Towards Energy Efficient Load Balancing for Sustainable Green Wireless Networks Under Optimal Power Supply

The enormous growth in the cellular networks and ubiquitous wireless services has incurred momentous energy consumption, greenhouse gas (GHG) emissions and thereby, imposed a great challenge to the development of energy-efficient sustainable cellular networks. With the augmentation of harvesting renewable energy, cellular base stations (BSs) are progressively being powered by renewable energy sources (RES) to reduce the energy crisis, carbon contents, and its dependency on conventional grid supply. Thus, the combined utilization of renewable energy sources with the electrical grid system is proving to be a more realistic option for developing an energy-efficient as well as an eco-sustainable system in the context of green mobile communications. The ultimate objective of this work is to develop a traffic-aware grid-connected solar photovoltaic (PV) optimal power supply system endeavoring the remote radio head (RRH) enabled heterogeneous networks (HetNets) aiming to minimize grid energy consumption and carbon footprint while ensuring long-term energy sustainability and energy efficiency (EE). Moreover, the load balancing technique is implemented among collocated BSs for better resource blocks (RBs) utilization and thereafter, the performance of the system is compared with an existing cell zooming enabled cellular architecture for benchmarking. Besides, the techno-economic feasibility of the envisaged system has been extensively analyzed using HOMER optimization software considering the dynamic nature of solar generation profile and traffic arrival rate. Furthermore, a thorough investigation is conducted with the help of Monte-Carlo simulations to assess the wireless network performance in terms of throughput, spectral efficiency (SE), and energy efficiency as well under a wide range of design scenarios. The numerical outcomes demonstrate that the proposed grid-tied solar PV/battery system can achieve a significant reduction of grid power consumption yielding up to 54.8% and ensure prominent energy sustainability with the effective modeling of renewable energy harvesting.


