Dynamic Routing and Coordination of Cluster for Unmanned Aerial Vehicle (UAV) Swarms

The ﬂying networks provide an eﬃcient solution for a wide range of military and commercial purposes. The demand for portable and ﬂexible communication is directed towards a quick growth in interaction among unmanned aerial vehicles (UAVs). Due to the frequent change in topology and high mobility of vehicles, routing and coordination becomes a challenging task. To maximize the throughput of the network, this study addresses the UAV swarm’s problems related to the coordination and routing and deﬁnes the proposed solution to solve these issues. For this, a network is assumed which contains an equal number of dynamic vehicles. It also presents the communication graph concept of UAVs and designs a ﬁxed-wing UAV model to improve the eﬃciency of the network in terms of throughput. Furthermore, the proposed algorithm based on Cauchy particle swarm optimization (CPSO) aims towards the better performance of UAV swarms and aims to solve the combinatorial problem. The simulation results show and conﬁrm the performance of the proposed algorithm.


Introduction
In recent years, unmanned aerial vehicles (UAVs) have gained interest and have been applied in the area of search and rescue, agriculture defence, transportation, monitoring, and surveillance [1]. ese vehicles can cover wider areas in the field. e coordination and monitoring system is needed for the fleet to determine their route and utilize this resource in a very efficient manner [2]. e UAV technology is also known as utilizing a network of UAVs. It attains sensing and actuating equipment to collectively perform the complicated task. Cooperation and coordination between the networks of UAV swarms will help to complete complicated missions across a large area. e decision making for the next position occurs at many stages of communication. Mostly many vehicles communicate via air to ground or air to air along with their control stations [3][4][5].
is technology also provides flexible configuration, improvement in the communication performance, and line-of-sight communication link on demand. Many studies show and provide a flexible solution for routing problems. A hierarchical structure is considered to have a UAV cluster. With the help of the proposed algorithm, routing takes place within the cluster and it is a basic tool for cluster communication [6]. Some other methods are used to improve the throughput of a channel by which UAV's applications expand vastly but still issues are present [7]. e UAV network information sharing is a more challenging problem limited to the throughput especially in disaster areas. It affects the overall effectiveness of the operation. For UAV swarms, different methods and techniques are designed for an optimal solution. By using these techniques, capacity of a channel is improved with an increase in the rate of the spatial multiplex.
With the development of technology, the spectrum deployment and communication ratio is highlighted. Previously, many scientists tried to solve the UAV swarm problem with different technologies, but still there are some issues to be solved. Many researchers had studied the cellular network of UAVs [8,9]. It is required to centralize the control due to the high intelligence level of unmanned vehicles. To address the need of UAV's network information distribution necessities, the challenges shall be added especially in distributed scenarios that affect the overall operation [10]. So, the effective techniques are specifically designed for UAV swarms. Cauchy particle swarm optimization (CPSO) has an effective result bearing key factors enhancing effectiveness that are topology structure, parameter selection, and swarm initialization [11,12]. But in PSO, the disadvantage is that it easily falls into local optimum. erefore, algorithm CPSO attains the best solution in mimimum time. e operators help in reducing the possibility of immature convergence by dealing with every ant individually.
In [13], the study provides the solution for the routing problem of the vehicle. e problem is split into a set called a cluster. is study finds the best possible distance of the route of the vehicle. e PSO technique is adopted to solve the routing issues. In [14], the improved version of PSO is called combinatorial particle swarm optimization which is used to find the number of clusters. It also increases the population variety and eliminates terminated particles. It also increases the speed of junction and solution's quality. In another study, a CPSO technique is used [15]. On the other hand, the inertia weight factor and chaotic maps are introduced to solve the clustering optimization.
is study combines CPSO with weight factor and chaotic maps to attain better performance. In [16], clustering particle swarm optimization (CPSO) is proposed for the efficiency of search.
is algorithm cannot guarantee convergence due to the sensitivity of the original cluster. e CPSO algorithm finds the best possible solution. Similarly, in another study [17], topology-based routing protocols improve the adeptness of the system. It also focuses on a detailed review of routing protocols that are topology-based. Furthermore, simulation evaluates the topology-based routing protocols. In [18], particle swarm optimization (PSO) with software-defined networking-based communication protocol is proposed which guarantees network management. Novel particle selection criteria are proposed for coordination purposes, which aims to guarantee network manageability of UAV formations, thus being able to guarantee service persistence in case of node failure occurrence. In [19], the article presents a review on the major research areas of unmanned vehicles, i.e., 3-degree-of-freedom (3DOF) routing algorithm and routing protocols.
is article also aimed to provide a comparison between the methods and algorithms to find the best ones among these algorithms for a particular application. On the other hand, in a dynamic environment, the stability of topology is presented. e algorithm used in that article is the cluster routing algorithm used for the stability of topology and proper transmission. e proposed algorithm in [20] is paralleled with ant colony optimization (ACO) and particle swarm optimization (PSO). In [21], a new method has been proposed due to a change in the topology of a network. A dynamic routing method for a group of UAVs using SDN technology with dynamic topology was proposed. Simulation results show the comparison of the proposed method's performance with other algorithms. e simulation results show that the proposed method has better results than traditional algorithms. All these aforementioned studies use different algorithms and methods to solve the issue. But still there are issues in the topology structure, swarm initialization, and parameter selection in every algorithm. Our proposed CPSO shows effective results in solving all these issues. e only disadvantage of using PSO separately is that it falls into local optimum. So, the proposed algorithm is the best choice to solve the problem with minimum time.
Our contributions target the design of a CPSO algorithm for the better performance of UAV swarms. In this study, a swarm communication model along with a network mobility model is presented. Similarly, the fixed-wing UAV model is also presented with a communication graph concept of UAV. is article also offers the problem of UAV's swarms to maximize the throughput of the network. e network modelled is composed of multiple UAVs to form dynamic routing topologies. e advantage of the algorithm is that it improves overall fitness. Lastly, the MATLAB simulations verify the reliability, performance, and fitness of the CPSO.
is paper is organized as follows. e introduction is presented in Section 1. e state of the art is given in Section 2.
e problem statement and its proposed solution are given in Section 3. Section 4 gives the preliminaries of UAV communication and its routing mobility model. In Section 5, the proposed solution is presented. e simulations are given in Section 6. Section 7 presents the conclusion of this paper.

