Joint Task Scheduling,Resource Allocation,and UAV Trajectory under Clustering for FANETs

Wenjing You,Chao Dong,Qihui Wu,Yuben Qu,*,Yulei Wu,Rong He

1 The Key Laboratory of Dynamic Cognitive System of Electromagnetic Spectrum Space,Ministry of Industry and Information Technology,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China

2 AVIC Chengdu Aircraft Design and Research Institute,China

Abstract: This paper establishes a new layered flying ad hoc networks (FANETs) system of mobile edge computing(MEC)supported by multiple UAVs,where the first layer of user UAVs can perform tasks such as area coverage, and the second layer of MEC UAVs are deployed as flying MEC sever for user UAVs with computing-intensive tasks.In this system, we first divide the user UAVs into multiple clusters, and transmit the tasks of the cluster members(CMs)within a cluster to its cluster head(CH).Then,we need to determine whether each CH’ tasks are executed locally or offloaded to one of the MEC UAVs for remote execution(i.e.,task scheduling),and how much resources should be allocated to each CH (i.e., resource allocation), as well as the trajectories of all MEC UAVs.We formulate an optimization problem with the aim of minimizing the overall energy consumption of all user UAVs, under the constraints of task completion deadline and computing resource, which is a mixed integer non-convex problem and hard to solve.We propose an iterative algorithm by applying block coordinate descent methods.To be specific, the task scheduling between CH UAVs and MEC UAVs,computing resource allocation, and MEC UAV trajectory are alternately optimized in each iteration.For the joint task scheduling and computing resource allocation subproblem and MEC UAV trajectory subproblem,we employ branch and bound method and continuous convex approximation technique to solve them,respectively.Extensive simulation results validate the superiority of our proposed approach to several benchmarks.

Keywords:flying ad hoc networks(FANETs);successive convex approximation; clustering; mobile edge computing(MEC)

I.INTRODUCTION

Unmanned aerial vehicles(UAVs), or say drones, are generally known to have the advantages of low cost,high flexibility, and bird’s-eye view, and widely used in many critical applications such as reconnaissance,surveillance, and data collection tasks [1-4].Nevertheless, due to the limited energy and load of a single UAV, the coordination of multiple UAVs to complete complex tasks has become a research hotspot.With the development of electronics and communication technologies, UAVs are showing a trend of integration and miniaturization.Following this, UAV ad hoc networks,i.e.,flying ad hoc networks(FANETs),consist of dozens or even hundreds of small and micro UAVs,which has received extensive attention from industry and academia.However, it will bring communication problems such as routing overhead.Clustering[5,6]is an effective networking and management technology for UAV self-organizing networks.UAV clusters could perform tasks more efficiently under the clustering architecture.

Recently,with the development of communications and artificial intelligence (AI), computing-intensive applications are triggering a revolution in mobile applications.Since UAVs used for data collection usually have relatively weak computing power and limited battery capacity,it becomes a challenge to quickly process local dense data on device.The development of mobile edge computing(MEC)enables UAVs to offload computing-intensive tasks to nearby MEC servers through wireless transmission links, which will effectively alleviate the above challenge.Considering that the traditional MEC servers on the ground are restricted by the fixed geographical location, a UAVs-assisted MEC architecture [7-11] that integrates medium and large sized UAVs with strong computation abibilities into the MEC network came into being.Following that architecture, in this paper, we consider two roles of UAVs in the studied FANET,i.e.,user UAVs who generate computing-intensive tasks and MEC UAVs who serve as flying MEC servers.

In this work,we study a FANET surveillance system composed of multiple user UAVs and multiple MEC UAVs, where the user UAVs are divided into several clusters,each one consisting of one cluster head(CH)and several cluster members(CMs).Each user UAV is equipped with sensors for data collection and moves in a specified area to cover the task point following the predefined trajectory.And within each cluster,the collected data by all CM UAVs is gathered together at the CH UAV.While each CH UAV can execute the tasks of its CM UAVs by itself,it is energy consuming,which might be a concern for user UAVs with relatively limited energy.Accordingly,the computing tasks of user UAVs could also be offloaded to those MEC UAVs for remote execution.

