Several previous studies paid attention to BS topology management considering traffic redistribution during off-peak hours. In [11], the authors reported the cell zooming method based on the traffic demand fluctuations without investigating the data transmission quality. Then, in [12], the authors suggested a suboptimal BS on/off method that is distributively implemented to save overall energy consumption. However, this study ignored the dynamic fluctuations of traffic load and received signal quality. In addition, the authors in [13] pointed out transmission quality issues, where the transmission power and cell coverage adaptions are based according to the traffic load, but [13] analyzed for a single-cell case. Furthermore, authors in [14] proposed energy procurement between cellular operators as a single group, where the BSs share the traffic arrivals by turning off the BS into a sleep mode during low traffic hours.

To reduce the carbon footprints more effectively, the issues of energy harvested-enabled green BSs have been developed to incorporate the cell zooming scheme [15]. By taking advantage of hybrid energy supplies, BSs are designed to overcome the limitations of standalone RE-powered cellular networks. To mitigate the RE fluctuations among collocated BSs, the authors in [16] proposed an energy cooperation policy between BSs according to the traffic profile. The authors examined the joint optimization of RB allocation and BS on/off [17] and optimal packet scheduling [18] with hybrid supplies under user blocking constraints. The tradeoff between EE and traffic latency is investigated in [19], where the burst traffic patterns exhibit better EE performance enabling more BSs into a sleeping mode compared with uniform traffic distributions. All of the reported research works analyzed the system performances through sleep–wake up or cell zooming mechanisms by adjusting traffic load and transmission power. However, none of these works have analyzed the CoMP-based cell zooming technique considering traffic load fluctuations.

Authors in [20] presented the concept of cell breathing in the form of cell zooming, traffic offloading, and BS turning off for heterogeneous networks (HetNets) for EE BS topology management. Then, the authors extended this work contemplating the multi-metric UE-BS association approach with hybrid supplies [21]. A joint radio resource management and load balancing technique are proposed for HetNets based on traffic density-differentiated policies [22]. The authors in [23] provided insights on a coordinated small cell on/off schemes, providing offloading support for macrocell BSs. The authors proposed a green user association policy for cloud RAN network architecture based on the virtualized algorithms to measure energy availability information [24]. The authors in [14] studied energy-aware adaptive transmit power control mechanisms for energy cooperation-aided mmWave renewable-powered cellular networks. This work does not consider the stochastic nature of traffic diversity and coordination mechanisms. In the present study, we examined the EE and SE performance defining some performance metrics accounting for adaptive cell zooming policy and dynamic nature of traffic load.

Existing available research works focusing on improving transmission capacity, quality of experience, and SE ignoring the demand for energy-efficient aspects for mobile communications. To the best of the authors’ knowledge, this study first considers the CoMP-based cell zooming technique in the context of green cellular communications. Notably, the system only considers two BSs for user-BS connection algorithms under JT CoMP-enabled cell-adjusting technique. The major contributions of this study can be outlined as follows.

The rest of this paper is organized as follows. Section 2 presents the system model. Section 3 discusses performance analysis, which includes simulation setup and results. Finally, Section 4 concludes the study by highlighting the attained results.

A large-scale two-tier homogeneous cellular network comprising of N number of BSs (B) and baseband unit (BBU) pool is considered where B={B1,B2,…,BN} and covering an area A={A1∪A2∪…∪AN}⊂R2. Notably, A_i is the coverage area of BS B_i, ∀i∈{1,…,N}. All the BSs are assumed to be arranged in a remote radio head (RRH)-enabled hexagonal pattern with a 2/2/2 tri-sector. Notably, the OFDMA technique cancels out the intra-cell interference, and all the orthogonal RBs are reused among the BSs. The transmission process can take place from the baseband unit (BBU) to the RRH unit through an optical fiber cable and RRHs to users’ connection wirelessly. The BBU pool can decide on the data distribution-based coordination method. All the individual BS in the considered network is powered by an on-site solar energy harvester with an adequate battery bank and grid supply connectivity. A smart energy management unit prevents the storage device from overcharging/discharging and controls the key selection of primary energy sources. Notably, the grid electricity plays as the standby supply during the malfunctioning of green supply. Different JT CoMP-based cell zooming schemes are deployed to evaluate the best-received signal intensity to serve UEs, which is depicted in Fig. 1. The centralized control server (i.e., BBU pool) monitors the signal power, traffic demands, and UE-BS distances to perform zooming algorithms to maximize EE.

