Energy Efficient Resource Allocation for H-NOMA Assisted B5G HetNets

The resource allocation solution offered based on non-orthogonal multiple access (NOMA) and orthogonal multiple access (OMA) schemes are sub-optimal to address the challenging quality of service (QoS) and higher data rate viz-a-viz energy efficiency (EE) requirements in 5th generation (5G) cellular networks. In this work, we maximize the EE using user equipment (UE) clustering (UE-C) with downlink hybrid NOMA (H-NOMA) assisted beyond 5G (B5G) HetNets. We formulate an optimization problem incorporating UE admission in a cluster, UE association with a base station (BS), and power allocation assisted by H-NOMA, i.e., OMA and NOMA schemes in the macro base station (MBS) only and heterogeneous networks (HetNets) environments. The problem formulated is a type of non-linear concave fractional programming (CFP) problem. The Charnes-Cooper transformation (CCT) is applied to the formulated non-linear CFP problem to convert it into a concave optimization, i.e., mixed-integer non-linear programming (MINLP) problem. A two-phase $\epsilon $ -optimal outer approximation algorithm (OAA) is used to solve the MINLP problem. The simulation results show that H-NOMA with HetNets outperforms H-NOMA with MBS only in terms of UE admission, UE association, throughput, and EE.


I. INTRODUCTION
The rapid growth in the number of mobile user equipment (UE), and heavy data-driven applications, i.e., live video gaming, video streaming, social networking, etc are imposing challenging requirements like minimum delay, higher data rates, spectrum efficiency (SE), and energy efficiency (EE) on the beyond 5th Generation (B5G) cellular networks. This exponential growth in mobile UEs viz-a-viz mobile data traffic is adding to a significant increase in the energy consumption in cellular networks. Information communication technology (ICT) is consuming almost 2% of the world's total energy. Energy consumption emits carbon which causes the greenhouse effect. Thus, ICT offering higher data rates, low carbon emission, and EE are the prime considerations in B5G cellular networks. Thus, academia and industry in The associate editor coordinating the review of this manuscript and approving it for publication was Jie Tang . the wireless communication field must divert their research towards future energy-efficient green cellular networks in B5G networks [1], [2], [3].
Energy-efficient radio resource management techniques are required to satisfy the needs for quality of service (QoS) and higher data rates with minimum energy consumption. EE is the ratio of data rate to the total energy consumed [4], [5].
EE can be improved using heterogeneous networks (HetNets), which include a macro base station (MBS) and small base stations (SBSs) [6], [7], [8]. HetNets cover more geographical areas and offer higher data rates and are energy efficient than the conventional MBS-only networks. In Het-Nets, MBS is large with high transmit power, and SBSs (i.e., femtocell, picocell, etc.) are of small size with low transmit power [9]. A low transmit-powered SBS will also preserve energy because additional energy will not be required for cooling purposes compared to a high transmit-powered MBS. VOLUME 10, 2022 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ Additionally, EE can also be improved by efficient reply to the question that how resources will access the network? The conventional multiple access (MA) schemes, i.e., orthogonal multiple access (OMA) includes time division multiple access (TDMA), frequency division multiple access (FDMA), and code division multiple access (CDMA) [10]. As an orthogonal FDMA scheme assigns a single subcarrier (SC) to a single UE which makes these conventional OMA schemes unsuitable due to the availability of limited spectrum, requirements of high data rate, and EE.
Compared to OMA, the non-orthogonal multiple access (NOMA) scheme can increase the SE and EE because more UEs can be accommodated on a single SC at the expense of multi-UE interference (MUEI). Power domain-NOMA (PD-NOMA) based on power domain multiplexing is the most widely used type of NOMA [11]. In downlink PD-NOMA, the base station (BS) at the transmitter side uses superposition coding to transmit information. Successive interference cancellation (SIC) is used to receive information at the receiver side by removing MUEI at the expense of increased complexity [12]. OMA and NOMA are combined as a hybrid-NOMA (H-NOMA) technique that could be a better choice to manage the complexity compared to NOMA. UE-clustering (UE-C) with H-NOMA can further reduce the MUEI. UE-C is defined as the group of UEs assigned an orthogonal SC. In H-NOMA based on UE-C, an SC assigned to a cluster that contains only one UE is defined as OMA-SC and otherwise NOMA-SC. Therefore, the data rate and EE can be improved using a UE-C-based downlink H-NOMA in B5G HetNets. Therefore, the optimal resource allocation for H-NOMA with UE-C in the downlink can enhance the EE.
The resource allocation for H-NOMA has been applied in many scenarios related to this paper. H-NOMA with UE-C in downlink is used for optimal resource allocation i.e. number of UEs admitted in clusters, number of UEs associated with BSs, power allocation to UEs for maximization of throughput [7]. H-NOMA including power domain and code domain NOMA is applied to connect low power trillion UEs and to achieve higher spectral efficiency in the uplink for next-generation internet of things (IoT) networks [13]. H-NOMA as a combination of NOMA and TDMA is used in the uplink to transmit information to clusters UEs and it is applied for reflecting beamforming of intelligent reflecting surface (IRS) and time allocation among BS's power transfer and different UEs clusters' information transmission to maximize the throughput of the IRS-assisted wireless powered communication network [14]. The next section provides the literature review. Table 1 summarizes the past work on EE by using OMA, NOMA, H-NOMA, and UE-C techniques in HetNets and MBS-only networks.

