Planning of Dynamic Channel Allocation in HetNet

2015 IEEE International Conference on Advanced Networks and Telecommuncations Systems (ANTS)

Planning of Dynamic Channel Allocation in HetNet under IEEE 1900.4 Framework Ayan Paul

Mainak Sengupta

Madhubanti Maitra

BSNL Kolkata, India [email protected]

Electrical Engineering Jadavpur University Kolkata, India

Electrical Engineering Jadavpur University Kolkata, India Madhubanti.Maitra66@ hotmail.com

Ideally, the spectrum allocation decision must be in real time and should be aligned with the overall business strategy of the WSP. In this work, we have proposed a comprehensive spectrum allocation policy for the WSP in the context of DSA framework. IEEE 1900.4 working group has been formed to propose a standard architecture for rolling out the DSA concept in HetNet [4]. Hence, we have considered network architecture of the WSPs based on IEEE 1900.4 standard.

Abstract—Dynamic spectrum allocation (DSA) approach offers an effective means to wireless service providers (WSPs) for efficient utilization of their spectrum resources in order to manage their heterogeneous network (HetNet). The architecture for implementation of DSA has been proposed in IEEE 1900.4 standard. Under the IEEE 1900.4 framework, we envisage a scenario in which, the spectrum available with the WSP is inadequate to serve the aggregate demands of channels or the spectrum, from its radio access networks (RANs). Hence, a plausible solution has been sought in this work that employs and implements the notion of the Bankruptcy Game to formulate a policy decision which would certainly be agreeable to all the RANs. Maximization of proportional fairness (PF) is selected as the potential objective of the WSP. To arrive at an efficient solution that suits the objective of the WSP, we have proposed a novel heuristic namely, H_MPF and have compared the performance of the H_MPF with those of the Shapley value and simulated annealing based channel allocation (SACA) techniques considering solution quality and computational time as the performance indices. Keywords- IEEE 1900.4 standard; dynamic spectrum allocation; bankruptcy game; proportional fairness

I.

A. Business problem As demand for wireless services are growing in a rapid pace, it is quite likely that a WSP may face a situation in which, the cumulative demand of channels (spectrum) by the RANs is more than the amount of channels available to the WSP. Even if the available channels are adequate in terms of count, a RAN may be compelled to receive channels at inferior (higher) frequency band in lieu of its desired frequency band. As efficiency level of spectrum (because of propagation characteristics) varies across different frequency bands, change in frequency band affects the coverage area of the BSs. Let us illustrate this with an example. Suppose the GSM RAN demands 10 channels in 900 MHz band. However, due to spectrum scarcity in this band, it is allocated 10 channels in 1800 MHz frequency band. In the above scenario, the area of coverage of the RAN will be considerably reduced as channels at 1800 MHz band suffers more loss compared to the channels in 900 MHz band. In the case of this constrained spectrum scenario, the WSP may choose a policy in which, the demand of the RAN that garner highest revenue is met on priority basis. We refer this approach as revenue maximization approach (RMA). In our opinion, this strategy may not always be a wise choice for the following reason. In the HetNet setting, it is desirable to maintain consistent level of quality of service (QoS) across the composite wireless network (CWN) as its users enjoy seamless mobility across the HetNet. Hence, it may be prudent for the WSP to consider the fairness aspect at the time of spectrum allocation to its RANs [5]. In this work, we have used the concept of proportional fairness (PF) [6] to measure the fairness of the spectrum allocation solution. Thus, the challenge of the WSP is to arrive at an appropriate and efficacious policy for allocation of spectrum among its multiple RANs.

INTRODUCTION

Heterogeneous network (HetNet) consisting of radio access networks (RANs) of different radio access technologies (RATs) has become the fundamental architecture of the new age wireless communication systems such as LTE-Advanced [1] and the future fifth generation (5G) networks [2]. However, management of expensive spectrum resource poses significant challenge to the WSPs in managing their HetNets. Dynamic spectrum allocation (DSA) approach [3] has been proposed for more efficient spectrum utilization over traditional static spectrum allocation (SSA) scheme [3]. The SSA scheme exclusively assigns certain bandwidth of spectrum to spectrum users only for pre-specified purpose. On the other hand, in DSA approach, a pool of spectrum is maintained by the WSP and from the pool, spectrum is dynamically allocated to different users based on their requirements. Recent technological developments such as cloud RAN [2] and self-organizing network [1] are set to play key enabler in implementing the DSA techniques in cellular networks. In addition to this technological aspect, spectrum allocation policy of the WSP for distribution of the available spectrum among its RANs is another important focus area.

