Elastic Routing Mechanism for Flying Ad Hoc Network

Flying Ad-Hoc Network (FANET) is a hot topic in current research. The design of routing mechanism is challenging because when the scale of Unmanned Aerial Vehicle (UAV) nodes is large, vast amount of routing overhead may lead to network collapse. An elastic routing mechanism is proposed for large-scale small UAVs multitasking scenarios. Firstly, the New-Unifying Connected Dominating Set (N-UCDS) algorithm is proposed to construct a virtual backbone network based on the connected dominating set. The number of neighboring nodes, remaining energy and link duration are considered to influence the UAV network performance when electing backbone nodes. Secondly, by deploying and running the New Better Approach to Mobile Ad-Hoc Network-Advanced (NBATMAN-ADV) routing protocol on the backbone nodes, the link quality can be evaluated by using the received signal strength index and signal-to-noise ratio of the physical layer data. In this way, the change of the link can be quickly sensed while reducing the routing overhead. The simulation results show that the routing protocol proposed in this paper has significantly improved average packet delivery rate, end-to-end delay and received throughput compared with other traditional proactive routing protocols.

to achieve intercommunication of the entire network nodes. 97 The use of lightweight active routing strategy can reduce 98 the delay of route discovery and achieve fast routing. There-99 fore, this paper is divided into two parts: the construction The concept of Unifying Connected Dominating Set (UCDS) 105 was proposed in [11]. The UCDS algorithm is a distributed 106 algorithm which each node only needs to obtain two-hop 107 neighbor topology information to execute the algorithm cor-108 rectly. Each node in UCDS supports the maintenance of 109 routing information by playing an important role in relay 110 forwarding and routing distribution. Maintenance of routing 111 information can be done in UCDS. The virtual backbone 112 nodes in UCDS algorithm are members of Dominating Set 113 (DS) and Connected Set (CS). DS members are responsible 114 for route distribution and relay forwarding of common nodes, 115 and CS members are responsible for connecting the members 116 of the dominating set. The election of DS members is mainly 117 based on the domination factor, but the domination factor is 118 only determined by the number of node neighbors without 119 considering other factors of UAVs. The single metric makes 120 the backbone nodes change too frequently, leading to insta-121 bility of network topology. The CS membership election is 122 performed by the node itself with CS rule and CS exception 123 rule. Then the result of the judgment is broadcasted to the 124 neighbor DS members and finally the DS members elect 125 the CS members. This process makes the construction of 126 the virtual backbone network inefficient and the topology 127 convergence speed slow. 128 Therefore, this paper proposes an N-UCDS algorithm for 129 electing backbone nodes to construct a virtual backbone net-130 work. In the large-scale small UAVs multitasking scenar-131 ios, the energy of small UAVs is limited, so the impact of 132 energy consumption and link quality on the survival time 133 and stability of the UAV network is fully considered in the 134 DS member election process of the N-UCDS algorithm. The 135 N-UCDS algorithm proposes that when a node is elected as a 136 DS member, it automatically executes a new CS rule based on 137 the two-hop neighbor node information to make a judgment 138 and directly designate some Connected Set Candidate (CS') 139 nodes as CS members. This saves an update cycle time and 140 speeds up the topology convergence without affecting the 141 correctness of the algorithm, making it more applicable to 142 UAV networks. The BATMAN-ADV [12], [13] routing protocol is a proactive 145 routing protocol that works at the second layer of the OSI 146 model, the data link layer. Each node only needs to know 147 the best next-hop neighbor node to the destination node and 148 does not have to worry about global topology changes. This 149 makes the overall architecture of the protocol small and can 150 quickly adapt to changes in the network topology, which 151 is very suitable for UAV networks. Each node periodically 152 broadcasts OGM packets for informing other nodes of its own 153 existence. After receiving OGM packets from other nodes, 154 the node forwards them according to the policy, enabling the 155 OGM packets to spread to the whole network. 