A. LITERATURE REVIEW
The researchers have worked to maximize EE using NOMA in MBS-only networks from [15], [16], [17], [18]. In [15], the authors have maximized EE of cognitive radio (CR) inspired NOMA network using the suboptimal matching for sub-channel allocation (SOMSA) and difference of convex (DC) programming for power allocation with maximum transmit power and QoS constraints for each primary UE in MBS only networks. In [16], the authors proposed a sequential convex approximation (SCA) scheme to maximize the EE of CR-inspired NOMA network under each primary UE QoS constraint by solving the non-convex fractional programming (FP) problem. In [17], the authors proposed an iterative sub-optimal algorithm by adopting FP and DC programming to maximize EE under minimum rate and maximum transmit power constraints in downlink coordinated multipoint (CoMP) systems with NOMA. In [18], the authors proposed a novel energy-efficient matching scheme for subchannel assignment and a Lyapunov optimization scheme for power allocation in a single downlink NOMA network.
The researchers have worked to maximize EE using OMA in HetNets from [19], [20], [21], [22]. In [19], the authors proposed a heuristic algorithm to improve the spectral efficiency and reduce the overall power consumption of a mobile communication network based on two-tier deployment of device-to-device (D2D) communication subject to maintaining a minimum signal threshold. In [20], the authors maximized the EE by using a stochastic geometry-based model with random discontinuous transmission (DTX) mode under the finite local delay constraint in the downlink HetNets. In [21], the authors maximized the EE using a heuristic algorithm based on the cross-entropy (CE) and Karush-Kuhn-Tucker (KKT) conditions-based method in a downlink HetNets system under maximum transmit power and minimum rate constraints. In [22], authors presented a joint EE resource allocation (JEERA) algorithm to improve EE and limit interference in HetNets subject to QoS constraints.
The researchers have worked to maximize EE using NOMA in HetNets from [23], [24], [25], [26]. In [23], the authors maximized the EE by using a low-complexity subchannel matching algorithm and lagrangian duality algorithm for power allocation under the constraint of the limited supply of energy in NOMA HetNets with simultaneous wireless information and power transfer (SWIPT). In [24], the authors considered perfect and imperfect channel state information (CSI) to maximize EE by using a heuristic algorithm in 5G NOMA HetNets subject to QoS and maximum transmit power constraints. In [25], authors presented a spectrumefficient, energy-efficient, and delay-constrained heuristic algorithm to reduce energy consumption subject to the constraints of delay and QoS for full-duplex self-backhauled (FS) HetNets. In [26], the authors proposed a fair power allocation (FPA) approach to maximize the EE subject to the constraint of transmit powers for UEs in a two-tier downlink ultra-dense HetNet with NOMA.
The researchers have worked little to maximize EE using H-NOMA in MBS only networks from [27], [28], [29], [30], [31], [32]. In [27], a heuristic algorithm is proposed by collective optimization of the power allocation, and channel assignment that maximizes the EE in an MBS-only network subject to QoS and maximum transmit power constraints. In [28], the authors proposed Dinkelbach's algorithm and SCA method to maximize the EE with a minimum transmission data rate in a cooperative CR network. In [29], EE is maximized in downlink NOMA systems by using the heuristic resource allocation (Heur-RA) algorithm and optimal resource allocation (Opt-RA) algorithm to solve mixed integer non-linear programming (MINLP) problem under QoS and minimum transmit power constraint. In [30], the authors have proposed an H-NOMA scheme for throughput improvement considering desired user traffic volume as well as channel conditions. In [31], a game theory-based joint VOLUME 10, 2022 energy-efficient optimization of resources in cognitive radio networks (CRNs) consisting of secondary networks and primary networks through H-NOMA is studied. In [32], a SE-EE trade-off-based technique for H-NOMA, i.e., a combination of TDMA and NOMA systems is proposed.