978-1-5090-0293-1/15/$31.00 ©2015 IEEE

184


B. System model The proposed network architecture of IEEE 1900.4 consists of four network entities namely, operator spectrum manager (OSM), network reconfiguration manager (NRM), RAN measurement collector (RMC) and RAN reconfiguration controller (RRC) [4]. In this case, each RAN houses two entities viz. RMC and RRC. RMC collects the context information about its respective RAN and forwards it to the NRM, which is unique to the CWN of the WSP. OSM frames the spectrum allocation policy of the CWN to optimize the usage of spectrum resources of the WSP. Further, OSM communicates the policy to the NRM for its implementation. Following the policy, the actual spectrum allocation decision is taken by the NRM based on the inputs from the RMCs. Finally, RRC configures the radio nodes of the RAN with the spectrum assigned to it.

works have taken simplistic assumption that the RANs would be able to operate at any channelizations. Also, the works do not differentiate efficiency of spectrum at different frequency bands. Therefore, we have incorporated these realistic constraints in the present work. Although, several works [1516] have been carried out based on IEEE 1900.4 standard, the problem of policy decision for scarce spectrum resources has not been addressed earlier. In [9], we have considered the present problem and have proposed simulated annealing based channel allocation (SACA) technique to solve the problem. However, SACA technique consumes significant amount of computational time. Therefore, in the current work, we have proposed more computationally efficient technique that would be more effective in real time setting. Hence, this work is unique in its kind. B. Cooperative game theory A cooperative game has two main concepts: coalition and coalition value. In this case, RANs are the players in the cooperative game and a group of RANs form coalition in order to receive higher payoff in the game [8]. Let ℕ = {1,2, … , } be an index set of n number of RANs of the WSP. However, the RANs belonging to coalition  ℕ have an agreement among them to act as a single entity. Set ℕ is called grand coalition that denotes coalition of all the RANs. Value of the coalition ( ( )) indicates the worth of the coalition. If RANs 1 and 2 collectively receive a spectrum amount of ĝ unit by forming a coalition, the value of the coalition {1, 2} is represented by v({1, 2})=ĝ. A cooperative game is uniquely defined by the pair (ℕ, ) [8].

C. Our contribution In this work, we have mapped the spectrum allocation problem of the WSP as n-player cooperative bankruptcy game [7-8]. Subsequently, we have transformed the problem as an integer linear programming (ILP) problem. We have designed novel heuristic namely, H_MPF to find out the most suitable spectrum allocation solution as far as objective MPF is concerned. We have studied the performances of the heuristics with that of conventional cooperative game theory solution concept Shapley value [8] and simulated annealing based channel allocation (SACA) [9] techniques on different performance indices viz. solution quality and computational time. Finally, we have proposed an efficient spectrum allocation policy for the OSM such that the WSP can maintain uniform QoS within the CWN. The paper is organized as follows. A brief background of the work is presented in Section II. In Section III, the problem formulation is presented. Section IV is devoted to the solution methodologies employed in this work. The simulation results have been provided in the Section V. Section VI concludes the work. II.

C. Bankruptcy game Bankruptcy situation arises when the cumulative claims of the creditors on a company is more than the worth of the company [7]. When the available spectrum of the WSP is insufficient to cater the aggregate spectrum demand of the RANs, the situation is similar to the bankruptcy situation. Here, amount of spectrum is identical to worth of the bankrupt company whereas, bids of the RANs are similar to the claims of the creditors.

