Journal of Network and Computer Applications 34 (2011) 18611868

Contents lists available at ScienceDirect

Journal of Network and Computer Applications

journal homepage: www.elsevier.com/locate/jnca

A cross-layer resource allocation scheme for ICIC in LTE-Advanced

Zhaoming Lu n, Yan Yang, Xiangming Wen, Ying Ju, Wei Zheng
Beijing University of Posts and Telecommunications, Beijing 100876, China

a r t i c l e i n f o
Article history: Received 30 June 2010 Received in revised form 17 December 2010 Accepted 26 December 2010 Available online 18 January 2011 Keywords: Cross-layer Potential games Particle swarm optimization Inter-cell interference coordination Power allocation

abstract
As a new technology, inter-eNB coordination has been included in LTE-Advanced study items. Moreover, the network architecture in LTE-Advanced system is modied to take into account coordinated transmission. In our study, we explore the problem of jointly optimizing the power level and scheduling of resource blocks for LTE-Advanced network based on orthogonal frequency division multiplexing (OFDM). We propose a distributed optimization scheme based on evolutionary potential games, and in the process of objective function modeling we employ the Lagrangian multiplier method to solve the constraint objective optimization problem. Then particle swarm optimization (PSO) method is adopted to nd the optimal power allocation and scheduling for each resource block in the multi-cell framework. Numerical results prove that proposed algorithm notably improves the overall throughput, while user fairness is guaranteed. Importantly, additional computation and communication cost introduced by cross-layer optimization is also evaluated. & 2011 Elsevier Ltd. All rights reserved.

1. Introduction Inter-cell interference (ICI) makes a main dent in system throughputs for cellular mobile networks based on OFDM. Intercell interference coordination (ICIC) that is a research highlight currently has the task to manage radio resources such that intercell interference is kept under control (3GPP TS 36.300, 2009). Through effective resource allocation, ICIC technology reduces the probability of time slot and spectrum resource conict of users from different cells, and decreases the interference level, improves signal to interference plus noise ratio (SINR) as well as system throughput and spectrum efciency for users. There are many achievements in this area, such as power allocation method for ICIC (Hui et al., 2009; Castellanos et al., 2008; Pischella and Belore, 2008), soft frequency reuse (Nortel, 2007), intelligent scheduling based on SINR (Elayoubi et al., 2008; Simonsson, 2007), opportunity scheduling technology (Svedman et al., 2007; Hu et al., 2008), and MIMO (Boudreau et al., 2009). ICIC is inherently a multi-cell radio resource management (RRM) function that needs to take into account information (e.g. the resource usage status and fairness of different users) from multiple cells. Inter-eNB coordination is proposed in LTE-Advanced, in which eNBs communicate with each other through ber connected enhanced X2 interface. Rapid and dynamic information exchange can be achieved via this enhanced X2 interface. Hence, cooperative games theory can be explored in

Corresponding author. Tel.: +86 18901381605. E-mail address: lzy_0372@163.com (Z. Lu).

inter-eNB coordination based wireless networks for ICIC. This is because the resource allocation strategies of cells affect each others performance and game theory provides an effective mechanism for this interplay situation. In addition, enhanced X2 interface supplies a rapid information communication channel for eNBs, which brings the perfect information game into practice. Actually, there are dependencies between most of the strategies for ICIC. It is almost impossible to optimize the applications performance by determining the strategies just in sight of single layer without knowledge of other layers in the time-varying and resourcelimited wireless environment. An effective approach to gure out this problem is using cross-layer design and optimization technique (Srivastava, 2005). However, it is a great challenge to nd the optimal solution for the cross-layer optimization problem. It is difcult to solve the optimization problem regarding the throughput and fairness as objective functions, because these functions are nondeterministic and nonlinear. As a simple evolutionary algorithm with rapid convergence, PSO method is an applicable scheme for objective function solving. In this paper, we draw concepts from potential game theory to develop a cross-layer ICIC framework that converges to a desirable state in wireless networks. Specically, we design a cross-layer optimization scheme including power allocation in physical layer and resource block (RB) scheduling in media access control (MAC) layer. A potential game model was also used (Neel, 2006) to design dynamic frequency selection algorithms, and paper (Candogan et al., 2010) showed that the power control game could be approximated by a potential game. However, these approaches are only for single strategy such as power allocation or resource scheduling. In Lacatus and Popescu (2007), convex