In this system, we should make the following decisions.Firstly, we need to determine the association of CH UAVs and MEC UAVs (also task scheduling or offloading decision),i.e.,the task of each CH UAV should be offloaded to which MEC UAV.Secondly,after offloading, each MEC UAV should allocate how much computing resources to each task(i.e.,resource allocation).Lastly, since all user UAVs are flying under predefined movement trajectories, the trajectories of all MEC UAVs should also be optimized (i.e.,UAV trajectory).We mathematically formulate the problem of joint task scheduling, resource allocation,and UAV trajectory under clustering for FANETs as a mixed integer non-convex problem, which is hard to solve.Leveraging block coordinate descent methods,we propose an iterative algorithm to solve the aforementioned problem.Specifically, for the joint task scheduling and computing resource allocation subproblem, we use branch and bound method to obtain its optimal solution, while for the MEC UAV trajectory subproblem, we employ continuous convex approximation technique to obtain its locally optimal solution.The main contributions of this paper are summarized as follows.

• We first study the problem of joint task scheduling, resource allocation, and UAV trajectory under clustering for FANETs, which jointly optimizes task scheduling between CH UAVs and MEC UAVs, computing resource allocation, and MEC UAVs trajectory, to minimize the user UAVs’ energy consumption under computing resource and task completion deadline constraints.

• To solve the aforementioned mixed integer nonconvex problem, an iterative algorithm based on the decline of block coordinates is proposed,which decomposes it into two subproblems.The branch and bound method is used to solve the joint task scheduling and computing resource allocation subproblem, and the continuous convex approximation technique is employed to solve the MEC UAV trajectory subproblem.

• The performance of the proposed algorithm is evaluated through extensive simulations.Compared with two other benchmarks,when the CPU cycles of user UAVs workload change, the proposed algorithm reduces the total energy consumption of user UAVs by 71.90% and 27.38%,respectively.With the change of the maximum number of CHs that can access to MEC UAVs,the overall energy consumption of user UAVs is reduced by 60.52%and 41.56%,respectively.

The rest of this paper is organized as follows.Section II gives the review of the related works.Section III introduces system models and formulates an optimization problem.Section IV presents the proposed algorithm that decouples the problem into two subproblems,then solves it iteratively.Section V presents the simulation and performance evaluation.Section VI concludes this paper.

II.RELATED WORK

2.1 UAV Trajectory Planning

From the perspective of communication, compared with traditional fixed base station deployment [12],the use of UAVs can greatly improve communication performance, such as providing a higher data rate or having a larger coverage area.In order to improve communication performance, in the UAV static network,the main research is to optimize the deployment of UAVs.In the UAV dynamic network, the research mainly focuses on the UAV trajectory planning.To maximize the sum-rate of the two-user network, [13]derived the optimal power allocation and UAV placement.[14]proposed to optimize the altitude of UAVs to provide maximum coverage on the ground.[15]proposed an adaptive deployment scheme for a UAVaided communication network.The UAV adapts the displacement direction and distance for serving randomly moving users’ instantaneous traffic in the target region.[16] studied a communication system assisted by multiple UAV-mounted base stations (BSs),aiming to minimize the number of required UAVs and improve the coverage rate by optimizing the threedimensional (3D) positions of UAVs, user clustering,and frequency band allocation.[17] investigated the UAV trajectory planning to reduce the time required for UAVs to fly from one IoT device to another IoT device,thereby maintaining the freshness of data collection information in the UAV-assisted IoT networks.In[18]and[19],a UAV acts as a MEC server to perform the task of offloading from the ground terminal.[18]jointly optimize the user association, UAV trajectory,and upload power of each terminal, to maximize the total amount of unloading of all terminals.[19] aims to minimize the average weighted energy consumption of SMDs and the UAV.[20] considered a multi-UAV enabled wireless communication system,where multiple UAVs serve as aerial base stations to serve a group of users on the ground.For the non-convex UAV trajectory problems, solving it by applying the successive convex optimization techniques.In the application of UAVs-assisted MEC, the trajectory planning of the UAVs can effectively guarantee the timeliness of the collected IoT data processing, and can reduce the energy consumption of data transmission and calculation.However, the problem of user association and UAV trajectory optimization is actually a mixed integer non-convex problem, which is challenging to solve.