BACKGROUND OF WORK

A. Related works Majority of the works [10-11] related to DSA are based on competitive framework in which, multiple WSPs vie for common spectrum. However, the problem of distribution of spectrum among multiple BSs of different RATs has been considered in [12]. The management of resources in large scale HetNet consisting of reconfigurable devices has been addressed by Amin et al. [13] by striking a balance among the conflicting objectives of minimizing the energy consumption, maximizing the spectral efficiency and maximizing the fairness among the end users with regard to the data rates allocated to them. However, the problem of spectrum allocation of a WSP among its RANs has not been considered in these works.

III.

PROBLEM FORMULATION

Let ℕ be an index set of RANs managed by WSP in a particular service area. A RAN within set ℕ is denoted by index ∈ ℕ. We have assumed that the RANs are capable of operating only in selected frequency bands. Let R be the index set of operating frequency bands. As per our system configuration, the channel bid of RAN includes two items namely, the number of required channels and index of preferred frequency band for the channels i.e. = 〈 , 〉. For example, if a particular BTS is capable of operating on multiple bands such as 900, 1800 and 1900 MHz bands then, R={900, 1800, 1900}. Thus, = 1 if the preferred operating band of the BTS is 900 MHz. The WSP divides its available spectrum to finite number of channels of equal bandwidth. However, the width of the channels is decided by the WSP based on characteristics of the RANs under its control.

Previously, the spectrum distribution problem of the WSP has been studied in [5] [14]. However, the authors in the above

185


Moreover, in case of bankruptcy games, the ({ }) may be expressed by the number of channels that are not claimed by the RANs not belonging to coalition [8]:

Let the total number of available channels of the WSP is L in a certain area at a certain point of time. A channel is indicated by index . Variable ∈ ℜ denotes the allocation of channel to RAN where, ℜ is the solution space. The variable = 1 when, channel is allocated to RAN , otherwise 0.

({ }) = max 0, − ∑

One of the possible objectives of the OSM/WSP is: maximizing the proportional fairness (MPF) [17] in which, the aggregate of the logarithmic utilities is maximized. Hence, the objective function is expressed as: = ∑ log( ) = ( ) (10) In the above equation, corresponds to the utility function of RAN i. The design of the utility function is described in section IV. The term is the utility value of RAN i in case it receives amount of channels.

In this work, we envisage a scenario in which, the aggregate amount of demanded channels in band is more than the number of available channels at the band. Number of allocated channels to RAN can be expressed by where: =∑ (1) In spectrum constrained situation, RAN may be allocated channels at region = 2 when it has demanded channels at region = 1. In this case, is also expressed as: =∑ (2)

To summarize, the objective of the work is to: Maximize

In the above equation, indicates the number of channels allocation to RAN at spectrum band . In this context, the main challenge of the WSP is to distribute the available channels among the RANs considering the conditions below: ∑ =1 (3) ∑

=

(9) ∈

=

(11)

Subject to the constraints: ∑ =1 ∑ ∑ = ≤∑ ≤ ∑ ∑ ≥ ({ }) ∑ ∑ ≥ ({ })

(4)

≤ ≤ (5) The constraint (3) signifies that a channel can be allocated to single RAN at a time. This is to avoid interference among the RANs. The constraint (4) indicates the condition that all the available channels must be entirely distributed among the RANs. Condition (5) points out to the fact that the number of allocated channels must be within the range [ , ] where, is the minimum number of channels required for RAN to be operational.

∀ ,∀ ∀ ∀ ∀ ⊂ℕ

(12) (13) (14) (15) (16)

∈

As is the decision variable in our problem formulation, function transforms the objective function (9) in the terms of variable . Hence, optimal values of is requires to be determined to maximize the objective function value. IV.