We consider log-normally distributed shadow fading channel model, where path loss in dB can be expressed as

(1) γ(d)=γ(do)+10αlog(ddo),

where d is the separation between the transmitter and receiver. γ(d0) is the free-space reference path loss in dB at a distance d₀, and α denotes the path-loss exponent.

The received signal power in dBm for kth UE at a distance d=di,k from ith BS B_i under JT CoMP transmission is given by

(2) PRi,k=Ptx1,k+Ptx2,k-γ(d)+Xσ,

where PTi,k is the transmitted power in dBm and Xσ is the shadow fading parameter modeled as a zero-mean Gaussian random variable with a standard deviation σ dB. Considering that two coordinated BSs serve a single UE simultaneously in JT CoMP transmission, two different transmit powers are considered in the calculation. However, the inter-cell interference can be expressed as

(3) Ik,inter=Pinteri,k=∑ m≠1,2Ptxm,k.

Notably, that intra-cell interference is zero because of the orthogonal condition. P_N is the additive white Gaussian noise power given by PN=-174+10log10(BW) in dBm, where BW is the bandwidth in Hz. The SINR Ψi,k at kth UE from BS B_i can be given by

(4) Ψi,k=PRi,kIk,inter+PN.

The solar energy generated at BS i∈{1,…,N} during time slot t is a random variable denoted by ri(t)∈[0,rimax], where rimax is the maximum available solar energy generation by B_i. However, the annual solar energy generation can be calculated by

(5) EPV=CPV×δr×gamma×ζ×365days/year,

where C_PV the rated PV array capacity in kW, Δr is the average daily solar radiation intensity in kWh/m2/day (4.65 as shown in Fig. 2), gamma is the dual-axis tracking factor (typically 1.3), and ζ is the derating factor that considers the effects of dust, wire loss, and temperature fluctuations on the solar power production.

The total estimated power consumption by a macrocell BS is function of traffic intensity (χ) and number of transceivers (N_TRX) is given by [3]:

(6) Pin={NTRX[P1+ΔpPTX(x-1)],if0<x<1NTRXPslpifx=0 .

P1=P0+ΔpPTX is the the maximum power consumption of a BS sector, and P₀ is the consumption at the idle state. Tabs. 1 and 2 show the parameters of BS power consumption. However, P₁ can be computed as

(7) P1=NTRXB10MHz(PBB′+PRF′)+PTXηPA(1-σfeed)(1-σDC)(1-σcool),

where PBB′ and PRF′ are the baseband and RF power consumption, respectively. B represents the system bandwidth, and ηPA defined the power amplifier efficiency. σDC, σfeed, and σcool represent the loss incurred by DC power conversion, feeder cable, and active cooling, respectively.

Energy consumption gain (ECG): ECG metrics account for the energy required to send a request in data transmission over a particular duration. ECG in green communications can be defined as the ratio of the energy consumption ratio metrics of the proposed system to the reference schemes under the specified network settings [24]. For instance, the network architecture without cell zooming is considered a reference baseline, and the cell zooming-enabled green-powered cellular system is the proposed scheme recognized as a more energy-efficient architecture. In other words, ECG quantifies the improvement of energy consumption with respect to the reference baseline scheme. A system with lower ECG is identified as more energy-efficient as this system consumes lower power to transmit the same amount of data transmission.

(8) ECR=ECRproposedECGref×100%.

Energy-saving index (ESI): ESI metrics measure the energy-saving gain introducing the cell zooming concept under different zoom-out ranges. ESI can be defined as

(9) ESI=Eρ-EμEρ×100%

where Eρ indicates the power consumption without incorporating the cell zooming technique and Eμ denotes the power requirement with cell zoom out for the entire cellular architecture respectively.