1084-8045/$ - see front matter & 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.jnca.2010.12.019

1862

Z. Lu et al. / Journal of Network and Computer Applications 34 (2011) 18611868

non-cooperative games were used to investigate the convergence and Nash equilibrium (NE) of combined power and waveform adaptation games that allowed players to obtain desired target SINR in the networks. However, the convergence of formulation adaptations could only be established in centralized networks. There are three contributions in our paper. Firstly, a dynamic resource allocation method is proposed, by which we can adjust the resource allocation dynamically according to the channel conditions. At the same time, ICI can be coordinated and system throughput can be improved. Secondly, through potential game, a distributed method for cross-layer optimization with potential to reduce complexity is proposed. Meanwhile, rapid convergence of iteration can be achieved in an interplay scenario. Thirdly, the most important, we adopt a cross-layer design to allocate resource such as transmitted power and RBs, and the dependence between protocol layers in OSI is exploited. Scheduler can allocate system resource in line with time-varying channels in more effective way in cross-layer architecture and multi-layer diversity gain can be achieved. The rest of the paper is organized as follows. We analyze the ICI model in Section 2. The denition of potential game is provided in Section 3, as well as the existence and uniqueness of NE. We transform a constraint objective function to an unconstraint objective function in Section 4 by means of Lagrangian multiplier method. In Section 5 a cross-layer resource allocation scheme based on potential games and PSO is proposed, and the convergence of this method is also analyzed. In Section 6 the performance of our algorithm is simulated and the price is evaluated concretely. We conclude and point out future directions of research in Section 7.

transmitter, respectively. pn j is the transmitted power in subband j from cell n, and PN is additive white Gaussian noise (AWGN). Apparently, each RB attempts to tune the operating frequency and transmitted power in order to maximize the SINR and hence the throughput. In our work, we investigate the issue of maximizing the overall throughput of the network, dened as the summation of throughput of all RBs with same subband, by nding the optimal frequency and transmitted power allocation strategy. Also, due to the concern of scalability and computational complexity, we prefer a decentralized solution to the throughput maximization problem. Let pj and sj denote the power level and user allocation of the jth RB, respectively. Therefore, for each RB, the value of SINR is computed by the two variables of pj and sj. To solve the problem, two different kinds of dependencies should be considered; one is the dependencies between strategies of RBs with same subband in diverse cells, and the other is the dependencies between the strategies in physical and MAC layers. Therefore, the throughput maximization problem is coupled and cannot be solved straightforward.

3. Potential games The concept of potential games is originally proposed by Monderer and Shapley (1996). Potential games have received increasing attention recently. Various notions of potential games are introduced and studied in the literature. Generalized ordinal, ordinal, exact, and weighted potential games are introduced in Monderer and Shapley (1996). Voorneveld et al. have studied ordinal potential games and characterized several properties of these games. The notions of best-response potential games and pseudo-potential games are studied in Voorneveld (2000) and Dubey et al. (2006), respectively. A strategic game is a potential game if it admits a potential function that can be interpreted as deviations among different players, or rather potential functions quantify the difference in the payoff due to unilaterally deviation of each player. Potential games possess several special properties. The existence of the potential function guarantees these properties. Under some conditions, all potential games have pure strategy NE. More interestingly, under some conditions not too strict, every learning process based on best-response of the players converges to an NE. In other words, starting from an arbitrary point, the sequence of unilaterally best-responses of players reaches to an NE after nite steps. 3.1. Denition Let G /N , Y , uS be a game in strategic form with a nite number of players that is N. The set of strategies of player i

