Particle Swarm Optimization Video Streaming Service in Vehicular Ad-Hoc Networks

Owing to the increasing capabilities of mobile devices and the development of mobile communication techniques in vehicular ad hoc networks (VANETs), mobile multimedia services have focused on supporting high Quality of Service (QoS) and Quality of Experience (QoE) for the subscribers in video streaming services. In VANETs, high-quality video streaming services aim to provide subscribers with safety and various infotainment applications. Due to the dynamic topology and frequent connectivity changes in moving vehicles, video streaming services require elastic and continuous vehicle information updates to present interactive real-time views of nontrivial road scenarios. The QoS and QoE are affected due to obstacles, re-location tracking, network traffic, and bandwidth factors that occur due to mobility problems and significantly influence the ability of vehicles with mobility to provide video streaming services. To achieve high QoS and QoE of video streaming services in VANETs, this paper proposes a novel protocol named PSOstreaming based on a particle swarm optimization (PSO) which is one of the mainstream and nature-inspired algorithms for swarm intelligence (SI). PSOstreaming calculates the PSQ score, which is the scoring method of the node in the topology in terms of the data communication capability measurement to analyze and optimize topology information in real-road circumstances. In addition, PSOstreaming utilizes a 3D vector mobility prediction algorithm for the mobility prediction method for the vehicles to address the characteristics of VANETs. In the topology, PSOstreaming defines Global PSO members and Local PSO members to disseminate video packets for video-streaming services to the service requester by computational equations. Experimental results indicate that PSOstreaming achieves high-quality video streaming services with a flexible response to dynamic topology changes and a high frame delivery ratio in terms of QoS and QoE.


I. INTRODUCTION
As the era of advanced development of vehicles and vari- 24 ous infotainment content enjoyed in vehicles is approaching, 25 (VANETs) are a very realistic and practical way to meet these 26 needs [1]. In VANETs, vehicles assume the role of network 27 nodes with special characteristics, such as high mobility, self- Third, data caching should be efficiently conducted to store 80 the video data content on the calculated route until commu-81 nication is complete. Previous studies [18], [19], [20] have 82 a caching mechanism that does not consider video packet 83 distribution for real-time video streaming services. 84 Mobile Ad-hoc Networks (MANET) technologies for mul-85 timedia data forwarding have recently been proposed in 86 diverse research areas [21], [22]. MANET has a critical 87 energy consumption issue for the multimedia streaming ser-   In this paper, we propose a novel protocol named 98 PSOstreaming to provide video streaming services with sup- 99 porting QoS and QoE in VANETs by efficiently addressing 100 the three challenging issues. We first adopt Particle swarm 101 optimization (PSO) algorithm to solve the load-balancing 102 problem in dynamic topology changes. We then formulate 103 short-term mobility predictions to react to the mobility of 104 the vehicle immediately. The dependent long-term mobil-105 ity prediction algorithm based on the vehicle's trajectory to 106 the destination may cause resource wastage and take time 107 to recover the forwarding route when the user vehicle is 108 off the calculated path. The proposed protocol uses particle 109 swarm optimization information from the topology and the 110 user vehicle's mobility to establish the forwarding path via a 111 3D mobility short-term prediction algorithm, rather than the 112 long-term predictions researched previously [14], [23], [24]. 113 The long-term prediction algorithms have to use the trajectory 114 of the user vehicle to predict the user vehicle's future location 115 to forward the streaming data. However, using the vehicle's 116 trajectory information to the destination to predict the vehi-117 cle's future location cannot immediately handle unexpected 118 circumstances. When the previous studies faced unforeseen 119 circumstances due to the free will of the drivers, the mobility 120 prediction algorithms took more time and more resources to 121 rebuild the data forwarding path to the requested vehicles, 122 which often happens in real road scenarios. Additionally, 123 it constrains the topology resource waste in the streaming 124 services. Otherwise, the short-term mobility prediction in the 125 PSO streaming can avoid the mentioned problems. However, 126 the PSOstreaming immediately reacts to topology changes 127 because of the service user number, which makes the resource 128 balance of APs in the topology. Owing to the dynamic routing 129 decisions for user vehicles and the closed cooperation of 130 different methods of PSO and 3D mobility prediction, the 131 proposed protocol is capable of adaptively coping with rapid 132 topology changes in real-time video streaming scenarios. The 133 main contributions of this paper are as follows: 134 • We propose a PSO topology information searching 135 mechanism for APs and calculate the score of APs' 136 resource availability. Compared to previous protocols 137 that search for static availability data by end-to-end 138 routing, the proposed algorithm has flexibility and great 139 adaptability in real-time video streaming. Furthermore, 140 the proposed protocol leads to high load balancing in 141 the topology because all APs organically share their 142 resource information.

