P.M Papazoglou (1), D.A. Karras, Assoc. Member IEEE (2), and R.C. Papademetriou, Member IEEE (3)* I. INTRODUCTION

Transcription

1 On new dynami hannel assignment shemes and their effiient evaluation through a generi simulation system for large sale ellular teleommuniations P.M Papazoglou (1), D.A. Karras, Asso. Member IEEE (2), and R.C. Papademetriou, Member IEEE (3)* Abstrat - The rapid evolution of ellular tehnology and the augmentative user demand for advaned mobile servies leads the industry to develop more effiient network strutures. The inreasing number of ellular users and the demand for broadband mobile ommuniations (3 rd and 4 th generation) drives to the researh of new methodologies for the design of ellular networks and servies. Simulation environments more than ever offer the opportunity to develop and study with low ost new strutures and methods for the implementation of new servies. This paper thoroughly reviews entralized and distributed Dynami Channel Assignment (DCA) shemes in terms of simulation harateristis and needs and proposes a omprehensive simulation system for evaluating them, whih inorporates their priniples, entities and onepts involved. The main goal, however, is to present three new DCA variations and prove - through experimental evaluation using the proposed simulation system - that the suggested shemes outperform traditional DCA shemes. The proposed simulation system is implemented in Java, in order to reate a high performane generi simulation environment with the future apability of internetworking. The system is designed with the goal to be effiient in simulating large sale generi ellular teleommuniation systems and the assoiated DCA shemes. Moreover, another goal of the proposed system is to serve as a test bed for the evaluation and development of DCA shemes (espeially for eduational purposes) involved in suh large sale ellular systems towards their planning as effetive ellular mobile radio networks. Keywords - ellular networks, DCA variations, simulation models, hannel alloation shemes. I. INTRODUCTION The onept of ellular networks The ellular priniple divides the overed geographial area into a set of smaller servie areas alled ells. Eah ell has a base station and a number of mobile terminals (e.g. mobile phone, palms, laptops, or other mobile devies). The base station is equipped with radio transmission and reeption equipment. The mobile terminal within a ell ommuniates through wireless links with the base station assoiated with the ell. A number of base stations are onneted to the Base Station Controller (BSC) via mirowave links or dediated leased lines. The BSC ontains logi for radio resoure management of the base stations under its ontrol. It is also responsible for transferring an ongoing all from one base station to another as a mobile user moves from ell to ell. In order to establish a ommuniation with a base station, a mobile terminal must first obtain a hannel from the base station. A hannel onsists of a pair of frequenies: one frequeny (the forward link/downlink) for transmission from the base station to the mobile terminal, and another frequeny (the reverse link/uplink) for the transmission in the reverse diretion. An alloated hannel is released under two senarios: the user ompletes the all or the mobile user moves to another ell before the all is ompleted. (1) Lamia Institute of Tehnology, Greee and ECE Dept., University of Portsmouth, UK (e_mail: papaz@teilam.gr) (2) Chalkis Institute of Tehnology, Greee and Helleni Open University, Greee (e_mails: dakarras@teihal.gr, dakarras@ieee.org) (3) ECE Dept., University of Portsmouth, UK (e_mail: rallis.papademetriou@port.a.uk) 1