156 VOLUME 10, 2022 Therefore, this paper proposes a NBATMAN-ADV rout-157 ing protocol that needs to be deployed and run only on the     Chowdhury [15] studied the problem of efficient data 211 dissemination between mobile nodes in wireless networks, 212 MCDS is commonly used to reduce redundant transmissions 213 in broadcasts. The problem of constructing MCDS is dis-214 cussed, using MCDS as a starting point for constructing con-215 tention aware connected dominating set (CACDS) algorithm 216 to optimize the network competition problem with large node 217 size. Qi et al. [16] proposed to construct CDS in FANET to 218 solve the joint optimization problem of node transmission 219 power and location. It proposed a topology control mecha-220 nism based on CDS and a time-discretized topology construc-221 tion maintenance algorithm, which outperforms the general 222 particle swarm optimization (PSO) algorithm in terms of net-223 work overhead and network stability in UAV cluster systems. 224 Wang et al. [17] proposed that a swarm of UAVs can build a 225 virtual backbone network (VBN) based on graph-theoretical 226 d-hop DS, where each UAV outside the VBN can send the 227 collected data within distance to the VBN. an adaptive ADMS 228 algorithm was proposed to maintain a stable VBN, which can 229 achieve a better trade-off between routing overhead, response 230 time, and maintenance cost. 231 Liang et al. [18] used CDSs as the virtual backbone of 232 WSNs, but due to the actual environmental factors, the trans-233 mission radius of nodes in the network is unstable, so the 234 robustness of VBs in WSNs is considered, and the corre-235 sponding algorithm is proposed to construct d-robust CDSs in 236 WSNs with unstable transmission range. But the paper does 237 not consider the construction of virtual backbone networks 238 for 3D scenarios, which can be extended to 3D spatial FANET 239 scenario. Mao et al. [19] proposed an efficient distributed 240 routing algorithm based on connected dominating sets to 241 build a virtual backbone network can effectively mitigate the 242 broadcast storm problem in mobile self-assembly networks, 243 which is more applicable to dynamic self-assembly networks. 244 By broadcasting only through selected network nodes, the 245 same effect as flooding can be achieved and the broad-246 cast storm problem is avoided. Considering the transmission 247 range, residual energy and mobility of nodes, this algorithm 248 can significantly reduce the network construction overhead, 249 ensure network connectivity, improve energy efficiency, and 250 prolong network survival time.     Fig.1 shows the large-scale small UAVs multitasking sce-324 narios. The ground is divided into multiple areas. The UAV 325 cluster system operates in the air and needs to perform various 326 tasks. The trajectory of a single UAV can be represented 327 by the set {(X 0 , X 1 , S 1 ) , · · · , (X n−1 , X n , S n )}, where S n ∼ 328 N (µ s , σ 2 s ) denotes the hovering time of the UAV performing 329 the mission at the path point X n . Assuming that the displace-330 ment length l n of the projection of the path points X n−1 and 331 X n on the x − y plane conforms to the Rayleigh distribution 332 [36], when σ = 1 2π λ , the distribution function and the 333 probability density function can be expressed as follows: where µ(l n ) = σ π 2 = 1 4λ and D(l n ) = 4−π 2 σ 2 = 4−π 4π λ .
The angles of the projected displacements of path points 338 X n−1 and X n in the x − y plane, f (θ) ∼ U (0, 2π ) conform 339 to the uniform distribution [31], and the altitude of each path 340 The lengths of the displacements projected by the UAV on 343 the x − y plane and the flight altitude are independent of each 344 other, so the joint probability density of both can be expressed 345 as [37]: To simplify the analysis, it is assumed that all UAVs have 349 the same communication capability and all use omnidirec-350 tional antenna. d max indicates the maximum communication 351 VOLUME 10, 2022 The N-UCDS algorithm is improved from the UCDS algo-  Table 1.
where for node i, the dominance factor d ij of its neighbor node  The steady state of the network within the two-hop neigh-390 borhood of the node i can be expressed as: where N i CDS denotes the number of backbone nodes in the 393 two-hop neighborhood of node i and V i CDS_change denotes the 394 rate of node attribute change in the two-hop neighborhood 395 of node i. The node attribute change refers to the transition 396 between backbone node state and non-backbone node state. 397 The coefficients θ A , θ B , θ C are calculated using the mul-398 tivariate gradient descent method in machine learning. The 399 calculation is shown in Algorithm 1 and where α is the 400 learning rate.