2.2 Task Scheduling in MEC

Task scheduling or offloading in traditional terrestrial MEC such as [21, 22] do not consider the trajectory planning of UAVs when UAVs are served as MEC servers.Wuet al.[23] jointly optimized the trajectory and user scheduling of a single UAV base station to maximize the minimum average upload rate of the terminal, without considering the UAV battery capacity.Since limited energy is an important issue in the design of UAV-assisted MEC system, [24]and [25] focused on energy-efficient resource allocation and computational offloading issues,respectively.Zhouet al.[26] studied a UAV-enabled MEC wireless powered system under partial and binary computational unloading modes.Baiet al.focused physicallayer security, and proposed an energy-efficient computation offloading technique for UAV-MEC systems.References[27-29]conduct research on the provision of MEC services by multiple UAVs.Yanget al.[27]considered the problem of minimizing sum power in the MEC network with multiple UAVs, via jointly optimizing user association, power control, computation capacity allocation, and location planning.Considering both computation bits and energy consumption, [29] formulated a computation efficiency maximization problem in a multi-UAVs assisted MEC system.The above mentioned studies are based on the assumption that each user can only offload the task bits to one UAV.[28]conceived a multi-UAV-assisted multi-access MEC model, which allows each user to offload different parts of task bits to multiple UAVs simultaneously for parallel computing.However, in most existing researches, UAVs enabled mobile-edge MEC servers systems provide computing services for ground users with low mobility.When a group of UAVs need computing services after performing tasks such as area coverage, existing researches cannot be directly applied due to the high mobility of UAVs.For UAV clusters, how to effectively provide them with computing services to reduce the energy consumption of user drones is a question worth considering.

III.MODELS AND FORMULATION

3.1 Network Model

We consider a UAV-enabled MEC surveillance system, which is composed ofNuser UAVs andMMEC UAVs, as shown in figure 1, where the former equipped with camera for regional monitoring,and the latter equipped with MEC servers to provide computing services for the former.Here,the UAVs with cameras are called user UAVs, and the UAVs with MEC servers are called MEC UAVs.The user UAV set is denoted asN={1,2,...,N},and MEC UAV set is denoted asM={1,2,...,M}.We assume that there areStask points in the research area,and the task set is denoted asS={1,2,...,S}.Assuming that the user UAVs have been divided into clusters, in whichH ⊆Nis the set of cluster head (CH) UAVs, andCMj ⊂Nis the set of cluster member (CM) UAVs associated with CH UAVj ∈H.The user UAVs visit theseS knowntask points in units of clusters,and genetic algorithm is used to optimize the trajectory of CH UAVs to ensure that the user UAVs cooperatively visit the task points with the shortest flight distance.Therefore,the mechanical energy consumption of user UAVs is fixed during the task execution,and not considered in the total energy consumption of user UAVs in the later problem formulation.We divide the mission period intoTtime slots with an equal slot lengthτ,where the set of the time slots is denoted asT={1,2,...,T}.The positions of CH UAVs is denoted by (), wherej ∈H.The positions of CM UAVs are denoted by(),wherek ∈CMj.The positions of MEC UAVs at timetis expressed as.We assume that MEC UAVs and user UAVs are flying at different constant altitudes to avoid collisions.

3.2 Task Execution Model

As for user UAVs, we assume that each of them constantly generates taskat time slott, wheret ∈T.Specifically,can be described as= (,wheredenotes the total number of CPU cycles required to calculate task,anddenotes the data size of taskin whichl ∈(H ∪CM).As for MEC UAVs, we assume that they are equipped with directional antennas of fixed beam widthθ, and the maximum computation resource that the MEC UAVican provide in each time slot is.In this paper,we apply the binary computation offloading mode,introducing the discrete binary offloading indicator variables,i ∈M,j ∈H,t ∈T}of CH UAVs, where= 1 denotes that at time slott, MEC UAViprovides computing service for CH UAVj, and= 0denotes that the CH UAVjconducts task itself without offloading at time slott.We assume that each CH UAV can offload task to at most one MEC UAV,that is

Table 1. A list of major symbols used in this paper.

