Clustering Algorithms and Comparisons in Vehicular Ad Hoc Networks

Vehicular Ad hoc Network (VANET) is a new era in the transmission of dynamic information across communities. Intelligent Transportation Systems is only one of the many applications for VANET (ITS). The topology of VANET is extremely dynamic, and connections are irregular. These features cause information transmission in the VANET to be unreliable. Vehicle clustering is a successful strategy to increase the network's scalability and connection dependability. Characteristics of the VANET have an impact on clustering performance as well. An extensive explanation of VANET clustering algorithms is given in this article. A complete evaluation of clustering in VANETs is provided based on the clustering procedure. Most methods examine the clustering process in terms of Cluster Head selection metrics, formation, and its maintenance. The clustering methods are contrasted based on factors such as stability, convergence, overhead, and latency. There is also discussion of some of the most typical issues and the solutions used. Also, a summary of the performance metrics used to assess clustering algorithms is provided.


Introduction
Future generations of transportation technology will largely consist of the Intelligent Transportation System (ITS), that combines all types of vehicle communications.Services from ITS include traffic management, safety software, emergency alerts, and driving assistance [1].A self-organizing network built from moving vehicles is known as a VANET.Mobile ad hoc network (MANET) is a subset of VANET.MANET is a self-configuring network of nodes that move without a fixed infrastructure.When automobiles are used in place of mobile nodes in MANETs, the network then adopts fixed pathways, such as roadways.In a VANET, nodes move and accelerate on average relatively quickly, which causes the network's topology to change quite quickly.The On-Board Units (OBUs) and Roadside Units (RSUs) of VANET are its components.A RSU, that is set up alongside the road, record all vehicle data, which is then sent to other OBUs.RSUs have no control over any operations requiring the transmission of information in OBUs or vehicles.OBUs are additional components that are built into dynamic vehicles to help with information sharing between RSUs and vehicle components.The two main VANET communication types are vehicle to vehicle (V2V) with vehicle to infrastructures (V2I) communication.OBU-equipped vehicles are able to interact with each other within their radio ranges while V2I communication, along in addition to the placing of infrastructure along roadside the medians as well as the use of different applications which can improve the excellence of service offered by infrastructure to vehicles, aren't practical.Communications between a vehicle to an object (V2X) can include V2V and V2I.The name of the communications system designed exclusively for use in cars (DSRC) is Dedicated Short-Range Communication.It has a transition range of 100 to 1000 meters and is designed for information transfer and vehicle-tovehicle communication [3].Like Wi-Fi, the DSRC technology operates in a similar manner.The United States' Federal Communication Commission (FCC) has designated a higher spectrum band with a 75 MHz bandwidth [5], [6].DSRC supports both V2V and V2I communications.The primary purposes of VANET are to ensure vehicle safety, manage traffic, and communicate precise vehicle information.To get the most data for communication in VANET, the latest clustering techniques are provided.The primary contributions of this paper are as follows: First, we provide a summary of the observed and researched evolution of clustering methods in VANETs from 2010 to 2022.Moreover, the majority of these techniques were not listed in past studies.Second, we compile a list of the current clustering tactics and group them into three categories: Cluster Head selection, cluster formation, and cluster maintenance.Then, we evaluate these algorithms against a variety of criteria.Thirdly, various problems are presented together with the methods employed to solve them.After that, a thorough examination of the most popular metrics for assessing the effectiveness of clustering methods is presented.Network performance metrics and cluster efficiency parameters make up the performance metrics.The simulation tools for each clustering approach are also presented.

Clustering in VANET
Clustering is a well-liked VANET technology that offers a practical approach to improving network services and management.It performs significantly better in a range of applications than the conventional flat structure.Network nodes can be organized into distinct groups known as clusters [9].A group of neighbouring automobiles often forms as a result of numerous significant details and measures.The names of the node in this group include • Cluster Head (CH) -This node performs the role of coordinator or cluster leader.The major responsibility of the CH is to enable communication and information sharing among cluster members and other CHs.Several factors are taken into consideration before choosing the CH.• Cluster Member (CM) -The nodes in the cluster are the CMs.By sending broadcast messages to one another, these nodes communicate with one another.• Gateway Node (GW) -It is not necessary for this node to provide RSU to every cluster in order to aid communication with it.The cluster-based communication structure used by the VANET is shown in Figure 2. The CH is solely responsible for internal cluster communication.Inter-cluster and intra-cluster communication are the two distinct channels that separate a cluster's internal communication.Predicting node-to-node failure links during cluster maintenance improves the cluster's stability [6].