2 The need of Simulation systems in ellular networks The performane and the behavior of a real ellular network an be evaluated using simulation systems without the need to perform field experiments and develop prototypes. The simulation solutions give us the opportunity to develop ustom protools, hannel alloation shemes, network strutures, et, towards to a desired ellular network. Due to the omplexity of real ellular networks the simulation software development strategy beomes a very important fator that influenes the resulted network model. The struture of the simulation environment affets the performane of simulated ellular network and for that reason we try to study the design and the development of suh system. Many approahes to simulation systems have been made using general purpose languages, general simulation languages and speial purpose simulation pakages. General purpose languages have low ost, are portable and no additional training is needed but every model starts from srath, verifiation is diffiult, ode reusability is low and the model development yle is long. General simulation languages offer standardized features for modeling, shorter development yle and easy verifiation. Suh languages have higher ost, limited portability and additional training is needed. Languages suh as Simula[1], Parse[2], GPSS[3], Simsript[4], modsim[5], slam[6] and other of speial purpose offer high flexibility for simulation systems development but introdue lak of portability and low adaptability to speifi appliation models. Other software tools like Matlab, have a plethora of sientifi libraries and apabilities for flexible sript ode development but suffer from low performane exeution. A Large sale simulation system requires a flexible network environment with advaned harateristis suh as portability, internetworking and high adaptability to the real network behavior. The invention of new simulation languages introdue important drawbaks whih are: New languages are domain-speifi and are rarely adapted by sientifi ommunity The orresponding libraries impose the designers to adapt their appliations in speifi requirements Designers an not ahieve high adaptability of the simulation kernel in their appliations Java language is the most suitable for building flexible, portable and high performane network appliations and is adapted by the majority of sientifi ommunity. With this simulation program, we attempt to build a generi simulation system for ellular ommuniation systems of 3 rd towards 4 th generation in order to perform simulations not in ad-ho systems as most software tools do. With a generi simulation system, lear and general onlusions an be extrated ompared to ad-ho wireless networks. II. AN OVERVIEW OF CHANNEL ALLOCATION SCHEMES & GOS CONSIDERATION The apaity of a ellular system an be desribed in terms of the number of available hannels, or the number of users the system an support. The total number of hannels made available to a system depends on the alloated spetrum and the bandwidth of eah hannel. The available frequeny spetrum is limited and the number of mobile users is inreasing day by day, hene the hannels must be reused as muh as possible to inrease the system apaity. The assignment of hannels to ells or mobile is one of the fundamental resoure management issues in a mobile ommuniation system. The role of a hannel assignment sheme is to alloate hannels to ells or mobiles in suh a way as to minimize: a) the probability that the inoming alls are dropped, b) the probability that ongoing alls are dropped, and ) the probability that the arrier-to-interferene ratio of any all falls below a prespeified value. In literature, many hannel assignment shemes have been widely investigated with a goal to maximize the frequeny reuse. The hannel assignment shemes in general an be lassified into three strategies: Fixed Channel Assignment (FCA) [7,8, 9,10,11], Dynami Channel Assignment (DCA) [7,12,13,14,15], and the Hybrid Channel Assignment (HCA) [7,16]. In FCA, a set of hannels are permanently alloated to eah ell based on a pre-estimated traffi intensity. In DCA, there is no permanent alloation of hannels to ells. Rather, the entire set of available hannels is 2

3 aessible to all the ells, and the hannels are assigned on a all-by-all basis in a dynami manner. One of the objetives in DCA is to develop a hannel assignment strategy, whih minimizes the total number of bloked alls. The FCA sheme is simple but does not adapt to hanging traffi onditions and user distribution. Moreover, the frequeny planning beomes more diffiult in a miroellular environment as it is based on the aurate knowledge of traffi and interferene onditions. These defiienies are overome by DCA but FCA outperforms most known DCA shemes under heavy load onditions [8]. To overome the drawbaks of FCA and DCA, HCA was proposed by Kahwa et al.[7], whih ombines the features of both FCA and DCA tehniques. DCA shemes an be implemented as entralized or distributed. In the entralized approah [7] all requests for hannel alloation are forwarded to a hannel ontroller that has aess to system wide hannel usage information. The entral ontroller then assigns the hannel by maintaining the required signal quality. I distributed DCA [12], the deision regarding the hannel aquisition and release is taken by the onerned base station on the basis of the information from the surrounding ells. As the deision is not based on the global status of the network, it an ahieve only suboptimal alloation ompared to the entralized DCA and may ause fored termination of ongoing alls. Many hannel alloation shemes have been proposed. There are many variations of the FCA, DCA and HCA shemes. FCA variations Channel borrowing and non borrowing shemes In hannel borrowing shemes [17], a ell an borrow a hannel from an adjaent ell only if there is no free hannel in initial ell. After the use of the borrowed hannel, the hannel returns to its initial ell. Known shemes in hannel borrowing ategory [17] are the Simple borrowing (SB), Borrow from the Rihest (BFR) and Borrow First Available (BFA) sheme. In SB, after the initial hannel assignment in eah ell if there is no free hannel for a new all, the needed hannel an be borrowed from a neighboring ell. In the ase of the BFR sheme, the available hannel is borrowed from a neighboring ell that holds the greatest number of free hannels (for borrowing) ompared to other neighboring ells. Without any optimization when borrowing, BFA sheme is the simplest beause the borrowing is based on the first available hannel [17]. When the ell hannels are divided in two groups standard hannels and borrowable hannels we have the Simple Complex Channel Borrowing Sheme (SCCB). In Borrowing with Channel Ordering (BCO) we use hannel priorities to define the borrowing order. Other FCA variations are Sharing with Bias (SHB), hannel Assignment with Borrowing and Reassignment (CABR), and Ordered Dynami Channel Assignment with Rearrangement (ODCA) [18]. Non borrowing shemes are mostly based on hand-of strategies where an available hannel is diffiult to loate. DCA variations As we mentioned before, entralized and distributed versions are the main variations of the DCA sheme. Centralized DCA shemes The First Available (FA) is the simplest method where the first available hannel within the reuse distane enountered during a hannel searh is assigned to the new all. Future bloking probability is a very important fator for the hannel seletion in a new all. Loally Optimized Dynami Assignment (LODA) examines this probability in order to assign a new hannel. The Mean Square (MSQ) sheme selets an available hannel that minimizes the MSQ between ells that use the same hannel. 3