B. MOTIVATIONS AND CONTRIBUTIONS
After going through the past literature [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32] and looking at Table. 1, to the best of the author's knowledge, B5G challenges, i.e., maximum UE accommodation, higher data rate, and EE, etc using an H-NOMA, i.e., NOMA and OMA, etc and UE-C modeled in a HetNets has not been explored in the past. We consider these challenges and formulate an optimization problem incorporating UE admission in a cluster, UE association with a BS, power allocation, and H-NOMA scheme in MBS only and HetNets environments. Then, we gauge the performance edge of H-NOMA in HetNets over traditional H-NOMA in MBS only. The main contributions of this work are listed below: • We formulate an optimization problem incorporating UE admission in a cluster, UE association with a BS, power allocation, and H-NOMA, i.e., OMA and NOMA schemes in MBS only and HetNets environments.
• The problem formulated is a type of non-linear concave fractional programming (CFP) problem. The Charnes-Cooper transformation (CCT) is applied to the formulated non-linear CFP problem to convert it into a concave optimization.
• The MINLP problem is solved using an -optimal outer approximation algorithm (OAA). This is a two-phase algorithm, i.e., primal stage and master stage. In the primal phase, the MINLP problem is transformed into non-linear programming (NLP) problem to get the upper boundary of the optimal solution. In the master phase, the MINLP problem is transformed into a mixed-integer linear programming (MILP) problem to get a lower boundary of the optimal solution.
• The simulation results show that H-NOMA with Het-Nets outperforms H-NOMA with MBS only in terms of UE admission and association, throughput, and EE. The presentation sequence of the paper is as follows: Section II includes the system model and problem formulation. Section III and IV propose algorithm and simulation results, respectively. The conclusion is given in section V.

II. SYSTEM MODEL AND PROBLEM FORMULATION
A. SPATIAL MODEL Figure 1 shows the H-NOMA scheme based MBS-only and HetNets architecture. The perfect known CSI is assumed at MBS in the MBS-only network and additionally at SBS in HetNets. UEs are uniformly distributed and combined into various clusters based on the best channel gains. The UEs admitted in macro base station clusters (MBCs), i.e., clusters of MBS, and small base station clusters (SBCs), i.e., clusters of SBS, are served by an orthogonal SC rationed to each cluster that evades inter-cluster interference. Symbols used in problem formulation are defined in Table 2.
Let N = {1, 2, . . . , N }, where N is the number of UEs assumed in a network. K = {1, 2, . . . , K } represents the number of clusters. B = {m, s} denotes the BSs, i.e., m = MBS and s = SBS. A downlink H-NOMA scheme is used in the clusters. More clearly, if a cluster is assigned an orthogonal SC that contains only one UE, then SC is defined as OMA-SC, and if a cluster is assigned an orthogonal SC that encompasses more than one UE, then SC is defined as NOMA-SC.

B. BINARY VARIABLES
A binary admission variable i n,k is defined to represent n th UE's admission in k th cluster, and is given below: A binary association variable j b n,k is defined to represent n th UE's association with b th BS, i.e., MBS in MBS-only network and HetNets, already admitted in k th cluster, and is given below: C. CHANNEL MODEL Assume channel gain between n th admitted UE in k th cluster and associated with b th BS is given below: where rayleigh fading is indicated ash b n,k , antenna gain is G o , α is path loss exponent, d b n,k is the distance between n th admitted UE in k th cluster and associated with b th BS, ξ is gaussian random variable with zero-mean and standard deviation σ in log-normal shadowing, and d o is the reference distance [33].