143
• We derive mathematical models for the resource 144 arrangement of APs in the topology before the video 145 streaming service is initiated.  [18] proposed an 235 algorithm to support adaptive bit rate (ABR) based on video 236 characteristics and efficiently using caching in radio access 237 networks. Yashuang et al. [19] proposed a dual time-scale 238 dynamic cache scheme in base stations to support ABR 239 streaming under the condition of high channel variations to 240 achieve high QoS and QoE for video streaming services 241 in VANETs. Zhao et al. [20] proposed a scheme to cache 242 OTT multimedia streaming content in future connected RSUs 243 using vehicle mobility prediction. However, existing proto-244 cols that support the QoS and QoE of multimedia streaming 245 services only consider static destination vehicles. Moreover, 246 they exploit mobility prediction and caching with the mobility 247 information of normal vehicles on roads instead of the trajec-248 tory information of the destination vehicles.

249
The video streaming service application aims to provide 250 a high-quality infotainment service to the user vehicle. Info-251 tainment enables the smart cars to provide informational and 252 entertainment services, enhancing the experience of drivers 253 and passengers. Video streaming applications in infotain-254 ment services are the primary technology to achieve the 255 goal. In recent years, the vehicle infotainment system has 256 captivated significant attention from automobile industries 257 discussed in [30]. For the high-quality media content pro-258 vided in infotainment services, the services have a topology 259 that minimizes the video file data loss with the minimum 260 delay to the destination vehicle. Indipendented infotainment 261 service without considering QoS and QoE in the multimedia 262 data forwarding process [30] cause the challenging issues 263 mentioned before in multimedia service in VANETs.

264
As mentioned previously, the study timeline has pointed 265 out QoS and QoE improvement solutions in the routing pro-266 cess. However, these protocols cause a high error rate and 267 quality decline, owing to the unintended mobility behavior 268 of the requested vehicles. These challenges arise from two 269 main factors. One is the low refresh rate of the topology infor-   Particle Swarm Optimization (PSO) [37] is a metaheuristic 327 SI technique that uses a stochastic population and achieves 328 optimization. PSO adopts real-life social behavior of diverse 329 animals, such as fish schooling and birds flocking while 330 performing the movement for food search. The PSO concept 331 analyzes the physical movements of an individual in a swarm, 332 where each particle is guided by its own best position and 333 that of the entire swarm. Therefore, the individual particles 334 in PSO work properly in group movement situations such as 335 VANETs. PSO also has high adaptability in various network 336 environments in VANETs. In VANETs with video streaming, 337 multiple characteristics users are excited in the topology. 338 The users have different network features in their network 339 devices. Therefore, the devices participating in the network 340 service have to use optimized different algorithms to for-341 ward the requested data to the destination. PSO is adaptive 342 algorithm that can optimize the network status of the nodes 343 adapted using a metaheuristic technique. PSO algorithm sat-344 isfies the dynamic topology challenge issues in VANETs due 345 to the reason mentioned [38]. 346 The proposed protocol uses the Particle Swarm Optimiza-347 tion (PSO), which is the most suitable algorithm compared 348 with others in the diverse circumstances in VAENTs. The 349 proposed protocol aims at two primary goals. The first is 350 topology load balancing using a Particle Swarm Optimization 351 algorithm, and the second is vehicle video streaming service 352 using a 3D vector mobility prediction algorithm and video 353 data caching mechanism. The mentioned techniques for the 354 video streaming service in VANETs lead low delay in data 355 forwarding time and minimize the data loss in terms of 356 the diverse conditions. The mentioned previous research in 357 VANET for video streaming services and multimedia proto-358 cols has struggled the challenge issues in VANETs discussed 359 before. In the next section, we describe the two main tech-360 niques of the proposed protocol to solve the challenge issues 361 in VANET streaming service.

