Clustering in vehicular ad hoc network: Algorithms and challenges

Highlights

• A comprehensive classification of VANET clustering algorithms: strength and weakness.

• Intelligence-based VANET clustering including machine learning and fuzzy logic.

• The distinction between vehicle mobility as well as network mobility strategies.

• A detail study of VANET multi-hop clustering strategies.

Abstract

Clustering is an important concept in vehicular ad hoc network (VANET) where several vehicles join to form a group based on common features. Mobility-based clustering strategies are the most common in VANET clustering; however, machine learning and fuzzy logic algorithms are also the basis of many VANET clustering algorithms. Some VANET clustering algorithms integrate machine learning and fuzzy logic algorithms to make the cluster more stable and efficient. Network mobility (NEMO) and multi-hop-based strategies are also used for VANET clustering. Mobility and some other clustering strategies are presented in the existing literature reviews; however, extensive study of intelligence-based, mobility-based, and multi-hop-based strategies still missing in the VANET clustering reviews. In this paper, we presented a classification of intelligence-based clustering algorithms, mobility-based algorithms, and multi-hop-based algorithms with an analysis on the mobility metrics, evaluation criteria, challenges, and future directions of machine learning, fuzzy logic, mobility, NEMO, and multi-hop clustering algorithms.

Introduction

Vehicular communication for intelligent transport systems (ITS) is a rapidly growing research area. Wireless access in vehicular environments (WAVE) is used for wireless communication in vehicular ad hoc network (VANET) on the dedicated short-range communications (DSRC) frequency bands by IEEE802.11P and IEEE1609. DSRC/WAVE is currently used to satisfy the low latency requirement for safety and control messages for vehicle-to-vehicle (V2V) communication and long-term evolution (LTE) is used for vehicle to infrastructure (V2I) communication. 5 G millimeter waveband is under research to provide ultra-low-latency for vehicular communication. During V2V communication, each vehicle acts as a mobile router and an on-board unit (OBU) is used in each vehicle to communicate with the other vehicles. In this paper, vehicle, car, and node are used interchangeably to mean vehicle.

VANET has some common features with mobile ad hoc network (MANET); however, VANET has its unique features, such as high mobility that differentiates it from MANET. The vehicles in VANET do not suffer from energy deficiency but faces many new challenges due to their high mobility. When the number of vehicles increases, scalability becomes an important issue. In the absence of any central infrastructure, VANET suffers high packet loss due to a large volume of message dissemination among the vehicles for V2V communication. VANET also suffers from issues such as the hidden terminal problem, high latency for safety message transmission, message security, broadcast storm problem, quality of service (QoS), packet routing, congestion control, and resource management. To solve these issues, a hierarchical structure can be used. In a hierarchical structure, two or more nearby vehicles, who have some common features, join in a group which is called clustering. Clustering concept is widely used in data mining, and machine learning. Clustering is also used in MANET, which is the predecessor of VANET, to cluster the mobile nodes. In a clustered vehicular environment, a large network of vehicles is considered as a network of some small networks or clusters.

In VANET clustering, cluster head (CH) plays a key role in the formation process of a cluster, as shown in Fig. 1. A cluster can be created in various ways based on the input metrics. The member vehicle of a cluster is called cluster member (CM). Other than CH and CM, some algorithms use two CMs to communicate with other clusters on behalf of the CH are called cluster gateways (CGs). Unless specified as CG, all members of a cluster are termed as CMs. One CH, zero/one/two CGs, and any number of CMs can be present in a cluster. In VANET clustering, CH acts like a mobile router and CM acts like a mobile node. The role of CG lies between CH and CM. The cluster is formed based on the metrics such as the average relative velocity of the vehicle, acceleration, position, direction, the degree of the vehicle, the density of the vehicles, transmission range, etc. CH is selected from the vehicles which is most stable among the participating vehicles. The rest of the vehicles join the cluster as CMs. Therefore, CH selection is a part of the cluster formation process and no separate CM selection criteria need to be presented. CH and CMs maintain a routing table containing information of the CH and CMs of the cluster for intra-cluster communication. However, CM does not maintain any routing table for other clusters, which is maintained by the CH, if necessary. Hence, a large network is considered as a group of some small networks or clusters.

The coverage of a cluster is limited by the transmission range (TR) of the CH, as shown in Fig. 2. Since the distance covered by a CH is limited by its TR, a vehicle which is located at the edge of a cluster has a high probability of losing connection with the CH. The relative speed of two vehicles can vary all the time depending on the speed of two vehicles. When a vehicle's position is at the edge of a cluster, the vehicle may enter and exit the TR of the CH frequently due to the change in relative speed and the vehicle will lose connection with the CH more frequently. As a result, data loss will be very high when a vehicle remains at the edge of a cluster. For this reason, some algorithms prefer a geographically center vehicle as the CH for reliability.