I. INTRODUCTION
Over the last decade, the high-tech advancement in mobile devices together with the increasing popularity of intensive multimedia applications has accelerated the astounding rise The associate editor coordinating the review of this manuscript and approving it for publication was Cunhua Pan . of data demand. A recent study on global mobile data traffic forecasted that the unique mobile subscriber's number would increase to 6.2 billion by the end of 2023 [1]. To satisfy the higher demand of users, telecom operators are increasingly deploying more BSs and hence, leading to an increase in a vital portion of the energy utilization in mobile communication. The international energy agency (IEA) in its VOLUME 8, 2020 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ latest global energy projection, the World Energy Outlook 2018 (WEO2018), has reported that global energy demand is expected to grow by about 27%, or 3,743 million tons oil equivalent worldwide from 2017 to 2040 [2]. Authors in [3] mentioned that the annual energy consumption by the mobile industry has increased from 219 TWh in 2007 to 519 TWh in 2019, and it is speculated that the demand rises at a yearly rate of 10%. Consequently, the electricity bill of telecom networks globally amounts to above ten billion per year. It is expected that the GHG outflow emitted by the information and communication technology (ICT) sector will increase from 1700 quintals in 2014 to 235 quintals by 2020, with 51% of CO 2 emissions generated from the telecom industry [4], [5].
Another downside, BSs in the radio access networks (RAN) contribute almost 60-80% of entire energy expenditure and place a heavy burden on the electrical grid (EG) supply to deal with the question of sustainable power issue [6], [7]. In response to the request for reducing grid energy consumption and environmental aspects, cellular operators are widely focusing on green communications concentrating on the environmental implications. Recently, access networks in green radio communication systems supplied by sustainable and reliable renewable energy have brought a remarkable result, which has drawn great attention in both industry and academia [8]- [11]. For instance, Huawei [10] designed a green BSs under 3G network in Bangladesh powered by solar PV to reduce the sole dependency on the electric grid.
It is commonly believed that an electrical grid system is enough as a power source for BS where the grid connection exists. The use of grid energy emits harmful carbon contents which significantly causes air pollution, the rise of world temperature, and damage of the ozone layer. Furthermore, the lower efficiency and routine operation and maintenance make the system much less reliable and resulting in outage during grid failure. Being motivated by the aforementioned issues, the utilization of RES such as solar PV at the BS site providing operators the option to develop green access cellular networks through the deduction of harmful greenhouse gas discharge. Numerous studies [12]- [15] have been carried out for curbing down the energy costs by adopting renewable energy sources to power large-scale cellular network. Still, research in this field is not saturated. Most of the authors have not analyzed the cost-effective solution for long term evolution (LTE) cellular networks with real traffic patterns.
Nowadays, modern cellular systems are required to manage the excessive number of linked equipment (millions per macrocell), excessive bit rate ( tens of Gbps), improved spectral efficiency, enhanced energy efficiency, strong reliability (∼99.999%), intensely low latency (∼1 ms) for data-intensive, mission-critical and time-critical applications [16], [17]. Consequently, investigations for finding ways to develop a more competent cellular network have been carried out in recent times [18]. Typically conventional macrocell architectures are designed to cater to larger spectral areas that may often fail to attain the desired throughput to support seamless broadband service to uplink users. Besides, users move far away from the macro BS and often cause severe inter-cell interference. The weak indoor penetration and the appearance of dead spots considerably reduce the indoor coverage under the traditional macrocell scenario. As a potential solution to overcome these issues, heterogeneous networks (overlay of small cell networks like micro/pico/femto cells with macro BS) offer significant network performance leap by means of higher network capacity and better coverage [19]- [23]. HetNets are the multi-tier cellular network, where the macro BSs act as a mother BS and provide a uniform coverage area. On the other hand, the micro and pico/femto cells act as supporting BSs which consume low energy due to the smaller amount of coverage area. Hence, the installation of HetNets potentially improve spatial reuse, and the capacity in the hotspot can minimize the coverage hole of the cellular network in dead spots which can save a significant amount of energy as well as extend the coverage while ensuring seamless connectivity and mobility in future generation cellular systems [24].
In this trend, load balancing has appealed much consideration as a talented way out for higher resource allocation, better system enactment, and reduced operational expenditure as well. Load balancing is an efficient technique for adjusting the load and relieving the congestion among heterogeneous networks in the upcoming 5G networks. With the introduction of load balancing techniques, a cellular network can gain in different ways, such as the effective exercise of frequency bands, improvement in coverage for cell-edge users, and boosting in overall network throughput [25], [26]. However, the concept of load balancing can be applied for improving the spectral efficiency of mobile access set-ups. On the other hand, the load balancing mechanism flexibly adjusts BS coverage depending on experienced traffic density and enhances the service quality of an entire cellular network through balancing the traffic among surrounded BSs in a cluster. Authors in [27] proposed a distributed user association based load balancing algorithm to balance the load among the neighboring BSs. Reference [28] introduces an empirical load balancing algorithm that can be used to optimize the range of the HetNets. Therefore, the load balancing techniques are used for redistributing the load among the active BSs and can benefit in many ways, such as efficient utilization of frequency bands, enhancement in coverage for cell-edge users, and increment in overall network throughput. In this paper, the concept of load balancing is employed for improving the energy efficiency performance under optimal power supply scenarios. Throughput aware load balancing based on the effective data rate of users [29] and energyaware resource management [30] under hybrid power supplies is comprehensively analyzed focusing on minimum grid power consumption. Instead of the conventional throughput maximization technique, authors [31] proposed a joint user association and power control algorithm for load balancing in modern telecom networks. On the other hand, a cluster-based load balancing policy is adopted to maximize the network throughput (achieved 6.42% better performance compared to the non-load balancing technique) in reference [32].
In a cellular network, the non-uniform traffic density due to the tempo-spatial variations of mobile subscribers has incurred an unnecessary use of energy [32]- [34]. To overcome this problem, telecom operators are dedicated to developing green mobile communication, which adaptively adjusts the cell size according to the user profile. This condition is supported by a conceptual strategy called ''cell zooming'' to optimize the energy consumption of base stations according to the user condition. In summary, cell zooming is the increment or decrement of cell size from its originally designed size under the traffic condition of the cell. This reduces the traffic load in its neighboring cells and some cells can even go switch-off, which will result in energy saving. Recently, various techniques have been proposed by turning off the BSs to minimize the energy consumption of the cellular network [35]- [37]. For saving energy, LTE proposes switching off the evolved node base station (eNBs) at low-traffic times [38]. Reference [37], proposed a cell zooming concept to minimize energy consumption by integrating BS switching. Authors in [39], [40] proposed a multipoint coordinated cell zooming concept, where the actual size of the cell does not change but the cell size is virtually extended. It is commonly known as virtual cell zooming or virtual MIMO system. A relay assisted cell zooming concept is presented in [40], [41], where a moveable or fixed relay station is placed at the edge of the original cell to extend the coverage area. References [42], [43] introduced a physically adjustable cell zooming technique, where the cell size is adjusted by changing the different parameters such as antenna height, antenna tilt, transmitted power, etc. Energy efficiency and spectrum aware cell zooming are conducted to tackle the upsurge in traffic load in [44], [45]. The wireless network performance enhancement contemplating adaptive cell zooming and sleeping mode mechanism for ultra-dense HetNet with green power supply is pointed out in [46], [47]. Reference [48] proposed a data-driven energy minimization problem implementing a cell zooming algorithm leveraging the tradeoff between blockage ratio and energy saving issue. Reference [49], introduces a theoretical optimization method for saving energy by switching off the BSs. Implementation of the cell zooming concept for the heterogeneous network is presented in [37] without taking into account actual traffic arrival rate and average energy savings for green cellular infrastructure. However, some of the schemes are failed to capture the load-dependent power utilization in BSs resulting in overestimations. Moreover, many of them presented either very basic algorithms ignoring the actual locations of users, or no algorithm at all.
The prime concern of this work is to develop a traffic-aware energy-efficient and sustainable cellular network that inherently minimizes the pressure on the public utility grid and reduces the carbon contents. The optimization of renewable energy utilization is done using a HOMER software taking into account different types of constraints. Moreover, Matlab based Monte-Carlo simulations are carried out for measuring the performance metrics of the wireless network under a wide range of system scenarios. The key contributions of this work can be summarized as follows: • To propose a grid-tied solar PV/battery system for powering the LTE macro/micro-cellular base stations with guaranteed continuity of services. Notably, the solar PV panels play as a primary supply whereas the electrical grid system supplies energy as a secondary source for operating BSs during the shortage or absence of solar PV system. Moreover, the battery bank ensures guaranteed continuity of supply by supplying the backup power.
• To examine the effectiveness of the proposed system in terms of optimal system architecture, techno-economic criteria, and carbon footprints considering the intermittent behavior of solar intensity and traffic rate. In addition, the impact of BS transmits power, and different LTE system bandwidth on the performance metrics are critically illustrated.
• To evaluate the throughput, energy efficiency, and spectral efficiency performance of the wireless network taking into account the inter-cell-interference and wireless channel model including shadow fading.
• A load balancing technique is developed for maximizing the SE and EE as well through optimal utilization of green energy, while the entire system is still envisioned by hybrid power sources as outlined above. After that, a heuristic algorithm for load balancing mechanism and resource block allocation technique is proposed. The rest of the manuscript is outlined as follows: Section II contains the system architecture and modeling of the proposed system. Section III illustrates the cost modeling and optimization formula. Section IV demonstrates the performance analysis with insightful comments. Section V represents the conclusions of the manuscript highlighting the key outcomes.