And we suppose that the maximum number of CH UAVs served by each MEC UAViin thetslot isthat is

In each time slot, due to the limited computing resource that MEC UAVican provide,we have

where,is the computing resource provided by the MEC UAVifor the CH UAVjin slott, andis the maximal computation capacity in terms of CPU frequency cycles at MEC UAVi ∈M.

The Euclidean distancefrom CM UAVkto CH UAVjis expressed as

The data transmission rate from CM UAVkto CH UAVjis expressed as

whereBis the channel bandwidthis the transmit power of CM UAVk,α=g0G0/σ2,G0≈2.2846,g0is the channel gain per unit distance, andσ2is the noise power.Note that there exist other channel models such as the A2A model capturing free-space path loss model and additional attenuation factors added to the free space propagation model for the LoS link[30],and we would like to study the effect of more A2A channel models in the future work.For all user UAVs,we assume the MAC protocol is based on orthogonal frequency division multiplexing access(OFDMA),so the mutual interference between them is not considered.

If the CH UAVjdecides to offload the task to the MEC UAViin slott,then the cluster head UAV should be within the communication range of MEC UAVi,that is

whereis the communication coverage of MEC UAVi.The Euclidean distancebetween CH UAVjand MEC UAViin slottis expressed as

The uplink data rate from CH UAVjto MEC UAViis expressed as follows

whereis the transmit power of CH UAVj ∈H.

Since the time of returning calculation results is usually considered to be negligible, the task completion time is divided into communication time and calculation time.Specifically,if CH UAVjoffloads tasks of the whole clusterjto MEC UAViin slott,communication time is composed of two parts,according to(9)and(10),one is the maximal transmission time among CM UAVs to the CH UAVj,and the other is transmission time that CH UAVjto MEC UAVi.If the CH UAVs execute tasks locally, the communication time only includes the first part,i.e.,

Then, the communication energy consumption from CM UAVkto CH UAVjis

The communication energy consumption from CH UAVjto MEC UAViis

According to (13) and (14), the calculation time is divided into local computing time of cluster headjand MEC computing time of MEC UAVi.

Then, the energy consumption for CH UAVjlocal computing is

where,kj ≥0 is the effective switched capacitor,vjis a constant.

Therefore,the total task completion time of the clusterjcan be expressed as

The total energy consumption for user UAVs of clusterjin the slottis expressed as

Suppose that each task needs to be completed within the maximum time slot lengthTmax,that is

3.3 Problem Formulation

In this work, we consider jointly optimizing the task scheduling between CH UAVs and MEC UAVs,computing resource allocation, and MEC UAVs trajectory to to minimize the total energy consumption of user UAVs under computing resource and task completion time constraints.MEC UAVs trajectory set is defined asU=, the task scheduling set between MEC UAVs and CH UAVs is defined asA={}, computing resource allocation set is defined asF={}.The problem of minimizing the energy consumption of user UAVs can be formally formulated as follows:

whereU,A,andFare optimization variables.(19)-(b)indicates that each CH UAV could offload its task to one of the MEC UAVs or execute it by itself.(19)-(c)means that each MEC UAVican only serve at mostCH UAVs at the same time.(19)-(d) means that the total amount of required computing resources cannot exceed the amount of computing resources that MEC UAVican provide.(19)-(e) indicates that any CH UAV can offload its task to some MEC UAV,only if it is within the communication coverage of this MEC UAV.(19)-(f)means that any task must be completed within the maximum deadlineTmax.It is not difficult to obtain that the above problem is a mixed integer non-convex problem, due to the involved mixed integer variables and non-convex expressions by the UAV trajectory variable.

IV.SOLUTION

In this section, an iterative algorithm is proposed for solving the formulated optimization problem.The problem (19) is decomposed into two subproblems.Specifically, for given MEC UAVs trajectoryU, we first determine computing resource allocation variableFaccording to the maximum allowable completion time of the task, then we optimize task scheduling variableAby solving a multiple-choice multidimensional 0-1 knapsack problem (MMKP).For given task scheduling variableAand computing resource allocation variableF, MEC UAV trajectoryUis optimized based on successive convex approximation(SCA)techniques.And the block coordinate descent(BCD)method is used to solve the two subproblems iteratively.Since the task scheduling and resource allocation subproblem is the subproblem of the original problem in(19)under given MEC UAVs trajectoryUand the MEC UAVs trajectory subproblem is the subproblem of the original problem under given task schedulingAand computing resource allocationF,solving the original problem subject to(A,F,U)is equivalent to solving these two subproblem.