401
Algorithm 1 θ A , θ B , θ C Selection Algorithm Hypothesis: Output: As shown in Fig.2, the network reaches the most stable 402 state when θ A = 0.3, θ B = 0.5, θ C = 0.2 and W i takes the 403 maximum value.

405
Small UAVs energy consumption includes propulsion and 406 hovering energy consumption and UAVs communica-407 tion energy consumption. This paper assumes that 75% of 408 the energy is used for propulsion and hovering and 25% of 409 the energy is used for communication. All UAVs have 100% 410 energy at the initial moment. Whether the UAV node is a 411

Algorithm 2 Communication Link Duration Algorithm
When the virtual backbone network is introduced, the routing 437 mechanism can obtain many benefits. Deploying and running 438 a lightweight active routing protocol on backbone nodes can 439 achieve a better balance between routing overhead and route 440 discovery delay. In the BATMAN-ADV routing protocol, the 441 routing metric criterion mainly depends on the TQ value. 442 The TQ value is divided into two parts: PTQ (Path Transmit 443 Quality) represents the global link quality value after multiple 444 hops through a path; LTQ (Link Transmit Quality) represents 445 the local link quality value with neighboring nodes. The 446 protocol uses the sliding window mechanism for OGM packet 447 statistics and the average TQ value as the routing criterion. 448 When the link quality changes significantly, the change rate 449 of the average TQ value is slow and does not reflect the rapid 450 change of the link well, resulting in poor convergence of the 451 protocol. After the network topology changes, the protocol 452 has poor ability to sense network fluctuations.

453
NBATMAN-ADV routing protocol evaluates link quality 454 using received signal strength metrics and signal-to-noise 455 ratio of physical layer data. Each network node is very easy 456 to obtain physical layer data. This paper believes that the link 457 quality with high signal-to-noise ratio and strong received 458 signal strength should be better. metric and metric min are 459 two routing metrics in the NBATMAN-ADV routing proto-460 col. metric is the link quality value after multiple hops, and 461 metric min is the link quality value of the worst path quality 462 among multiple hops. To reduce the impact of temporary 463 fluctuations in the network, the link metric is smoothed with 464 the data from the previous time. The initial values of both 465 metric and metric min are 1. The values of metric and metric min 466 are updated when the node receives OGM packets. In the 467 NBATMAN-ADV routing protocol, the factors affecting the 468 optimal next-hop node selection are changed from TQ to 469 metric and metric min , and the neighbor node with the max-470 imum metric is selected as the next-hop forwarding node 471 subject to metric min > τ min . τ min is the minimum threshold 472 of metric min . The value of τ min can be adjusted according to 473 the network type and network environment.

474
The received signal strength of the node can be expressed 475 as: The node received signal strength indicator can be 484 expressed as: VOLUME 10, 2022 The signal-to-noise ratio of the signal transmitted by the 487 UAV node from i to j can be expressed as: where P denotes the maximum transmit power of the UAV remaining UAV nodes, which can be expressed as: where RSSI t is the received signal strength indicator at the

530
In the BATMAN-ADV routing protocol, each node needs 531 to broadcast OGM packets periodically. While all nodes need 532 to forward OGM packets sent by other nodes. Assuming that 533 the broadcast period is T interval and the OGM packet size 534 is Size OGM . The number of OGM packets generated by the 535 network in the time (0, t) is N 2 t T interval and the total data volume 536 is N 2 Size OGM t T interval .