II. SYSTEM ARCHITECTURE AND MODELING
The schematic diagram of the solar PV/EG/battery-powered cellular network system is shown in Fig. 1. An LTE macro BS usually consists of transceivers (TRXs), power amplifiers (PAs), a radio-frequency (RF) unit, a baseband (BB) unit, a DC-DC power supply, and a cooling system [50], [52], [53]. The BS is powered by solar PV plus public utility grid and backup power is provided by the battery bank. BS consumes energy from the electrical grid system in case of the deficiency of solar-generated power. We consider three different BS configurations namely macro 2/2/2, macro omni, and micro. Modern macro cellular BSs have widely adopted the concept of a distributed base station in which the baseband signal processing unit is separated from the radio-frequency unit that is defined as a remote radio head.

A. SOLAR PV PANEL
Solar PV panels are arrays consist of series/parallel connected multiple solar cells to transform sunlight into electricity. The amount of solar energy generated by the solar PV panel is mostly affected by the location, materials/ tilt of the solar panels, and energy harvesting technology. Moreover, the internet of things (IoT) based smart monitoring system can be incorporated to control the generated power of the solar PV system [51]. The annual energy produced by the solar PV panel can be determined as follows [52] where R PV expresses rated capacity for solar PV panel (kW), PSH means peak solar hour calculated from the average value, and η PV is the panel efficiency. The monthly statistic of solar radiation and the clearness index are presented in Fig. 2.

B. ELECTRICAL GRID
Generally, the electrical grid is an interconnected system that is used for supplying power to the load. In this work, the proposed solar PV/battery system has been connected to the utility grid for sharing the excess electricity to or from the grid, which ensures the maximum utilization of renewable energy by maintaining the quality of services (QoS).

C. BATTERY BANK
A battery bank is an auxiliary device which stores surplus electricity during the peak generation and supplies the BS in case of energy shortage or outage. The battery state of charge (B SOC ) is an important factor in selecting storage devices. The minimum state of charge is the minimum discharge limit of the battery. The depth of discharge (B DOD ) implies the maximum energy supply limit from the battery bank determined from the equation below [52] However, the autonomy of the battery system (B aut ) indicates the probable period limit up to that the storage device can supply the essential electricity to run BS load if the solar PV array is in failure. The battery bank autonomy can be represented as follows [52] where N batt is the total battery number used in the battery bank, V nom is the theoretical voltage value of a single battery (V), Q nom is the nominal capacity of a single battery (Ah) and L BS is the average daily BS load in kWh. The lifecycle of the battery (L batt ) is directly related to the replacement costs of the proposed project can be calculated using (4) [52] where T batt is the lifetime of a single battery (kWh), T a is the annual battery throughput (kWh/year) and R batt,f is the battery float life (year).

D. CONVERTER
Among the different types of converters, an inverter is an instrument which is used to convert the AC voltage into DC voltage. The capacity of the inverter mostly depends on the size of the load and inverter efficiency and can be calculated as follows [52] where L AC represents the peak AC load in kW, is the efficiency of the inverter in %, and σ sf is the safety factor.

E. BS POWER CONSUMPTION MODEL
Modeling and optimal sizing of the proposed system are directly related to BS energy consumption. Table 1 summarizes the power consumption of every single component of the LTE macro/micro BS for the highest traffic intensity. On the other hand, Table 2 represents the BS key parameters. In a practical cellular network, the incoming traffic arrival rate is time-varying and the BS energy demand is traffic sensitive. An approximate traffic pattern can be predicted by utilizing the Poisson distribution model as follows [55] where λ(t) is the normalized traffic distribution, p(t, α) is the Poisson distribution function of traffic demand at a particular point of time, and α is the mean value. The total power utilization of a BS as a function of traffic intensity (χ) can be represented as follows [55]  where P 1 = P 0 + p P TX is the maximum power consumption of a BS sector, N TRX is the total number of the transceiver, p is the load dependency power gradient and is the consumption at idle state.
The maximum power consumption of the base station as a function of different types of losses is determined as follows [55] where σ DC , σ MS , and σ cool respectively represent the loss associate with the DC-DC regulator, main supply, and cooling system. N TRX is the total number of antennas in a base station and is the maximum transmission power in W. The P BB and P RF are the baseband and RF power consumption respectively for the given 10 MHz transmission bandwidth. The total BS energy consumption according to (8) is listed in Table 3. The amount of power drawn by the different BSs configurations is shown in Fig. 3. Fig. 4 demonstrates the daily traffic load profile which has been calculated using (8).