VANET clustering algorithms
Due to mobility of VANETs, clustering algorithms that were previously employed in MANETs cannot be employed.Due to the length of time required to complete the clustering phases, additional control overheads can be necessary.In order to maintain the cluster structure dynamically while without dramatically raising network costs, an effective clustering technique should only create a small number of clusters.Three MANETs clustering algorithms-Mobility Based Clustering (MOBIC), Weighted Clustering Algorithm (WCA), and Distributed and Mobility Adaptive Clustering (DMAC)-were created to satisfy the particular needs of vehicular communications.Moreover, the VANET's clustering strategies were mostly borrowed from older MANETs.Since 2010, when the VANET began to expand and develop, several clustering techniques for VANETs have been developed.

Clustering in VANETs
To finish this process, there are two stages: A. First phase-(Cluster Formation): During the cluster formation and CH selection phases, nodes send messages to choose the principal CH and CM; thereafter, regular data packets are sent between them.In order to create a stable cluster, numerous techniques may be used between the transmission of the marketing message and the CH selection.

B. Second phase-(Cluster Maintenance):
In this stage, stable cluster merging, secondary CH selection, re-clustering, and cluster splitting take place.These phases have each been explored separately by several researchers in literature.This section describes the steps and standards used in each clustering process, such as CH selection, cluster building based on hop count, and cluster administration.C. Cluster generation phase.In order to complete the clusters that have been constructed, this step passes through two processes: cluster creation process and CH selection process.With the VANET, clusters can be created in a variety of ways, including centrally located and dispersed, single-hop and multi-hop, location service and user information based, etc.The development of clusters based on topology is covered in this section.A cluster architecture in VANETs can be modelled utilising the distance that existing between the CH with its other members, the communication range, or its cluster radius.As a result, only the single-hop and multi-hop categories of algorithms are distinct.(see Figure 3).

Single-hop Clustering Algorithm
This is the technique that builds clusters in which every node and its CH are only one hop apart.This implies that each node has a direct connection to the CH.Several clustering techniques directly produce single-hop clusters based on the CH's transmission range or the limited cluster radius.The performance of single-hop clustering technique is enhanced in terms of security, connection, and stability.Single-hop clustering methods offer CHs very efficient coordination and more dependable intra-cluster communication.Due to the small coverage area of this type of cluster, there are a lot of clusters and a high maintenance cost.Collisions can happen when the number of cars is highly dense, which results in a low PDR.Because the cluster performance will suffer, these two scenarios should be avoided.In summary, low latency and strong cluster stability are provided by single-hop techniques, although clustering coverage still has to be improved.The highest and lowest number of vehicles in a cluster may also be restricted in order to address the problem of both high and low density.

Multi-hop Clustering Algorithm
Every node in a cluster is a multi-hop distance from their CH when it is built using the multi-hop distance method.The number of clusters can be decreased, the cluster coverage area can be increased, and cluster stability can be improved with multi-hop clustering techniques.The multi-hop approaches provide exceptional cluster stability and coverage, especially in terms of the amount of CM re-affiliation, CH modifications, and cluster endurance.Although the formation of multi-hop clusters is more challenging, it will take a long time for the cluster to form, which can cause a delay in the data transmission.Also, there is room for improvement in the cluster overhead.Also, certain simulation results show that when there are more than three hops, the cluster performance suffers.In other words, cluster performance will suffer as the hop count rises.Table 3 compares the clustering techniques in terms of transmission distance, vehicle density, vehicle speed, hop count, and traffic scenario.

F. Cluster maintenance:
There is high packet loss due to VANET's dynamic topology and frequent vehicle reconnections and disconnection.The cluster management process maintains strong connectivity and also provides a consistent link lifetime through CH by reducing often occurring vehicle re-clustering.The technique involves cluster merging, vehicle acceptance, vehicle leaving, and additional cluster maintenance processes.When a new vehicle transmits a signal to the CH, then a new vehicle is assigned to cluster then takes over as the CM of that particular cluster.Then CH sends signals often as vehicles join and leave.A local database change will then be made by the CH.The data for a member car is erased from the CH's local database when it can no longer communicate with it.Cluster merging, which can lower the number of clusters and improve clustering efficacy, happens when two or more clusters can be represented by a single merged cluster.The cluster merging requirements varies for each strategy.