363
In this section, we present the proposed protocol, PSOstream-364 ing for supporting video streaming services in VANETs. 365 First, we explain the network model and the overview of our 366 protocol. Subsequently, we describe in detail of the process 367 of PSOstreaming in time order.   to the nearest AP on the topology. Then, the request packet 405 is forwarded to the global PSO member already selected for 406 the video streaming service. The response packet is sent back 407 to the first requester vehicle using the same path the request 408 packet used. The first request packet has the information of 409 the vehicle and the request content data identification, which 410 uses for searching from the backbone server for the video 411 streaming service.

412
The proposed protocol can be divided into PSQ measure-413 ment, PSO group decision, 3D vector mobility prediction, and 414 link connection with the delivery process. In the PSQ mea-415 surement phase, the PSQ value is estimated, which unifies 416 the resource status and availability of the APs in all topology 417 areas. Furthermore, using PSO, the proposed protocol indi-418 cates how to maintain the real-time data of the APs. In the 419 PSO group decision phase, every AP gets its own invested 420 roles following the basic concept of PSO. The roles comprise 421 global and local PSOs. In the 3D vector mobility prediction 422 phase, the proposed protocol employs a short-term mobil-423 ity prediction model to manage the unsuspected mobility 424 changes of the vehicles. This mobility prediction compen-425 sates for the defect of a long-term mobility prediction model 426 based on the trajectory of the vehicles. In the link connection 427 with the delivery process, a one-hop link is connected along 428 with the selected end node and requested vehicle. It utilizes 429 the 3D vector mobility prediction algorithm for a seamless 430 In the equation 1 and 2, where y k denotes the output corre-466 sponding to the kth particle input, Y kd is the desired particle In standard PSO, diverse particles distributed in the field 484 are gathered in groups around the best particles as the iter-485 ation progresses. The particle swarm optimization algorithm 486 assumes that a group of animals searches for their goals, such 487 as food and destinations. However, not all animals in a group 488 ensure their distance and location from the goal because 489 they are unaware of the specific coordinates or geographical 490 locations of the goal. The fastest way to find a goal is to search 491 for the area around the animals closest to the goal. In PSO, 492 each particle has a direction for optimal particles and searches 493 for the communication area to determine the best particles. 494 In the learning phase, APs are the particles in the particle 495 swarm optimization algorithm.

496
Moreover, in the proposed protocol, the topology is divided 497 into several sectors to distinguish the area and choose a global 498 PSO member in the sector. APs compute their PSQ score 499 and share the score with their neighbors to determine the 500 best global PSO member in the sector. A group of n APs is 501 broadcast in a divided sector-search space. Each AP in the 502 search process considers its search history and the best score 503 within the group of other APs, and position changes are based 504 on this. The position of the Aps, score, and location of the 505 best AP changes according to the following equation in the 506 standard PSO algorithm: In the equation 3 and 4, where X i is the position vector of 511 the i-th AP and S i is the PSQ score of the AP. Consider P i as 512 the best position of the i-th AP during its search process in 513 the communication range, and P g as the position of the best 514 global PSO member during the current search in the sector. 515 ω determines the current communication speed of the AP to 516 its neighbors. c 1 and c 2 are learning factors that make the 517 AP have the process of self-summary and learning to the best 518 of the sector, and determine the location of the best position 519 of the global PSO member in the sector. rand() is a random 520 number distributed within [0,1]. The communication speed of 521 the AP is limited to the maximum range S max .

522
In the equations 5, 6, and 7, p ij is the probability that the 526 jth bit is sign I in N APs. E(N ) denotes the average entropy 527 VOLUME 10, 2022 N P t represents the mobility prediction results from the 567 equation 8. Where x t denotes the vehicle's latitude when the 568 time t, where y t denotes the vehicle's longitude when the time 569 t. v x t denotes the vector prediction factor that gives the loca-570 tion information for the moving vehicle. O P t represents the 571 position changes of the moving vehicle due to the vector 572 position changes in time t. Where L i denotes next 10 min-573 utes location information prediction results using previous 574 5 minutes vector direction information from the mobility pre-575 diction architecture. Using the PSO optimization algorithm, 576 the PSOstreaming minimize the change value θ for better 577 performance in delay and data loss.

578
Using the equations 9, 10, and 11, the proposed protocol 579 can predict the mobility of vehicles in the topology in a short 580 time. The predicted data of all vehicles are then forwarded to 581 the AP in the communication range; hence, the AP collects 582 and manages mobility data from all vehicles within its range. 583 All mobility information of the vehicles in the sector is super-584 intended in real time because the APs containing the vehicle's 585 mobility information forward the information to global PSO 586 members.