4.1 Task Scheduling and Resource Allocation Subproblem

Given MEC UAV trajectoryU,the subproblem of task scheduling variableAand resource allocation variableFcan be expressed as

If the task is unloaded to MEC UAV for calculation,according to (9), (10), (14), and (16), (19)-(f) can be written as

If the task is calculated locally by CH UAV,according to(9),(10),(14),and(16),(19)-(f)can be written as follows:

Then problem(20)can be rephrased as

whereG=is the trajectory vector of the CH UAVs.Actually, we consider that the task will be completed in the maximum allowable time, that is, the constraint (23)-(e) is equal to determineF.So, the computing resources of the MEC UAVs can be allocated to more CH UAVs under the task completion time constraint.With fixed MEC UAVs trajectoryUand computation resource allocationF, we can see that the the problem (23) is linear toA, and it is to solve the matching problem of multiple MEC UAVs and multiple CH UAVs, which is a classical multiple choice multi-dimensional 0-1 knapsack problem(MMKP).Although MMKP is usually a NP-hard problem,the standard optimizer MOSEK can be used to obtain the optimal solution by using Branch and Bound (BnB) method for the problem with relatively small size.

4.2 MEC UAV Trajectory Subproblem

Given the task scheduling variableAand resource allocation variableF, the problem of solving MEC UAVs trajectoryUcan be expressed as

where the objective function and constraints are continuous function on the MEC UAV trajectory variableU.However, the subproblem (24) is non-convex because the objective function and the second constraint are both non-convex onU.So,the subproblem(24)is difficult to solve directly.

Lemma 1.The subproblem(24)is not a convex optimization problem since both the objective function and constraint(24)-b are non-convex.

Proof.We define

Then,f(x)finds the first derivative with respect tox,that is

Andf(x) finds the second derivative with respect tox,that is

We can seef(x)′′is not always greater than 0 for allx ∈domf,sof(x)is a non-convex function.Both the objective function and constraint conditions of subproblem(24)include this non-convex component,thus the subproblem(24)is a non-convex problem.

In order to solve this non-convex subproblem, we apply SCA technique to convert it into a convex optimization problem, thus decreasing the complexity of this subproblem.First, we introduce a continuous slack variableφ={),∀i ∈M,j ∈M,t ∈T}

So problem((24))can be converted to problem(29).

Notice that the objective function of problem (29) is convex aboutφ.And the constraints are convex functions aboutandφ,however,they are nonconvex with respect to theU.So, we leverage SCA approach to deal with constraints to obtain the locally optimal solution of problem(29).

Lemma 2.Employ SCA to handle non-convex constraints a and b, thus transforming problem (29) into a convex problem in U.To be specific, the could be approximated bye))in the e-th iteration of SCA.The strict lower bound of)is

After applying SCA according to Lemma 2, problem(29)can be rewritten as

Problem(36)is a convex optimization problem,which can be solved effectively by CVX.The iterative algorithm based on block coordinate descent is shown in Algorithm 2,where the MEC UAV trajectory subproblem is solved by Algorithm 1.

?

Lemma 3.The time complexity of Algorithm 2 is O(2MHT+ImaxM3T3), where M is the number of MEC UAVs, H is the number of CH UAVs, T is the number of time slots,and Imax is the number of maximum iterations in the SAC-based algorithm (i.e., Algorithm 1),respectively.

Proof.Note that the complexity mainly exists in the step 2 and step 3, since the solution of F can be directly obtained by (23)-(e) in the step 1.In the step 2, as we employ BnB method to solve the MMKP problem about A, it needs exponential time in the worst case and the time complexity isO(2MHT),whereMis the number of MEC UAVs,His the number of CH UAVs, andTis the number of time slots, respectively.In the step 3, we utilize SCA to solve the non-convex problem about U, which takesO(Imax(MT)3), whereImaxis the number of maxi-mum iterations.Summarizing the above results concludes this lemma.

