Power Allocation Optimization and Green Energy Cooperation Strategy for Cellular Networks with Hybrid Energy Supplies

Abstract

Energy harvesting is an increasingly attractive source of power for cellular networks, and can be a promising solution for green networks. In this paper, we consider a cellular network with power beacons powering multiple mobile terminals with microwave power transfer in energy beamforming. In this network, the power beacons are powered by grid and renewable energy jointly. We adopt a dual-level control architecture, in which controllers collect information for a core controller, and the core controller has a real-time global view of the network. By implementing the water filling optimized power allocation strategy, the core controller optimizes the energy allocation among mobile terminals within the same cluster. In the proposed green energy cooperation paradigm, power beacons dynamically share their renewable energy by locally injecting/drawing renewable energy into/from other power beacons via the core controller. Then, we propose a new water filling optimized green energy cooperation management strategy, which jointly exploits water filling optimized power allocation strategy and green energy cooperation in cellular networks. Finally, we validate our works by simulations and show that the proposed water filling optimized green energy cooperation management strategy can achieve about 10% gains of MT's average rate and about 20% reduction of on-grid energy consumption.

1. Introduction

The continuous development of wireless communication technology combined with the deep optimization of the network has resulted in an exponential growth of mobile data traffic, which has contributed to a rapid increase in the energy consumption and carbon emission of the information and communication technology (ICT) sector. It is estimated that, in 2020, the annual carbon emission of the ICT industry will be 235 Mto [1] and the electric energy consumption will be 414 Twh [2]. This not only leads to enormous network operation cost, but also places heavy burdens on grid and the environment. Thus, reducing power consumption of the infrastructure within the system is crucial to cellular networks.

Multiple studies are providing strong support that the use of harvested renewable energy reduces the carbon footprint and on-grid power consumption of the cellular networks [3]-[5]. That because exploiting renewable energy to power facilities of cellular networks can reduce on-grid energy consumption. Energy harvesting (EH) from renewable energy sources (e.g., solar, wind, vibration, ambient radio power, etc.) is emerging as a practically appealing solution to prolong the lifetime of energy-constrained wireless networks [6], [7]. At the same time the renewable energy is more economical and green than conventional energy that generated from fossil fuels etc. [8], [9]. By introducing the harvested renewable energy to power the next-generation cellular networks, potentially 20% of CO2 emission can be reduced [10].

To enjoy the environmental friendliness and low-cost of the renewable energy, EH is an ideal solution. However, owing to the space-time instability and non-uniformity of green energy, it may not guarantee sufficient power supplies for facilities in the cellular networks. Thus, to overcome the unreliability of the renewable energy source, hybrid cellular networks powered by multiple types of energy supply (e.g., the on-grid energy, the solar energy, and the wind energy), where EH and the grid coexist, will be an ideal solution [4], [11]. In such cellular networks, the system is powered by renewable energy if the green energy collectors can gather enough power; otherwise, the facilities switch to on-grid energy.

With the development of energy harvesting circuit, power transfer in wireless communication systems has drawn significant attention. [12] propose a new network architecture for radio frequency charging stations, overlaying with an uplink cellular network. The paper described an energy transfer wireless communication system, in which energy can be delivered via microwave power transfer (MPT). The capability allows the wireless devices to harvest energy from microwave signals for information processing and transmission. We can use MPT to charge mobile terminals (MTs) and get rid of power cords. The study extended to a harvest-then-transmit architecture for power transfer in wireless broadcast system [13]-[15], and energy cooperation systems [16]-[25]. The works are studied in different system model and scenarios. In [13] the multi-antenna access point first broadcasts wireless power to all users via energy beamforming in the downlink, and then, the users send their independent information to the access point simultaneously in the uplink using their harvested energy. [14] lets the source and relay harvest energy from the access point in the downlink and work cooperatively in the uplink for the source’s information transmission. [15] characterizes the spectral efficiency of an uplink radio frequency-powered macrocell network that the users transmit in the uplink while replenishing energy from base station in the downlink. [16] proposes an energy cooperation save-then-transmit scheme in a point-to-point wireless communications system, where energy is allowed to flow between the transmitter and receiver. A practical coordinated multipoint system with clustered base stations cooperatively communicating with mobile terminals is considered in [17] and [23]. [18] studies a model for energy cooperation between cellular base stations with hybrid power supplies, limited energy storages, and connected by resistive power lines for energy sharing. A cooperative mechanism for wireless energy harvesting and spectrum sharing in 5G networks is studied in [24], where secondary users harvest energy from both ambient signals and primary user’s signals and can act as relays. [25] analyzes a cooperative energy efficiency model in cellular network under different cooperative transmission scenarios, interference levels and wireless channel conditions. Assuming the future channel side information is available, the maximum throughput of point-to-point EH fading channels can be achieved by the directional water filling algorithm [26]. For wireless networks with hybrid energy sources, a power allocation algorithm was proposed in [27] to achieve energy cost minimization. [28] shows us an optimal power allocation policy to minimize the conventional energy consumption with delay-constrained data traffic requirement in heterogeneous cellular networks. Transmission protocols for cellular networks with hybrid power supplies have also been studied in [4], [9], [11], [28], [29]. However, as the harvested green energy is snatchy and sporadic, the design of the network setting is challenging. Energy management decision in EH networks should be based on both the channel side and the energy side adaptively. Thus more decisions must be made, and more information will be needed.

