rural area that characterized by a uniform initial load, and relatively low and uniform traffic load.

The results obtained for the average values ( aC , fC , and E) and their associ-ated standard deviations (σa, σf, and σe) are computed using the DDCA scheme and summarized in Table 1. It demon-strates that the scheme has an excellent performance in a suburban and rural area as it responses positively to almost

Figure 2. Topology of a cellular net-work

4

0 3

13

17

16

6 2

15

5 12

11

7 18 1

8

14

10 9

30 31 29

28 27

26

25

24

22 23

36

35

19 20

21

34

33 32



all requests. It provides an efficiency of 100% (i.e., the number of dropped calls is 0) for all cells when α=0.2 and β=8. However, when β increases to 10 channels/cell, then the number of free channels are reduced, which causes channel allocation failure and in some cells some calls are dropped.

For this scenario, when β=8 chan-nels/cell, the scheme succeeds to al-locate resources to all incoming calls, and the number of calls that are denied services is zero for all cells. Thus, the scheme achieves an average allocation efficiency of 100% as all arrived calls granted services. When β=10 channels/cell, the scheme still offers an excellent performance as it provides an average allocation efficiency of 96%. In this scenario, the results obtained show that, when β=10 channels/cell, all cells are allocated the required channels, except cells number 0, 5, 14, and 31.

Scenario #2: Non-Uniform Traffic Load

The second scenario represents a GSM network with a nonuniform traffic load.

It simulates a “hot-spot” in cell 8, in which the traffic load (α) is varied from 0.4 to 1.0 calls/sec in step of 0.2. Each cell around cell 8 (cells 1, 7, 9, 20, 21, and 22) has traffic load of 0.4 calls/sec. The traffic load of all other cells is 0.2 calls/sec. This scenario may represent a GSM in an urban downtown area, where the traffic load is centralized in shop-ping or business districts. Although, the values of Ca,i, Cf,i, and Ei are computed using our simulator, only the results obtained for aC , fC ,E are presented in Figures 3 to 5, respectively. Figure 3 shows that as β increases, the average numbers of channels that are success-fully allocated are decreased for all values of α. This is because there are a limited number of channels that can be allocated for the demanding cells and the traffic load is very high.

Finally, in Figure 6, we summarize the values of Ei for the hot-spot cell (cell number 8). The results clearly show that the efficiency reaches a very low value when α=1.0 calls/sec (high traffic load or rate of incoming calls) and β=10 chan-nels/cell. This is because the high traffic load, the limited number of channels

Values of aC , fC , E, and their respective standard deviations (σa, σf, and σe) for uniform traffic load of α = 0.2 calls/sec and uniform initial load of β

= 8 and 10 channels/cell)

β aC σa fC σf E(%) σe

8 3.16 0.93 0.00 0.00 100 0.00

10 2.43 0.65 0.14 0.42 96 0.13

Table 1. Results for scenario #1



Figure 5. E versus β for scenario #2 Figure 6. Ei versus β for cell 8 for scenario #2

75

80

85

90

95

100

4 6 8 10Initial Load ( ) (Channels/cell)

Ave

rage

all

oca t

ion

e ffi

cien

cy,

E (%

)

0

10

20

30

40

50

60

70

80

90

100


All

ocat

ion

effi

cien

cy E

i (%

) fo

r cel

l 8

.

Figure 3. aC versus β for Scenario #2 Figure 4. fC versus β for Scenario #2

0.0

1.0

2.0

3.0

4.0

5.0

6.0


Ave

rage

allo

cate

d ch

anne

ls,

Ca

(C

hann

els)

0.0

1.0

2.0

3.0

4.0

5.0

6.0

7.0

8.0


Ave

rage

fail e

d ch

a nne

ls,

Cf

(C

hann

els)



available, constraints imposed by the network to avoid all forms of interfer-ences, and the constraints imposed on the algorithm for not changing any of the allocated channels. The algorithm only manipulates the currently available channels in the best way to achieve the optimum performance.

cONclUSION

The DDCA scheme presented an excel-lent performance to allocate and reuse channels efficiently under different net-work operation environments without violating the co-channel and adjacent channel interference constraints. This is mainly because the AWC algorithm could satisfy cells needs by intelligently allocating and reusing channels. Their results obtained showed that an aver-age allocation efficiency of 100% was achieved when the initial and the traffic loads are low and uniform as in suburban or rural areas, and the average allocation efficiency was reduced to 96% when the initial load is increased by 25% from 8 channels/cell to 10 channels/cell, and the traffic load remains unchanged at 0.2 calls/sec throughout the network.

Furthermore, the DDCA scheme presented an excellent performance even at a heavily loaded network opera-tion environment in an urban downtown area, where the traffic load is centralized as in shopping or business districts. The minimum average allocation ef-ficiency achieved was 86% when the initial load at the hot-spot cell reaches

10 channels/cell and the traffic load is 1.0 calls/sec.

The results obtained also showed that the performance of the DDCA scheme was highly affected by the num-ber of unused (free) channels remained after initialization, and the performance decreases as the number of free channels after initialization is decreased.