F. PERFORMANCE METRICS
In this work, the throughput, energy efficiency, and spectral efficiency performance of the two-tier LTE cellular network are critically evaluated taking into account the tempo-spatial VOLUME 8, 2020 variation of traffic intensity. According to Shanon's information capacity theorem, total reachable throughput (in kbps) of the cellular network can be estimated by the following equation [57] where SINR i,k is the received signal-to-interference-plusnoise-ratio at k th UE located in BS B i . N is the number of transmitting BSs and U is the total number of UEs in the network.
The energy efficiency of the wireless network is one of the vital factors for assessing performance. The term ''energy efficiency'' denotes the number of bits transmitted per joule of energy. The performance metrics energy efficiency is denoted as η EE and can be expressed as follows [57] η EE = C total P net (12) where P net is the total power consumed in all the BSs at time t and can be calculated by using (8). The spectral efficiency (SE) is represented in bps/Hz and can be determined by the following formula [25] where 0 ≤ ξ ≤ 1, SINR min , ψ max and SINR max are the attenuation factors accounting the implementation loss, minimum SINR, maximum SE, and the SINR at which ψ max is achieved. Then the number of required RBs for UE k of class j can be estimated by [25] β (j) where R j i,k is the required data rate in bps, and W RB is the bandwidth per RB in Hz (e.g., 180 kHz in LTE).
The number of total resource block (n RB ) that is required for the desired system bandwidth can be calculated as follows Hence, the total number of simultaneous users (N SU ) connected to the network can be determined as follows The throughput, energy efficiency, and spectral efficiency metrics of the wireless network are determined using the Monte-Carlo based Matlab simulation. The simulations are accomplished by averaging around 10,000 iterations bearing in mind the time-varying traffic density. It is also considered that every user is associated with at least one resource block (RB). The technical parameters and their values for the basic parameters of the Monte-Carlo simulation setup are summarized in Table 4. To minimize the number of active BSs without sacrificing QoS can be attained through demand shifting from lightly loaded BSs to the surrounded BSs. Based on the statistical data, the BSs experiencing low traffic arrivals transfer its load to neighboring BS and thereby enter into sleep mode. It is to be noted that the surrounded BSs are handling the new arrivals in such a way that it would not lose their performance by means of data transmission services and also reserve 5% backup resource blocks in order to ensure robust QoS during peak periods. The key difference from the cell zooming concept is that it does not need to depend on the input power supplies over the zooming hours. Owing to the stochastic nature of RE production, wireless network performance might have conflicts. In this paper, we first designed the dimension of optimal power supply capacity and then measured the additional load carrying capability implementing the load-balancing technique.

G. CELL ZOOMING CONCEPT
The traffic density of a cellular network varies over time and space due to the mobility of the user and the dynamic 200640 VOLUME 8, 2020 nature of many data applications. For instance, the traffic density in the office areas is higher and lower in the domestic areas during the daytime. This event is going to reverse in the evening. As a result, there exists always some higher density BSs and some lower density BSs in a cluster. In this case, the load balancing and cell zooming techniques can be applied to adjust the load, which will significantly increase the energy efficiency ensuring the service quality of the user equipment.
In this subsection, the basic concept of the traffic-aware load balancing and cell zooming is illustrated, which has the potential to handover the peak traffic demand among the neighboring base stations to attain a greater level of system performance. The fundamental concept of traffic steering based cell zooming technique is presented in Fig. 5. In this work, we consider a two-tier LTE cellular network with a hexagonal layout formed by 19 BSs. Where one BS is placed at the center of the hexagonal and the other eighteen BSs are deployed encompassed the center cell as shown in Fig. 5(a). All the BSs are connected to the control server for coordination. According to the traffic loads, received signal power, and energy availability of the BSs, the control server decides to zoom in, zoom out, and sleep mode of the cells. For making simple, the cell zooming principle is presented for seven BSs as shown in Fig. 5.
If the traffic intensity of the central BS is higher than the predefined value, the control server provides information to reduce the cell size by zooming in as shown in Fig. 5(b). On the other hand, if the mobile users (MUs) move out from the central BS and make the neighboring BS congested, the control server broadcasts the necessary information to zoom out the cell to avoid overflow of the neighboring BS as shown in Fig. 5(c). Moreover, if the traffic intensity of the central BS is very low and the collocated BSs has the capability to take all the MUs, then the low-density center BS will go to sleep mode to reduce the energy consumption as shown in Fig. 5(d). In this case, the neighboring BSs can be zoom out if they are in low density otherwise serve the left MUs by transmitting cooperatively as in Fig. 5(e). However, the sleep mode provision is suitable for the low traffic density areas during the off-peak hours. The amount of load received by the low traffic density cell depends on their unused resource blocks and renewable energy availability. For better understanding, the cell zooming algorithm is presented in a flowchart form as shown in Fig. 6.
In the cell zooming technique, the channel condition and user requirements in the network are collected by the control server, while resource allocation and cell zooming operations are performed in a centralized way. The details of the resource allocation is as follows:

1) RESOURCE ALLOCATION
A resource block (RB)is the lowest unit of physical resources that are assigned to every user. In the time domain, the resource block is one slot long. In the frequency domain, the resource block is 180 kHz wide and either 12 × 15 kHz subcarriers or 24 × 7.5 kHz subcarriers wide. For most of the channels, the number of subcarriers associated per RB is 12.
As an example, a 5 MHz downlink signal can be expressed as 25 resource blocks wide or 301 subcarriers wide (DC subcarrier is not involved in an RB). In the LTE network, 10% of the available bandwidth is used as a guard band for avoiding overlapping [58]. The standard bandwidth and corresponding resource block and subcarriers for both uplink and downlink conditions are presented in Table 5. The number of resource blocks allocated for the individual BS is determined based on the operating bandwidth. The individual BS use their allocated RBs according to the user requirement. In every time slot, this decision has been taken by the allocation algorithm. If one RB cannot fulfill the target requirement of a UE throughput then another RB is allocated according to the availability of free RBs. So, the total bandwidth allocated for a single user can be calculated from the multiplication of total RB allocated for the user and the BW of a single RB. The detailed procedure of the resource block algorithm is presented in Table 6.
In the resource block algorithm, k = UE 1 , UE 2 ,. . . UE m = set of m users that have to be served, R t = total throughput target for UEs, n RB = number of resource blocks, W RB = BW of a single RB, β i,j = bandwidth assigned by BS i to UE j . Binary variable χ i,j signifies the association policy between BSs and UEs, as in the following