State of the Art
A dynamic routing protocol is projected in [22] for safe and consistent transmission. e algorithm proposed in the study works on two phases that are security and routing. It finds the path which is cost-effective among the source and destination node and focuses on the security of transmitted data. In [23], a geographic position-based routing protocol for UAV networks is proposed. is is used for the dejected protocols and position-based protocols. As compared with the hop by hop routing protocols, this method fully utilizes the transmission opportunities of all the links. Similarly, in [24], UAV swarm network is proposed and the optimal number of vehicles is analysed. Moreover, a low-latency routing algorithm is designed based on the connectivity of the network and partial location information. e results verified the proposed algorithm which decreases the link average and improves the packet delivery ratio.
Likewise, in [25], a very cheap adaptive routing protocol is proposed which exploits learning algorithm and topology features for UAV swarm. is algorithm represents the most suitable parts of swarm formation which are geometric addressing module, leaf-like routing pipe, and low-complexity learning model. In [23], OLSR algorithm is proposed in terms of delay, throughput, and the load of the network.
is routing protocol depends upon the link-state algorithm. Furthermore, the performance of the protocol is examined and analysed under different scenarios. e simulation results show that the link-state routing protocol has improved key indicators. Lastly, in [26], a dynamic UAV positioning method is proposed to maximize the value of sensors attained from multiple UAVs. To support real-time sensor data monitoring, the data should be delivered to the ground base station using an ad hoc network. In this article, particle swarm optimization is used to derive the optimum UAV location.