4 Distributed DCA shemes The distributed DCA shemes are based on signal quality omputations around the initiated alls. More preisely, are based on o-hannel distane, signal strength measurement and signal to noise interferene ratio. Loally Optimized Least Interferene DCA algorithm (LOLIA) [19] is a known variation and omputes the interferene among users that use the same hannel and examines the signal quality in omparison with a defined threshold. One of the most important harateristi of a ellular network is that a ertain number of alls will fail to establish an initial onnetion. We an state two measures for dealing with unsuessful all attempts: the never servied new alls, and the delayed suessful alls (waiting for initial onnetion). The servie quality in the busy period of the day may be expressed as the grade of servie [20] or quality of servie. Two requirements an be set in order to measure GoS level or Quality of Servie [21]: GoS, loss system : probability of bloking < x% GoS, delay system : probability (waiting time > z se) < x% Another quality metri an be the signal to noise interferene C/(N+I) ratio ompared to a predefined threshold for the aeptability of a all. Carrier to noise ratio is a key performane for the networks and represents the signal strength against the hannel noise. Another fator that degrades the arrier signal exept the noise is the interferene from other onneted users that use the same frequeny hannel. If we inrease the signal power to improve ommuniation quality we inrease in the same time the interferene to other onneted users. Thus, we introdue Carrier to Noise plus Interferene Ratio, C/(N+I), that represents the signal strength (quality) against noise and interferene from other onneted users.[22, 23] III. THE PROPOSED GENERIC SIMULATION SYSTEM Our system is a generi simulation system for ellular ommuniations of 3 rd and towards 4 th generation in order to perform simulations not only in ad-ho wireless networks but in every ellular network. The number of ommuniation hannels, the number of users and other simulation parameters an be onfigured using a speial onfiguration text file. The supported hannel alloation shemes for a given ellular network are distributed DCA and entralized DCA. As we mentioned in previous setion, the whole program was built in Java in order to have the apability of future extensions in any onstrains. In order to build a suffiient simulation system it is neessary to take in onsideration some basi aspets of the operation of the ellular network whih are: User harateristis - user attributes inluding onnetion status, urrent position, all holding time, alloated hannel, ommuniation signal strength and interferene (if neessary), et. Network parameters - Cells number, hannels per ell, ell positions, base station positions, et. Traffi and QoS - new all arrival shemes, user movement profile, all dropping onditions, hand-of onditions, et. Channel alloation shemes hannel alloation strategy aording to network onditions, ongestions ases, resoures limitations, et. Our simulation system onsists of one main module - the simulation module, and nine additional modules. The additional modules are used for the neessary omputations, the network model struture and simulation program exeution. 4

5 Traditional DCA shemes that are used in our simulation system In distributed DCA, the all management is based on the desired Carrier to noise interferene C/(N+I) ratio. For the entralized DCA sheme the first available hannel is alloated to the new user in the ell only if the hannel is not used in any other adjaent ell. The network struture The ellular network onsists of N ells and eah of the ells inludes a base station in the enter position (fig.1). Around the N ells there are some opied ells (from real ells) in order to generate the expeted behavior when users exist in boundary ells Figure 1 - struture of the ellular network model Eah ell is divided in a mesh of spots (fig.2) where the users may exist from the beginning of the all or due to hand-of situation. The number of spots per ell mesh an be defined from the simulation onfiguration. The basi parameter for the mesh onstrution is the fineness whih means the distane between the spots. Figure 2 - ell mesh (possible user positions in a ell) Aording to user onditions, the simulation program saves the relevant information in the user attribute table. These attributes summarize the user onditions and updated in every user new state (Table 1). User attribute Desription X position Current X position in the network Y position Current Y position in the network Connetion 0=no onneted, 1=onneted status Call holding Call termination time (se) time Alloated Alloated hannel (if onneted) hannel Path loss Path loss of the user (if onneted) Table 1 - user attributes 5