H. LOAD BALANCING CONCEPT
All the nineteen BSs in a two-tier cellular network will be sorted according to their arrival traffic rate, received signal power, and available energy in ascending order. As a consequence, seven cells act as a central BS in a cluster that is surrounded by the neighboring BSs. For instance, a center BS1 is surrounded by the neighboring BS2, BS3, BS4, BS5, and BS6 base stations who are serving as acceptors allowing BS1 to switch into a dormant mode for energy saving according to Fig. 5. Under this scenario, acceptor BSs are adjusting their transmitted power for the extended coverage areas while the donor BS is fully unaware of new traffic arrivals. In the case of high traffic density, the center BS will search the neighboring BSs for sharing the load and a percentage of the load can be transferred to the low traffic density BSs based on the proposed heuristic algorithm. Hence, the neighboring BSs will serve as acceptor BSs by receiving the shifted load and the center BS will act as a donor BS. In the case of very low traffic density, the center BS will shift the load to the neighboring low-density BSs and will go to sleep mode for saving the energy. If the traffic density is high again, the active BSs will send a message to the center BS to wake-up to reduce the load of the neighboring BSs. Notably, this coordination process can be accomplished by a control server by adjusting the coverage area and transmitting the power of the acceptor BSs. In this paper, the idea of network cooperation is governed by the traffic state of the donor BS itself and its acceptor BSs. In a load balancing technique, the adjacent BSs are capable of coordinating each other in a cluster. Employing this dynamic load balancing, energy efficiency performance can be significantly improved through adaptively adjusting the active BSs. For the traffic distribution, the donor BS selects the best combination of candidate space that can maintain expected QoS. Therefore, different acceptor BSs selection schemes are possible for selecting the best combination of candidate space. Consider a selection scheme C n , which implies a candidate space containing a set of candidate BS combinations from acceptor BSs (N a ) up to N s neighbor BSs. Then the selection scheme can be defined as C n (N a ≤ N  The detailed procedure of the load balancing algorithm is demonstrated in Table 7. In the load balancing algorithm, S i,j = S 1 , S 2 , . . . S Zi = set of BSs in operating mode whose traffic is sharing, C i,j = C i,1 , C i,2 . . . C i,Mi = candidate BS. C a i,j = acceptor candidate BS set, and Z i,j = traffic distribution it owns set of operating BSs. On the other hand, C s i,j = β i goes to sleep mode only when it can redistribute its traffic to all BSs in S i,j , χ i,j = χ 1 , χ 2 . . . χ k each BS in active mode, F i,j = β i searches best combination BSs (F ij ) in C a ij or C s ij .

1) NUMBER OF SIMULTANEOUS USERS
In order to determine the number of simultaneous users connected to the LTE base stations, we assume that voice data is 64 kbps for full rate and 32 kbps for half rate. It is also considered that the internet data is 256 kbps and video data is 512 kbps. Moreover, the bandwidth required for each number of resource blocks (RBs) is 180 kHz. The number of simultaneous users of LTE eNodeB is calculated as follows: VoLTE Packet length with all Header = Codec Bits + RTP Header + UDP Header + IP Header.
In the case of the RoHC enabled system, around 300 bits are required for performing a VoLTE voice call under the air interface [58]. For long term evolution, the smallest unit of radio resources is commonly recognized as a physical resource block (PRB). In LTE, one physical resource block has 12 subcarriers and 14 symbols over a 1-millisecond time length, or 12 × 14 = 168 resource elements (REs). Some of the REs are engaged by the control symbols (PDCCH) and pilot symbols (RS), which provide around 120 REs available for data broadcast. LTE downlink modulation provisions diverse modulation coding schemes (MCS) for example QPSK, 16 QAM, and 64 QAM for PDSCH which means that each resource can carry 2bits, 4 bits, or 6 bits per symbol respectively.
If we consider that UE is reporting channel quality indicator (CQI) 15, then eNodeB (i.e., LTE BS) use a 64 QAM modulation scheme and a 948/1024 = 0.926 effective coding rate is applied, which means that each RE holds 6 × 0.926 = 5.55 data bits on average. As a consequence, a single PRB can carry 120 × 5.55 = 666 data bits which are equivalent to 2 VoLTE voice samples. But LTE scheduler can't assign less than one PRB per user, so this will count that one PRB is required per VoLTE call. In VoLTE, no re-transmission is desired and voice data is produced every 20 ms to ensure that about 20 VoLTE calls can share the same set of PRB one after other. The maximum amount of VoLTE call that can be supported is calculated as follows where N PRB is the number of available PRB, and N PRB /VoLTE is the number of PRB per VoLTE call.

III. COST MODELING AND OPTIMIZATION
In this work, different types of costs can be evaluated using the HOMER optimization software under different network conditions aiming to minimize the net present cost.  (18) where the total annualized cost (TAC) and capital recovery factor (CRF) is calculated by (19) and (20) (20) where N is the lifetime of the project and i is the annual real interest rate. The salvage value of the system can be calculated by [52] SV = rep( rem comp ) (21) where rep, rem, and comp respectively are the replacement cost, remaining lifetime, and lifetime of the component. For ensuring guaranteed continuity of services, a significant amount of excess electricity is required that can be expressed as follows where E l comprises yearly losses of battery and inverter. One of the major aims of this work is to minimize energy deficiency through maximum utilization of solar energy in conjunction with energy consumption from the grid which in terms reduces NPC. The objective function of the system can be expressed as [55] minimize NPC (23a) subject to E PV + E EG > 0 (23b)  [52], [55].
where E BS is the annual BS load consumption as obtained from Table 3 and E batt is the energy afforded by the battery bank.
The constraint (23b) and (23c) guarantees that the energy generated by the grid-connected solar PV system carries the annual BS consumption. The constraint (23d) represents that the total generated energy including battery backup can satisfy the BS total energy demand and associated losses of the BSs. After fulfilling the energy demand of the BSs, a minimum of 10% of the additional energy is stored in the battery bank (excess electricity) for use in the critical condition and the other 90% energy is allowed for sharing which is represented in constraint (23e).

