Energy Harvesting in Heterogenous Networks with Hybrid Powered Communication Systems

Transcription

1 Energy Harvesting in Heterogenous Networks with Hybrid Powered Communication Systems invited paper Ahmad Asharoa, Abdukadir Ceik, Ahmed E. Kama Iowa State University ISU, Ames, Iowa, United States, Emai: {asharoa, akceik, Abstract In this paper, we investigate energy efficient and energy harvesting EH in heterogeneous networks HetNets where a base stations BSs are equipped to harvest energy from renewabe energy sources, e.g., soar. We consider a hybrid power suppy of green renewabe and traditiona micro-grid, such that traditiona micro-grid is not expoited as ong as the BSs can meet their power demands from harvested and stored green energy. Therefore, our goa is to minimize the networkwide energy consumption subject to users certain quaity of service and BSs power consumption constraints. As a resut of binary BS seeping status and user-ce association variabes, proposed is formuated as a binary inear programming BLP probem. Two cases based on the knowedge eve about future renewabe energy RE statistics are investigated: i an onine knowedge case where future RE statistics are unknown, ii an offine knowedge case where future network s statistics are a priori perfecty estimated. A green communication agorithm based on binary partice swarm optimization is impemented to sove the probem with ow compexity time. Index Terms Energy harvesting, seeping strategy, binary partice swarm optimization. I. INTRODUCTION Energy saving is considered as one of the critica research probems that has been discussed in green communication over the ast few years. In recent years, energy efficiency has emerged as a major concern in the operation of ceuar heterogenous networks HetNets. Dynamic base station BS ON/OFF switching, aso known as BS seeping strategy, is shown to be highy usefu in reducing energy consumption of ceuar HetNets [1], [2]. The BSs are turned off during periods of ow traffic and the sma number of active users are offoaded to a nearby BS. As a resut, the power consumption of ighty oaded BSs can be reduced or competey eiminated depending on the seep state of the turned off BS. In [2], the impact of turning off macroce BSs on the energy efficiency of the HetNet is studied whie keeping the sma ce BSs active. Severa robust and efficient schemes for BS ON/OFF switching have been proposed in iterature [3], [4]. For instance, in [3], three different approaches for sma ce BS switching in HetNets are discussed. The ON/OFF status of the sma ce BSs is controed by either the detection of active users by the sma ce BSs, wake-up signas by the core network, or wakeup signas by the users. In [4], the authors have introduced two modes to cater for the short and ong ide periods of the users. It is shown that dense HetNets can be used t achieve higher capacity and performance whie simutaneousy reducing energy consumption. BS seeping strategies for singe tier ceuar networks are investigated in [5], [6]. In [5], an energy saving agorithm, that turns off the BSs one by one and measures the network impact considering the oad increments of the neighboring BSs, is proposed. In [6], an agorithm based on simuated anneaing search is shown to provide considerabe energy savings with insights on the depoyment of sma ce BSs. In [7], the authors presented a compete framework for a smart-grid powered LTE system based on evoutionary agorithms. Repenishing a new battery or recharging it using traditiona wired charging method is not feasibe aways e.g., sensors ocated on mountains or in forests. Therefore, energy harvesting EH has been considered as one of the most effective and robust soutions to protract the ifetime and sustainabiity of wireess networks [8]. Many promising practica appications that use EH nodes have been discussed recenty, such as, emerging utra-dense sma ce depoyments, point-to-point sensor networks, and far-fied microwave power transfer [9]. One of the imitation of the EH is the discontinuity of the power generation which affects reiabiity of the service. In [10], the authors consider hybrid powering BSs connected to different micro-grids that cooperate to minimize the tota power cost by optimizing their resources aocation. The authors assume that each micro-grid can purchase back-up power from the main grid when needed, in order to ensure a reiabe service to the the users. In this work, we consider a downink EH HetNets system where each BS is equipped to harvest from renewabe source. The contribution of this work can be summarized as foows Considering a hybrid power suppy sources consisting of green renewabe and traditiona micro-grid, such that traditiona micro-grid is not expoited as ong as the BSs can meet their power demands from harvested and stored green energy. Formuating an optimization probem aims to minimize the network-wide energy consumption over a certain time sots. The goa is to optimize the BS seeping and userce association variabes under BS s maximum power constraint, maximum BS s storing energy constraint, and user s quaity-of-service QoS constraint. Two cases depending on the knowedge eve about future RE generation are investigated: 1 The onine case: in this case, future RE generation statistics are unknown. a binary inear programming BLP probem is formuated to optimize the BS seeping status and user-ce association. 2 The offine case: this case assumes that the future statistics of the network are perfecty and estimated.