6 The simulation takes plae during the simulation time. In this time, several proedures are exeuted: all termination hek realloation hek new all arrival user movement There are also additional proedures for the normal operation of the simulation program, like onfiguration file proessing, initialization, et. The basi network parameters for the simulation proedure are GSM like and are summarized in table 2. Network Parameters Desription CNR on ell edge (db) Carrier to Noise Ratio in the ell edge CNIR threshold (db) Aepted Carrier to Noise Ratio for a suessful all average all holding time (se) Call duration hannels per ell Maximum available number of Channel pairs (upload/download) per ell path loss fator(alpha) Signal attenuation due to distane (signal path) standard deviation of Shadowing attenuation shadowing( sigma) ellmesh fineness Number of possible loations of a user in a ell max users per ell Maximum onneted users per ell No. of ells Number of ells New all arrival rate Rate of new all ourrene User movement rate Rate of user movement ourrene Table 2 Network parameters for simulation Main Simulation/Network proedures Call termination. The program uses an exponential funtion that generates the all duration for eah new user. The all holding time is added to urrent simulation time for later examination. The assoiated proedure sans the network ells and searhes for any onneted users and examines the progressive all time. If this time is expired the user attribute table is updated with the new onnetion status of the partiular user both in distributed and entralized DCA. In addition, in entralized DCA the new free hannel after the all termination returns in the hannel pool. Figure 3a, shows the logi diagram of that algorithm. 6

7 Figure 3a all termination proedure Figure 3b Realloation hek (distributed DCA) Figure 3 New all arrival (distributed DCA) Figure 3d New all arrival (entralized DCA) Realloation hek (Distributed DCA ) The realloation algorithm searhes all the ells for onneted users and performs hek for eah user (fig. 3b). The omputations are based on signal strength and how is affeted from other onneted users in neighbor ells. If a user signal does not fulfill the C/(N+I) threshold (Carrier to Noise and Interferene ratio) the proedure tries to find another appropriate hannel. The Carrier to Noise and Interferene ratio an be derived using the type R ni = N AP d + α 0 0 n i 1 A P d i 10 ξ 0 10 α i ξ i Where n is the number of base stations and users, ξ i is the distortion due to shadowing from user to base station, A is a proportional oeffiient, P o is the transmitted power of a referene point, P i is the transmitted power of the i user and d i is the distane between user i and referene base station. Firstly, the algorithm alulates the signal strength between user and base station and in later time alulates any interferene from around onneted users. If an aepted hannel is found, is alloated to the new user, otherwise the all is dropped. The program searhes for a hannel aording to the seleted DCA sheme. In entralized version, the realloation proess searhes for a free hannel in the entral hannel pool. New all arrival based on user profile New alls are resulted from a random or a Poisson distribution with regard to a predefined daily model. The algorithm searhes all the ells for not onneted users (aording to user attribute table). Eah user belongs to a user profile with regard to the time of the day. When a new all arrival ours with regard to Poisson distribution, the program gets a random number x from Poisson distribution and alulates the orresponding P(x) whih is : x e λ λ Px ( ) = (2) x! (1) 7