The following scheme do not consider hand-off that may occur due to nodes movement between cells, therefore, it is important to modify the DDCA scheme to consider hand-offs. It is important to modify the DDCA scheme to consider total channels re-distribution if there are many cells ask-ing for more channels and they can not served, due to interference constraints, while there are many unused channels in neighbouring cells. We believe the DDCA scheme could solve this situation by redistribute both the allocated and unused channels in a way that satisfies all cells needs in the system.

reFereNceSAbeysundara, G. (2005). Dynamic cellular channel allocation using intelligent agents. School of Computer Science, University of Catleton.

Al-Agha, k. (2000). Resource manage-ment in wireless networks using intelligent agents. International Journal of Network Management, 10(1), 29-39.

Bejar, R., Domshlak, C., Fernandez, C., Gomes, C., Krishnamachari, B., Selman, B., & Valls, M. (2005). Sensor networks



and distributed CSP: Communication, computation and complexity. Journal of Artificial Intelligence, 161, 117–147.

Faltings, B., & Yokoo, M. (2005). Introduc-tion: Special issue on distributed constraint satisfaction. Journal of Artificial Intelli-gence, 161(1-2), 1-5.

Kostic, Z., Maric, I., & Wang, X. (2001). Fundamentals of dynamic frequency hop-ping in cellular systems. IEEE Journal on Selected Areas in Communications, 19(11), 2254-2266.

Kostic, Z, & Sollenberger, N. (2002). Per-formance and implementation of dynamic frequency hopping in limited-bandwidth cellular systems. IEEE Transactions on Wireless Communications, 1(1), 28-36.

Modi, P. J., Jung, H., Tambe, M., Shen, W. M., & Kulkarni, S. (2001). Dynamic distributed resource allocation: Distrib-uted constraint satisfaction approach. Intelligent Agents VIII Proceedings of the international workshop on Agents, theories, architectures and languages.

Prakash, R., Shivaratri, N. G., & Singhal, M. (1999). Distributed dynamic fault-tolerant channel allocation for cellular networks. IEEE Transactions On Vehicular Technol-ogy, 48(6), 1874-1888.

Salmenkaita, M., Gimenez, J., & Tapia, P. (2001). A practical DCA implementation for GSM networks: Dynamic frequency and channel assignment. IEEE Vehicular Technology Conference, 4, 2529– 2533.

Stallings, W. (2005). Wireless Communi-cations and Networks, Prentice-Hall, 2nd Edition.

Tanenbaum, A. (2003). Computer Net-works, Prentice-Hall, 4th Edition.

Yang, J., Jiang, Q., Manivannan, D., & Singhal, M. (2005). A Fault-Tolerant Dis-tributed Channel Allocation Scheme for Cellular Networks. IEEE Transactions On Computers, 54(5), 616-629.

Yokoo, M., & Hirayama, K. (2000). Algo-rithms for distributed constraint satisfac-tion: A review. Autonomous agents and multi-agent systems, 3(2), 198-212.

Yokoo, M., Durfee, E. H., Ishida, T., & Kuwabara, K. (1998). The distributed constraint satisfaction problem: Formaliza-tion and algorithms. IEEE Transactions on Knowledge and Data Engineering, 10(5), 673-685.

Yokoo, M. (1995). Asynchronous weak-commitment search for solving distributed constraint satisfaction problems. Proceed-ings of the 1st International Conference on Principles and Practice of Constraint Programming, 88–102.

Yokoo, M. (1994). Weak-commitment search for solving constraint satisfaction problems. Proceedings of the 12th Interna-tional Conference on Artificial Intelligence, 313–318.

Yubin, S., Huan, H., Hua, L., & Shaowen, Y. (2000). Performance analysis of the minimum call blocking probability for dynamic channel allocation in mobile cel-lular networks. Proceedings of the IEEE International Conference on Communica-tion Technology, 1, 269-273.

Zhang, M., & Yum, T-S P. (1989). Com-parisons of channel assignment strategies in cellular mobile telephone systems. IEEE Transactions on Vehicular Technology, 38(4), 211-215.

Zhang, W., Wang, G., Xing, Z., & Wit-tenburg, L. (2005). Distributed stochastic



search and distributed breakout: Properties, comparison and applications to constraint

Hussein Al-Bahadili is an associate professor at the Arab Academy for Banking & Financial Sciences. He earned his MSc and PhD from the University of London (Queen Mary College) in 1988 and 1991, respectively. He received his BSc in engineering from the University of Baghdad in 1986. He has published many papers in different fields of science and engineering in numerous leading scholarly and practitioner journals, and presented at leading world-level scholarly conferences. Al-Bahadili is also working as an IT and communication consultant in support of the IT and Wireless Communication Division of the Middle-East Advanced Semiconductor Inc. (MEASI), the leader in communications in Iraq. His research interests include parallel and distributed computing, wireless communication, data communication systems, com-puter networks, cryptography, network security, data compression, image processing, data acquisition, computer automation, electronic system design, computer architecture, and artificial intelligence and expert systems.

Arafat Abu Mallouh is currently a computer lab supervisor at the University of Hashemite. He earned his MSc degree in computer science from Amman Arab University for Graduate Studies in 2008. He received his BSc in computer science from the University of Hashemite in 2001. His primary research interest is in computer networks, artificial intelligence, experts systems and fuzzy logic, data mining.

optimization problems in sensor networks. Journal of Artificial Intelligent, 161(1-2), 55-87.

Documents

A Distributed Dynamic channel Allocation Scheme in cellular … · 2011. 1. 6. · in cellular communication networks (CCNs), such as the global system for mobile com-munication (GSM)