Table 2. Simulation setting.

V.PERFORMANCE EVALUATION

In this section, we evaluate the performance of the proposed joint MEC UAV trajectory planning, task scheduling, and computing resource allocation optimization algorithm (i.e., Algorithm 2, labeled as“AFU” in all simulation figures) through extensive simulations.Considering the UAV-assisted mobile edge network under clustering architecture, the number of ground task pointsSis 70, which are randomly distributed in the area of 150m×150m.The number of MEC UAVs, CH UAVs, and CM UAVs areM,NhandNm, respectively.The data size of intensive computing resources generated by UAV in each time slot is[100,1000]KB, and the number of CPU cycles required is[108,109]cycles.Maximum computation capacity of each MEC UAV[5,10]GHz.The initial horizontal positions of the two MEC UAVs are (0,0) and (0,200) respectively,and the final horizontal positions are(200,200)and (200,0) respectively.In order to avoid conflicts between UAVs,we set the UAVs to different altitudes.The channel gaing0of the reference distance per unit distance is 1.42×10-4.Note that in the AFU algorithm,the convergence is achieved when the following inequality about the objective values in two successive iterations holds:|Obje-Obje-1|/Obje-1≤ϵ,whereObjeis the objective function value after iterationeandϵis set as 0.001 in the simulations.We consider the following benchmark algorithms in the simulation.

•GGU.The optimization problem is solved following the approach in [31].Similar to [31], the problem is solved in an iterative way, where the task scheduling subproblem(a binary integer programming problem)is solved by the greedy algorithm(Algorithm 1 in[31]),and the joint resource allocation and MEC UAV trajectory subproblem is solved by the steepest descent method similar to Algorithm 3 in[31].

•FixU.MEC UAV moves with a fixed trajectory,only optimizingAandF.

•Local.All computing tasks are performed locally in the cluster head.

Figure 2 shows the trajectory planning results of 4 cluster head UAVs and 2 MEC UAVs.The ground mission is distributed in an area of 150m×150m,CH UAVs start from different initial positions, and visit task points in the region according to the optimized trajectory, and finally returns to initial positions.The solid lines represent the optimized trajectory of CH UAVs,and the dotted lines indicate the optimized trajectory of MEC UAVs.The initial horizontal positions of the two MEC UAVs are (0,0) and (0200) respectively,and the final horizontal positions are(200,200)and (200,0) respectively.The altitude difference between the two MEC UAVs is 50 m.If the trajectory of the MEC UAVs are not optimized,the MEC UAVs may fly diagonally, thus losing more opportunities to provide computing services to CH UAVs.In AFU,the MEC UAVs’trajectory are optimized according to the trajectory of the CH UAVs to reduce the transmission energy consumption and the local calculation energy consumption of the CH UAVs.

Figure 1. An illustration of a FANET surveillance system with UAV clustering.

Figure 2. UAV trajectory planning.

Figure 3. The effect of user UAV’s workload.

Figure 3 shows the total energy consumption of user UAVs versus the number of CPU cycles required by user UAVs’workload.With the increase of CPU number required by user UAVs, due to limited computing resources of each MEC UAV, user UAVs may have to perform tasks in the local CH UAVs, resulting in the increase of overall energy consumption of all user UAVs.AFU is better than the other three benchmarks,and Local is better than FixU and GGU.This is because the computing tasks of the Local scheme are all executed in the CH UAVs, resulting in high total energy consumption of user UAVs.In the FixU scheme,MEC UAVs fly in a fixed diagonal line, which can provide computing services for user UAVs.Because MEC UAV trajectory is not optimized, it cannot provide computing services for more user UAVs for a long time,so the energy consumption performance is worse than AFU.More specifically, when the workload of each user UAV changes from 4×108 CPU cycles to 109 CPU cycles, AFU reduces the overall energy consumption of user UAVs by 71.90%and 27.38%respectively compared with Local,FixU and GGU.