Clustering algorithms comparison
Several distinct parameters are frequently considered when comparing clustering methods.Any clustering algorithm can be built and specified using these parameters.Some crucial factors include cluster security, latency, convergence, and overhead.Table 4 contrasts the benchmark techniques based on these settings.High stability, little overhead, latency, and convergence are features of a successful clustering technique.

Challenges and techniques used for solution
There are numerous studies using various clustering methods available to improve the performance of the network.As stated in Table 5, we examine some of these issues in this section along with the methods used to overcome them.To find solutions, the researchers have looked at a range of problems and applied a range of clustering algorithms.

Performance evaluation metrics
Cluster efficiency and network efficiency are both metrics that are most usually used to assess how effectively clustering techniques function.Any clustering technique's effectiveness can be tested and evaluated using a range of parameters.

Cluster performance parameters
Cluster performance indicators demonstrate the effectiveness of clustering solutions and represent the reliability of the network's core nodes.These variables are used to evaluate the cluster's overall performance and stability.Several metrics for cluster performance include: Cluster/CH Stability: It shows how frequently a particular vehicle has been chosen as a CH overall.Cluster number: It refers of the quantity of clusters that develop throughout network operation.With fewer clusters, the clustering process becomes more successful [6].
Cluster/CH lifetime: The vehicle has led the cluster for the longest time ever.CM lifetime: That is how long a node may remain CM.We split the total amount of instances the CM has switched states by the CM's lifetime to obtain its average.CH change rate: It shows the typical progression of the CH number over the course of time.Cluster change rate: Over the duration of a single time unit, each vehicle's average cluster number changes.Cluster size: The total number of vehicles.High cluster shapes and sizes, lengthy CH and CM a lifetime, limited cluster numbers, and moderate cluster and CH shift rates are characteristics of a good and trustworthy clustering technique.However, these traits fall short in their ability to describe the precise nature of communication links between networked cars.

Network performance parameters
The performance of the entire network is determined by these variables, which include the following.Throughput: It is the quantity of bits sent over any network every second.The performance of the network is improved when throughput is increased.

Ratio of packet loss or collision:
The frequency of transmission-related packet losses.Packet Delivery Ratio (PDR): It is the proportion of packets received by the destination to all packets received.Overhead: The vehicle receives an average amount of control messages.. E2E Delay (End to End Delay) or Latency: It measures how long a packet takes to travel from its source to its destination.The dependent on context clustering algorithms for congestion prediction, routing, and data dissemination are estimated using all of these qualities.A reliable and effective clustering strategy results in high PDR, low packet loss rate, short E2E delay, high throughput, and minimal overhead.The factors that were investigated and the simulation tools used for each strategy are shown in Table 6.
Table 6.Clustering algorithms evaluation parameters

Conclusion
VANETs have seen a number of applications in recent years.The primary purposes of VANET are to ensure vehicle safety, manage traffic, and communicate precise vehicle information.The VANET topology is dynamic because of the fast moving vehicles.Clustering represents one of the efficient solutions to the scalability problems caused by the constantly changing characteristics of the VANET.In order to address various VANETs challenges, this paper gave a thorough survey of the most clustering strategies.We discussed the clustering mechanism in VANETs.An overview of clustering methods over the course of 20 years, along with the quantity of citations they received, was first presented.The techniques and criteria for each clustering phase were then presented, together with the metrics for choosing the CH for each approach, cluster construction using hop distance, and cluster maintenance.We also compared these algorithms based on several important factors to determine how well they performed.Then, we discussed some of the difficulties faced by VANETs, along with the clustering strategies used to address them, and we evaluated the effectiveness of these strategies.We concluded by introducing some of the most used measures for assessing the effectiveness of clustering methods.According to our survey, cluster stability is one of the main problems in VANETs and the majority of clustering techniques are made for roads.Future work will use hyper-graph theory to create a novel clustering method for VANET that is appropriate for urban settings with the goal of improving clustering stability.

Fig. 1 .
Fig.1.Vehicular communications types[6] Few clustering algorithms start by choosing the CHs before starting to build clusters.Other clustering algorithms do the opposite.D. Cluster head selection.The resilience and scalability of the network are significantly impacted by CH stability.Communication within and between clusters is assured by the stable CH.To boost VANET security, a reliable vehicle can only be a CH.Table2is a list of some of these techniques and the parameters employed for CH selection.

Table 1
lists numerous VANET clustering techniques that were presented between 2010 and 2022.

Table 2 .
CH selection metrics E. Cluster formation.