2. Inter-cell interference analysis As shown in Fig. 1, we consider a synergic eNB downlink system including N cells. In order to coordinate the ICI and optimize the resource allocation schemes, each eNB allocates time-frequency resource and adjusts transmitted power dynamically, through exchanging data and signaling in real time by enhanced X2 interface. Assume that the available bandwidth is 10 MHz, and the total spectrum is divided into J subbands. One RB is made up by one subband and one time slot. When ICI from other N 1 eNBs around cell k is taken in account, SINR of the jth RB in cell k can be expressed as (suppose this RB is allocated to the user m) SINRj, k PN gk jpk j N P gn jpn j

n 1, n a k

where gk j is the power gain factor from cell k to user m, which is rx dened as g p ptx , where prx and ptx are the power of receiver and

Fig. 1. Inter-eNode B coordination.

Z. Lu et al. / Journal of Network and Computer Applications 34 (2011) 18611868

1863

is Yi, and the payoff function of player i is ui : Y -R, where Y Y1 Y2 YN that denotes the set of strategy proles, and Rdenotes the set of real numbers. Then we dene the exact potential function P : Y -R, if for every i A N and for every yi A Yi , ui yi , xui yi , z P yi , xP yi , z, 8x, z A Yi

where Rj is the raw data rate, and R j is the throughput of the RB. Pe is the error probability of the transmission channel, which is a function of SINR value providing the modulating and coding scheme. The total throughput of the spectrum for multi-cell system can be expressed as T p, s
N X j1

where Yi denotes the strategy proles except the strategy of player i. G is called an exact potential game if it admits an exact potential. Here the potential function can reect the change in the utility accrued by every unilaterally deviating player. It is easy to show that the necessary and sufcient condition for an exact potential game is @ui y @Py , @yi @yi 8i A N , y A Y

R j

j 1, 2, . . ., N

where p p1 , p2 , . . ., pN , s s1 , s2 , . . ., sN , and N is the number of RBs in one cluster.

4.2. Fairness criterion From an operator view, fairness is very important since users expect to have the same experience regardless whether they are close to an eNB or at the cell edge. Total throughput and spectrum efciency gures make sense only if a fairness criterion is fullled. Since maximum system capacity may be obtained by providing low throughput to some users, it is important that all users be provided with a minimal level of throughput. Let Tm be the throughput for user m. The normalized throughput with respect to the average user throughput for user m, Tm , is given by Tm Tm avgj Tj j 1, 2, . . ., N

3.2. Existence and uniqueness of NE The following two lemmas characterize the existence of NE in exact potential games. Lemma 1. If P is the potential function of exact potential game

G /N , Y , fui gi A N S , then the equilibrium set of G coincides with the equilibrium set of coordination game G /N, Y , fP gi A N S (Monderer
and Shapley, 1996). Lemma 2. Every nite exact potential game possesses a pure strategy Nash equilibrium (Monderer and Shapley, 1996). Consequently, every y A Y that maximizes P(y) is a pure strategy equilibrium of G. However, in general the converse is not true. There might be pure or mixed strategies NE that are just local maximum points of P(y). Some conditions are required to promise the NE be unique. Lemma 3. If the strategy space is convex and P(y) is continuously differentiable on the strategy space, then the NE of G is a stationary point of P(y). If P(y) is concave, then every NE of G is a maximum point of P(y). Such a NE is unique if P(y) is strictly concave. In exact potential games, potential function is continuous if and only the utility functions are continuous. In practice, a large family of potential functions is continuous and the strategy space is compact.

The fairness is evaluated by determining the normalized cumulative distribution function (CDF) of the user throughput (NGMN Alliance, 2008). The interpretation of fairness criterion is that at least 90% of the users should have at least 10% of the average user throughput. Then the constraint for fairness can be expressed as F p, s NTm Z 0:10:9 Z 0 where NTm Z 0:1 denotes the number of users whose normalized throughput is bigger than 0.1.