Figure 4 shows the total energy consumption of user UAVs versus the number of clusters.We can observe that as the number of user UAV clusters increases,the amount of computing tasks generated by user UAVs increases.So the total energy consumption of user drones under all algorithms increases.When the number of clusters does not exceed 3,the total energy consumption in AFU, FixU and GGU increases slowly.When the number of user UAV clusters is greater than 3, the GGU energy consumption increases the most,and the AFU performance is the best among these algorithms.Compared with Local, FixU and GGU,AFU reduces the total energy consumption by 60.52%and 41.56%respectively.

Figure 4. The effect of number of CHs.

Figure 5 shows the total energy consumption of user UAVs versus the maximum number of CH UAVs that MEC UAVs are allowed to access.It can be seen from the figure that with the increase of the maximum number of cluster head UAVs that MEC UAVs are allowed to access,the total energy consumption of user UAVs in other algorithms is reduced except for the algorithm“Local”.This is because the stronger the communication access ability of MEC UAVs, the more tasks the cluster head UAVs can offload to MEC UAVs.We can also observe that when the maximum number of CH UAVs allowed to access is greater than 1,the total energy consumption of user UAVs in AFU and FixU algorithm decreases significantly,while GGU decreases sharply when the maximum number of CH UAVs allowed to access is greater than 2.Compared with Local,FixU and GGU,AFU reduces the total energy consumption by 60.52%and 41.56%respectively.

Figure 5.The effect of MEC UAV’s maximum accessed user UAVs.

Figure 6. The effect of number of MEC UAVs.

Figure 6 shows the total energy consumption of user UAVs versus the number of MEC UAVs.From figure 6, except for algorithm “Local” that all tasks are executed locally on CH UAVs, the total energy consumption of user UAVs in AFU, FixU and GGU decrease with the increase of MEC UAVs, and the energy consumption of AFU is lower than that of FixU and GGU.When the number of MEC UAVs increases to 4, the total energy consumption of user UAVs in AFU and FixU are similar.This is because with the increasing number of MEC UAVs,CH UAVs have more opportunities to be within the communication coverage of MEC UAVs.That is to say, CH UAVs can offload more computation tasks to MEC UAVs, thus reducing the total energy consumption of user UAVs.When the number of MEC UAVs increases to a certain value, even if the trajectories of MEC UAVs are not optimized,multiple MEC UAVs with fixed tracks can also cover a large area and provide computing services for more CH UAVs.Therefore,when the number of MEC UAVs increases to a certain value,the energy consumption performances of AFU, FixU and GGU are similar.When the number of MEC UAVs increases from 1 to 4, AFU reduces the total energy consumption of user UAVs by 81.17%and 49.46%respectively compared with Local,FixU and GGU.

Figure 7. Convergence of the proposed algorithm.

Figure 7 presents the convergence of our proposed algorithm.We can observe that both AFU and GGU converge within only five iterations when number of MEC UAVs is two and number of CH UAVs is four.The fast convergence of the algorithm proves the effectiveness of the algorithm.which confirms the effectiveness of the proposed algorithm.Figure 7 also shows the AFU is better than GGU in terms of user UAVs’ total energy consumption.Specifically, after five iterations, the user UAVs’ total energy consumption of AFU decreases from 44.08 J to 30.6827 J,while that of GGU decreases from 44.08 J to 33.2691 J.

VI.CONCLUSION

In this paper, we investigated a UAVs-enabled MEC surveillance system under clustering FANETs.Specifically, the task scheduling of cluster-head UAVs and MEC UAVs,computing resource allocation and MEC UAVs trajectory are jointly optimized to minimize the user UAVs’ overall energy consumption under computing resources and task completion time constraints.An iterative algorithm(AFU)is proposed to solve this problem.The proposed algorithm currently only optimizes the horizontal trajectory of the MEC UAV,and the optimization of the height of the MEC UAVs as well as considering more UAV communication models[32,33]and distributed channel access[34]will be studied in the future.

ACKNOWLEDGEMENT

This work is supported in part by the National Natural Science Foundation of China under Grant No.61931011, in part by the Primary Research & Developement Plan of Jiangsu Province No.BE2021013-4,in part by the National Natural Science Foundation of China under Grant No.62072303, in part by the National Postdoctoral Program for Innovative Talents of China No.BX20190202, in part by the Open Project Program of the Key Laboratory of Dynamic Cognitive System of Electromagnetic Spectrum Space No.KF20202105.