In the prior works, the network’s parameters can’t be changed dynamically and a suitable strategy can’t be chosen according to the variation of the channel state information (CSI) of the system. Moreover, in [12] all power beacons (PBs) have to deliver energy to MTs with the same transmission power, no matter how far the distance between PB and MT is. Much energy has been wasted, and due to the high energy transfer the addictive white noise is increased, both of them do harm to the performance of the system. To achieve agile response to the changes of the CSI, and overcome the shortage of non-adjustable transmission power of PB, in this paper, we introduce a dual-level control architecture to detect and manage the system dynamically. In the architecture, controllers are responsible for gathering information for a core controller (CC), and the CC grasps the entire system. After processing the collected data, the CC can adjust the parameters of the network according to the conditions of the system and send commands to PBs dynamically. We propose a power allocation optimization approach for cellular networks with hybrid energy supplies as shown in Fig. 1, where stations named PBs are deployed in an existing cellular network for recharging MTs via MPT wirelessly. Each PB is equipped with a solar panel and connected to grid by the CC, so PBs can be powered by hybrid power supplies. With the CC, PBs can get conventional energy from grid if the renewable power is insufficient, and achieve green energy cooperation.

Fig. 1.System model

In this work, we will develop effective power allocation optimization and green energy cooperation algorithm to improve the performance of cellular networks powered by hybrid energy sources. Our contributions can be summarized as follow:

The paper is organized as follows. Section 2 defines the system model. Section 3 formulates the uplink rate maximization problem. Section 4 proposes the WFOGECMS algorithm for the uplink rate maximization problem. Section 5 provides simulation results, and conclusions are presented in Section 6.

 

2. System Model

This section describes the definitions and assumptions of the network, and explains how the system deals with wireless power transfer and communication. The system model is illustrated in Fig. 1, and the notation is summarized in Table 1.

Table 1.Summary of notations

In our model, multiple base stations, PBs and MTs are independent with each other, and MTs have been divided into clusters corresponding to PBs. We focus on maximizing the uplink rate by designing the renewable energy and CSI aware scheme. In the dual-level control architecture, controllers are configured on base stations, gathering information relying on the base stations’ coverage of the area. Information collected by controllers consists of: the position of PBs and MTs, the amount of harvested green energy, each MT’s access status with PB and base station, and the CSI information. The CC, which connects to controllers (by its connection with base stations) and PBs by cables, responsible for complex calculation, and has a global grasp of the system. After analyzing the gathered data, the CC sends commands to PBs. According to the commands of the CC, PB can optimizes power allocation among its antennas, and green energy cooperation can be implemented by PBs locally injecting/drawing power into/from other PBs via the CC. The CC controls the direction of energy flow that exchange among PBs. Under the priority use of renewable energy, the shortage is supplemented by grid.

Furthermore, we assume para-static time-slotted model for both renewable energy and equivalent channels (there are two types of wireless channels in the network: the energy link and the data link, we collectively name them as the equivalent channel). The energy harvesting rate and the channel coefficient remain constant in each slot and may change from one slot to another. Without loss of generality, we choose one communication block as the reference time slot. We focus our study on one communication slot. In the following, we will present the energy model and the uplink cellular network model.