Generally, a cluster of vehicles means a 1-hop cluster where a CH can reach all its CMs directly because the CMs are within the range of the CH; however, some clustering algorithms are based on multi-hop strategy. When a vehicle cannot reach the CH of a cluster directly but can reach a member of the cluster, then the new vehicle joins to the cluster through a CM. Hence, a CH can cover CMs in a multi-hop manner which is termed as multi-hop clustering, or N-hop clustering, or k-hop clustering. The value of N or k depends on the number of hops the CH can cover. In the Fig. 3, the 2-hop CM cannot reach the CH but can reach a CM of the CH. As a result, the 2-hop CM joins the cluster through a CM of the CH creating a multi-hop cluster.

In machine learning and data mining, many algorithms have been developed for efficient clustering. Many VANET clustering techniques are based on machine learning algorithms such as k-means clustering, hierarchical clustering, etc. Another domain for VANET clustering is fuzzy logic where future movement of the vehicle is predicted using fuzzification and defuzzification. Many hybrid architectures are also proposed for VANET clustering where machine learning-based algorithms are integrated with fuzzy logic to create efficient and stable clusters by selecting a more suitable vehicle as the CH. Previous surveys [1] on VANET clustering did not take into account intelligence-based clustering algorithms such as machine learning and fuzzy logic-based algorithms. Besides intelligence-based strategies, many algorithms emphasize the mobility parameters to provide stable clusters. Network mobility (NEMO) concept is also used for vehicular communication since NEMO has similarity with the clustering technique where the mobile router (MR) moves from one place to another place with its mobile nodes (MNs) and MNs communicate through MR only. This scenario can be compared with the CH and CMs in the clustering algorithm. Hence, MR, MN, and access router (AR) in NEMO are equivalent to CH, CM, and RSU in the VANET. Additionally, multi-hop clustering strategies are also used in VANET clustering where the CH can cover more than one hop area to reduce the number of clusters. Therefore, our aim is to classify all three major types of VANET clustering algorithms: intelligence, mobility, and multi-hop to stimulate the research of efficient and stable clustering algorithms for VANET.

The first attempt to study clustering algorithms for VANET was presented without any classification. A detail discussion on the existing survey can be found at arXiv [1]. The first classification of the VANET clustering techniques is performed based on position, destination, and medium access, etc. Few of the algorithms presented are based on intelligence, vehicle mobility, and multi-hop strategies; however, the absence of NEMO strategies and presentation of a very few intelligence and multi-hop algorithms without any further classification are not enough to study VANET clustering. Some mobility-based clustering techniques are presented in a paper along with ID-based, degree-based, direction-based algorithms, etc. Some machine learning and fuzzy logic-based strategies are discussed in a paper without classification.

Beacon message, density, direction, etc. are considered to classify the existing clustering approaches in a survey; however, only a single technique is discussed from each group. A detail classification of VANET clustering is presented based on the cluster application where a flow of the clustering techniques starting from its MANET origins is discussed. CH selection criteria, CG and CM selection metrics, etc. are described in detail with a discussion on simulators used for VANET clustering. A detail description of mobility-based clustering strategies is presented; however, no classification is presented for intelligence-based or multi-hop-based strategies. Machine learning concepts are described for the vehicular environment including few machine learning clustering techniques. A comparison of the existing surveys is shown in Table 1.

Some of the review papers discussed the machine learning-based strategies in a narrow scope while none of them give any concentration on fuzzy logic or hybrid strategies of machine learning and fuzzy logic. Vehicle mobility is covered by some papers while neglected the NEMO issue. Some reviews included multi-hop strategies but lack any detail classification.

In VANET clustering, many fuzzy logic-based algorithms have been proposed along with machine learning-based algorithms. Some NEMO algorithms are also used for VANET clustering along with vehicle mobility-based algorithms. Moreover, many multi-hop strategies have been proposed for VANET clustering to reduce the number of clusters. Therefore, we need to study all these research works extensively to study VANET clustering comprehensively, which is absent in the literature survey. During the classification of the VANET clustering algorithms, all the existing papers lack some important strategies such as machine learning, fuzzy logic, NEMO, and multi-hop algorithms. Therefore, a comprehensive study of machine learning, fuzzy logic, mobility, NEMO, and multi-hop strategies is still not present in the literature. Hence, we get three broader categories to classify VANET clustering algorithms: intelligence, mobility, and multi-hop-based clustering algorithms.

The contributions of this paper can be summarized as follows:

• The paper studied intelligence-based VANET clustering extensively, classifying them into machine learning-based and fuzzy logic-based along with a comparison among the algorithms in terms of strengths and weaknesses.

• Comparison among hybrid architectures which combine machine learning and fuzzy logic algorithms to exploit the advantages of both the schemes presented.

• Apart from vehicle mobility-based strategies, network mobility-based strategies studied separately.

• Details study of multi-hop strategies presented in a separate section.

The rest of the paper is organized as follows. In Section 2, we presented the classification of VANET clustering. Intelligence-based VANET clustering algorithms are presented in Section 3. Mobility-base algorithms are classified in Section 4, where multi-hop-based algorithms are presented in Section 5. Present challenges and future research directions are presented in Section 6 with a conclusion in Section 7.