2 Proposing a ow compexity green optimization approach based on binary partice swarm optimization BPSO agorithm to find a near optima soution and compare it with the we known genetic agorithm GA [11]. II. SYSTEM MODEL In this paper, we investigate a time-sotted system of a finite period of time divided into b = 1,.., B, time sots of equa duration T b. A. Network Mode We consider a haf dupex downink transmission of threetiers HetNets consisting of a macroce tier, microce tier, and smace tier with a tota of L+1 BSs i.e., a singe macroce BS and L combined BSs of micro base stations MBSs, and sma base station SBSs. The ocations of a BSs are modeed by an independent homogeneous Poisson Point Process PPP. We consider a hybrid power suppy microgrid sources consisting of a green grid GG and a traditiona grid TG. The former uses renewabe sources to generate the eectric power, whie the atter uses cassica sources to generate the eectric power. Each BS is connected to the GG so that can provide hep in energy when needed. The GG has the abiity to purchase a back-up power from traditiona grid TG that is controed by contro unit CU when needed as shown in Fig. 2. Macro ce Micro ces Sma ces Green Grid GG Fig. 1: Contro Unit Traditiona Grid TG System mode of hyprid EH. Denoted U b as the tota number of users in the network during time sot b. We denote by Ū, the maximum number of users that can be served by a BS, where index = 0 for macroce BS and 1 for other BS tiers, such that Ū Ū0. These numbers refect the BSs capacities due to avaiabe number of frequency carriers and/or hardware and transmit power imitations. In order to avoid the cochanne interference, we assume that a the channes shared the spectrum orthogonay between the BS. Finay, we assume that the a user is served by at most one BS either macroce BS, MBS, or SBS. In genera, we assume that the communication channe between two nodes x and y at time sot b is given as foows h b xy = d α xy h b xy, 1 where d xy is the Eucidean distance between the nodes x and y, α is a pathoss exponent, and h xy,b is a fading coefficient with a coherence time sot T b sec. Without oss of generaity, a channe gains are assumed to be constat during T b. B. Base Station Power Mode Since the energy arrivas and energy consumption of the BSs are random and their energy storage capacities are finite, some BSs might not have enough energy to serve users at a particuar time. Under such scenario, it is preferred that some of the BSs are kept OFF and aowed to recharge whie their oad is handed by the neighboring BSs that are ON. On the other hand, dynamic base station switching-on/off can hep in ensuring power saving of HetNets by reducing the traditiona non-renewabe power consumption of BSs that have a heavy energy usage mainy during ow traffic period. Each BS can be set in either of two operationa modes: active mode AM and seep mode SM. The decision to togge the operationa state from one to another is taken centray i.e., the decision is taken by some centra entity based on the current oad offered to the network. In the AM, the BS is serving a certain number of users, thus, the BS radiated power can be expressed as P BS = U P,u, 2 that corresponds to the sum of the radiated power over a users U connected to a certain BS. In the SM, the BS consumes power equa to γ. The seep mode is a reduced power consumption state in which the BS in not competey turned off and can be readiy activated. Athough the BS is not radiating power in this mode, eements such as power suppy, baseband digita signa processing, and cooing are sti active. Therefore, the BS keeps consuming power uness it is in a state of compete shutdown. For simpicity, the tota power consumption of BS can be approximated by a inear mode as foows [12] { α P P = BS + β, for AM, 3 γ, for SM, where a corresponds to the power consumption that scaes with the radiated power due to ampifier and feeder osses and b modes an offset of site power which is consumed independenty of the average transmit power. We denote by ɛ b a binary matrix of size L+1 U. Its entries ɛ b,u is equa to 1 if user u is aocated to BS at time b and 0 otherwise.on the other hand, a dynamic ON/OFF switching mechanism is considered to turn off redundant MBSs and SBSs whenever it is possibe. More specificay, BS can be turned off during ow traffic periods and the sma number of active users are offoaded to nearby BSs A binary vector π b of size L 1 is introduced to indicate the status of each BS. Its entries π b equa to 1 if BS is in AM during time sot b and 0 otherwise. Note that in order to ensure that the users can not be connected to a BS in the SM, then, the foowing condition shoud be respected ɛ b,u π b, = 1,., L, u = 1,., U, b = 1,., B. 4