Problem Definition.
is section defines the problem statement and the proposed solution. Considering a given scenario having three clusters may vary according to the requirement of the task. Each of them contains an equal number of UAVs having a cluster head (CH). Accordingly, every vehicle is required to be located at different positions or stations by communicating with each other through coordination among them. Swarms are assigned with an independent motion to form a dynamic topology. When routing and coordination of swarms takes place, the intersection points remain unchanged. e development of the flight or plan takes a little bit of time even if the inputs change concerning time.
e flow comprises three steps which are as follows. (1) Start-CH, (2) CH-Another CH, (3) Another CH-End. Figure 1 shows these parts in black, red, and green colours, respectively. UAV attains flight time which is equal to the sum of the time consumed in transition between the base of fleet and interest points and elapsed time during the execution of the task at interest points. During the flight, UAV needs to share the information which is a problem to the throughput in the network.

Proposed Solution.
e aforementioned problems are solved by using our proposed algorithm based on CPSO in this study. It creates an efficient strategy by fixing the parameters and different variables to reflect the current situation. It aims towards the better performance of UAV's swarms and to solve the combinatorial problem. is solution is feasible for choosing the proper cluster head and for the problem of the group of routes that may arise when the number of vehicles increases. Each route starts and ends at a station that is chosen among all available stations without overhead of the allowed time for the flight.

Preliminaries of UAV Communication and Its
Routing Mobility Model } is a directed graph when drones are flying in a formation [27], where V(t) � V 1 , . . . , V n is called the vertex set, E(t) is called the edge set, and W(t) is called the weight adjacency matrix. e directed graph edge is denoted by e ab � (V a , V b ), where V a is the tail of the edge and V b is the head of the edge. e weight adjacency matrix W(t) � [w ab ], whereas the matrix elements show the adjacency weight. e communication node a gets information from b node when w ab > 0. e diagonal matrix is represented by D � diag d a a � 1, . . . , n has elements of matrix is an undirected graph when w ab � w ba . e undirected graph is defined as if two nodes in the graph are connected by the edges G(t). e matrix L is diagonalized to its min value of matrix Ζ which satisfies (1) e eigenvalues of matrix L are ζ 1 ∼ ζ max . e communication topology of UAVs consists of drones in an undirected connected graph. e second-order equation below defines the dynamic system model of UAVs.
where x a (t) is defined as the position status, v a (t) is defined as the speed status, and c a (t) is the control input of member a. e following control protocol is espoused to make sure that the formation members are consistent with the expected movement status.
where the control gains of the system are n 1 , n 2 , and n 3 . e time required for information to transfer from member b to a is denoted by T. r ab is the relative position, and v s is the expected speed. e ability index is denoted by h a which is used to attain the expected speed information of the formation. e aforementioned control protocol matrix is defined as where C(t) � [c 1 (t), . . . , c n (t)] T is the vector formed by the formation members control input, x is comprised of position status of formation members e control protocol of the above system is

Swarm Communication and Mobility
Model. Suppose a network of UAVs forms a swarm. A swarm includes a set of aerial vehicles that can fly together and perform complicated tasks with little changes in topology.
us, routing and coordination of swarm takes place. Similarly, swarms form a dynamic topology due to the assignment of independent motion trajectories. So, for internal swarm communication, routing is used. e network structure has a human in loop and self-organization control with some other features [28,29]. Furthermore, assume a control station (CS) that is answerable for providing communication to clusters. All vehicles are equipped with an antenna. A hierarchal structure network is designed. e proposed technique is implemented due to its high performance, high signal processing complexity, and so on. In the swarm communication framework, one tier is for inner and the other is for outer communication [30][31][32]. According to the requirement of the mission, the vehicles are divided into subgroups for swarms. e subgroup adopts aerial vehicle (AV) which is used for sharing intelligence. e structure of the cluster adopts a cluster head member (CHM). e information flow is defined in the problem statement with the help of a diagram. e communication process of swarms especially for inner and outer communication among the subgroups of swarms takes a little bit of time. Each vehicle operates in two types of mode: out of the band and in-band full-duplex. CS wirelessly communicates with AV and CHM in the same period. e swarm's communication is controlled by the clustering of CS assuming time-division mode activated by CS. Figure 2 shows the system model of swarms in which blue and green arrows show the frequencies. Similarly, orange lines show some interloping among the aerial vehicles.