8 Where λ represents the new all arrival rate. This rate is analogous to the time of the day. A fixed set of simulation time steps represents a whole day. Eah day is divided in to five zones aording to traffi onditions. Eah zone represents a speifi perentage period of the day. The parameter λ, hanges with regard to eah ative zone in simulation time. Table 3, illustrates the five zones with orresponding λ parameter. zone Zone perentage λ Desription (approximately) 1 34% 1 Least busy zone 2 21% % 5 Most busy zone 4 17% % 2 Table 3. Daily zones Zone 1, may represents the hours 12-9am (λ=1) and zone 3 the hours 1-4pm (λ=5). Finally, the number of users to be onneted in eah ell is analogous to x, it is based on P(x) and never exeeds the maximum limit of users that is alulated by the daily model. If proper onditions exist for a new all arrival, a random position in the ell mesh is alulated. The user position attribute is updated with the mesh spot position in ombination with the position of the base station in the enter of the ell. Channel Alloation The signal strength is alulated between the new user and the base station. Channel alloation is suessful if a free hannel is found in the ell and the C/(N+I) onstraint is fulfilled (fig.4). After a new all arrival, several ations take plae in turn (fig. 4a): a) Searh for unonneted users in eah ell b) Calulate a random user position in the mesh ) Plae the new user aording to ell s base station position and mesh spot d) Calulate the signal strength between base station and new user Firstly we obtain the shadow attenuation [22,23] as σ n sh = (3) where σ is the standard deviation of shadowing and n is a number from the normal distribution. Using the shadow attenuation and distane between user and base station, we an derive the distane attenuation dw. Finally, we alulate the CNIR (Carrier to Noise Interferene Ratio) between user and base station [22,23] nedge 10 n = 10 dw (4) where nedge is the CNR on ell edge (db). e) San the other ells and alulate interferene among new user and other users that use the same hannel f) Chek if C/(N+I) ratio is aeptable aording to predefined threshold g) If C/(N+I) is aeptable, establish the new all and update the table of user attributes. New all arrival and hannel alloation (entralized DCA) A new all arrival ours under the same onditions in distributed DCA. After the alulation of new user position, the algorithm searhes for an available hannel that is not used in urrent ell or in any adjaent ell. The searh is performed using the hannel pool that keeps information for hannel usage all over the ellular network. The hannel alloation riterion is based only in the hannel usage 8

9 in adjaent ells. This method requires less proessing power than the method in distributed DCA beause is not based to mathematial alulations. User movement (distributed & Centralized DCA) The simulation program supports two types of user movement. The first type of user movement is based on random number generation like new all arrival and the seond on Gaussian distribution. The algorithm loates the onneted users and hanges their urrent positions. The realloation proedure in distributed DCA heks the new user positions and performs new hannel assignment if needed. In entralized DCA, after user movement, the user is disonneted and the algorithm searhes for a proper free hannel from the entral hannel pool. The Gaussian distribution density funtion is 1 ( x µ ) 2 /2σ f( x) = e 2 σ 2π (5), where σ is the standard deviation and µ is the mean value. The Gaussian distribution in our program has mean value zero and standard deviation one. Hene, the density funtion beomes 1 x 2 /2 f( x) = e (6) 2π In the user movement proedure, firstly we generate a Gaussian number e.g. x 1 and the orresponding f(x 1 ) when x1 [ 0.5,0.5]. If another Gaussian number e,g x 2, where x2 f ( x1) a user move is generated. III. THE PROPOSED DCA VARIATIONS We have used our simulation program to evaluate the new proposed DCA variations whih are: Balaned version (based on least-ongested algorithm around neighbors) Best CNR version (based on strongest CNR around neighbors) Round bloking version (based on unbalaned DCA version around neighbors) Firstly, the program tries to assign a hannel within the initiated ell of the new all and afterwards ativates any of the above methods. The above mentioned proedures suh as realloation, new all arrival and user movement are adapted eah time to the seleted DCA variation (e.g. balaned, unbalaned, round bloking, et) Balaned variation (min ell ongestion) The DCA unbalaned algorithm assigns the first hannel that fulfills the onditions and the DCA balaned uses the least ongestion algorithm for the ompletion of hannel assignment. As we mentioned before, the balaned method uses the least ongested algorithm. Initially, we assume that the least ongested ell is the ell where the new all is ourred. The algorithm sans all the ells sequentially in the neighbor region and finds the least ongested ell (fig. 4a). For example, ell number 7 (figure 4b), ontains only 4 users ompared to other ells in the neighbor and so is the least ongested ell. 9

10 Figure 4a The least ongestion algorithm (Balaned variation) Figure 4b The least ongestion Logi Best CNR variation The best CNR method goes around the six neighbor ells of the initiated ell and alulates the CNR between user and neighbor ell base stations. After the strongest CNR is found, the program tries to assign a new hannel from the orresponding ell. The algorithm initially assumes that the strongest CNR exists from the first neighbor ell and ontinues to alulate CNR from the rest neighbor ells sequentially until the strongest CNR is found. Figure 5a and 5b, illustrates the algorithm struture and logi respetively. Max_CNR Max_CNR_ell = First Neighbor Cell Current Cell=First Neighbor Cell Max_CNR=CNR of Current Cell 2 7 Get first/next Neighbor ell Current Cell= Neighbor ell Compute CNR for Current Cell 4 5 Current CNR > Max_CNR NO YES Max_CNR_Cell= Current Neighbor Cell Max_CNR=CNR of Current Neighbor Cell NO Are all neighbor ells sanned? YES Return Max_CNR_Cell Figure 5a The Best CNR algorithm Figure 5b The Best CNR Logi 10