4.3. A Lagrangian multiplier method As described above, the objective function of system throughput should be optimized under the constraint of fairness, which can be formulated as ( max T p, s s:t F p, s Z 0

4. A Lagrangian multiplier method for constraint objective function optimization 4.1. System throughput Automatic modulation and coding methodology is employed in LTE-Advanced. Entities such as UEs and eNBs are capable of dynamically choosing the modulation and coding schemes and adjusting the transmission rates according to the measured channel condition which is indicated by transmission bit error rate (BER). Given a particular modulation scheme, BER is uniquely determined by the SINR experienced by the receiver of the link. Generally speaking, higher SINR value yields lower BER and higher transmission data rate. Therefore, the mutual interference dramatically degrades the transmission rate of each cell and the aggregated throughput of the whole network. As we know, RBs tune the physical parameters such as operating frequency and transmitted power in order to maximize the throughput, which is dened as (Song et al., 2008) R j SINRj maxRj 1Pe SINRj , Rj
Rj

Let T p, s T p, s, then the maximum problem can be translated into a minimum problem ( min T p, s s:t F p, s Z 0

This is a nonlinear optimization problem with inequality constrains, and Lagrangian multiplier method can be used to solve this problem. The augmented Lagrangian function can be dened as

Fp, s, o, s T p, s

1 max0, osF 2 o2 2s

By Fp, s, o, s, a nonlinear optimization with inequality constraints can be converted to an optimization problem without constraint min Fp, s, o, s

1864

Z. Lu et al. / Journal of Network and Computer Applications 34 (2011) 18611868

5. Cross-layer optimization algorithm based on potential games 5.1. Cooperation games modeling The throughput maximization problem is challenging, as the scheduling and power allocation strategies of one RB affect the SINR of other RBs. Fortunately, inter-cell information can be shared through inter-eNB coordination technologies. The framework of cooperation games can be used to study the optimal resource allocation problem, which means every player in this game attempts to make the social welfare maximum by cooperating with other players, rather than just attempting to maximize their own utility with selsh behavior. First of all, we need to map a resource scheduling problem to a game, and the relationship between elements in a game and those in a LTE-Advanced network can be expressed as players3RBs with same subband in different cells strategy3Joint resource scheduling schemes Here, RBs with same subband is mapped to players in game, which is a decision maker in the interactive decision process. RBs can suppress ICI through choosing optimal scheduling user and allocating appropriate transmitted power, and then the performance of LTE-Advanced networks can be improved. Specically, for game G /N, Y , fuj gj A N S, N denotes the number of cells of cluster, and Y denotes the strategy prole of RBs, which can be expressed as Y fp, sg, where p is the alternative transmitted power prole for RBs, and s is the set of alternative users to be scheduled for RBs. In practice, transmitted power is a series of discrete values with an interval of 0, Pmax , and the value of Pmax is up to operation devices. uj denotes the utility function for player j, which can be expressed as uj Fp, s, o, s In order to make the utility function minimum, RBs will continuously choose the optimal scheduling user and adjust the transmitted power. As the game G /N , Y , fuj gj A N S is a cooperative game, the potential function can be dened as P Fp, s, o, s Apparently, the relationship between potential function and utility function is @uj y @Py , @yj @yj 8j A N, y A Y

evolutionary algorithms, PSO exploits a population, called a swarm, of potential solutions, called particles, which are modied stochastically at each iteration of the algorithm. However, the manipulation of swarm differs signicantly from that of evolutionary algorithms, promoting a cooperative rather than a competitive model (Bui and Alam, 2008). Let V be a 4-dimensional search space, f : V -R be the objective function, V fx1 , x2 , . . ., xM g, and M be the number of particles. Then the ith particle is a point in the search space, and its position can be expressed as xi xi1 , xi2 , . . ., xi4 p, s, o, s, i 1, 2, . . ., N