D. UE-C WITH H-NOMA
According to the PD-NOMA principle, power domain multiplexing is used at the BS. UEs, decode and receive information by eliminating MUEI using SIC. The arrangement sequence of total t admitted UEs in k th cluster associated with b th BS is from lower channel gain UE to higher channel gain UE can be written as: Accordingly, the received powers by total t UEs admitted in k th cluster associated with b th BS can be given as: Assume P b is the total power at b th BS, (i.e., MBS or SBS) is divided among all the MBCs or SBCs, respectively. A weighting factor w b n,k is adopted that allocates higher power to UE which has lower channel gain, and allocates lower power to UE which has higher channel gain in k th cluster as given: where, b ∈ B and no UE is admitted and associated if w b n,k = 1. The value of weighting factor can vary between 1 ≥ w b n,k > 0. The received power of UE is computed as: where p b k is the power allocated to k th cluster in association with b th BS, i.e., MBS or SBS.

E. SIGNAL TO INTERFERENCE PLUS NOISE RATIO (SINR) MODEL
In SIC, n th UE first detects the message of UE n and then removes the detected message from the desired information in a consecutive way. In an MBS-only network, the n th UE's SINR admitted in the k th cluster in association with b th BS, (i.e., MBS) can be written as: where b ∈ B and B = {m}, and p b n,k is the received power of n th UE, p b n ,k is the received power of UE n , N o is the noise power spectral density. In any cluster, n th UE can apply SIC if the received SINR of n th UE is greater than or equal to the received SINR of UE n .
More specifically, the following inequality is necessary to be satisfied for successful decoding and removing the UE n signal from n th UE's signal admitted in k th cluster in association with b th BS, (i.e., MBS or SBS) is given as: In HetNets, in addition to the interference in the MBS-only network using UE-C-based downlink H-NOMA, UEs also experience interference between BSs (i.e., MBS and SBS). The n th UE admitted in k th cluster and association with b th BS experience interference from b BS. The SINR of n th UE admitted in k th cluster in association with b th BS is written VOLUME 10, 2022 as: The factor h b n,k P b represents the added interference from other BS to the UE associated with one BS.

F. THROUGHPUT MODEL
The data rate achieved by UE is defined as throughput. The throughput of n th UE admitted in k th cluster and associated with b th BS is given as: where b ∈ B = {m} in MBS only network and b ∈ B = {m, s} in HetNets.

G. PROBLEM FORMULATION
In this subsection, We formulate an optimization problem incorporating UE admission in a cluster, UE association with a BS, and power allocation while considering H-NOMA, i.e., OMA and NOMA schemes in MBS only and HetNets environments. The objective function and constraints to model the optimization problem are defined below: 1) A utility function for maximization of EE is defined using Eq. (1), (2), (7), and (11) as below: 2) Using Eq. (1), the constraint to ensure that n th UE admits in only one k th cluster is given as: 3) Using Eq. (2), the constraint to ensure that n th UE associates with only one b th BS is given as: 4) Using Eq. (1) and (2), the constraint to ensure that n th UE admitted in k th cluster must be associated with b th BS is given as: 5) The constraint to ensure power allocation to each MBC or SBC is given as: 6) Using Eq. (1), (2), and (11), the constraint to ensure QoS data rate for n th UE admitted in k th cluster, and association with b th BS, i.e., MBS or SBS is given below: Maximization of EE is achieved when resource allocation results based on constraints defined above are fed to the objective function. Incorporating objective function and constraints defined above, the optimization problem based on H-NOMA, i.e., OMA and NOMA schemes for resource allocation in MBS only and HetNets can be formulated as:

C1
: The objective of the function in (18) is the maximization of EE (Mbits/sec/watt) considering admission, association, and power allocation to UE subject to the constraints C1 to C7. The constraint C1 indicates that at a time, n th UE can only admit in one k th cluster. The constraint C2 indicates that at a time, n th UE can only associate with one b th BS (i.e., MBS in MBS only network and MBS/SBS in HetNets). The constraint C3 indicates that n th UE associated with one b th BS must be admitted to one k th cluster. Constraint C4 allocates power to MBC or SBC such that entire power at MBS is allocated to MBCs and entire power at SBS is assigned to SBCs, respectively. The constraint C5 is related to the minimum rate requirement, i.e., QoS, of n th UE admitted in k th cluster in association with b th BS.