3 In this paper, we aways keep the macroce BS active i.e., π b 0, b = 1,.., B to ensure coverage and minimum connectivity in this typica HetNet i.e., one macroce BS surrounded by mutipe of MBSs and SBSs. In the case of mutipe macroce BSs covering a bigger geographica area, macroce BSs coud be turned off and ce breathing mechanisms can be empoyed to ensure connectivity [13]. C. Energy Harvesting Mode In this paper, we assume that each BS can harvest from RE in both AM and SM. We mode the RE stochastic energy arriva rate as a random variabe Φ Watt defined by a probabiity density function pdf fϕ. For exampe, for photovotaic energy, Φ can be interpreted as the received amount of energy per time unit with respect to the received uminous intensity in a particuar direction per unit soid ange. In genera, the energy consumption of the BS during time sot b can be expressed as U E0 b = T b ɛ b 0,uP 0,u + β 0, = 0 5 E b = T b π b [ α 0 U α ɛ b,up,u + β ] + 1 π b γ, 1, By using 4, we can re-write 6 as foows U E b = T b ɛ b,up,u + π b β + 1 π b γ, 1, α The harvested energy in BS and GG at the end of time sot b, are given respectivey by 6 7 H b = T b η ϕ b, 8 H b g = T b η g ϕ b g, 9 where η and η g are the energy conversion efficiency coefficient of the RE at BS and GG, respectivey, where 0 η, η g 1. Notice that the current stored energy in BS and GG depend on both the current harvested energy during sot time b and the previousy stored energy during previous sots. Therefore, he stored energy in BS at the end of time t is given by S b = [ S b 1 + H b E b E e ] +, 10 where E e is the eakage energy during T b. [x] + = max0, x. III. PROBLEM FORMULATION AND SOLUTION In this section, we formuate and sove optimay two optimization probems, based on the knowedge eve of the RE generation, aiming to minimize the network s energy consumption during the B time sots. The first optimization probem corresponds to the onine case where the mobie operator manages its BSs time sot by time sot without any prior information about the future RE generation. The second one corresponds to the offine case with fu information about the future RE generation where a the decisions variabes are simutaneousy optimized for the B time sots. The offine case is a not reaistic case. In this study, it is used as a benchmark scenario for comparison with onine case or as an approximation of the case where RE energy uncertainty is amost negigibe. The achievabe data rate of user u served by BS at a given time b is given by R,u b = og P,u h t,u 2 11 N 0 where N 0 is the noise power density. A. Onine Optimization Probem In this case, we assume that the mobie operator is not aware about the future RE generationi.e., ϕ b and ϕ b g are known during b ony. Therefore, optimization probem that aims to minimize the tota consumed energy at each time sot b is formuated as foows minimize π b,ɛb,u 0 E b c = subject to: E b π b, ɛ b,u S b π b 1, ɛ b 1,u 12 U ɛ b,up,u P, = 0,., L, 13 ɛ b,ur,u b R 0, u = 1,., U, 14 S b 1 π b, ɛ b,u + H b S, = 0,., L, 15 U ɛ b,u Ū, = 0,., L, 16 ɛ b,u 1, u = 1,., U, 17 ɛ b,u π b, = 1,., L, u = 1,., U, 18 where constraint 13 and 14 represent the maximum aowabe transmit energy of BS and user QoS, respectivey. Constraint 15 forces the tota energy stored in the battery of a BS during the time sot b to be ess than the battery capacity denoted by S. Constraints 16 and 17 to satisfy the backhauing condition and to ensure that each user is served by at most one BS, respectivey. Notice that, this optimization probem wi be soved at the beginning of each time sot. Hence, the optima soutions for such a probem can be determined using simpex method with Gurobi/CVX interface [14]. B. Offine Optimization Probem In this case, we assume that the mobie operator can perfecty predict the future RE generation ahead of time. This case can be considered as a usefu benchmark to compare with the onine case. Therefore, the objective function becomes the minimization of the tota energy consumption of the network during a B time sots.