Load factor (δ): δ represents the ratio of the number of RBs occupied (RB_O) to the available RB (RB_T) in a given system bandwidth. Without loss of generality, we presume that one RB is occupied by a single active user, and the occupancy rate of the RB allocation is linearly proportional to the incoming traffic profile. As the number of demand request increases, the number of occupied RBs to satisfy the QoS also increases. For example, a cellular system operating at 10-MHz bandwidth has 50 RBs that can be allocated simultaneously at a particular time. Under the proposed framework, when the total available RBs are occupied in a particular BSs, the next available users can be associated with the neighboring BSs without a call service drop. However, δ can be expressed as

(10) δ=RBoRBT.

Energy efficiency (ηEE): ηEE cab be defined as the ratio of total achievable throughput per grid power consumption.

(11) Ttot=∑ k=1U∑ i=1NBlog2(1+SINRi,k),

(12) ηEE=TtotPgrid,

where Pgrid=∑ i=1NPin(i,t)-∑ i=1NPG(i,t) is the net grid energy consumption in BS B_i at time t, P_in(t) is the total power supply in BS B_i at time t, and P_G(t) is the harvested RE at time t. Moreover, the EEI is an equipment level metric that quantifies the enhancement of EE performance and is considered by the proposed framework as the reference system. EEI can be defined as follows

(13) ηEE=TtotPin.

The idea of cell zooming allowing new incoming users by adjusting the cell size depends on different user association mechanisms. The cell zooming technique has the potential to increase EE by turning off lightly loaded BSs in a cluster. The centralized control server in the BBU pool coordinates the traffic load and makes zooming decisions accordingly. The zooming server broadcasts zooming in when the traffic intensity is higher and vice versa. The provisioning of the proposed sleep mode mechanism governed by the cell zooming method substantially reduces additional energy dissipation over the conventional user-centric zooming technique. Fig. 4 depicts the cell zooming strategy. Moreover, for better understanding, Figs. 4 and 5 show the heuristic cell zooming algorithm for SINR-traffic-enabled JT CoMP-based schemes.

SINR-traffic-aware cell zooming: According to CoMP theory, a BS provides the maximum SINR, offering the best-received signal intensity to users. Moreover, lightly loaded BSs could either zoom out to serve more incoming users or zoom in to go into sleep mode for energy saving depending on the network settings. Under JT CoMP transmission, the BSs are sorted in a two-tier cluster in descending order of SINR and ascending order of traffic demand. In other words, a user is served jointly with two BSs that offer peak SINR and the BS having the closest distance. Notably, the closest BS does not always provide the best signal quality due to severe attenuation through shadow fading. The zooming procedure that incorporated this technique is known as the SINR-traffic-aware scheme. If the current traffic is less than the threshold limit (χ<χth), then the load should be released, and UE access must be reallocated. Moreover, the other BS that provides the topmost SINR value is selected. Fig. 5 shows the pseudo code of the proposed JT CoMP-based cell zooming algorithm. Similarly, the algorithms for other methods can be described.

We presume the same power profiles for all 19 cells in two tiers, and UEs are assumed to be distributed randomly over the given geographical area. In addition, every single active user that occupies one RB is considered throughout the simulations. Tab. 3 shows a set of input parameters for simulations in the MATLAB environment.

Fig. 6 depicts the variation of ECG with the cell zoom-out level for different UE-BS association schemes. All the user association methods follow a similar pattern to reach their minimum value with the increment of the zoom-out level. This finding signifies the improvement of energy expenses for the higher value of the zooming level. The distance-based user connection policy shows inferior performance compared with others. By contrast, the SINR-based scheme outperforms others. Notably, the ECG gap is apparently less significant between SINR and SINR-traffic-based methods. The reason is that the two BSs offering the best SINR and minimum existing traffic show almost close performance with the SINR technique at varying zooming levels. Notably, the SINR-based method resembles SINR–SINR, and distance–distance schemes can be represented as the distance only scheme; therefore, they are used interchangeably in this manuscript.

The energy-saving performance follows a similar pattern for all UE-BS association schemes with the zoom-out options as shown in Fig. 7. All the curves are up trending with the increment of zooming level, and the energy-saving gap is insignificant because of the similar ECG behaviors. The energy-saving curve demonstrates the opposite trend of ECG with the percentage of zooming. However, the SINR-based scheme has excellent energy-saving capability over the conventional distance-based cell zooming mechanism. Similarly, in the ECG graph, the separation between SNIR and SNIR traffic-based schemes lies close together comparatively because both methods offer the best signal quality. Thus, a hybrid SINR–SINR-enabled zooming system is the preferred choice among others.