SOLUTION METHODOLOGY

A. Utility design Utility function measures the benefit of allocated spectrum to the RANs. RAN receives different amount of benefit depending upon specification (band) of the received spectrum. The utility of RAN , may be expressed as [9]:

Now, we investigate the possibility of existence of spectrum allocation solutions that are acceptable to all the RANs. Naturally, a solution will be agreeable to a RAN only if, the solution offers at least similar payoff (i.e. number of channels) compared to the payoff it receives individually i.e. ≥ ({ }) (6) The solutions that satisfy constraints (4) and (6) are called imputations in the jargon of cooperative game theory. However, imputation does not consider the possibilities of formation of coalitions other than the coalition of all the RANs. Hence, in case of imputations, it remains a possibility that a group of RANs collectively receive more channels by forming coalition among themselves by breaking the grand coalition i.e. ({ }) > ∑ (7)

=

+∑

×

∀ℎ ≠ ; ℎ, ∈ (17) In the above equation, the utility function is divided into two parts. The first part represents utility value of RAN because of its received spectrum at its desired spectrum band . Value of = 1 when at least one channel is allocated to RAN in its desired spectrum band otherwise, the value is 0. The notation δ signifies the gain in spectrum efficiency of a RAN with higher amount of spectrum allocation. Here, spectrum efficiency refers to the traffic carrying capacity (measured in Erlang/MHz) of the RAN. In a study, it has been found that value of is greater than 1 for GSM network of matured WSPs operating different cities in India [18]. The observation indicates that spectrum efficiency of RAN increases with higher spectrum availability. On the other hand,

∈

Therefore, the grand coalition does not become stable if, the above condition (7) is true. Hence, all the valid channel allocation solutions must satisfy the following condition in addition to the constraints indicated in equations (3-6): ∑ ≥ ({ }) (8) ∈

186


the second part of the utility function indicates the benefit of RAN from allocation of number of channels at inferior spectrum band ℎ compared to its desired spectrum band . Hence, the term signifies that at least one channel is allocated to RAN at inferior spectrum band ℎ. Further, indicates the inefficiency of channels at higher spectrum bands due to different propagation characteristics at different spectrum bands. The detail derivation of the utility function is available in [9].

Suppose RAN A is more spectrum efficient than RAN B and both have demanded 20 number of channels at 900 MHz band. Say, under first phase of our algorithm, RAN A is allocated 12 nos. of channels at 900 MHz and 4 nos. of channels at 1800 MHz whereas, RAN B is given only 5 nos. of channels at 900 band and 8 nos. channels at 1800 band. Therefore, RAN B receives less utility at 900 band due to fewer number of channels allocated to it. Alternatively, the higher efficient RAN (A) may be allocated more channels at the inferior band (higher band such as 1800 MHz) and less channels at the superior band (900 band). Under this arrangement, the RAN B will gain utility at the cost of RAN A that would lose its utility at the superior band. Naturally, if the gain of utility of the RAN B is higher than the loss of RAN A then, the overall objective value (PF) would be maximized. Following our earlier example, if we allocate 12 number of channels of 900 band to RAN B and RAN A is allocated 12 number of channels at 1800 band then, utility of RAN B at 900 band would increase at the cost of RAN A. On the other hand, gain of RAN B at 1800 band will be reduced as more number of channels are diverted to RAN A which, stands to gain at 1800 band. Hence, we have explored the latter option to search better quality of solution with respect to objective MPF.

B. H_MPF Algorithm For the objective MPF, we have designed a novel two-phased technique namely, H_MPF_PH1 and H_MPF_PH2 (Fig. 1). The first phase of the technique i.e. H_MPF_PH1 is a greedy method. Here, a RAN is identified in which, allocation of one channel would yield the maximum utility to the HetNet. Subsequently, the channel is allocated to the RAN considering the constraints (12-16). The process is repeated until all the available channels are allocated to the RANs. In H_MPF_PH2 technique, we have explored the possibility to improve the solution of the greedy technique. The logic behind our second phase technique is as follows. As per our utility design, the final utility of RAN is aggregate of its utilities at each band (vide (17)). If the availability of channels at the superior band (i.e. lower frequency band such as 900 MHz) is limited then, our initial algorithm tends to allocate majority of the channels of the lower band to the RAN having maximum spectrum efficiency. Naturally, the RAN that has relatively less spectrum efficiency receives even less utility as it is allocated very few channels in its desired band.