C1
: The problem in (19) is non-deterministic polynomial-time (NP)-hard in nature and categorized as MINLP. The intensification of discrete variables in polynomial time imposes restrictions on finding an optimal solution with any algorithm. This problem contains binary variables i n,k , j b n,k as well as continuous variables u b n,k . There will be an exponential increase in the search space of (19) if the total number of UEs (N) will increase. An optimal solution can be obtained using an exhaustive search algorithm (ESA) on binary variables. But its complexity is high because we will have to solve 2 |N | optimization problems if the search space for binary variables is 2 |N | . That's why we have proposed OAA to reach a near-optimal solution to guarantee convergence [33]. A detailed implementation of OAA [34] for the MINLP problem in (19) is given in the next section.

III. OUTER APPROXIMATION ALGORITHM
The MINLP problem in (19) contains binary, integer, and continuous variables. The binary variables are UEs admission and UEs association. The integer variables are the numbers of UE. Continuous variables are related to the received power by UE. The medley of all mentioned variables and their non-linear behavior shapes the problem in (19) significantly challenging and complex. Nevertheless, the special design of the problem directed us to employ OAA to solve this problem. OAA delivers -optimal values of u b and v, employing these values in (18), we obtain -optimal values for EE. Figure 2 presents flow diagram of OAA.

A. DESCRIPTION OF ALGORITHM
Assume objective function as for C1 to C8, set of constraints as ψ C1−C8 , π = {p b n,k } where b ∈ B, and χ = ℵ∪π in (19). The problem in (19) satisfies the following propositions.
1) π is convex, compact, and non-empty. Objective function and the constraints ψ C1−C8 are convex in π, if the values of χ are fixed. 2) Objective function and constraints ψ C1−C8 are continually differentiable once. 3) Fixing χ solves each nonlinear continuous subproblem for satisfying the constraints. 4) Fixing χ makes the possibility for the absolute solution of the NLP problem. All four propositions for the problem in (19) are satisfied as to the sequences of non-decreasing lower bounds and non-increasing upper bounds for MINLP and are computed iteratively with OAA. OAA is converged in an infinite number of iterations with convergence capability [35].
Problem (19) is divided into the master and primal problems for finding the lower and upper bound sequences, respectively. The primal problem is described by fixing χ and χ y is integer variable's value at the y th iteration for problem in (19) is given as: The solution to the primal problem in (20) is provided by exercising outer approximation (successive linearization) with fixed values of binary variables. The problem in (20) is solved to find the values of π y used in the master problem. The solution of the primal problem provides upper VOLUME 10, 2022 bounds and the solution of the master problem provides lower bounds. Primal solution π y helps in deriving the master problems. The master problem is derived around the primal solution π y by using OAA to make the constraint functions ψ C1−C8 and the objective function linear [36], [37].
The master problem is solved to find the integer variables that are applied in the next iteration χ y+1 . The difference between the upper and lower bounds becomes minimum if the algorithm continues. When the difference between these two values becomes less than , then the algorithm quits. The problem in (19) is modified as below: The problems in (21) is revised as below: The projection of (19) on χ space is provided in the problem (22). All constraints are contained in primal problem (20) for all χ y . Therefore, the solution of the problem (22) is written as below: min ψ min π − χ y , π y − ∇ χ y , π y π − π y χ − χ y subject to ψ C1−C8 χ y , π y − ∇ψ C1−C8 χ y , π y π − π y χ − χ y ≤ 0 (23) The equivalent minimization problem with the introduction of a new variable κ can be given as below: The master problem in (24) results in lower bound values. Upon satisfaction of the corresponding propositions 1, 2 and 3, the master problem in (24) becomes equivalent to (19). The branch and bound algorithm is used to find the solution to the MILP problem, i.e., master problem in (24) [38].

B. ALGORITHM CONVERGENCE AND OPTIMALITY
OAA converges linearly [35], [36]. OAA optimally solves concave optimization problems having convex constraints and objective function by fixing values of χ . The branch and bound architecture is used to provide an optimal solution within = 10 −3 by fixing values of χ in OAA. If an optimal solution is found within = 10 −3 after satisfying all four propositions optimally for fixed values of χ then it results in termination of OAA in a finite number of steps. If κ is greater than (χ y , π y ) for any feasible point in (24) then it shows the optimality of π in (24). For master problem, feasible solution does not exist for given value of χ when κ is less than (χ y , π y ). If there exists no feasible solution for any value of χ y in (24), it will not be included for subsequent master problems. Hence, the algorithm's convergence is achieved.