2.1 The Energy Model

We consider a system with N PBs, M MTs, S base stations. Each PB is equipped with H(H ∈ Z&H>1) antennas. Because of multiple antennas PB has, it can charge multiple target MTs simultaneously. PBs are distributed uniformly, and are independent with base stations and MTs. The number of MTs is no larger than the total number of PBs’ antennas, i.e., M ≤ NH. We denote the set of PBs and that of MTs as N = {1,...,N} and M = {1,...,M}, respectively.

As show in Fig. 1, PBs are powered by grid and green energy simultaneously. MT is powered by the nearest PB with energy beamforming. Base stations are powered by grid and just receive information from MTs. For convenience, we neglect the static energy consumption of PBs, base stations, MTs, controllers, and CC as [17] did. We consider the power management in PBī as shown in Fig. 2. BP uses renewable energy that is locally harvested or transferred from other PBs preferentially, otherwise, it consumes the on-grid power. The harvested renewable energy at PBī (i∈[1, N]) is denoted by Eī, and

Where is Q the sum harvested renewable energy by PBs. We assume that E1, E2, ...,EN are independent and uniformly distributed in [0, Q], each with an equal mean of Q/N. Note that the independent energy distribution may correspond to the case in which the PBs are in different position.

Fig. 2.Energy management schematics at PBī

The power of PBi consisted of four parts: the locally harvested renewable energy Ei, the drew/ injected green energy from/into other PBs via the CC ζEi+/Ei-, and the power from grid PiG. PBi injected its locally harvested renewable energy into other PBs, only when its green energy is surplus. Set the number of MTs served by PBi is Mi (Mi ∈ Z, M ≥ Mi ≥ 0), the power consumption of PBi can be given by

where ζ∈[0,1] is the efficiency coefficient of energy transfer, and Eq. (3) should satisfy the following constraints:

The proof of Eq. (4) is as follow.

By satisfying the constraints of Eq. (5) and Eq. (6), at least one of Ei+ and Ei- is zero. Because in real-time applications only three cases will be happening to PBi:

Where Ei - Pi > 0, Ei - Pi = 0, and Ei - Pi < 0 corresponding to three cases that the green energy of PBi overflowing, exactly right, inadequate respectively. Above all, we can get Ei+ Ei- = 0, and Eq. (4) is proved.

Without loss of generality, a typical MT is denoted by U0, the nearest PB that serves U0 is named the typical PB and denoted by PB0. The energy link between PB and MT is characterized by path loss but no fading. The number of MTs served by PB0 is M0 (M0 ∈ Z, M ≥ M0 ≥ 0), assume U0 is the k-th MT served by PB0 (1 ≤ k ≤ M0). The transmission power from PB0 to U0 is denoted by P0,k, the received power of U0 is GmP0,kd0,k-α, where d0,k is the Euclidian distance between U0 and PB0 in meter, Gm is the amplify factor in energy beamforming, α > 2 is the path-loss exponent in energy link. Considering the power sensitivity of EH [30], we obtain the boundary condition Eq. (9) of the expression. So the received power of U0 can be written as

where ε is the power sensitivity of EH. (If we consider the side-lobe gain of other PBs, when one PB out of PB0 denoted by PBj is working, it may have side-lobe gain to U0. But the amplify factor of the side-lobe Gs is almost one percent of Gm, meanwhile, the distance dj,k between PBj and U0 is larger than d0,k, so the interference of the side-lobe gain GsPj,k[max(dj,k, 1)]-α. is much less than the main-lobe that we ignore it).

We suppose the efficiency factor of energy conversion is equal to one (ideal). Thus the transmission power of U0 can be given as

2.2 The Uplink cellular network Model

In our model, base stations and MTs are distributed uniformly. MTs in the same cell are independent and the carriers with the corresponding base station are orthographical to each other, the environment is interference-limited. The nearest base station serves U0 is named the typical base station and denoted by B0. The date link is characterised by pass-loss and fading, and the channel fading f is assumed to follow exponential distribution (Rayleigh fading), i.e. f ~ exp(1). Let U0 sends data to B0 with power , the received signal by B0 is , where D0,0 is the Euclidian distance between U0 and B0 in meter, and β > 2 is the path-loss exponent. It can be expressed as

The interference comes from MTs out of UB0：

Where Mv is the v-th (v∈[1, M-nu0]) interfering MT to the data link between U0 and B0. UB0 is the MTs set that served by B0, nu0 is the number of MTs within UB0. Pv is the transmission power from PB to Mv, dv is the Euclidian distance between PB and Mv, Dv is the Euclidian distance between B0 and Mv.