IV. PERFORMANCE ANALYSIS A. SIMULATION SETUP
In this simulation setup, the considered project duration is 20 years and the yearly interest rate is 6.75%. The average yearly solar radiation of the selected location is 4.59 kWh/m 2 /day. Fig. 2 demonstrates the average yearly solar intensity profile of the selected area. The dual-axis tracking mode of the solar PV panels is designed and 10% back power is reserved to support the BS load in any critical condition. However, HOMER optimization software takes decides at each time slot to fulfill the BS energy demand at the lower NPC, according to the predefined constraint.
The key component and their technical specifications for the HOMER simulation setup are shown in Table 8. Moreover, the key parameters and their specifications for the Monte-Caro simulation setup are presented in Table 4. The technical parameters that are related to the cell zooming option are    [56]. summarized in Table 9 and Table 10 respectively for macro and micro BS.

B. OPTIMAL SYSTEM ARCHITECTURE
The HOMER layout of the proposed system for different base station configurations are shown in Fig. 7(a), (b) and (c) under 10 MHz bandwidth. As refer to Table 1, a greater amount of DC load is found for the macro 2/2/2 configuration to support the six transceiver antenna. The optimal system architecture of the macro 2/2/2, macro omni, and micro configurations considering the average solar radiation (4.59 kWh/m 2 /day) is summarized in Table 11. Moreover, Table 12 compares the optimum system architecture for the various solar intensity of the selected location. In line with our expectation, the macro 2/2/2 configuration required higher solar PV panel capacity to satisfy the higher energy demand. It is also found that the capacity of the solar PV panel decreases with the increment of solar intensity. This is happening because the lower capacity of the solar PV panel can produce higher energy during the higher solar intensity time. On the other hand, the optimal size of the electrical grid, battery, and converter remains constant for all BS configurations which imply that the grid tide solar PV system can be implemented easily for different network configurations without changing the size of the components.
The amount of power contributed by the solar PV panel and electrical grid systems under 10 MHz bandwidth is shown in Fig. 8, Fig. 9 and Fig. 10 respectively for the macro 2/2/2, macro omni, and micro enabled base station systems. As is seen, the solar PV panel generates a greater amount of renewable energy for all conditions due to the higher value of sunlight intensity in the selected location. On the other hand, the electrical grid system contributes the additional energy due to the deficiency of solar energy. The solar PV panel and electrical grid system complement each other. In the case of higher solar intensity, solar PV panel contributes higher energy. Moreover, the backup power is provided by the battery bank in the case of energy shortage by the solar PV and electrical grid system.

C. ENERGY YIELD ANALYSIS 1) SOLAR PV ENERGY
The optimal size of the solar PV panel to meet the BSs load requirement is presented in Table 11 and 12. As refer to Table 11, the optimal size of the solar PV panel for 10 MHz macro 2/2/2 configuration is 3 kW. Thus, 12 Sharp ND-250QCs solar modules are required for LTE macro   2/2/2 BS configuration: as connected 4 in series and 3 in parallel. The amount of energy harvested by the solar PV panels is calculated using (1) as follows: 3 kW × 4.59 × 0.90 × 365 days/year = 4,523 kWh. Moreover, a dual-axis tracker boosts the energy by one fourth which is 6,486 kWh. In a similar process, the annual energy generated and required number of solar PV modules are calculated for other configurations.
On the other hand, the annual energy generated by the solar PV panel for different BSs conditions is shown in Fig. 11 and Fig. 12. According to Table 1, the macro 2/2/2 BS required a higher amount of energy as compared to macro omni and micro BS. As a consequence, the annual energy generated by the solar PV panel is comparatively higher for macro 2/2/2 configuration. It is also noticed that the amount of energy generated by the solar PV panel is rising linearly with the increment of system bandwidth to meet the higher load requirement of the BS.

2) GRID ENERGY
The amount of energy contributed by the electrical grid system is shown in Fig. 13. All the curves are increasing in nature which implies that a higher value of system BW involves a higher amount of grid energy consumption. Moreover, macro 2/2/2 base stations consumed more grid energy as compared to the others because of the higher amount of energy demand.

3) EXCESS ELECTRICITY
The annual energy breakdown for the grid-connected solar PV system is illustrated in Fig. 14. It is always desirable to develop a system that produces a greater amount of renewable energy by utilizing locally available renewable energy sources. In line with our expectations, the solar PV panel generates a significant amount of energy for all network configurations. On the other hand, the electrical grid system contributes a small amount of backup power for ensuring the continuity of power supply. HOMER optimization software  calculates the total amount of energy contributed by the different types of sources and decides at each step to meet the base station energy demand with maximum utilization of renewable energy. However, the total yearly BS energy consumption is 5,042 kWh and the battery and inverter losses are 376 kWh/year and 98 kWh/year respectively. Thus, the yearly excess electricity for 10 MHz bandwidth can be estimated using (22) as follows: (E PV ) 6,486 kWh + (E EG ) 656 kWh -(E BS ) 5,042 kWh-(E l ) 474 kWh = 1,619 kWh/year (32.02%). These values are evaluated for LTE macro 2/2/2 BS considering real traffic patterns and average solar intensity. Similarly, the annual energy generated for all other BSs is calculated.