588
When a live streaming service is requested, the first process 589 is to choose a local PSO member, where the AP has the 590 highest PSQ score in the requested vehicle's communication 591 range. The selected AP then forwards the requested packet 592 to the global PSO member in its sector. The global PSO 593 member downloads the requested content and determines 594 the optimal path to the local PSO member connecting the 595 requested vehicle, using pre-learning phase information. The 596 optimal path to the first local PSO member follows the PSQ 597 score to reach its destination. From the pre-learning data by 598 particle swarm optimization, the global PSO member recog-599 nizes the PSQ scores of every AP in the topology within the 600 communication range. Following the direction of the route 601 that requested the used packet, the global PSO member sets 602 the discovery route table to the first local PSO member using 603 the pre-learning topology information. The requested content 604 data are forwarded based on the calculated path to the local 605 PSO member, and mobility support according to the user 606 vehicle's mobility is described in the following local PSO 607 group decision.  Figure 3 shows the operation of local PSO members in a 610 section. After the Global PSO members have selected in the 611 sectors, the Local PSO selection algorithm is processed to 612 help the streaming data forward to the requested vehicle. 613 The local PSO member candidates are the random AP in the 614 sector with a high PSQ score to the requested vehicle to serve 615 the high QoS and QoE in the streaming services. The first 616 local PSO member is selected, and the mobility information 617 proposed protocol initiates the handover process to the other 646 global PSO member which belongs to the following sector 647 where the vehicle is located. The handover process begins 648 when the local PSO member recognizes that the sector of 649 the next local PSO member does not match. The previous 650 local PSO member connects the link to the next local PSO 651 member in the next sector and completes the received content 652 data from the global PSO member. After the content data 653 are forwarded to the next sector, the new local PSO member 654 will be the first local PSO member in the next sector. The 655 new member sends a packet that includes the progress of the 656 content file transmission and data request of a new global 657 PSO member. The new global PSO member receives the 658 requested packet, then forwards the content data using the 659 PSQ score and follows the local PSO decision process. Optimization algorithm in VANET using adaptive QoS-based 671 routing. The comparison protocols for performance evalu-672 ation are for VANETs with single data transmission. The 673 recent works of the video streaming service using Swarm 674 Intelligence in VANETs have not been considered due to 675 the computing resource problem in the previous. However, 676 the network devices in the vehicle and the access points 677 get improved enough to compute the swarm intelligence 678 protocols for optimization recently. This paper proves how 679 properly the proposed protocol shows better results in video 680 streaming services using the Swarm intelligence called PSO 681 compared with other SI protocols with the previous video 682 streaming services. To better explain the results of the pro-683 posed and compare protocols, we assume that the SI pro-684 tocols have a pre-learning task to recognize the topology 685 information, such as the number of vehicles in the topology 686 the number of intersections in the map times. We set this 687 assumption because the SI protocols require time to predict 688 the random mobility of vehicles in the topology to develop 689 their algorithms. Otherwise, CLONE does not have time for 690 pre-learning because it has no SI algorithms.  Table 1.  by copying the payload from the last received frame before 737 it. The metric SSIM is utilized as an image/video metric to 738 measure the received frame quality based on its structural, 739 luminance, and contrast similarity. SSIM, measured using 740 MSU tools, improves the MOS and PSNR metrics by reveal-741 ing the perceived quality of the received video sequences. The 742 SSIM differs from the PSNR because it approximates struc-743 tural distortion, instead of pixel-by-pixel errors, to evaluate 744 useful information for the human eye. shows other weak points. AQRV uses ant colony optimization 755 (ACO), which only uses the first searched path to the user 756 vehicle for transmission until the path has a lower level than 757 the other path they find. Although the first path is much 758 longer than the different paths they find later, ant colony 759 optimization in root selection of the video streaming service 760 only chooses the first path for transmission and initiates the 761 data forwarding. If the first path has a low energy level than 762 the other path they find, they move the forwarding path to 763 another path. AQRV is based on the ACO method in SI. 764 Therefore, AQRV takes several hops in data forwarding for 765 video streaming because the large packet size of the data in 766 the transmission costs a lot. Hence, AQRV has to keep chang-767 ing the path to the user vehicle. It leads to multiple hops and 768 gives a disadvantage to PDR. BSOGR indicates the lowest 769 result in PDR because it does not consider the used path in the 770 data forwarding. BSOGR abandons the used path to the user 771 vehicle when the energy level gets lower, and never considers 772 the used path even though it recovers from the low energy 773 level. Hence, BSOGR cannot choose diverse options for the 774  because the short path to the user vehicle is abandoned. 