A Joint Channel Allocation and Power Control Scheme for D2D Communication in UAV-based Networks

With the increasing application of unmanned aerial vehicles (UAVs), UAV-based base stations (BS) have been widely used. When there is no ground BS or BSs being out of service, such as mountainous areas, isolated islands and disaster areas, UAV-based networks may be rapidly deployed for communications. In this paper, we propose a framework of power control and channel allocation for an underlying device-to-device (D2D) communication in UAV-based networks. The objective of this study is to maximize the energy efficiency of D2D pairs while ensuring the quality of service of networks. Four kinds of interference scenarios in the system are considered, including intra-cell interference and inter-cell interference generated by cellular users and D2D transmitters, respectively. In addition, we assign UAVs to D2D pairs and formulate the uplink power control and channel allocation problem. Then, the quantity-weight adaptive salp swarm algorithm (QWASSA) is proposed, which adopts dynamic leader-follower numbers and positions update form. Finally, the performance of the QWASSA is evaluated by simulations, and results show that the QWASSA has excellent global searching ability and local mining ability, which is not only superior to other benchmark schemes but also close to the optimal performance of D2D pairs in terms of energy efficiency.


I. INTRODUCTION
Integrating unmanned aerial vehicles (UAVs) into the fifth generation (5G) and beyond cellular networks is a promising technology. UAVs can be used as aerial communication platforms to assist ground communications, such as traffic unloading, recovery after natural disasters, emergency response, Internet of Things (IoTs), and so on [1]. To solve the problem of wireless connectivity for devices that are without infrastructure coverage, such as mountainous areas, isolated islands, especially interruption of communication caused by disasters, many efforts have been made to study wireless communication with UAVs [2]- [5]. Compared with terrestrial communications, UAVs are generally easier to deploy, more flexible to reconfigure, and better in communication channel performance [2]. Meanwhile, [4] studies the measurement and modeling of air access channels between UAVs and base stations (BS) in typical urban and rural macrocell scenarios, which is helpful to develop more reliable and more efficient communication technologies for UAVs to access the territorial cellular networks. On the other hand, the authors of [3] focus on maximizing the coverage of UAV-based BSs under the constraint of the transmitting power. In [5], a unified framework for a UAV-assisted emergency network in disasters is established. However, existing literature have rarely investigated on the power control and channel allocation of D2D pairs in a D2D-abled UAVs-aided system. The purpose of this paper is to study a scenario lacking ground BSs (GBS), such as mountainous areas, isolated islands, disaster areas, and so on. Several UAVs, served as aerial BSs, can provide communication for areas lacking infrastructures with the help of D2D communications. However, in these emergencies, energy-efficient schemes are also required along with improving services.
Previous studies have mainly focused on energy efficiency (EE) on UAV-aided networks. In [6], the authors propose an energy-efficient multi-UAV coverage model based on spatial adaptive strategy. In [7], the authors try to maximize the energy efficiency of UAVs by designing the trajectory path of UAVs, taking into account both throughput and propulsion energy consumption of UAVs. Literature [8] studies the joint optimization of the UAV track, aiming at minimizing the total energy consumption of each node under the constraint that the estimated mean square error is less than the target threshold. With the exponential growth of data traffic, the use of caching and D2D communication has been recognized as an effective approach for mitigating the backhaul bottleneck in UAV-assisted networks [9]. The authors of [10] study a scenario that UAV-BS replaces destroyed traditional GBSs when a natural disaster occurs, a large-scale UAV is deployed to provide wireless service to ground devices with the support of D2D-enabled links. However, one UAV can only provide a small limited range of signal coverage. Meanwhile, in [11], the optimization of power control for D2D communication underlying UAV-assisted access systems is investigated, but channel allocation is not considered. In order to improve the above deficiencies, in this paper, we consider a D2D-abled UAV-aided system with multiple UAVs and propose a channel allocation and power control method aiming to maximize EE of D2D pairs.
To date, existing literature have investigated EE on an underlay D2D-enabled UAV-aided network. The main contributions of this paper are as follows. First, we aim to optimize EE of D2D pairs while considering QoS constraints in a D2D-abled UAVs-aided system. Second, by introducing adaptive quantities and weights, the quantity-weight adaptive salp swarm algorithm (QWASSA) is proposed to solve the optimization problem, such that the adaptive leader-follower numbers can have a better balance between exploration and development capabilities. Meanwhile, adaptive weights updating strategy makes it easier for QWASSA to jump out of the local optimum. Thus, we propose an efficient iterative resource allocation algorithm, the problem of optimizing D2D pairs' EE can be solved by the proposed QWASSA.
The rest of this paper is organized as follows. In Section II, the system model, the interference modeling and the problem formulation are introduced. The proposed method of power control and channel allocation is given in Section III. Simulation results are shown in Section IV, and conclusions are drawn in Section V.