4 minimize π b,ɛb,u 0 E c = subject to: B b=1 E b π b, ɛ b,u S b π b 1, ɛ b 1,u 19 U ɛ b,up,u P, = 0,., L, b = 1,., B, 20 ɛ b,ur,u b R 0, u = 1,., U, b = 1,., B, 21 S b 1 π b, ɛ b,u + H b S, = 0,., L, b = 1,., B, 22 U ɛ b,u Ū, = 0,., L, b = 1,., B, 23 ɛ b,u 1, u = 1,., U, b = 1,., B, 24 ɛ b,u π b, = 1,., L, u = 1,., U, 25 Notice that the constraints are simiar to the constraints except that they have to be satisfied for a time sots b = 1,.., B. The offine probem can be aso soved using simpex method with Gurobi/CVX interface [14]. C. Specia case The communication channe is assumed to be a bock fading channe with a coherence time T c second. Therefore, the scheduing and user-ce association can be assumed to be taken over a short time scae. Whie, the operationa state of the switching ON/OFF of the BSs can be taken over a ong time scae, where each ong time sot consists of mutipe short sots. Hence, the probem can be soved by optimizing ony ɛ b,u at the beginning of the short time sot and optimizing both π b and ɛ b,u at the beginning of the ong time sot. IV. LOW COMPLEXITY ALGORITHM The formuated BLP optimization probems given in Section III is considered as NP-hard probem due to the existence of the binary variabes, hence, we propose to empoy a metaheuristic agorithm, namey BPSO. The BPSO agorithm was firsty deveoped in 1997 by J. Kennedy and R. Eberhart [15]. The idea is inspired from swarm inteigence, socia behavior, and food searching by a fock birds and a schoo of fish. The main advantages are summarized as foows: i BPSO presents a simpe search process and is easy to impement with few parameters to manipuate e.g., such as the number of partices and acceeration factors for BPSO, ii it requires ow computationa cost attained from sma number of agents, and iii it provides a good convergence speed [16]. Then, we propose to compare its performances with the we known evoutionary GA [11]. A. Binary Partice Swarm Optimization BPSO The BPSO starts by generating N partices λ = [π1, 1.., πl B,.., ɛ1 1,1,.., ɛ B L,U ] ;n = 1,.., N of size L + L + 1U 1 for onine case soved for each time sot b and LB +L+1UB 1 for offine case to form an initia popuation S. Then, it determines the minimum energy consumed by each partice that satisfy the QoS by soving the optimization probem. Then, it finds the partice that provides the best soution for this iteration, denoted by λ best. In addition, for each partice n, it saves a record of the position of its previous best performance, denoted by λ n,oca. Then, at each iteration i, BPSO computes a veocity term V m n corresponding to eement m in λ as foows: V m n i = ΩV m n i 1 + ψ 1 i λ n,oca m i λ n m i + ψ 2 i λ best m i λ n m i, 26 where Ω is the inertia weight and ψ 1 and ψ 2 are two random positive numbers ψ 1, ψ 2 [0, 2] generated for each iteration i [15]. Then, it updates each eement i of a partice λ n as foows: λ n m i + 1 = { 1 if r rand < Ψ 0 otherwise. V n m i, 27 where r rand is a pseudo-random number seected from a uniform distribution in [0, 1] and Ψ is a sigmoid function for transforming the veocity to probabiities and is given as: 1 Ψ x = e x Agorithm 1 Proposed Soution using BPSO Agorithm 1: i = 1. 2: Generate an initia popuation S composed of N random partices λ n, n = 1 N. 3: whie not converged do 4: for n = 1 N do 5: Compute the corresponding consumed utiity function i. 6: end for 7: Find n j, i j = arg min E c n i i.e., n j and i j indicate n,i the index and the position of the partice that resuts in the E n c minimum energy consumption. Then, set Ec best = E n j c i j and λ best = λ n j i j. 8: Find i = arg min i E n c i for each partice n i.e., i indicates the position of the partice n that resuts in best oca utiity. Then, set λ,oca = λ n i. 9: Adjust veocities and positions of a partices using : i = i : end whie These steps are repeated unti reaching convergence by either attaining the maximum number of iterations or stopping the agorithm when no improvement is noticed. Detais of the proposed optimization approach are given in Agorithm 1. B. Genetic Agorithm The performances of the proposed BPSO agorithm is compared to those of the we-know GA. In our genetic based approach, we generate randomy N partices λ n, n = 1 N