11 Round Bloking variation The round bloking algorithm searhes in neighbor ells and when a suessful hannel assignment is made the algorithm stops. If the all establishment is not suessful in the last neighbor ell, the all is bloked. Figure 6a and 6b shows the round bloking algorithm and the logi respetively. Round Bloking Get first/next neighbor ell Compute CNR from urrent BS Find available hannel 2 7 Compute Interferene from other users CNIR > CNIR Threshold? 4 5 Sueed=1 Sueed=0 Last neighbor reahed or Suessful onnetion? Return Sueed Figure 6a The Round Bloking algorithm Figure 6b The Round Bloking Logi Unbalaned variation Unbalaned variation represents the traditional DCA sheme where only one attempt (in all initiating ell)for onnetion is done based on the Carrier to Noise plus interferene riterion (fig. 7). Unbalaned Compute CNR from urrent BS Find available hannel Compute Interferene from other users CNIR > CNIR Threshold? Sueed=1 Sueed=0 Return Sueed Figure 7 Traditional DCA algorithm (Unbalaned) 11

12 IV EVALUATION CRITERIA FOR DCA ALGORITHMS AND EXPERIMENTAL STUDY Basi Statistial metris The bloking probability is one of the most important harateristis for the performane of a ellular network. When a new all arrival ours and the network an not alloate a hannel then we say that this all is bloked. The bloking probability P bl is alulated from the ratio number of bloked alls Pbloking = number of alls (7). If the reeived power of eah user is high enough, we an make the assumption that the interferene from other users an be ignored. Thus, we an ompare the simulated bloking probability with the theoretial whih is n 1 s ( vh) s Pbloking _ theoretial = s n 1 i ( vh) i= 0 i (8) where n is the number of users, s is the number of hannels, v is the average all arrival rate (for no onneted user) and h is the average all holding time. The dropping probability is also an additional and very important harateristi of the ellular network. When a all is in progress and the required quality onditions are not met then this all is obligatory driven to termination. The dropping probability P f is alulated from the ratio number of fored alls P = fored alls number of alls number of bloked alls (9) We have used several basi statistial metris suh as Standard deviation, Mean Value and Variane in the experimental setion of this paper. Experimental results The system produes the simulation results in matlab files (.m files) for later proess in matlab. Eah result file ontains the simulation data and the appropriate matlab ommands for graph viewing and statistial analysis. The simulation results inlude data for alulating important measures suh as bloking probability, dropping probability, hannel reuse, ell ongestion, number of realloations, hannel reuse per ell, et. The generated matlab files inlude additional ommands for alulating several statistial metris of simulation results like Standard deviation, Covariane and Average values. 12

13 Indiative results As we mentioned before, the simulation system produes a variety of graphs. Figure 8, shows the hannel reusability per network ell of an example network. The example network onsists of 19 ells, 100 users per ell, 32 hannels per ell and uses the distributed DCA sheme. 350 Channel RE-USE per CELL 300 b i lit y a u s e R e l n C han Cell Number Figure 8 hannel reusability The daily model that we have used divides the whole simulation time in five days respetively. The bloking probability varies among the hours of eah day (fig.9) BLOCKING,USERS=100 y b i lit a b o p r g n o k i B l Sim time Figure 9 - bloking probability Next figure (fig. 10) shows the dropping probability with regard to traffi onditions based on mentioned daily model for the example network DROPPING,USERS=100 y b i lit a b o p r g n p i D rop Sim time Figure 10 - Dropping probability 13