II. SYSTEM MODEL
In this section, the system model, the interference modeling, and the problem formulation are illustrated in subsections A, B, and C, respectively.

A. System model
The system model is shown in Fig. 1. In some remote areas and extreme cases, such as mountainous areas and isolated islands where there is no GBS, as well as disaster areas where GBSs are out of service, but the rapid provision and recovery of communication are significant for people's production, life and disaster relief. Therefore, in this paper, we consider using several UAVs as aerial BSs to provide temporary communication connections for users, so as to solve the communication interruption caused by GBS absence. In a short time, we assume that UAVs are still in the air, which can provide stable communication for users within the coverage area. We consider reusing uplink resources because compared with the downlink spectrum resources, the utilization rate of the uplink resources is lower [12].
We consider a single-cell uplink scenario in which the ground base station does not exist. The UAVs are temporary aerial BSs, and the interference is limited. The users are distributed randomly in cells. Assume that each UAV is stationary in a short time, in this case, each UAV-BS controls several cellular UEs (CUEs) and surrounding D2D pairs, and each CUE or D2D pair can only be controlled by one UAV-BS. The cellular transmissions under the control of a UAV-BS are orthogonal while CUEs in disparate UAV-BSs can reuse the same channels. Similarly, a D2D pair can reuse only one CUE channel of the same UAV, and one CUE channel can be reused by multiple D2D pairs. The UAV-BS contains the complete information of the instantaneous channel state information of links to all UEs.

B. Interference modeling
As described in [13], we analyze several complicated interference scenarios in the system. Fig. 2 shows four different interference scenarios to D2D receivers (D2D R ), those are intra-cell interference of CUEs towards D2D R , intra-cell interference of other D2D transmitters (D2D T ) towards D2D R , inter-cell interference of CUEs towards D2D R and inter-cell interference of other D2D Ts towards D2D R .
Intra-cell interference of CUEs towards D2D R : As shown in Fig. 2(a), in a UAV-BS, the CUE creates interference towards D2D R , which reuses the same channel.
Inter-cell interference of CUEs towards D2D R : As shown in Fig. 2(b), this kind of interference, towards D2D R , is generated by CUEs in different UAV-BSs.
Intra-cell interference of D2D Ts towards D2D R : As shown in Fig. 2(c), the interference is generated by other D2D Ts , which reuse the same channel of D2D R in the same UAV-BS.
Inter-cell interference of D2D Ts towards D2D R : As shown in Fig. 2(d), this scenario shows the inter-cell interference where D2D Ts comes from other UAV-BSs interference with

C. Problem Formulation
Let k ∈ K = {1, 2, . . . , K} and m ∈ M = {1, 2, . . . , M } index the kth channel and the mth UAV, respectively. In the mth UAV-BS, let Dr m x and Dt m x denote receiver and transmitter of the xth D2D pair D m x , respectively. The intra-cell interference towards Dr m x is given by The inter-cell interference towards Dr m x is given by where X' means the set of X D2D pairs in the nth UAV-BS. a k,j represents the channel state indicator variables between D n j and D m x . It should be emphasized that D n j and D m x are in two different UAV-BSs. p n j and p n c denote the transmit power of Dt n j and CUE that shares the same channel k but in another UAV-BS. l n x,j and l n x,c are the channel gain, regardless of the link being cellular or D2D, and β denotes the path loss exponent.
Furthermore, the instantaneous signal to the signal-tointerference-plus-Noise ratio (SINR) of Dr m x can be expressed as where p m x denotes the transmit power of Dt m x , and l x is the propagation distance of D2D pair D m x . h m x represents the channel gain and σ 2 is the aggregate power of noise.
Therefore, the spectral efficiency (SE) [bit/s/Hz] of Dr m x is given by The total power consumption of D m x , which concludes transmit and circuit power, is given in the following where p 0 is the circuit power of D2D pair D m x . The circuit power needs to be multiplied by 2, because both Dt m x and Dr m x must be taken into account.
The EE of D2D pairs [bit/J/Hz] is given by Our objective is to maximize the EE of all D2D pairs in the system. Combining (1)-(6), we formulate the joint power control and channel allocation problem as an optimization problem as follows where SE min in (7a) is the minimum SE of D2D R , which satisfies the QoS requirement. Eq.(7b) and (7c) mathematically model our assumption that one D2D pair can only reuse one channel. Eq.(7d) ensures that the transmit powers of D2D Ts cannot go beyond the maximum limit.
We define A as the channel allocation matrix for all D2D pairs, in the same way, P as the power control matrix. This optimization problem is a complicated NP-hard problem, and no efficient polynomial-time solutions exist, as the complexity may increase. Therefore, we propose the quantity-weight adaptive salp swarm algorithm to devise a solution.