5 of size L + L + 1U 1 for onine case soved for each time sot b and LB + L + 1UB 1 for offine case to form an initia popuation S. Then, it determines the minimum energy consumed by each partice that satisfy the QoS by soving the optimization probem. After that, the agorithm seects τ1 τ N strings that provide the minimum consumed energy and keeps them to the next popuation whie the N τ remaining strings are generated by appying crossovers and mutations to the τ survived parents. Crossovers consist in cutting two seected random parent strings at a correspond point which is chosen randomy. The obtained fragments are then swapped and recombined to produce two new strings. Then, mutation i.e., changing a bit vaue of the string randomy is appied with a probabiity p [17]. This procedure is repeated unti reaching convergence or reaching the maximum number of iterations. After soving the optimization probem, the tota cost of the non-renewabe energy consumed is equa to the cost of the energy consumed by a BSs that exceeding the avaiabe harvested energy stored at time b and given by C b = [ L [ E b S b 1 ] + S b 1 g ] + 29 where Sg b 1 is the stored energy at the GG at the end of time sot b 1. Therefore, the tota cost over mutipe time sots is given by C = B C b. b=1 V. SIMULATION RESULTS In this section, seected numerica resuts are provided to evauate the performance of the EH HetNets systems. Seected BSs transmit their messages periodicay every T b = 60 sec. A the fading channe gains adopted in the framework are assumed to be independent and identicay distributed i.i.d Rayeigh fading gains. The efficiency transmission and conversion ratios are set to η = η g = 0.3, respectivey. The target data rate user R 0, the number of MBSs and SBSs are 10 bits/s/hz, 4 and 8, respectivey, uness otherwise stated. The noise power is taken to be N 0 = N W, where N = 174 dbm/hz and W = 180 KHz. The power consumption parameters are seected according to the energy aware radio and network technoogies EARTH mode for macroce BS, MBSs, SBSs, are given, respectivey [12] as foows: α = {4.7, 2.6, 4} W and β = {130, 56, 6.8} W. The other power consumption parameters for MBSs and SBSs are given respectivey by γ = {39, 2.9} W. The maximum transmit power eves for the for macroce BS, MBSs, SBSs, are set, respectivey, to P = {46, 38, 20} dbm. At each BS, RE is assumed to be generated foowing Gamma distributions Γ20, 2, Γ12, 2, and Γ3, 1 for macroce BS, MBSs, and SBS, respectivey, where in Γx, y, x is the shape parameter and y and scae parameter. Whie for GG, RE is assumed to be generated foowing a Gamma distribution Γ25, 2. The tota stored energy at macroce BS, MBSs, and SBSs cannot exceed S = {50, 12, 6} KJ, respectivey, and the battery eakage is set to be E e = 10 mj every T b. The BPSO is executed with the foowing parameters: N = 20 and Ω [0, 1] is a inear decreasing function of the BPSO iterations expressed as foows: Ω = 0.9 t I, where I = 200 is the maximum number of iterations. Average energy cost [KJ] Without EH With EH Without seeping strategy With seeping strategy Tota number of users per sot U b Fig. 2: Average energy cost of B = 20 time