14 DCA omparative results We have simulated a ellular network of 19 ells apaity with 10, 50, 80,100,200,400,750 users per ell and 5, 32 hannels per ell respetively. The simulation results are divided in two ategories : Unbalaned Distributed and Centralized DCA (traditional Channel Alloation) Distributed DCA with the three proposed variations In Centralized DCA sheme, the new all arrival and user movement proedures are based in random distribution. In addition, in distributed DCA sheme, the new all arrival is based on random distribution and user movement is based on random or Gaussian distribution. Figure 11 shows the bloking probability for Distributed and Centralized DCA. The Centralized sheme is very inflexible as the number of users is inreasing ompared to other Distributed shemes. With user movement or not, Centralized DCA an not be adapted to the user load. 1 BLOCKING,hannels per ell=5 0.9 y b i lit a b o p r g n o k i B l DCA-rr,h5,nm DCA-drr,h5,nm DCA-drg,h5,nm DCA-rr,h5,m DCA-drr,h5,m DCA-drg,h5,m USERS Figure 11. Bloking probability of various DCA shemes (5 hannels per ell) nm/m=no user movement/user movement, rr=entralized DCA, random all arrival, random user movement (in m ategories) drr=distributed, DCA, random all arrival, random user movement (in m ategories), drg= distributed DCA, random all arrival, Gaussian user movement (in m ategories) Figure 12 illustrates that even with 32 hannels per ell, the entralized DCA shemes are still inflexible to higher user load. 1 BLOCKING,hannels per ell= y b i lit a b o p r g in o k B l DCA-rr,h32,nm DCA-drr,h32,nm DCA-drg,h32,nm DCA-rr,h32,m DCA-drr,h32,m DCA-drg,h32,m USERS Figure 12. Bloking probability of various DCA shemes (32 hannels per ell) 14

15 Figure 12, onfirms the previous note about the performane of Centralized DCA in small networks and shows that even with more available hannels per ell the Centralized DCA is not satisfatory for small networks. The above figures also show that the user movement and user movement distribution plays important role for the network behavior. In Centralized DCA the number of ell ongestions is smaller in ontrast with Distributed DCA (fig.13 & 14) and that means that the network management is easiest is entralized DCA CELL CONGESTION,hannels per ell=5 n o ge sti C on C el DCA-rr,h5,nm DCA-drr,h5,nm DCA-drg,h5,nm DCA-rr,h5,m DCA-drr,h5,m DCA-drg,h5,m USERS Figure 13 - ell ongestions (5 hannels) 500 CELL CONGESTION,hannels per ell= n o ge sti C on C el DCA-rr,h32,nm DCA-drr,h32,nm DCA-drg,h32,nm DCA-rr,h32,m DCA-drr,h32,m DCA-drg,h32,m USERS Figure 14 - ell ongestions (32 hannels) The three proposed DCA variations, perform better than the traditional DCA under heavy user load (fig.15). In traditional DCA sheme only one attempt for onnetion is done within the urrent ell. On the ontrary, in all the three proposed variations several attempts for onnetion are made among the adjaent ells. The best CNR algorithm it seems to be the most appropriate for hannel alloation beause ensures the strongest CNR between user and base station and so higher shield from interferene is ahieved. 15

16 1 Mean Value, BLOCKING 0.9 y b i lit a P rob g n k i B lo Unbalaned Best CNR Round Bloking Balaned (min ell ong) No of Users Fig 15. Bloking probability of the traditional DCA sheme and the three proposed Shemes ty b i li a P rob g n p i D rop Mean Value, DROPPING Unbalaned Best CNR Round Bloking Balaned (min ell ong) No of Users Fig 16. Dropping probability of the traditional DCA sheme and the three proposed Shemes The dropping probability (fig. 16) with regard to proposed variations is lower than the traditional DCA sheme. Although, in high user loads, all the DCA shemes seem to have the same behavior beause there are a large number of users ompared to available hannels in the ell. V. CONCLUSIONS & FUTURE WORK There is no doubt that ellular network simulation is a very ompliated proedure. From the point of view of network designer the development and researh of various hannel alloation shemes must be faed from the lowest starting point. The type and the number of parameters that the designer should take in onsideration depend on the type of the network and the proedure fousing inside the network. The divergene between simulation results and real network behavior is affeted by the struture of the implementation algorithms and the orresponding researh of real network onditions, like new all arrival shemes, user movement, et. A deeper researh requires full ustomizable software with high level of adaptation in real network speifiations and behavior. In this paper we have presented a generi simulation environment for GSM ellular networks and also several DCA variations for effiient hannel assignment. We have learly shown through thorough experimental results that our proposed DCA variations perform better than the traditional DCA 16