The SINR of the data link is , in which σ2 is the background additive white Gaussian noise (AWGN) variance. The rate of the data link from U0 to B0 is given by

Where B is the bandwidth of the base station.

 

3. Problem Formulation

In this section, we first introduce the uplink rate maximization problem, then solve the problem with the WFOPAS algorithm.

3.1 The Uplink Rate Maximization Problem

Considering the energy side, M MTs have been divided into clusters corresponding to PBs. The number of MTs served by PBi is . Assume the k-th (1 ≤ k ≤ Mi ≤) MT served by PBi is served by BSj，with Eq. (13), The sum-rate of the i-th MT cluster is . We will use the Uplink Rate (UR) as the performance metric, which is the MT’s average rate and defined as

To minimize the on-grid energy consumption, consider the situation that the system powered by renewable energy only, and green energy cooperation has not been implemented yet. Consequently, the uplink rate maximization problem can be formulated as

In P1, the peak summed transmit power constraint and the power non-negativity constraint are imposed in Eq. (16) and Eq. (17), respectively.

Since M, M, B, nuj, Mi is given, when the sum-rate of each cluster is maximum so as the sum-rate of the network. Thus, P1 can be decomposed to maximize the rate of each cluster, so we equivalently introduce a modified version of P1 as follows:

s.t. (16), (17)

and the version of P2 is expressed as:

s.t. (16), (17)

Hence, let i = 1,2, ..., N P1 is the sum of P2, any feasible solution for P2 is also feasible for P1.

3.2 The WFOPAS Algorithm

We solve problem P2 for the optimal power allocation solution among each cluster’ MTs. It can be verified that P2 is a convex problem since the objective function is concave. At the same time, all the constraints are linear. Thus, the optimal solution satisfies the KKT conditions, and the Lagrange duality method can be applied to solve this problem optimally [31].

Let φi ≥ 0, bi,k ≥ 0, ∀i ∈ [1,N], ∀k ∈ [1,Mi] be the Lagrange operators associated with each of the power constraints of P2 given in Eq. (16) and Eq. (17). Then, the Lagrange function of P2 can be defined as

where φi, bi,k meet the mutual slack conditions, which is

thus, we can solve P2 by solving Eq. (20) equivalently. Solving ∂Li / ∂Pi,k = 0, the optimal power allocation can be expressed as a function of Lagrange multipliers

When the system powered by renewable energy only, and green energy cooperation has not been implemented yet. The green energy has been used up to maximize the sum-rate of the network, it is easy to obtain PiG = 0 and , so . With Eq. (21) the Lagrange operator φi can equal to any nonnegative value. With Eq. (17) and Eq. (22), it can be easily verified that bi,k = 0. Thus, the power distribution reaches a fixed level 1/φi, the optimal solution can be obtained as

And how to obtain is given in the WFOPAS algorithm as shown in Table 2.

Table 2.Algorithm 1: The WFOPAS algorithm for Solving Problem P2

As a result, we obtain for all MTs finally, that maximize the sum-rate of the network. Thus P2 have been solved completely.

In the following, we compare the performance of the proposed WFOPAS algorithm and the average power allocation strategy (APAS) used in [12] in a system powered by green energy only, and no green energy cooperation among PBs. During comparing, we consider a two-cell simple network. We set N = 2 PBs, S = 2 base stations, M = 3 MTs. The path-loss exponents are α = 3 and β = 3. We assume the background AWGN variance in every data link are the same and equal to -110 dBm. The amplify factor in MPT beamforming is Gm = 10 [12]. For simplification, the bandwidth of base station B = 1 Hz. The radius of the small cell cellular network is R = 500 meter, simulating 100000 times.

In Fig. 3, it is visible that the MT's average rate increases with the total harvested renewable energy. Under different SINR threshold ϑ, the performance with WFOPAS is much better than that of APAS. The MT's average rate with WFOPAS is improved about 7%~20% than APAS. This behavior can be explained as follows. With WFOPAS, the transmission power in every PB can be allocated adaptively based on the CSI. In each cluster, more power is distributed to the channels whose CSI are good, and less energy is allocated to the MTs with poor channel conditions. But with APAS, power allocation can’t be optimized.

Fig. 3.Average rate VS The total harvested renewable energy

 

4. The Water Filling Optimized Green Energy Cooperation Management Strategy

So far, we have solved P2, and the WFOPAS algorithm can optimize the MT’s average rate of the system. In the following, we focus our study on maximize the rate of the system by use green energy cooperation and WFOPAS algorithm jointly in cellular networks, which powered by green energy and grid together. We will show how to implement green energy cooperation with low complexity.