sots versus tota number of users. Tabe I: MBSs and SBSs status during mutipe time sots Number of Active MBSs Active SBSs users per b m 1 m 2 m 3 m 4 s 1 s 2 s 3 s 4 U 1 = U 2 = U 3 = U 4 = U 5 = U 6 = 220 U 7 = U 8 = U 9 = U 10 = Fig. 2 pots the tota average energy cost, which is equa to C B, for B = 20 versus number of users U b, b = 1,.., B, for onine case. This figure investigates the impact of RE with two scenarios: 1 with the proposed EH i.e., hybrid of RE and TG energy, 2 without EHthe energy depends on the TG energy ony. It aso investigate the impact of the seeping strategy i.e., optimizing π on the system performance. we can see that the proposed scheme with EH and with seeping strategy offers a significant amount of energy saving switching over the other scenarios. It shoud be noted that the seeping strategy is very usefu speciay for ow traffic period with a considerabe energy cost gap. Indeed, for U b = 100 users, the average energy cost can be reduced by around 30% for the EH scenario by going from 13.5 KJ to around 9.5 KJ. However, this gap reduces when number of users increases. This can be justified by the fact that, when the number of users are reativey high, most of BSs shoud be in the AM in order to satisfy the user QoS. Tabe I confirms the seeping strategy resuts in Fig 2. In genera it can be noted that, activating the MBSs and SBSs essentiay depends on the traffic and BS s battery eve. For exampe, as shown in Tabe I, during ow traffic periods e,g., b = {2, 4, 7, 10} i.e., U 2 = 40, U 4 = 80, U 7 = 80, U 10 = 60, the seeping strategy activate some of BSs and keeps the others in the SM in order to harvest some energy. On the other hand, when the network is more congested e.g., during sots b = {3, 6, 8, 9} i.e., U 3 = 200, U 6 = 220, U 8 = 160, U 9 = 160, most of the BSs are in AM.

6 Energy cost [KJ] U b = 160 U b = 80 GA BPSO Agorithm Optima soution Number of time sots Fig. 3: Comparison between optima soution with BPSO agorithm and GA. Energy cost versus number of time sot Under the same setup of Fig. 2, Fig. 3 compares between the optima soution obtained using BLP with BPSO agorithm and the we known GA for different tota number of users U b = {80, 160}. It can be seen, that the BPSO achieves better performance than GA and cose to the optima soution in both ow and high traffic periods. We can notice that both agorithms are cose to the optima when the network is more congested. This can be expained, by knowing that during high traffic period, the network needs to keep most of the BSs in AM, hence, optimizing ony the association variabe i.e., ɛ. It is aso worth to note that optimizing π has more weigh in saving energy that optimizing ɛ due to the high vaues of offset power parameter β compare to the ampified power parameter α. Energy cost [KJ] Fig. 4: Number of users per sot U b Onine case Offine case Comparison between onine and offine cases. Finay, Fig. 4 compares the onine case to a benchmark case i.e., offine case. Fig. 4 pots the tota energy cost of the network for both cases versus different numbers of users. Since activating the BSs depends on their battery eves and the traffic status, the offine case can manage the avaiabe resources gobay and more efficienty. For exampe, during b = 7 i.e., U 7 = 80, the offine case consume more energy by forcing some BSs to be in SM and activate them where the network is more congested, i.e., U 8 = U 9 = 160. Athough it consumes more energy than the onine case, which is around 0.1 kj, when b = 7, the offine case saves more energy, which is around 0.6 kj, during the next two time sots b = 8 and b = 9. VI. CONCLUSIONS In this paper, we proposed a downink energy harvesting