17 sheme and, moreover, they ould be effiiently evaluated through the proposed generi simulation system. Our future goal is to develop a simulation system for large sale generi ellular networks using other implementation arhitetures in order to extrat learer and reliable onlusions for real world DCA algorithm behavior and performane. With the internetworking apability of Java language, our simulation system might be, also, used for eduational purposes too. In order to develop an effiient and useful simulation environment we have to study the evolution, the speifiations and the onstraints of real ellular networks as well as to investigate the algorithmi shemes that might be used, like multi-threading, for the user and network model, et. REFERENCES [1] Osar Nierstrasz. A survey of objet-oriented onepts. In W. Kim and F. Lohovsk, editors, Objet-Oriented Conepts, Databases and Appliations, pages ACM Press and Addison Wesley, (1989). [2] R.L.Banrodia, Et al, Parse: A parallel simulation environment for omplex systems, IEEE Computer, 21(10):77-85, Ot.1998 [3] Karian, Z. A.,Dudewiz, E. J.: Modern Statistial, Systems, and GPSS Simulation - The First Course; New York (1991). [4] Russell, E. C.: Building Simulation Models with SIMSCRIPT II.5; C.A.C.I., In., Los Angeles (1983). [5] Joel West and Alasdar Mullarney. ModSim: A language for distributed simulation. Proeedings SCS Multionferene on Distributed Simulation, pages , San Diego, California, February (1988). [6] J. Miller, O. Weyrih and D. Suen, A Software Engineering Oriented Comparison of Simulation Languages, Proeedings of the 1988 Eastern Simulation Conferenes: Tools for the Simulationists, Orlando, FL (April 1988). [7] M.Zhang, and T. S. Yum, Comparisons of Channel Assignment Strategies in Cellular Mobile Telephone Systems, IEEE Transations on Vehiular Tehnology, vol.38,no.4,pp ,1989. [8] W.K. Lai and G.C. Coghill, Channel Assignment through Evolutionary Optimization, IEEE Transations on Vehiular Tehnology, vol.45,no.1,pp91-96,1996. [9] V.H. MaDonald, The ellular Conepts, The Bell System Tehnial, Journal, vol.58, pp.15 42, [10] S. M. Elnoubi, R. Singh, and S.C. Gupta, A New Frequeny Chan-nel Assignment Algorithm in High Capaity Mobile Communiation Sys-tems, IEEE Transations on Vehiular Tehnology, vol. VT-21, no. 3,pp , [11] Z. Xu and P.B. Mirhandani, Virtually Fixed Channel Assignment for Cellular Radio- Telephone Systems: A Model and Evaluation, IEEE International Conferene on Communiations, ICC 92, Chiago, vol. 2, pp , [12] L.J. Cimini and G.J. Foshini, Distributed Algorithms for Dynami Channel Alloation in Miroellular Systems, IEEE Vehiular Tehnology Conferene, pp , [13] D. C. Cox and D. O. Reudink, Inreasing Channel Oupany in Large Sale Mobile Radio Systems: Dynami Channel Reassignment, IEEE Transantions on Vehiular Tehnology, vol.vt- 22, pp , [14] E. Del Re, R. Fantai, and G. Giambene, A Dynami Channel Alloa-tion Tehnique based on Hopfield Neural Networks, IEEE Transantions on Vehiular Tehnology, vol.vt-45, no.1, pp.26 32, [15] K.N. Sivarajan, R.J. MEliee, and J.W. Kethum, Dynami Channel Assignment in Cellular Radio, IEEE 40th Vehiular Tehnology Confer-ene, pp , [16] T.J. Kahwa and N.D. Georgans, A Hybrid Channel Assignment Shemes in Large-Sale, Cellular Strutured Mobile Communiation Sys-tems, IEEE Transations on Communiations, vol.26, pp ,

18 [17] I.Katzela and M.Naghshineh, Channel assignment shemes for ellular mobile teleommuniation systems: A omprehensive survey, IEEE Personal Communiations, pp.10-31,1996 [18] S.Jordan, Resoure Alloation in Wireless Networks, Journal of High Speed Networks, vol.5,no.1,pp.23-24,1996 [19] P.Cherriman, F.Romiti and L.Hanzo, Channel Alloation for Third-generation Mobile Radio Systems, ACTS 98, vol.1,pp ,1998 [20] T.S. Rapaport, Wireless Communiations Priniples and Pratie, Prentie Hall, 2002 [21] M.Salamah and H.Lababidi, Dynamially adaptive hannel reservation sheme for ellular networks, Elsevier Computer networks, artile in press, 2005 [22] T.S. Rappaport, Wireless Communiations Priniples and Pratie, Prentie Hall, (2002) [23] William C.Y.Lee, Mobile Cellular Teleommuniations, MGraw-Hill, (1995) 18