4) BATTERY BANK
The required number of battery units for macro 2/2/2 BS is 64: 8 batteries connected in series and 8 batteries in parallel. Total time that the battery bank can independently support the BS is 166 hours, which can be found from (3); (64 batteries ×V nom = 6 V × Q nom = 360 Ah ×B DOD = 0.7 × 24h)/daily average BS load 13.91 kWh. Furthermore, the nominal lifetime of the battery bank is 10 years calculated by (4) for the annual throughput and lifetime throughput of 2,400 kWh and 68,800 kWh respectively. Likewise, HOMER calculates the battery bank autonomy to be 759 hours, battery losses to be 22 kWh/year and the normal battery life is 10 years when the micro configuration is used.  An extensive comparison of battery bank autonomy and throughput for the different network configurations are presented in Fig. 15 and Fig. 16 respectively. As is seen, the battery bank autonomy is inversely proportional to the system bandwidth. However, battery autonomy is more than enough to fix the PV malfunctions or any other failures. On the other hand, the battery bank throughput is directly proportional to the system bandwidth. A higher amount of battery bank throughput is found for macro 2/2/2 configuration.

D. ECONOMIC ANALYSIS
For evaluating the economic feasibility, different types of cost that are associated with the grid-tied solar PV system such as capital cost (CC), operation and maintenance cost (OMC), replacement cost (RC), fuel cost (FC), and the salvage value (SV) incurred within the project lifetime are determined using data from Table 8.
The nominal cash flow summary of the grid-tied solar PV system for the macro 2/2/2, macro omni, and micro configurations are respectively demonstrated in Fig. 17, Fig. 18, and Fig. 19 under 10 MHz bandwidth. For all the configurations, the initial capital cost has the highest value and replacement cost is the second. It is also seen that among the different types of components battery banks incur a large capital cost, replacement cost, O&M cost, and net present cost due to the involvement of a large number of battery units for providing sufficient backup power.     $40,195, $37,850, and $37,961 respectively calculated using (18). In line with our expectations, all the configurations provide a satisfactory level of NPC value. Moreover, the macro 2/2/2 configuration has a higher NPC value due to the higher volume of load demand.
An extensive comparison of the cost of electricity for the aforementioned configuration is demonstrated in Fig. 21. It is found that higher system BW has a positive impact on the curves of electricity cost. It is also found that macro 2/2/2 configuration has the lowest value of COE, whereas macro omni configuration has the highest value. This is happening due to the high volume of energy consumption by the macro 2/2/2 configuration and higher system bandwidth.

E. CELL ZOOMING AND LOAD BALANCING ANALYSIS
The number of simultaneous users associated with the network under different applications of UEs is presented in Fig. 22. It is seen that among the different types of users, video call users are small in number, and voice call users are large in number. This is happening because the video call required a greater amount of resource block and voice calls required a smaller amount of resource block. A base station can support the higher number of users if the BSs are operated at higher system bandwidth. However, all the users are not connected at the same time and different types of users will go through different applications. As a result, if we consider that 50% voice users, 35% internet users, and 15% of video users are simultaneously connected to the network, we will get the total number of connected users. These values can increase with the increment of system bandwidth as shown in Fig. 23.   For VoLTE, the number of simultaneous users depends on the number of resource blocks required, which is directly VOLUME 8, 2020 related to the modulation technique used which depends on the radio condition of the UE. Different types of modulation schemes can be applied in the LTE system and UE reports RF conditions with channel quality indicator (CQI) to eNodeB and using this report, eNodeB decides the modulation for a particular resource block. The CQI values can be found from 1-15 where 15 provides the best channel condition. Under the best CQI, the best MCS can be used and the number of simultaneous users can be increased. Moreover, continuous improvement of simultaneous users can be obtained for higher system bandwidth for the eNodeB. The variation of energy efficiency performance concerning the effect of load factor for different system bandwidth is shown in Fig. 25. Where the term 'load factor' is defined as the ratio of maximum load to the actual value of the load in the network. Fig. 25 implies that the BSs operated at lower traffic density, provide lower energy efficiency. On the other, if the BSs are operated at higher traffic density than the average, it will offer relatively higher EE performance. For instance, under 10 MHz bandwidth, when the BSs are operated without applying a load balancing algorithm the energy efficiency becomes 254.4 Kbps/W. In contrast, if the BSs are operated by applying the load balancing algorithm, the EE can vary from 223.37 Kbps/W to 584.48 Kbps/W. Moreover, the load balancing algorithm saves energy by offering the sleep mode provision for the lower density BSs.
The phenomenon of diverse traffic intensity BSs offers that some of the lower density BSs can be switched off or sent to sleep mode by applying the load balancing algorithm. The amount of average energy saved by the sleep mode provision of BS is presented in Fig. 26. In line with our expectation, the energy-saving curves of the BS are increasing linearly with the increment of sleeping hours. Moreover, a higher system bandwidth exhibits a higher amount of energy saving. From the average energy saving curves, it is found that a 4 hours sleeping provision per day can attain 1,100 kWh energy saving annually in the case of 10 MHz bandwidth. Fig. 27 illustrates that a significant amount of energy and power can be saved by zooming in the cell coverage area. As is seen, the amount of energy saved is linearly and the power consumption is inversely related to the zooming level  of the cell. For an instance, 80% of cell zooming can save around 30 of the energy requirement. The cell zooming techniques also increase the EE performance of the base station. However, cell zooming is an effective way of developing a smart cellular network but, it is not practicable for the high traffic density areas.
The spectral efficiency of the wireless network can be improved by applying the load balancing algorithm. A detailed comparison of spectral efficiency for the different system bandwidth and different transmission power is shown in Fig. 28. Numerical results depict that the increment of spectral efficiency is directly related to the shifting of load and a 5% edge zone load shifting can increase the spectral efficiency of around 4%.