with the velocity and tness, respectively, vi vi1 , vi2 , . . ., vi4 , and fitnessi f xi1 , xi2 , . . ., xi4 Fp, s, o, s, i 1, 2, . . ., N i 1, 2, . . ., N

as well as its best position and the global best position, respectively, pi pi1 , pi2 , . . ., pi4 and pg pg 1 , pg 2 , . . ., pgN We can now construct the PSO based interactive method to solve the extreme problem of multivariate function. The procedure of this method is summarized as follows: 1. The best positions of particles, as well as their velocities, are randomly initialized in the search space, and the tness of every particle in current position is calculated, fitnessi f xi1 , xi2 , . . ., xi4 : 2. Start to iterate, update the vector of velocity vic wvic c1 r1 pic xic c2 r2 pgc xic Where ( vic
max Vc max Vc max vic Z Vc max vic r Vc

c f1, 2, 3, 4g,

i 1, 2, . . ., N

and the vector of position xic xic vic where ( xic

max Xc max Xc max xic Z Xc max xic r Xc

c f1, 2, 3, 4g,

i 1, 2, . . ., N

In other words, game G /N , Y , fuj gj A N S is an exact potential game. As potential function Py is strictly concave, in terms of Lemma 3, if P y is strictly concave, the NE of G /N , Y , fP gj A N S is unique, and the equilibrium leads to minimum of P(y). 5.2. PSO based method to solve extreme problem of multivariate function At each iteration of proposed algorithm, RBs need to solve the following extreme value problem of multivariate function: minFp, s, o, s Traditional methods are complex to solve the problem and easy to get the local optimal solution. However, compared with traditional methods, PSO method has less limitation and the solving process is simpler, which makes PSO method an efcient way to solve the extreme problem of multivariate function. Eberhart and Kennedy developed PSO as an expansion of an animal social behavior simulation system that incorporated concepts such as nearest-neighbor velocity matching and acceleration by distance (Kennedy and Eberhart, 1995). Similarly to

c1 and c2 are two positive constants called cognitive and social parameters, respectively. r1 , r2 are realizations of two independent random variables that assume the uniform distribution in the range 0, 1. w is a positive parameter called inertia weight, which is concerned with last velocity. When w is bigger, particles are inclined to global search; when w is smaller, particles are inclined to local search. Here, a linear decrement scheme proposed in Yuhui and Eberhart (1998), is adopted to obtain better global search performance. 3. Update pbesti and gbest. 4. If the criterion of stopping iteration is satised, that is reaching iterative number limit, jump out of circulate. Otherwise, go to step 2. 5.3. Convergence to Nash equilibrium A path in strategy prole Y is a sequence g y0 , y1 , . . . such 1 that yk yk i , x for some x A Yi . y0 is the initial state of the path, and player i is called the deviator in step k. Obviously, in each step

Z. Lu et al. / Journal of Network and Computer Applications 34 (2011) 18611868

1865

k, only one player is allowed to deviate. g is an improvement path with respect to G if for all k Z 1, inequality ui yk 4 ui yk1 is satised. G has the nite improvement property (FIP) if every improvement path is nite. Lemma 4. Every nite exact potential game has the FIP (Monderer and Shapley, 1996). Because game G is a nite potential game, according to Lemma 4, it will converge to NE after nite steps. Evidently, every improvement path of nite exact potential games must terminate in an equilibrium point. That is, the myopic learning process based on the one-sided better reply dynamic converges to the equilibrium set that is independent of the initial point. Note that the order at which players deviate to a better or best response can be deterministic or random and need not to be synchronized. It is the most interesting property of the potential games especially in order to nd the equilibrium of the self-optimization systems in distributed way. Decision criterion is important for convergence process, which may impact the speed and efciency of convergence of iteration. Here, best response criterion is used for convergence to protect its stability, which can be expressed as
Di yi fy i A Yi yi : yi argmaxui yi , yi g yi