The difference between WFOPAS and WFOGECMS is focused on two aspects: one is whether adopt green energy cooperation, and the other is the range of the water filling optimization algorithm applied. In the WFOPAS algorithm, green energy has no cooperation, the algorithm is applied in each MT cluster corresponding to PB, and it explains how to allocate Ei among antennas within PBi. In the WFOGECMS algorithm, green energy cooperation carried out via the CC, and the algorithm is implemented in all MTs of the network at the same time. The WFOGECMS algorithm explains how to allocate Q among all MTs.

4.1 The WFOGECMS Algorithm

Green energy cooperation can be implemented by PBs with extra/ inadequate renewable energy locally injecting/drawing power into/from other PBs via the CC. The direction of the energy flow between the CC and PB is controlled by the CC. The harvested renewable energy is efficiently utilized to maximize the rate, and energy shortage is supplied by grid. The WFOGECMS algorithm is expressed in Table 3. Whenever the CSI or ∀Ei changed, the CC recalculates and adjusts the parameters of the system immediately. And the rate of the system can be optimized by the WFOGECMS algorithm.

Table 3.Algorithm 2: The WFOGECMS Algorithm

4.2 The Mathematic Model of WFOGECMS

In the hybrid cellular network, renewable energy and grid co-exist to power PBs. The CC can get dj (the distance between MTj and its nearest PB), Dj (the distance between MTj and its nearest base station BSi) j ∈ [1,M], and the total harvested green energy Q from the system. The renewable energy is utilized efficiently to maximize the rate of the system. Consequently, the uplink rate maximization problem can be formulated as

where Pj is the transmission power from PB to MTj. We solve problem P3 for the optimal power allocation solution. Similar to P2, the Lagrange duality method can be applied to solve P3 optimally.

Let φ≥0, bj≥0, ∀j∈[1,M] be the Lagrange operators associated with each of the M power constraints of problem P3 given in Eq. (27) and Eq. (28). Since M, B, nui is given, the Lagrange function of P3 can be defined as

where φ,bj meet the mutual slack conditions, which is

Thus, we can solve P3 by solving Eq. (29) equivalently. Solving KKT conditions Lagrange operators ∂L / ∂Pj = 0, the optimal power allocation can be expressed as a function of Lagrange multipliers

When the system powered by renewable energy only, and the green energy has been used up to maximize the rate of the network, it is easy to obtain , with Eq. (30) the Lagrange operator φ can equal to any nonnegative value. With Eq. (28) and Eq. (31) it can be easily verified that bj = 0. Thus, the power distribution reaches a fixed level 1/φ, the optimal solutions can be obtained as

How to obtain is given in the WFOGECMS algorithm as shown in Table 4.

Table 4.Algorithm 3: The WFOGECMS Algorithm for Solving Problem P3

As a result, we have finally obtained for all MTs, which makes the average rate of network maximal. Thus P3 have been solved completely.

According to Table 3, the CC calculates each PB’s total transmission power Pi by summing Pj within the same cluster, then jointing Ei calculates ΔEi. After that, commands are sent to PBs. At last, PBs inject/draw redundant/scanty green energy into/from other PBs via the CC to implement green energy cooperation.

In the following, we will show the performance of the water level λ in the WFOGECMS algorithm. The results come from calculate. Suppose that there are N = 10 PBs, and S = 10 base stations in the system. The other parameters are the same as before: α = 3, β = 3, Gm = 10 [12], R = 500 meter, B=1 Hz, simulation 10000 times. The background AWGN variance in every data link are the same and equal to -110 dBm. In Fig. 4, the performance of λ = 1/φ is presented, which is the water level in the WFOGECMS algorithm. It is visible that λ increases with Q, and decreases with the density of MTs. The smaller the density of MTs and the larger the total harvested green energy, the higher the water level will be.

Fig. 4.The performance of λ VS the density of MTs and Q

 

5. Simulation Results

In this section, we will verify the theoretical results derived in Section 4 and evaluate the performance of the proposed WFOGECMS algorithm by simulation.

First, we consider a cellular network with N = 10 PBs, S = 10 base stations, M = 30 MTs unless otherwise specified. In addition, α = 3, β = 3, Gm = 10 [12], ϑ = 1, ζ = 0.9 [17], R = 500 meter. For simplification, B=1 Hz, and the background noise in each data link is -110 dBm, simulation 1000000 times.