heterogenous networks using hybrid power sources. A the base stations are equipped with a harvested source and can get some energy from green grid or/and traditiona grid when needed. We formuated a binary inear optimization probem aiming to minimize the consumed energy over mutipe time sots. The probem is soved optimay and compared with two ow compexity agorithms. After soving the probem, we investigated, via numerica resuts, the behavior of the proposed scheme versus various system parameters. Finay, we discussed the effect of seeping strategy to the system average energy cost. REFERENCES [1] C. Liu, B. Natarajan, and H. Xia, Sma ce base station seep strategies for energy efficiency, IEEE Transactions on Vehicuar Technoogy, vo. 65, no. 3, pp , Mar [2] Y. S. Soh, T. Q. S. Quek, M. Kountouris, and H. Shin, Energy efficient heterogeneous ceuar networks, IEEE Journa on Seected Areas in Communications, vo. 31, no. 5, pp , May [3] I. Ashraf, F. Boccardi, and L. Ho, Seep mode techniques for sma ce depoyments, IEEE Communications Magazine, vo. 49, no. 8, pp , Aug [4] L. Faconetti, P. Frenger, H. Kain, and T. Rimhagen, Energy efficiency in heterogeneous networks, in IEEE Onine Conference on Green Communications GreenCom 2012, Sept [5] E. Oh, K. Son, and B. Krishnamachari, Dynamic base station switchingon/off strategies for green ceuar networks, IEEE Transactions on Wireess Communications, vo. 12, no. 5, pp , May [6] G. P. Koudouridis, H. Gao, and P. Legg, A centraised approach to power on-off optimisation for heterogeneous networks, in Proc. of the IEEE Vehicuar Technoogy Conference VTC Fa 2012, Quebec City, Canada, pp. 1 5, Sept [7] H. Ghazzai, E. Yaacoub, M. S. Aouini, and A. Abu-Dayya, Optimized smart grid energy procurement for te networks using evoutionary agorithms, IEEE Transactions on Vehicuar Technoogy, vo. 63, no. 9, pp , Nov [8] J. Xu and R. Zhang, Throughput optima poicies for energy harvesting wireess transmitters with non-idea circuit power, IEEE Journa on Sected Areas in Communications, vo. 32, no. 2, pp , Feb [9] H. Tabassum, E. Hossain, A. Ogundipe, and D. I. Kim, Wireesspowered ceuar networks: Key chaenges and soution techniques, IEEE Communications Magazine, vo. 53, no. 6, pp , June [10] M. B. Ghorbe, T. Touzri, B. Hamdaoui, M. Guizani, and B. Khafi, Joint user-channe assignment for efficient use of renewabe energy in hybrid powered communication systems, in Proc. of the IEEE Goba Communications Conference GLOBECOM 2015, San Diego, CA, USA, pp. 1 6, Dec [11] M. Mitche, An Introduction to Genetic Agorithms. Cambridge, MA, USA: MIT Press, [12] G. Auer, O. Bume, V. Giannini, I. Godor, M. Imran, Y. Jading, E. Katranaras, M. Osson, D. Sabea, P. Skiermark, et a., Energy efficiency anaysis of the reference systems, areas of improvements and target breakdown, INFSOICT EARTH Energy Aware Radio and NeTwork TecHnoogies, Tech. Rep, Dec [13] Z. Hasan, H. Boostanimehr, and V. Bhargava, Green ceuar networks: A survey, some research issues and chaenges, IEEE Communications Surveys & Tutorias, vo. 13, no. 4, pp , Fourth [14] Gurobi optimizer reference manua, Avaiabe [onine]: [15] J. Kennedy and R. Eberhart, A discrete binary version of the partice swarm agorithm, in Proc. of the IEEE Internationa Conference on Systems, Man, and Cybernetics. Computationa Cybernetics and Simuation, vo. 5, pp , Orando, FL, USA, Oct [16] M. Cerc and J. Kennedy, The partice swarm-exposion, stabiity, and convergence in a mutidimensiona compex space, IEEE Transactions on Evoutionary Computation, vo. 6, no. 3, pp , Feb [17] D. Beasey, D. R. Bu, and R. R. Martin, An overview of genetic agorithms: Part 2, Research topics, University Computing, vo. 15, no. 4, pp , 1993.