537
The routing overhead in the NBATMAN-ADV routing pro-538 tocol is divided into: x the implementation of the N-UCDS 539 algorithm requires UAV nodes to broadcast HELLO packets 540 periodically, and HELLO packets need to contain information 541 about themselves and neighboring nodes; y DS member 542 nodes in the virtual backbone nodes broadcast OGM packets 543 periodically, and CDS member nodes forward OGM packets, 544 and OGM packets need to contain information about them-545 selves and neighboring nodes information. Then the number 546 of routing packets generated by the network in time (0, t) is 547 t T interval (N + N DS × N CDS ) and the total data volume can be 548 expressed as: where N πd 3 max 3L 2 (H high −H low ) denotes the average number of 552 neighbors of the UAV node, Size HELLO denotes the size 553 of the node's own information in the HELLO packet, 554 Size HELLO_neighbor denotes the size of one neighbor node's 555 information in the HELLO packet, N DS denotes the number 556 of DS members in the network, N CDS denotes the num-557 ber of CDS members in the network, and Size OGM _neighbor 558 denotes the size of one neighbor node's information in the 559 OGM packet. As the maximum communication distance d max 560 varies, the ratio between the number of routing packets and 561 the total data volume of routing packets for NBATMAN-562 ADV routing protocol and BATMAN-ADV routing proto-563 col is shown in Fig.3. When the maximum communication 564 distance is 1600 m, the average number of backbone nodes 565 for 100 UAV nodes to build a virtual backbone network 566 through the N-UCDS algorithm is 28.3157 within 1000s of 567 simulation time. Compared with the BATMAN-ADV routing 568 protocol, the NBATMAN-ADV routing protocol generates 569 4.85% of the routing packets and 65.43% of the total data 570 volume, so there is a significant reduction in routing over-571 head.

573
The simulation software in this paper uses MATLAB and 574 QualNet. MATLAB is used for algorithm and numerical 575 simulation with better performance, but it cannot simulate 576 the real network environment, so QualNet is used for net-577 work system simulation. In QualNet each network node is 578  significantly more CS members than the UCDS algorithm. 597 Fewer DS members and more CS members can make the 598 network more robust and can effectively cope with the effects 599 of network fluctuations.

600
When maximum communication distance is 1500 m, 601 Fig.5 represents the average value of the remaining energy 602 for 100 UAVs over time. The topology of the ad-hoc net-603 work is constantly changing, so the backbone nodes are also 604 changing. Therefore, the energy consumed by each UAV is 605 almost balanced. The simulation results show that the UAV 606 cluster system using the N-UCDS algorithm has a reduced 607 energy consumption rate, increased network survival time 608 and improved the duration of the UAV cluster system per-609 forming the mission.

610
100 UAV nodes are deployed in QualNet, the motion model 611 of UAV has been described in the third part of the article, the 612 VOLUME 10, 2022 activity space is 5 km × 5 km. 10 CBR services are randomly launched between nodes with a rate of 2 Mbps, interval of 614 0.2 s, packet size of 512 Byte, and simulation time is 1000 s.

615
The simulation scenario is shown in Fig.6. In this paper, three   Due to the high dynamic nature of UAV nodes, it leads 656 to frequent link failures, resulting in lower PDR. In the 657 NBATMAN-ADV routing protocol, the nodes use the number 658 of neighboring nodes, remaining energy, and link duration 659 as a combined factor to calculate the dominance factor to 660 select the backbone nodes. This reduces the rate of change 661 of the backbone nodes, increases the effectiveness of the 662 forwarding nodes and increases the connectivity of the net-663 work. Therefore, PDR is improved. In the NBATMAN-ADV 664   of the NBATMAN-ADV routing protocol has a significant 685 advantage over the other routing protocol. With the acceler-686 ation of the flying speed of the UAV, the network topology 687 changes more drastically while the structure of the backbone 688 network is constantly changing. However, in the NBATMAN-689 ADV routing protocol, the rate of change of backbone nodes 690 has decreased since the comprehensive factors are considered 691 in the selection of the backbone nodes.

692
Meanwhile, the UAV nodes do not need to worry about 693 the global topology change. The packets from the source 694 nodes only need to be delivered to the backbone nodes to 695 reach the destination nodes. This makes the architecture of the 696 protocol small and can quickly adapt to the network topology 697 change. Therefore, with the increase flight speed of UAV, 698 the performance of NBATMAN-ADV routing protocol has 699 significant advantages over other routing protocol.