C. COMPLEXITY OF -OPTIMAL ALGORITHM
The complexity of the -optimal algorithm is determined in this section. The complexity is determined by counting The complexity of OAA is given in (25). Likewise, the complexity of OAA is determined using Big O notation as O(A × E) + O(A × E × ). The numbers of UE are A, the numbers of BS are E, and the total constraints are .

IV. SIMULATION RESULTS
The key performance indicators (KPIs) of the proposed scenario in HetNets and the MBS-only network are evaluated using -optimal OAA. Basic open-source non-linear mixed integer (BONMIN) programming [40] is implemented in OAA. Admission of UE in a cluster, the association of UE with BS, network throughput, and EE are the considered KPIs. The simulation parameters used are given in Table 3.
It is assumed that the minimum numbers of UE are 5 with an increment of 10 UEs up to the maximum allowed 85 UEs. The maximum coverage area of MBS and SBS is 1000 m and 300 m, respectively. The minimum required rates for a UE (i.e., R min n ) are set to {0.5, 1, 2, 3} Mbps. The antenna gain (i.e., G o ) is 50. The far-field reference distance (i.e., d o ) is 10 m, path loss exponent (i.e., α) is 2, and Gaussian random variable in log-normal shadowing (i.e., ξ ) is 10 dB. The total  power for MBS (i.e., P m ) and SBS (i.e., P s ) is 43 dBm and 33 dBm, respectively. The circuit power (i.e., P c ) is −30 dBm. Figure 3 presents the UE admission in the clusters concerning the number of UEs in an MBS-only network and HetNets. Figure 3 shows that the H-NOMA scheme employs OMA when only one UE is admitted in MBC or SBC. However, the H-NOMA scheme employs NOMA when more than one UE is admitted in MBC or SBC. It is evident in Figure 3 that UE admission increases with the increase in the number of UEs in both the networks. But in the HetNets, the admission of the UE in clusters is higher than in the MBS-only network. This performance edge of the UEs admission in the HetNets over MBS only is because a UE based on the best SINR can be admitted either in MBC or SBC in the HetNets. The proposed scheme offers UEs to be admitted in UE-C using OMA and NOMA in the MBC and SBC in the HetNets. Here, SBC of SBS offers services to the UEs in the dead zones of MBC of MBS. However, H-NOMA in the MBS only could not reach the UEs in the dead zones of MBS in the MBS only. Therefore, the H-NOMA scheme in HetNets outperforms the H-NOMA scheme in MBS only in terms of UE admission in the clusters.   It can be seen that if we increase the number of UEs, then the numbers of associated UEs also increase in both networks. But UE association (i.e., based on best SINR) is more in HetNets than in the MBS-only network due to the availability of SBS in addition to MBS. Availability of diverse transmit power BSs, i.e., MBS and SBS, etc coupled with the H-NOMA scheme in HetNets offer multiple options to UEs for association (i.e., based on best SINR) using OMA and NOMA access schemes. This results in more UEs association in the HetNets as compared to the MBS only. Figure 5 shows UEs distribution/association in the Het-Nets. The HetNets offers diverse transmit power BSs, i.e., MBS and SBS, etc coupled with UE-C-based NOMA and OMA access schemes for UEs association in the network. In the HetNets, MBS extends seamless coverage and SBS covers dead zones not covered by the MBS in the network. Here, MBS overrides SBS in terms of transmit power, therefore, more UEs are associated with MBS and fewer UEs are associated with SBS in the HetNets. This results in more UEs association in the HetNets as compared to the MBS only. Figure 6 shows the plot of associated UEs versus required rate R min   numbers of associated UEs increase with the increase of the number of UEs in the network. However, the number of associated UEs decrease with the increase in the QoS required data rate. As the QoS required data rate is increased, fewer UEs qualify to maintain the higher QoS required data rates. Since, the HetNets offers diverse transmit power BSs, i.e., MBS and SBS, etc coupled with UE-C-based NOMA and OMA access schemes for UEs association in the network. Therefore, the HetNets outperforms the MBS-only network at higher QoS required data rates.    QoS required data rate is R min n = 1 Mbps. The result shows that EE and UE association both increase as the number of UEs increases up to a certain limit in the network. If we further raise the number of UEs, EE decreases because the system's maximum capacity is achieved. Hence, the performance regarding UE association and EE versus the number of UEs in HetNets is better than in the MBS-only network. Figure 9 shows the performance of UE-C based downlink H-NOMA strategy in HetNets for the required rate R min n = 0.5 Mbps, and the numbers of UEs N = {5, 15,25,35,45,55,65,75, 85} in terms of EE distribution versus the number of UEs. It depicts that if the number of UEs rises, then EE also rises to a certain limit, and after that if the number of UEs further increases then EE decreases due to the system capacity limit achieved. But MBS performs better compared to SBS in HetNets. Figure 10 shows the plot of throughput and EE versus the required rate R min n = {0.5, 1, 2, 3} Mbps for 85 UEs to evaluate the performance of UE-C based downlink H-NOMA in MBS-only network and HetNets. This plot shows the effect of an increase in QoS required data rate on average throughput and EE when we employ H-NOMA in MBS only and HetNets. It was found in figure 6 that UEs association starts dropping when QoS required data rate is increased. Effect of UEs association dropping at higher QoS required data rates also have a negative impact on throughput and EE when QoS required data rates are increased in the simulations as evident in the figure 10. However, this negative effect has a major effect in the MBS case and a minor effect in  the HetNets case. Thus, the H-NOMA scheme in HetNets outperforms the H-NOMA scheme in MBS only. Figure 11 shows the the performance of UE-C based downlink H-NOMA in HetNets and MBS-only network is analyzed for throughput and EE versus the numbers of UEs. The result shows that the throughput and EE in both HetNets and the MBS-only network increase when we increase the number of UEs up to a certain number. But if we further increase the numbers of UEs in the network then a slight decrease in EE appears but throughput increases. This figure also shows the better performance of HetNets over the MBS-only network regarding throughput and EE.