Fig. 5 shows the MT’s average rate of the system. With the practical energy transfer efficiency ζ = 0.9, the average rate increases with the total harvested renewable energy. It is visible that, with the proposed WFOGECMS algorithm, that ‘Joint WFOPAS and green energy cooperation’, the average rate is the best among‘Joint APAS and green energy cooperation’ scheme and ‘No green energy cooperation, WFOPAS only’ scheme that given in section 3.2. It is also observed that, when Q is little (smaller than 100 watt), the gap between them is large; with the increase of Q, the gap is getting smaller. It means when the total harvested renewable energy is big enough, the improvement brought by WFOGECMS algorithm is no longer obvious. Thereby, when the total harvested renewable energy is not enough, the WFOGECMS algorithm can bring a significant improvement. Third, the employment of green energy cooperation can bring up to 10% promotion in the MT’s average rate than without it, while WFOPAS can get 5.4% improvement than APAS, the benefit with green energy cooperation is more obvious. And we can obtain that the gain of green energy cooperation is more dominant over that of WFOPAS, and “Joint WFOPAS and green energy cooperation” strategy, whose name is WFOGECMS, is optimal.

Fig. 5.The average rate vs Q

Next, we evaluate the performance of the proposed WFOGECMS algorithm and another suboptimal average power allocation strategy with green energy cooperation (APASGEC) scheme in a practical system with N=10 PBs, S=10 base stations, and M=30 MTs. In our simulation, we assume that all PBs are deployed in different position with different renewable energy generation capacities. As shown in Fig. 4, the water level λ in the water filling optimization algorithm is described. In Fig. 6, we show the network’s sum-rate and on-grid energy consumption performance. It is observed that the WFOGECMS algorithm with ζ = 0.9 is better than APASGEC scheme; as the total harvested green energy Q increases, especially after Q=200 watt, the gap between them becomes smaller quickly. Which shows little gain in the sum-rate and approximate 20% decrease of on-grid energy consumption with WFOGECMS. The improvement of the performance mainly because the CC adjusts the renewable energy allocation among PBs dynamically and employ green energy cooperation adaptively.

Fig. 6.The sum-rate and grid power consumption VS the total harvest renewable energy

Finally, In Fig. 7, we show the sum-rate performance VS the density of MTs with N=10 PBs, and M=30 MTs. It is observed that, firstly, when S=10 base stations, with the density of MTs’ increases, the sum-rate performance in WFOGECMS and APASGEC almost the same, no matter the total harvest renewable energy Q=300 or 3000 watt. That means, when Q is big enough, the increase of Q cannot bring the sum-rate lift. Secondly, when Q=300 watt, the sum-rate performance in the WFOGECMS algorithm with S=20 base stations outperforms APASGEC scheme, but almost the same when S=10 base stations, no matter what the density of MTs is. That means, the increase of S brings a significant increase in the sum-rate. That mainly because: when Q is big enough, power is no longer the most important factor, the bandwidth provided by base stations becomes a bottleneck of the sum-rate lift. And the CC adjusts the renewable energy allocation among PBs dynamically and employ green energy cooperation adaptively. So the increase in the number of base stations brings a significant increase in the sum-rate with WFOGECMS algorithm.

Fig. 7.The sum-rate VS the density of MTs

 

6. Conclusions

In this paper, we proposed a new algorithm joint power allocation optimization and green energy cooperation for designing cellular networks with hybrid energy supplies. By introducing the CC, complexity calculation can be uploaded, and the system parameters can be adjusted by the CC dynamically. Thus, the redistribution of green energy can be implemented easily. With the newly proposed core controller-assisted energy cooperation mechanism among PBs, we formulate the network rate maximization problem in an uplink system. By applying the water filling optimization arithmetic, we develop an efficient solution to optimize the transmit power allocation with low complexity. Furthermore, we show the rate gains by jointly exploiting WFOPAS algorithm and green energy cooperation in cellular networks. It is revealed that, under a practical energy transfer efficiency value, green energy cooperation is beneficial when the harvested renewable energy among PBs is distributed unevenly. And the proposed water filling optimized green energy cooperation management strategy can achieve about 10% gains of MT’s average rate and 20% reduction of on-grid energy consumption, the performance of cellular networks with hybrid energy supplies is improved.