Table 1 Simulation parameters. Parameters Inter-cell distance Center frequency Number of antenna Path loss Thermal noise density eNB antenna pattern Value 500 2 GHz 44 128.1 + 37.6 log 10(R), R in km 174 dBm/Hz !  2 Ay min 12 y y , Am
3 dB

y3 dB 701, Am 20 dB
L2S Bandwidth of RB EESM 180 KHz

Lemma 5. Best response sequence nally converges to NE in nite exact potential games (Dubey et al., 2006). In other words, if choices of every player in each step are best regardless of the initial point of the path, it must terminate in NE after nite steps. On the basis of Lemma 5, provided RBs select strategies that make potential function minimum at each iteration, the game will converge to equilibrium. At the same time ICI brings least impact to quality of wireless channels, and the total system throughput is maximal under the constraint of fairness. The cross-layer potential game in this paper is a perfect information game. A signicant aspect of iteration of game is the mechanism by which players exchange information and the topology of players (Bertsekas and Tsitsiklis, 1989). In this paper all players were directly connected to each other via enhanced X2 interface, and just one is updated at a time. As the number of links rises with the square of number of players, in this paper the number of interconnection cluster is set as 19.

multiple users sharing a wireless channel (Andrews et al., 2001). Fixed power allocation is used to be cooperated with these two methods in the simulation. Total throughputs of these three schemes are compared in Fig. 2. Simulation results of 20 TTIs are plot in the gure. Obviously, the cross-layer resource allocation scheme improves the system throughput remarkably according to this gure. It is because the cross-layer method exploiting the dependence between protocol layers in OSI and the multi-layer diversity gain is obtained. RBs dynamically allocate the limited wireless resource (transmitted power, time slot, and subband) in more effective way according to the time-varying channel condition. Moreover, through PSO based potential games, optimal resource allocation strategy is obtained in the scenario of interplay. As a result, ICI can be restrained effectively, and the performance of system throughput is much better than non-cross-layer schemes. Meanwhile, fairness of users is regarded in this cross-layer architecture. In Fig. 3, we compare the cumulative distribution function (CDF) of the normalized user throughput of MAX C/I, round robin, PF, and cross-layer scheme. We can see that the number of user whose normalized throughputs are smaller than 0.1 is less than the 10% of total number of users, and the fairness of network is guaranteed with the promotion of network performance.

6.2. Cost analysis 6. Simulation analysis We test the cross-layer resource allocation scheme in LTEAdvanced simulation platform, in which a scenario of 19 cells is considered, and each cell is made up with 3 sectors. Hence, the interplay of RBs with the same subband can be computed adequately and the number of connection links is not excessive. There are 30 users in one sector. Users are distributed randomly in every sector. The frequency selective multipath channel is modeled as consisting of six independent Rayleigh multipaths, with an exponentially decaying prole. ITU PB3 model is adopted in this paper. The relevant parameters for simulation are given in Table 1. There are 50 RBs for scheduling simultaneously in one sector. The iterative number limit of PSO is 50, and the particle number is 20 in simulation. 6.1. Performance analysis We compare the performances of our scheme with two classical scheduling schemes including PF scheduling and Modied Largest Weighted Delay First (M-LWDF) scheduling. M-LWDF is an efcient way to support quality of service of Previous work on cross-layer method has succeeded in showing the potential performance gain over traditional ICIC schemes. However, the additional cost to be paid to gather the relevant parameters from multiple layers and network locations, and implement the cross-layer optimization has typically been neglected. Our cross-layer architecture has three additional types of costs compared with purely layered one. First, the time complexity; convergence of potential game is an iterative process, which requires substantial computation and may result in additional delay. Second, the communication overhead; in the process of potential game there are many imformation exchanges (transmitted power in different RBs) between eNBs, which may have a nonnegligible communication overhead. Third, the management complexity; network architectures with cross-layer optimizations are less modular and therefore more difcult to manage or recongure. The rst type of cost in our cross-layer architecture is measured below. As the enhanced X2 interface is used for interconnection of players in this cross-layer architecture, dynamic information exchange can be implemented between eNBs, and the communcation cost can be left out. The third type is not easy to quantify. However, it can be reduced by dening