Fixed-Wing UAV Model and Its Coordination.
e point mass system is used to model fixed-wing UAVs [33]. Consider a swarm of UAVs given below: where the forward displacement and horizontal displacement are denoted by x n and y n , respectively, _ o n is the height, V n is the ground velocity, Γ n is the flying path angle, Υ n is the heading angle, and ϕ n is the banking angle. rust is denoted by τ n , d k is the drag, lift is denoted by l n , and the gravitational acceleration and mass are denoted by g and M n , respectively. C and S denote sin and cos. e nonlinear model [34] by using feedback linearization can be linearized as where the simulated control inputs are c x n , c y n , and c o n . e actual control inputs are thrust, lift, and banking angle which can be written as τ n � M n c o n + g SΓ n + c x n CΥ n + c y n SΥ n CΓ n + d n , l n � M n c o n + g CΓ n − c x n CΥ n + c y n SΥ n SΥ n Cϕ n , ϕ n � tan − 1 c y n CΥ n − c x n SΥ n c o n + g CΓ n − c x n CΥ n + c y n SΥ n SΥ n ⎛ ⎝ ⎞ ⎠ .

(9)
Furthermore, V n , Υ n , and o n must meet the parameters which are as follows: where the minimum velocity and maximum velocity are denoted by V min and V max . Maximum lateral overload is represented by μ max . e minimum climbing speed and maximum climbing speed are denoted by ζ min and ζ max . Two coordinate coefficients are introduced that must be added in every UAV as a function for its route to meet spatial and temporal constraints for the group. e function is expressed as where the coefficients of temporal and spatial coordinates are denoted by F T n and F S n , respectively. t 1 and t 2 represent the expected arrival time of UAVs and J is a constant. t may vary depending upon the number of vehicles.

Cauchy Particle Swarm Optimization (CPSO).
is computation-based optimization is based on the study of natural behaviours. is technique provides random solutions and updates the velocity and position to attain the universal best solution. Each unit updates the searching position with the help of its best outcome. e updated position of the unit depends upon the present velocity and past position. Now, the new position becomes the best individual position and it is superior to all other positions. e equation below becomes Cauchy distribution if w meets the condition in the equation whereas w ε(−∞, +∞), the highest value of the function is denoted by w 0 , the width related to half of w 0 is δ. When w 0 and δ are 0 and 1, respectively, w meets the probability density function (PDF) and the above equation becomes Map operators and compass operators are used to find the global best outcome during search problems in the PSO algorithm. erefore, a particle's position and velocity are to be determined by the outcome. Cauchy operator K 1 increases the searching area and also avoids falling into local optimum. Operator K 1 can be written as where rand is 0 or 1. e rule that updates each particle ℘ n in every repetition is given as where the calculated position of n particle is represented by ℘ (n0) . e best global position is ℘ G e , and ℘ n ′ is the position of n after an update. e position of n th particle during the next repetition is given as

Mathematical Problems in Engineering
where ℘ n ′ will be far away from ℘ G e when K 1 is positive and vice versa. e most important advantage of using Cauchy operators is that particles scatter away from the center to find a better position. By equating updated position with earlier one, better fitness remains.
During the landmark operator phase, converging to the center of the swarm decreases swarm population. is problem is prevented by replacing the landmark operator with Cauchy which is used to update the particle based on its best position. e function of Cauchy is written as where w ∈ (0, + ∞). Cauchy operator K 2 is written as Each particle gets updated after each repetition according to the following rule: where the position of n th particle is ℘ σ max n0 . e best global position is given as ℘ G e . Due to K 2 , each particle moves towards the best global outcome. is operator guarantees the steady convergence of the algorithm.
Consider planning in three-dimensional space where the waypoint of each particle is D, position is ℘ n , and velocity vector is V n for n th particle as follows: where the q th waypoint position and velocity of n th particle in space are represented as q ∈ 1, . . . , D; (℘ x , ℘ y , ℘ z ) (n,q) , (V x , V y , V z ) (n,q) . e swarm is given as if total particles in it are P.
In this algorithm having particles P, there is one global position and another individual best position e which are given as   6 Mathematical Problems in Engineering where n ∈ 1, . . . , P is the number of particles. Furthermore, for multi-swarm function: ℘ (n,e) (t + 1) � ℘ (n,e) (t); iff ℘ (n,e) (t) ≤ F ℘ n (t + 1) , In a swarm, individual particle updates according to the following equation: where the accelerating coefficients are A 1 and A 2 and B 1 and B 2 are two random variables which are 0 and 1. By modifying the inertia weight σ, it can be improved, and it stabilizes the global and local outcome through search procedure. Global search can be enhanced by increasing σ. To enhance local search, decrease σ, which is given as where the current repetition is t, maximum repetition is W, and σ min and σ max signify the minimum and maximum values of σ.