814 AQRV and PSOR illustrate a similar graph because the two 815 mechanisms derive similar topology searching techniques. 816 However, CLONE indicates a better performance than other 817 SI protocols in the 300 m communication range, but the 818 results are reserved after the 400 m communication range. 819 The reasons for the results can be explained as two reasons: 820 If the density in the communication range gets low, CLONE 821 requires several hops to forward the streaming data to the user 822 vehicle, and it is difficult to determine the excellent condition 823 of nodes in low density and takes much time to search and 824 get a response about the near neighbor information. The 825 proposed protocol can handle the mentioned weak points 826 for the other protocols. PSQ searching technique helps to 827 determine high-quality APs for streaming forwarding to the 828 user vehicle. Even in a low density and large communication 829 range, the proposed protocol can realize the minimum time 830 for video streaming transmission owing to the pre-learning 831 of the topology information. Moreover, the other three SI 832 protocols do not have any short-term mobility support to 833 react to the mobility changes on the roads. Therefore, other 834 SI protocols have trouble with supporting the unexpected 835 mobility changes in the communication. Figure 6 illustrates the average delay depending on com-837 munication range change and density changes. The proposed 838 protocol indicates the optimal delay in the graph. Although 839 the three SI protocols had pre-learning time to figure out 840 the topology information, the three SI protocols do not have 841 a large data transmission way for video streaming services. 842 The mentioned SI protocols have trouble with supporting 843 video streaming services from moving vehicles. Only sin-844 gle packet techniques are included in the protocols. These 845 challenges lead to high packet failure, thereby increasing 846 processing recovery mode. CLONE has trouble handling the 847 short-term mobility changes owing to the random mobility 848 that is not based on the vehicle's trajectory. The vehicles 849 on real-road keep altering the direction and speed while in 850 motion. Even if they have trajectory information, the calcu-851 lated direction sometimes has to be changed by unpredictable 852 VOLUME 10, 2022  intersection the user vehicle is located. CLONE has link 892 connection, mobility support, and caching process to support 893 the video streaming service. Therefore, the results indicate 894 a better performance than other SI protocols. However, the 895 proposed protocol utilizes the 3D vector mobility support. 896 Therefore, the proposed protocol does not need a re-searching 897 message to find the user vehicle moving on the road. The pro-898 posed protocol contains the link with the user vehicle, owing 899 to the PSQ information with PSO data from the pre-learning 900 phase. It takes a low failure probability to link the connection 901 with the user vehicle. Hence, the proposed protocol indicates 902 the best results in average delay time. Figure 9 illustrates the packet delivery ratio depending 904 on the number of video streams. When the user requests 905 the video streaming services at once to the same AP, the 906 availability of the AP that requested the service from several 907 users decreases. A live video streaming service is required 908 to have a load balancing mechanism to handle the overload 909 circumstance to prevent the issue. BSOGR shows the low 910 level of the packet delivery ratio that it abandons the used path 911 if the energy level gets low. Moreover, there is no load bal-912 ancing mechanism in the protocol. AQRV and PSOR also do 913     it is affected by the mobility prediction method of the video 963 streaming services. BSOGR, PSOR, and AQRV optimize the 964 route to the destination for video streaming transmission, but 965 do not support the mobility of the user vehicle. Therefore, 966 the three compared protocols indicate lower QoE results than 967 CLONE and PSOstreaming. CLONE has a long-term mobil-968 ity prediction algorithm based on the trajectory of the vehicle. 969 The mobility data based on the vehicle's trajectory may not 970 react to the vehicle's unexpected mobility changes; therefore, 971 the unexpected mobility changes lead to the re-transmission 972 of the streaming video file. Because of the re-transmission, 973 the user QoE gets decreased, and it is critical for the video 974 streaming service. PSO streaming has a 3D vector mobility 975 user vehicle. This short-term prediction algorithm can react