Hence, we present the second part of the algorithm as follows. For a particular spectrum band , RANs are arranged ̂ in the descending order of the term is ̂ . Initially, set ̂ formed with the RAN having maximum value of ̂ (i.e. RAN ̂). Subsequently, channels of RAN ̂ (i.e. channels allocated to RAN ̂ in the initial phase of the algorithm) are distributed among the rest of the RANs (ℕ − ) that have demand on band following constraints (12-16). As a result, the number of channel allocated to RAN ̂ at band becomes zero.

Next, we perform a neighborhood search based on the following method [9]. The neighborhood solutions are generated by applying a single change in the current solution. An example has been shown in Fig. 2 where, 4 channels ( ) are to be allocated among RANs A and B. Here, the number of possible neighborhood solutions will be ∑ ∈ ( − 1). In the figure, 4 nos. of possible neighborhood solutions are generated through Move (1, B) (which indicates that channel 1 is assigned to RAN B), Move (2, B), Move (3, A) and Move (4, A). It may be noted that solutions generated through Move (1, B) and Move (2, B) have same objective function values. This is because the total number of allocated channels to RAN A and RAN B are 1 and 3, respectively in both the cases. Similarly, Move (3, A) and Move (4, A) produce same objective function values. Hence, we have considered any one move between Move (1, B) and Move (2, B). The chosen move is referred as feasible move. Using this feasible move approach, we have restricted the number of possible to move to ∑ ∈ × ( − 1) which is dependant on number of RANs interested at the spectrum band only. Thus, we have saved the computational time of the algorithm. Fig. 1: H_MPF algorithm

187


Fig. 3: Solutions from different techniques

To understand the characteristics of the solutions of different solution techniques, we have taken a case in which, the number of available channels in 900 and 1800 MHz bands are 30 and 70, respectively. The solutions are presented in Fig. 3. To highlight the effectiveness of H_MPF_PH2, results of H_MPF_PH1 and H_MPF_PH2 are shown separately. It is apparent from the figure that Shapley value allocates less number of channels to the RAN with less spectrum efficiency (i.e. RAN F). Also, it allocates relatively more channels to the RANs that are efficient and have more amount of bid (i.e. RANs A and E). As a result, Shapley value performs worst among its peers in respect to our chosen objective MPF. The performance of SACA technique is better than that of Shapley value solution as it allocates more number of channels to the efficient RAN having higher channel demand at the lower band (e.g. RAN A is allocated 16 channels at 900 band). Also, SACA technique allocates fewer amounts of channels to the less efficient RAN at 900 MHz band (i.e. RAN D is allocated only 4 numbers of channels). In H_MPF_PH1, RANs with higher spectrum efficiency are always preferred. Naturally, RANs with higher efficiency receives more channels in this technique. The H_MPF_PH2 technique explores a contrarian approach. Here, the RAN with higher channel requirement and efficiency is assigned channels at the inferior spectrum band. For example, RAN A is allocated 24 channels at 1800 MHz in H_MPF_PH2 compared to H_MPF_PH1 in which, it is allocated 14 channels at 900 MHz band. Subsequently, the channels of 900 MHz band allocated to RAN A in H_MPF_PH1 are redistributed to rest of the RANs (C, D and E) in H_MPF_PH2 technique. As a result, higher value of objective function has been achieved (i.e. -2.7325 in HMPH_PH2 vis-à-vis -2.9244 in H-MPH_PH1). Now, we have studied the performances of different solution techniques by varying the ratio of available channels with respect to the total demand in Fig. 3 while keeping the bid amount and other characteristics of the RANs unchanged. In the figure, it can be seen that the performance of Shapley value is worst as it allocates less channels to the least efficient RAN. The SACA technique finds better solutions compared to Shapley value solution in all the cases as it explores more number of solutions through its evolutionary search process. The H_MPF_ PH1 offers slightly inferior solution compared to SACA. This is because H_MPF_PH1 allocates a channel to a RAN having maximum marginal utility among all the RANs. However, our H_MPF_PH2 technique provides the best results in all the cases as it is based upon the basic insight that overall PF of the HetNet improves if RANs with lower spectrum efficiency are allocated channels at the higher spectrum bands. We have observed that H_MPF_PH2 technique particularly yields better results in the scenarios where, availability of channels at the lower (say, 900 MHz)

Fig. 2. Neighborhood solution generation

Through the generated solutions by feasible moves, we investigate the existence of solution in which, the value of the objective function is more than that of found earlier. If there exists such solution, we store the solution and move to the next cycle where, size of H is increased by adding the RAN with the second highest value of the term ̂ ̂ . Subsequently, we repeat the earlier procedure of distribution of channels of the RANs belonging to set H to the rests to search a better solution. We perform the above tasks for each spectrum band considering constraints (12-16). Finally, the algorithm returns the solution found to have the highest objective function value. V.