1866

Z. Lu et al. / Journal of Network and Computer Applications 34 (2011) 18611868

Throughput Differences among Strategies 130

120

Total Throughput (Mbps)

110

100

90

80 Cross-layer Scheme M-LWDF PF 2 4 6 8 10 TTI 12 14 16 18 20

70

60

Fig. 2. Throughputs comparison of different schemes.

CDF of Normalized Throughput 1 0.9 0.8 0.7 0.6 CDF 0.5 0.4 0.3 0.2 0.1 0 0 0.5 1 1.5 2 Normalised Throughput (bps)
Fig. 3. CDF of normalized user throughput.

MAX C/I Round Robin PF Cross-layer Scheme 2.5 3

proper interfaces between the layers and the cross-layer optimizer (Khan et al., 2006). In order to evaluate the time complexity of algorithm, the price of anarchy (Koutsoupias and Papadimitriou, 1999) is dened as the ratio between the objective function value of equilibrium and that of an optimal solution. POA Tobj P e Tobj P

If the price of anarchy of a game is 1, then its equilibrium is fully efcient. Ideally, as the model proposed in this paper adopts cooperative game framework, the solution of equilibrium and the optimal solution of objective function is equivalent. However, PSO method is applied in the process of optimization, which may not get the accurate optimal solution of the objective functions. That is, usually, the solution of PSO is an approximate optimal solution. Therefore, the price of anarchy of this cross-layer game is POA Tobj Pe %1 Tobj P

where Pe denotes the strategy prole under equilibrium, and P denotes the optimal strategy selection.

Z. Lu et al. / Journal of Network and Computer Applications 34 (2011) 18611868

1867

1 0.95 0.9 0.85 Price of Anarchy 0.8 0.75 0.7 0.65 0.6 0.55

4 5 6 Average Iterative Times

Fig. 4. Price of anarchy with average iterative times.

As shown in Fig. 4, after 2 or 3 iterations, cross-layer game could converge to a near optimal point. Hence, the time complexity brought by iteration is nite.

no. D08080100620802, and Important National Science & Technology Specic Projects under Grant no. 2010ZX03003-001-01.