1) CHALLENGES OF CELL ZOOMING
The cell zooming technique has many potential benefits but the implementation of the cell zooming technique involves some challenges. The major challenges of implementing the cell zooming technique are summarized below: 1. The implementation of the cell zooming technique requires tracing the actual traffic intensity profile and sharing the information to the control server (CS).
2. The implementation of the cell zooming technique requires a proper cooperation management system to control the antenna height, antenna tilt, transmitting power, BS cooperation, and relaying techniques. 3. The implementation of the cell zooming technique may increase the inter-cell interference and coverage hole. In a BS some users move from one BS to another BS or new users arrive. To satisfy the user demand, some cells need to zoom in and some cells need to zoom out which may cause this inter-cell interference and coverage hole problems.
From the aforementioned discussion, it is concluded that the implementation of the cell zooming scheme is challenging since the sleep and wakeup mechanism are not still considered a mature method. It is not so easy to practically construct an infrastructure due to the aforementioned issues. On the other hand, load balancing is a popular technique that can provide a similar benefit by finding the optimal traffic handover between the overloaded and nearby under loaded cells using the BS coordination mechanism. The key difference between the two mechanisms is that the lightly loaded BS goes to idle mode instead of sleep mode under load balancing concept. Although cell zooming and load balancing techniques exhibit similar performance in terms of energy efficiency, the load balancing technique draws more attention over the cell zooming technique since no sleep mode transition is provisioned.

F. ENERGY EFFICIENCY ANALYSIS
The throughput performance of the proposed system over 24 hours is shown in Fig. 29. The throughput curves for different network configurations follow a similar pattern of traffic demand profile. This is happening because the throughput is directly related to the resource blocks and resource block is a function of traffic demand profile. However, a higher number of RBs are allocated for operating the BS in higher system BW. On the other hand, the throughput performance for different system bandwidth is presented in Fig. 30. It is found that throughput values are increasing linearly with the increment of system bandwidth for all network configurations. Moreover, there is a little difference between the throughput values of macro 2/2/2, macro omni, and micro, but it is not seen clearly due to the high value of throughput. An extensive comparison of energy efficiency performance among the aforementioned network configurations considering the effect of dynamic traffic rate is shown in Fig. 31. The graph tells that the macro omni configuration has superior energy efficiency performance, and the macro 2/2/2 configuration has poor energy efficiency performance. Moreover, higher system bandwidth increases the EE performance of the system under all conditions. All the throughput and energy efficiency have been calculated considering the effect of inter-cell interference (ICI). A higher value of throughput and energy efficiency are always preferable for ensuring the guaranteed quality of services.

G. CARBON FOOTPRINTS
As an ideal renewable energy source, solar PV system does not emit any carbon content. On the other hand, VOLUME 8, 2020  a grid-connected solar PV system emits a significant amount of carbon footprints due to the burning of a huge amount of fossil fuel in grid energy generation. Fig. 32 depicts the comparison of carbon footprints generated by a grid-connected solar PV system under different network configurations. The figure illustrates that the macro 2/2/2 configurations produce a greater amount of CO 2 . Moreover, higher system BW contributes higher carbon footprints to cope up with the higher BS energy demand. This is happening because a significant amount of energy contribution takes place by the electrical grid system to maintain the continuity of the power supply.

H. FEASIBILITY COMPARISON
A detailed comparison of optimal criteria among the standalone solar PV, stand-alone grid, and grid-connected solar PV system along with sufficiency energy storage devices is shown in Table 13. It is found that the grid-connected solar PV system under macro 2/2/2 configuration can attain NPC and COE saving up to 8.8% and 10.43% respectively as compared to the stand-alone grid system. Moreover, it can save carbon footprints by around 88.64%.

V. CONCLUSION
This paper examined the feasibility of macro 2/2/2, macro omni, and micro BS provisioned to be powered by the gridconnected solar PV system with sufficient storage devices. The optimal system criteria including the technical and economic performance of the proposed system have been extensively evaluated under different system bandwidth. The numerical results depict that the grid-tied solar PV system with macro 2/2/2 configuration significantly saves 8.8% of the overall net present cost and achieves a substantial improvement in CO 2 savings yielding up to 88.64%. The battery bank can supply the macro 2/2/2 BS for 166 hours autonomously if the solar PV/grid system fails to support the BS energy demand. On the other hand, the wireless network exhibits a satisfactory level of throughput, energy efficiency, and spectral efficiency performance. Moreover, the incorporation of load balancing techniques can increase the energy efficiency performance up to 56.47% if BSs are effectively modeled with cooperation mechanisms. The load coordination technique gives better utilization of resources in case of average spectral efficiency over cell zooming condition and thereby makes the system more lucrative. In summary, the solar PV/grid-powered heterogeneous cellular network along with the load balancing technique is an excellent choice for developing an energy-efficient and sustainable green mobile communication which will minimize the grid pressure, carbon footprints, and net present cost. In future work, the authors will focus on a coordinated multipoint (CoMP) technique based user association algorithm for multi-tier networks considering a generalized stochastic geometry-based model.