Fig. 8 illustrates the comparison of throughput performance between JT CoMP-based cell zooming schemes with the conventional system, that is, non-CoMP-based systems. Notably, the throughput graph apparently follows the traffic demand shown in Fig. 3. According to the traffic load profile, the peak traffic arrivals occur at 8 PM when the maximum number of RBs is occupied. However, all the curves follow an identical pattern, and the distribution gap is substantial during peak traffic hours. Moreover, the gap is less significant for the SINR and SINR-traffic based schemes as explained beforehand. UE experiences better signal quality for the higher value of SINR resulting in higher throughput. The UE receives the best signal power under the SINR-based JT CoMP cell zooming scheme that exhibits better performance as shown in Fig. 8. Under the JT technique, the data received from two or more neighboring BSs to a particular UE lead to higher throughput performance.

Figs. 9 and 10 show a detailed comparison of EEI for different zooming schemes under 10 MHz. According to the definition of EEI presented in the performance metrics section, EEI is evidently a direct function of throughput. Notably, the SINR-based CoMP method provides the highest SE and EE, including the joint SINR enhancement and lower level of power consumption. The SINR-enabled zoom-out technique attains approximately 87.5% more EE than that of a non-CoMP-based tradition distance-aware system. Moreover, EEI shifts toward an upward direction with the increment of the zoom put level. As expected, the suggested framework follows the gradual improvement of EEI with the cell coverage level observed in Fig. 10. Thus, we can be safely inferred that the SINR CoMP-based method offers superior EEI performance.

Fig. 11 illustrates the EE performance with the system bandwidth for 2-kW installed solar capacity for each BS. In accordance with the definition, the scheme provides the peak throughput to uplift the EE. More RBs are allocated for the high system bandwidth. Therefore, EE curves start to increase in the upward direction to reach their maximum value identically as clearly seen from the figure. On the other hand, the SINR JT scheme offers the best utilization of RB in terms of maximum signal power to serve the associated UE.

Based on the aforementioned demonstration of Figs. 11, 12 illustrates the impact of cell zooming varying system bandwidth. A greater value of bandwidth and zooming level elevated EE performance significantly, that is, EE performance scaled with BW and zoom out level. The best system performance is obtained for the 20-MHz bandwidth and 100% cell zooming. However, the figure is derived for the SINR JT CoMP-based technique assuming that a 2-kW solar panel is installed in the BSs. Further analysis of the figure identifies that EE performance yields 63.7% better for the 20-MHz bandwidth than BW=5 MHz and achieved 64.8% enhancement for 100% cell zooming.

Fig. 13 presents the comparison of EE with the solar PV capacity demonstrating the performance of different user association schemes. As the solar capacity increases, the EE linearly scaled to reach the maximum point. Once again, the SINR-based scheme attains optimistic performance over all other schemes. For example, the SINR-based hybrid cellular system with 5 kW offers 32.4% more energy efficient than the PV=4 kW capacity. Moreover, the cell zooming level has a considerable impact on EE as shown in Fig. 10, which is obtained for a 10-MHz SINR-enabled hybrid system. The rising value of SPV and zoom out forces the EE to increase the desired quality satisfaction.

Fig. 14 depicts a comprehensive comparison of the ESI with a load factor. Fig. 14 is derived for BW=10 MHz and SPV=2 kW capacity, where the proposed scheme is compared with the non-zooming condition. In addition, Fig. 12 is drawn from the SINR–SINR-enabled method for the given network settings. All the curves follow a similar fashion to reach the least values with the increment of load factor. This figure again signifies the optimistic behavior of the SINR-based cell zooming method in which ESI is maximum. For instance, the ESI performance attains 17.5% better performance for the SINR-based cell zooming scheme compared with the non-zooming scheme when LF=0.5. In summary, the SINR-based JT CoMP cell zooming technique with a high bandwidth is the preferred choice in terms of energy-saving performance.