III. QUANTITY-WEIGHT ADAPTIVE SALP SWARM ALGORITHM
In this section, first, we introduces the salp swarm algorithm (SSA) and the adaptive salp swarm algorithm (ASSA). Then, we propose the the quantity weight adaptive salp swarm algorithm (QWSSA), which aims at solving the power control and channel allocation problem.

A. Salp Swarm Algorithm and Adaptive Salp Swarm Algorithm
In [14], the authors propose the first model of the salp chain. The salp chain can be divided into two groups: leader and follower. The leader salp is at the front of the chain, while the other salps are regarded as follower salps.
A two-dimensional matrix s is defined as the position of all salps, and an n-dimensional search space includes a food source F.
The leader salp position updates can be expressed where z = 1. s z w is the position of the leader in the wth dimensional. F w is the food position in the wth dimensional. ub w represents the upper bound of the wth dimension, and lb w represents the lower bound of the wth dimension. α 1 , α 2 and α 3 are random numbers in the interval [0, 1], respectively. Meanwhile, they dictate the next position in the wth dimension as well as the step size.
Furthermore, the update positions of follower salps are defined as where z ≥ 2 and s z w shows the zth follower position in the wth dimension.
Thus, the salp chain can be simulated by (8) and (9). It can be seen that there are two optimization matrices for the power control and channel allocation problem. Meanwhile, the channel allocation matrix A is discrete, while the power control matrix P is continuous. Thus, the continuous SSA can not be directly used to solve the optimization problem of power control and channel allocation for maximum energy efficiency in the D2D-abled UAVs-aided system. The authors in [15] propose the ASSA, which can solve the above problem.
The leader salp position updates can be described as and where P 1 and A 1 are two parts of the leader salp position. A best and P best are defined as the current optimal channel allocation and power control matrix in the iterative process, respectively. λ 1 , λ 2 and λ 3 are the same as α 1 , α 2 and α 3 in SSA. P max d represents the maximum power constraint. The follower salps positions are defined as and where A y and P y are two parts of follower salps positions. ω max and ω min are initial weight and final weight, respectively. F (A y , P y ) is represented for the fitness function, which is used to evaluate the current positions of follower salps. However, the number of leaders and followers is constant, which limits the searching abilities of these two algorithms. In order to overcome the above shortcomings and further improve the capability of global searching and local mining, QWASSA is proposed.

B. Quantity-Weight Adaptive Salp Swarm Algorithm
The numbers of leaders control the global searching ability, while the followers decide the local search ability. Thus, we introduce a leader-follower adaptive adjustment strategy based on the ASSA. When the population of the salp group is fixed, there are more leaders and fewer followers in the initial iteration, which means QWASSA focuses on the global search. As the iteration continues, the number of leaders decreases, and the number of followers increases. In the later stage of iteration, more followers make the QWASSA focus on local search, so the QWASSA is not easy to fall into local optimum.
The calculation formula of the proposed leader-follower number is learder salps : ρN f ollowerer salps : ρ means the weight and calculates by where µ is the scale factor of leader-follower, t and T represent the current iteration and maximum iteration, respectively. f is the perturbation deviation factor, and θ is an interval of [0, 1].
The leader salps positions are updated by and where 1 ≤ r ≤ ρN , A r and P r are two parts of leader salps positions. P best is defined as the current optimal power control matrix in the iterative process. λ 1 = 2e −(4t/T ) 2 , λ 2 and λ 3 are the constants of the interval [0, 1], respectively. P max d represents the maximum power constraint. The follower salps positions are defined as and where ρN ≤ y ≤ N , A y and P y are two parts of follower salps positions. µ 1 and µ 2 are the weight values, which µ 1 = 0.8 and µ 2 = 0.6. F (A y , P y ) is represented for the fitness function, which is used to evaluate the current positions of follower salps. In this paper, F (A y , P y ) = EE(A y , P y ). QWASSA is shown in Algorithm 1.
Algorithm 1 D2D Energy Efficiency Maximization Scheme based on the QWASSA in D2D-abled UAVs-aided system 1: Setting parameters of system and algorithm. 2: Input: the position A and P of salps. 3: while t ≤ T do do 4: //Calculate food fitness of salp F (A y , P y ) by using (6), and update F best , A best and P best . 5: if F best (t) ≤ F best (t − 1) then 6: F best (t) = F best (t − 1); A best (t) = A best (t − 1), P best (t) = P best (t − 1); update the position A and P of salps by using (7b), (7c) and (7d); update the numbers of leaders and followers by using (14) and (15). 7: for 1 ≤ r ≤ ρN do 8: update the position A and P of leader salps by using (16) and (17). 9: end for 10: for ρN ≤ y ≤ N do 11: update the position A and P of follower salps by using (18)  end if 14: end while 15: Output: F best , A best and P best .