Performance Evaluation
is section of the article weighs the enactment of the proposed CPSO with the help of simulations in MATLAB. e proposed algorithm is compared with PSO and evaluated. Table 1 presents the comparison of both algorithms in terms of the mean error for localization, cost, and efficiency.
ere is a total of 100 nodes having max and min errors. It clearly shows that the CPSO has attained an efficient output. e conclusion is that when nodes increase, the performance of CPSO also increases. e cost value for each particle is also presented which reveals the approach which still performs the best.

Mean Error for Localization (MEFL).
e simulation output figure below shows the mean error for localization with several nodes. Figure 3 clearly shows that in PSO, the mean error for localization decreases as the number of nodes increases. But in CPSO, the situation is the opposite. e proposed algorithm shows very low localization errors. e advantage of having a large number of nodes is that it localizes the node with very low error.

Accuracy of CPSO.
e accuracy in the percentage of the proposed algorithm CPSO is shown in Figure 4 along with the number of UAV nodes. It is clearly shown from the simulation output that when the number of nodes increases, the CPSO attains a higher accuracy. Both the algorithms attain a higher accuracy due to a suitable number of adjacent nodes. Our proposed algorithm has an accuracy of about 98% when several nodes are at their maximum value. Figure 5 provides a comparison of IDR with the number of UAVs and shows the effectiveness of CPSO in delivering data of approximately more than 85%. e proposed algorithm shows the best performance as compared to PSO. At some points, the ratio of both algorithms is the same, but when the number of vehicles increases, the ratio of the proposed algorithm also increases, so they have a direct relationship with each other as shown. Figure 6 shows the evaluation of average end-to-end delay with the number of UAVs of both algorithms. e simulation output clearly shows that CPSO has significantly less average end-to-end delay as compared to PSO. In high mobility networks, the delay increases because time is mandatory to find the node position. Whenever the nodes move rapidly, there is more route breakage.

Cluster Making Period (CMP)
. CMP is defined as the time necessary to form a cluster by the algorithm. It takes time to make a cluster generally owing to the complexity of the algorithm. Whenever a CMP is greater, the energy consumption is very high as UAVs consume high energy. As shown in Figure 7, CPSO significantly takes less time to form a cluster as compared to PSO. It is clearly shown that when no nodes increase, the proposed algorithm works proficiently and takes less time which increases its efficiency. CPSO performs clustering much better than PSO. e search space of the proposed algorithm is reduced which affects the selection of node and CMP. As compared to many algorithms, CPSO takes a short period to obtain the solution needed. In conclusion, the energy and delay decrease with the decrease in CMP.

Conclusion
Designing an algorithm is complicated due to the dynamic topology changes and high mobility of the UAV network. is study presents the CPSO algorithm for UAV network communication which includes the coordination and routing among them. is algorithm is applied to solve node localization of UAVs in space and to solve combinational issues. It also reduces the combinational time and improves the performance of the swarms. e proposed algorithm has a convergence ability, reduces the localization error and endto-end delay, and improves the accuracy and information delivery ratio. e simulation results show the performance of the proposed algorithm along with PSO. As a result, it is seen that CPSO is suitable for better performance in communication. CPSO has a greater ability to solve all the issues. Lastly, the overall fitness of the algorithm is increased. In the future, different algorithms based on AI will be studied to achieve high-speed localization and will be applied to solve different issues.

Data Availability
e data used to support the findings of this study are included within the article.