References 7. Conclusions and future work Cross-layer design is a new model that has great potential to change the design and management of wireless networks in the future. Potential games specify a set of games for which the convergence to NE is guaranteed in pure strategy. In this paper, we propose a cross-layer resource allocation method for ICIC and multi-layer diversity gain is obtained. We explored potential games theory to solve the cross-layer optimization problem, and the interplay of RBs in different cells are regarded adequately, as well as the convergence of iteration is guaranteed. In the process of objective function formulation Lagrange multiplier method is used to transfer the constraint objective function to unconstraint objective function. Then particle swarm optimization method is adopted to nd the optimal power allocation and appropriate scheduling user for each RB in the multi-cell framework. In addition, via parallel processing and enhanced X2 interface, time and communication complexity can be reduced effectively. The scope of our work may be extended in several respects. In our cross-layer design, only physical layer and MAC layer are considered; in the future work, application layer is expected to be added in the cross-layer design for more multi-layer diversity gain. In addition, time complexity of evolutionary potential game may be reduced further with a completely distributed way, which can be implemented through a self-motivated game rather than a coordination game. With a self-motivated game, there is little information exchange between RBs, and Jacobi iteration may be adopted to optimize users strategies in completely parallel way.
3GPP TS 36.300 v8.8.0. Evolved Universal Terrestrial Radio Access (E-UTRA) and Evolved Universal Terrestrial Radio Access Network Technical Specication; 2009. Andrews M, Kumaran K, Ramana K, et al. Providing quality of service over a shared wireless link. IEEE Communications Magazine 2001;39:1504. Bertsekas DP, Tsitsiklis JN. Parallel and distributed computation: numerical methods. 2nd ed. Athena Scientic Press; 1989. Boudreau Gary, Panicker John, Guo Ning, et al. Interference coordination and cancellation for 4G networks. IEEE Communications Magazine 2009:7481. Bui Lam Thu, Alam Sameer. Multi-objective optimization in computational intelligence: theory and practice. IGI Global; 2008. Castellanos C, et al. Performance of uplink fractional power control in UTRAN LTE. Singapore: VTC Spring; 2008. p. 251721. Candogan Utku Ozan, Menache Ishai, Ozdaglar Asuman, et al. Near-optimal power control in wireless networks: a potential game approach. In: Proceedings of the IEEE Infocom 2010. Dubey P, Haimanko O, Zapechelnyuk A. Strategic complements and substitutes and potential games. Games and Economics Behavior 2006;54:7794. Elayoubi SE, Haddada OB, Fourestie B. Performance evaluation of frequency planning schemes in OFDMA-based networks. IEEE Transactions on Wireless Communication 2008;7:162333. Hu H, et al. Radio resource management for cooperative wireless communication systems with organized beam-hopping techniques. IEEE Wireless Communication 2008;15:1009. Hui Zhang, Xiaodong Xu, Jingya Li, et al. Multicell power allocation method based on game theory for inter-cell interference coordination. Science in China Series F-Information Sciences 2009;52:237884. Kennedy J, Eberhart RC. Particle swarm optimization. Proceedings of the IEEE International Conference Neural Networks 1995;4:19428. Khan S, Peng Y, Steinbach E. Application-driven cross-layer optimization for video streaming over wireless networks. IEEE Communications Magazine 2006;1:12230. Koutsoupias Elias, Papadimitriou Christos. Worst-case equilibria. STACS 99; Lecture Notes in Computer Science, Springer-Verlag, Berlin; 1999. p. 40413. Lacatus C, Popescu DC. Adaptive interference avoidance for dynamic wireless systems: a game theoretic approach. IEEE Journal on Select Areas in Communications 2007;1:189202. Monderer D, Shapley L. Potential games. Games and Economics Behavior 1996;14: 12443. Neel JO. Analysis and design of cognitive radio networks and distributed radio resource management algorithms. Ph.D. Dissertation, Virginia Polytechnic and State University; 2006. NGMN Alliance. NGMN Radio Access Performance Evaluation Methodology; 2008. Nortel 3GPP RAN1 Contribution R1-072761. Uplink Power Control with Fractional Frequency Reuse for EUTRA. Orlando, FL; 2007.

Acknowledgments This work was supported by the Sci-tech Projects sponsored by the Committee on Science and Technology of Beijing under Grant

1868

Z. Lu et al. / Journal of Network and Computer Applications 34 (2011) 18611868

Pischella P, Belore JC. Power control in distributed cooperative OFDMA cellular networks. IEEE Transactions on Wireless Communication 2008;7:19005. Simonsson A. Frequency reuse and inter-cell interference coordination in E-UTRA. Dublin: VTC Spring; 2007. p. 30915. Svedman P, et al. Opportunistic beamforming and scheduling for OFDMA systems. IEEE Transactions on Communication 2007;55:94152. Srivastava Vineet. Cross-layer design: a survey and the road ahead. IEEE Communication Magazine 2005;S:1129.

Song Yang, Zhang Chi, Fang Yuguang. Joint channel and power allocation in wireless mesh networks: a game theoretical perspective. IEEE Journal on selected areas in Communications 2008;26:114959. Voorneveld M. Best-response potential games. Economics Letters 2000;66: 28995. Yuhui Y, Eberhart R. A modied particle swarm optimizer. In: Evolutionary Computation Proceedings. IEEE World Congress on Computational Intelligence; 1998. p. 6973.