SIMULATION RESULTS

We have studied the performance of H_MPF technique with that of Shapley value solution technique [8] and SACA [9]. We have executed the above solution methodologies in MATALB R2012a and run the simulation on PC environment. As a test case, we have considered 6 number of RANs namely, A, B, C, D, E and F; two frequency bands (i.e. 900MHz and 1800 MHz). The input parameters are listed in Table I. Here, we take a scenario where, threshold number of channels is already allocated to each RAN. Hence, we have assumed to be zero for simulation purpose. Therefore, is deemed to be the net demand of RAN . In the simulation, values of and for spectrum band are generated randomly. Assuming same antenna height in both the spectrum bands (900 & 1800 MHz), ratio of the radius of base station coverage in 900 MHz band (RBS900) and that of in 1800 MHz (RBS1800) is 1.675 as per Okmura Hata propagation model [19-20]. Hence, the value of (vide (17)) is assumed to be 0.597 (=1/1.675) in case, a RAN requests for channels at 900 MHz however, receives channels on spectrum band 1800 MHz.

RAT A B C D E F

Table I. Input parameter settings Required channel 30 1.9 50 1.3 25 1.4 30 1.1 40 1.7 25

1.2

Spectrum band 900 1800 900 900 900 1800

188


channel allocation decision to be taken on real-time is constrained by the computational time then, H_MPF_PH1 technique should be complemented by the OSM/WSP. The technique named as H_MPF_ PH2 should be considered when the availability of channels at the lower spectrum band is below 40% of the total demand at that band and the computational time is not a significant factor. Otherwise, traditional SACA technique that provides reasonably good solution in considerable computational time should be employed. REFERENCE [1]

[2] Fig. 3: Performance comparison of different solution techniques against objective of MPF [3]

spectrum band is very limited (up to 40% of the total demands) and relatively more number of channels are available at the higher spectrum bands (e.g. 1800 MHz). We have studied the computational time of the solution techniques as shown in Table II. It can be seen that Shapley value and H_MPF_ PH1 technique offers the solutions at very less time compared to the others. SACA technique takes maximum computational time for its explorative search process. However, average computational time of our proposed H_MPF technique (average of H_MPF_PH1 and H_MPF_ PH2) is less than that of SACA. Hence, we can conclude that our proposed technique H_MPF is more effective and efficient compared to the existing techniques such as SACA in real time settings. Sl. No. 1 2 3 4

[4]

[5]

[6] [7] [8] [9]

[10]

Table II. Input parameter settings Solution Technique Average Computational Time (in second) Shapley value 0.0074 SACA 0.1480 H_MPF_PH1 0.0624 H_MPF_ PH2 0.1488

[11]

[12]

[13]

To summarize, our proposed algorithm H_MPF_PH1 offers better solutions in terms of solution quality compared to Shapley value with similar computational time. Hence, H_MPF_ PH1 should be preferred over Shapley value as far as our chosen objective MPF is concerned. It can be found that SACA technique offers better result than H_MPF_PH1 technique. Nevertheless, computational time of SACA is far more than that of H_MPF_ PH1 technique. In a scenario where, the available channels at the lower spectrum band are less than 40% of the total demand, H_MPF_PH2 technique provides superior quality of solution than SACA technique while consuming similar amount computational time. VI.