IV. SIMULATION RESULTS
In this section, simulation results verify the theoretical analysis and the effectiveness of the proposed approaches. In the simulation, we consider a community of 1000 square meters without GBSs. Therefore, UAV-BSs are used to provide communication for users in the cell. In addition, it is assumed that in a short period time, the UAV-BSs are in a static state and can provide communication for users within the signal coverage range stably. D2D pairs are generated by the distance between two CUEs (i.e., if the distance between two CUEs is less than a constant value, the two CUEs conduct D2D communication). At the same time, UAV-BSs are deployed according to the K-means clustering algorithm. Default parameters used in the simulation are given in Table  II. Moreover, the SSA, the ASSA, and the particle swarm optimization (PSO) algorithm adopted as baseline to verify the superiority of our proposed QWASSA. Fig. 3 shows the fitness convergence curves of the SSA, the ASSA, the PSO and, our proposed QWASSA. Compared with the other three algorithms, QWASSA can significantly improve the efficiency of D2D pairs. After 140 iterations, the QWASSA rises steadily while the other algorithms remain the same. Therefore, QWASSA is not easy to fall into local optimum, and its search accuracy is higher. This is because that the QWASSA adopts adaptive followers positions updating and leader-follower numbers updating, and which has strong local mining ability and global searching ability. Fig. 4 compares the performance of D2D pairs EE of the PSO, the SSA, the ASSA, and our proposed QWASSA for different D2D pair distances, but the deployment of UAVs in the system keeps the same. Set the minimum rate to 2bit/s/Hz. With the increase of the distance between D2D R and D2D T , the EE of D2D pairs in the system increases monotonically due to the increase of the number of the D2D pairs. However, compared with the other three algorithms, the QWASSA can make the system more energy-efficient for different distances. Fig. 5 shows the average EE at different distances for the same UAV deployment. Although the total D2D EE increases, which is showing in Fig. 4, the average EE decreases. The more D2D pairs that each UAV-BS controlled, the greater inter-cell and intra-cell interference between D2D pairs, and the lower average EE. However, compared with the SSA, the ASSA and the PSO, the D2D EE of the proposed QWASSA decreases slower. Fig. 6 shows the relationship between D2D EE and the numbers of UAV in the system, and the distribution of D2D pairs keeps the same. As the number of UAVs increases, the D2D EE decreases monotonously. It means that when the numbers of D2D pairs are fixed and the numbers of UAVs increase, the D2D pairs will produce more inter-cell interference, resulting in a decrease of D2D EE in the system. Although the system D2D EE of the QWASSA scheme decreases with the increase of UAVs number, its performance remains superior to the other three baseline algorithms.

V. CONCLUSION
In this paper, we investigate the channel allocation and power control problem in the D2D-abled UAVs-aided system. We aim to maximize the energy efficiency of D2D pairs while guarantying the QoS of networks. To solve this problem, we propose the QWASSA algorithm. The adaptation of leaderfollower numbers leads to faster convergence speed and more excellent local mining capability of the QWASSA. Meanwhile, the position update mode of leader salps makes it impossible for iteration to fall into local optimization. The QWASSA performs the best in terms of D2D pairs energy efficiency in the D2D-abled UAVs-aided system compared with the other benchmark schemes.