V. CONCLUSION
This research work considers an optimization problem incorporating UE admission in a cluster, UE association with a BS, power allocation and H-NOMA, i.e., OMA and NOMA schemes in MBS only and HetNets environments. The Het-Nets offers to diversify coverage and transmit power by BSs as compared to MBS only where coverage and transmit power is fixed. In the HetNet, the SBS offers services to the UEs in the dead zones of the MBS. However, the services of the H-NOMA in the MBS only could not reach the UEs in the dead zones of MBS in the MBS only. The proposed H-NOMA scheme employs OMA when only one UE is admitted in MBC or SBC. However, the H-NOMA scheme employs NOMA when more than one UE are admitted to MBC or SBC. Therefore, the H-NOMA scheme in HetNets outperforms the H-NOMA scheme in MBS only in terms of UEs admission in the clusters, UEs association with the BSs, throughput and EE. Spectrum resource is becoming scarce in terms of demand and supply in B5G wireless communication. This research work will be extended in Space-Air-Ground Integrated Vehicular Networks (SAGVN) to address this challenging issue in B5G cellular networks.

APPENDIX. FRACTIONAL PROGRAMING AND CHARNES COOPER TRANSFORMATION
FP contains objective function as a ratio of two nonlinear functions generally. A FP can be described mathematically as in (26): where S ⊂ R n is a set of real values, described by x(t), y(t) and h n (t) where N ={1, 2, . . . , N }.
In (26), assuming S is convex set, x(t) is +ive and concave, y(t) is +ive and convex makes FP as CFP. CCT reduce a CFP to a concave programme [41], as given below in (27), (28): The concave problem for (26) equivalently can be described as below in (29): Only if the problem in (29) have optimal solution, then the problem in (26) can have optimal solution. Handler. He is also a Certified EC-Council Instructor and has conducted numerous trainings. He is currently the Chief Instructor at Engineering Wing, MCS, National University of Sciences and Technology. His research interests include mobile and wireless communication, MIMO systems, compressed sensing for MIMO OFDM systems, massive MIMO systems, M2M for mobile systems, cognitive radio networks, cyber security, and information assurance. VOLUME 10, 2022