[14]

[15]

[16]

[17]

CONCLUSION

[18]

In this work, we have addressed the problem of dynamic channel allocation to the WSP having HetNet under IEEE 1900.4 standard framework. When the available channels of the WSP are insufficient to cater to the demands of the RANs, the situation can be modelled as a bankruptcy game. If the

[19]

[20]

189

M. Peng, D. Liang, Y. Wei, and J. Li, "Self-Configuration and SelfOptimization in LTE-Advanced Heterogeneous Networks," IEEE Communications Mag., May 2013. B. Bangerter, S. Talwar, R. Arefi, and K. Stewart, "Networks and Devices for the 5G Era," IEEE Commun. Mag., vol. 52, no. 2, Feb. 2014. Q. Zhao and B. M. Sadler, “A survey of dynamic spectrum access,” IEEE Signal Processing Mag., vol. 24, no. 3, pp. 79-89, 2008. H. Harada et al., “IEEE dynamic spectrum access networks standards committee,” IEEE Communications Magazine, pp. 104-111, March 2013. A. Paul, M. Maitra, S.Mandal, S. K. Sadhukhanand D. Saha, “A dynamic spectrum management strategy based on cooperative bargaining framework,” Engineering Application of Artificial Intelligence, vol. 26, pp. 2495-2503, Nov 2013. B. Radunovi´c, and J. L. Boudec, “Rate performance objectives of multihop wireless networks,” INFOCOM’04, HongKong, 2004. I. J. Curiel, M. Maschier, and S. H. Tijs, “Bankruptcy games,” Zeitschrift far Operations Research, Vol. 31, pp. A 143-A 159. T. S. Ferguson, “Game theory text,” Mathematics Department, UCLA. M. Sengupta, A. Paul and M. Maitra, “Simulated annealing based solution for limited spectrum availability in composite wireless network,” in Proc. International Conference on Advances in Computing, Communications and Informatics (ICACCI), pp. 2571-2577, 2014. S. Sengupta and M. Chatterjee, “An economic framework for dynamic spectrum access and service pricing,” IEEE/ACM Trans. Networking, vol. 17, no. 4, pp. 1200-1213, Aug. 2009. S. Gandhi, C. Buragohain, L. Cao, H. Zheng, and S. Suri, “A general framework for wireless spectrum auctions,” in Proc. IEEE DySPAN, pp. 22–33, 2007. A. P. Subramanian, M. Al-Ayyoub, H. Gupta, S. R. Das and M. M. Buddhikot, “Near-optimal dynamic spectrum allocation in cellular networks,” IEEE DySPAN 2008, pp. 1-11, 2008. R. Amin et al., "Balancing Spectral Efficiency, Energy Consumption, and Fairness in Future Heterogeneous Wireless Systems with Reconfigurable Devices," IEEE Journal On Selected Areas In Communications, vol. 31, no. 5, may 2013. A. Paul, M. Maitra, S. Mandal, S. K. Sadhukhan and D. Saha, “A sustainable and collaborative strategy for dynamic spectrum management in next generation wireless networks,” Engineering Application of Artificial Intelligence, vol. 26, pp. 1620-1630, May 2013. E. H. Ong, J. Y. Khan and K. Mahata, "Radio Resource Management of Composite Wireless Networks: Predictive and Reactive Approaches," IEEE Transactions On Mobile Computing, Vol. 11, No. 5, May 2012. M. Sheng, J. Li, H. Li, W. Chen, J. Wen, “Resource Management Approach in Heterogeneous Wireless Access networks Based on IEEE1900.4 Architecture,” IEEE Wireless Information Technology and Systems (ICWITS), 2010. P.J.M van Laarhoven and E.H.L Aarts, “Simulated Annealing Theory and Applications,” Kluwer Academic Publishers, London, 1987. D. Manjunath, R.V. R. Kumar, R. Kathuria and R. Prasad, “Report on the 2010 value of spectrum in the 1800 MHz band”, Telecom Regulatory Authority of India, Jan. 2011. Y. Okumura , E. Ohmori , T. Kawano and K. Fukuda "Field strength and its variability in VHF and UHF land-mobile service", Rev. Elec. Comm. Lab., vol. 16, no. 9-10, pp.825 -873, 1968. M. Hata "Empirical formula for propagation loss in land mobile radio services", IEEE Trans. Veh. Tech., vol. 29, no. 3, pp.317 -325, 1980.