Intelligent Network Slicing in V2X Networks – A Comprehensive Review

I.INTRODUCTION

Intelligent network slicing is a technique used to optimize the performance of vehicle-to-everything (V2X) networks. It involves partitioning the network into multiple virtual networks, each of which is designed to handle a specific type of traffic. For example, one virtual network could be optimized for high-bandwidth video streaming, while another could be designed to prioritize low-latency communications for safety-critical applications. The connectivity needs are ever-increasing, so it is demanding rapid action toward consideration of the network capacity in the existing infrastructure, especially with the advent of the Internet of Things (IoT) [1]. Every few years, new technology emerges and adds traffic to the network. For instance, intelligent transportation systems have emerged recently and are focusing on making autonomous cars ubiquitous. This entails the connectivity demands from millions of sensors for context-aware driving [2]. This type of intelligent transportation falls into the category of V2X communication [3]. The system resides on multiple sub-communication networks like vehicle to vehicle (V2V), vehicle to network (V2N), vehicle to infrastructure (V2I), and vehicle to pedestrians (V2P) [4]. In the last decade, V2X networks have been capturing the interest of researchers in the domain of IoT and communication systems in general. The widespread of such sophisticated systems is seen as a landmark that will transform the future of vehicular networks for good [5]. V2X networks offer safe roads, efficient transportation, and many side applications are expected to emerge [6].

At present, the V2X networks comprise two types of technologies: the vehicular networks based on Dedicated Short-Range Communication (DSRC), which is founded on IEEE 802.11p and IEEE 1609.1.4 standards, and those based on the vehicular cellular networks such as long-term evolution (LTE) [7]. DSRC has long been seen as the only V2X technology; nevertheless, identifying the associated limitations and drawbacks, including the coverage, latency, and insufficient throughput and data rate, made the development of advanced and more sophisticated networks inevitable [8]. The higher throughput and data rates enable the development of heterogeneous networks (HetNets) that allow large coverage cells (LTE-Macrocells) as well as small coverage cells (direct link communication DSRC). The LTE macrocells help attaining higher throughput rates and wide coverage by having access to a large bandwidth in the applications like infotainment. The small cells meet a short range and direct coverage [9]. LTE cells can be used to broadcast priority signals using direct DSRC (short coverage) for transmitting information related to road safety conditions, such as accidents or giving way to ambulances to keep neighboring vehicles aware [10].

The capacity of V2I links and the underlying ultra-reliability requirements are catered to through the ergodic capacity of these networks while ensuring reliability across all links. The channel dynamics emanating from higher speeds depend on spectrum sharing efficiency, allocation of power, and catering to the large fading effects in a channel [11]. This shifted the focus toward using cellular communication networks called C-V2X [12]. This standard entails communication enablement across individual V2X entities like V2V, V2I, and V2P. In 2017, a bi-air interface was used, which includes a wide area network and a direct communication interface. The focus here was to ensure enhanced road safety mechanisms through awareness messages and decentralized notifications [13].

Fifth Generation (5G) networks are seen as the potential enablers of such V2X communication networks having the innate ability to meet the high throughput and low-latency demands [14]. HetNet is one of the 5G network topologies which contains multiple radio access technology (Multi-RAT) in the same network, as shown in Fig. 1. This scheme is considered to be a fundamental one while addressing networking slicing. The network slicing view differed from the users’ and network administrators’ perspectives. In the former, network slicing refers to using different RAT for meeting specific user needs, including V2X and safety. In the latter, network slicing refers to the user some RAT for massive machine-type communications (mMTC) applications, RATs for ultra-reliable low-latency communications (uRLLC) applications, and RATs for enhanced mobile broadband (eMBB) applications [15]. Massive machine-type communications RAT high-density connectivity requirements, including some specific applications in urban areas, etc. uRLLC RATs are used for low-latency and high-reliance communication, including autonomous driving. eMBB RAT is used for infotainment and other high-bandwidth application.

Fig. 1. V2X HetNet configuration.

With the increase of HetNets, the network control and management demands across the base stations also increase [16]. Additionally, the controller must address vehicle control demands, which can be managed through software and flexible networks.

Although a plethora of benefits is offered by the software-enabled V2X networks, the control and modeling of such network slices are highly complex. A typical slice of network demands meticulously worked modeling of the existing network and taking timely decisions [17]. Therefore, using intelligent and autonomous systems that rely on self-learning and decision-making becomes indispensable. To cater for such issues, machine learning (ML) algorithms are commonly used and reside on an underlying AI scheme. These systems contrast the traditional modeled networks, where a system designer needs to manually incorporate all the parameters to create network slices [18]. The ML methods such as deep learning (DL) and neural network can learn dynamic network parameters and make predictions based on available information while incorporating past events and learned parameters. These methods are widely used in network applications such as data analysis, network congestion analysis, and resource management [19]. Using existing methodologies and tailoring their use in V2X communication are expected to be a highly efficient network slicing technology.

ML’s popularity in network slicing is due to its ability to extract patterns and features from complex network structures and parameters [20]. Various ML/DL algorithms shall be used, which are trained initially on the general network conditions and later used for making the allocations on their own. Additionally, they encompass RL, where the system dynamics are continuously changing. Therefore, a static ML algorithm that has been exposed to only a finite set of conditions cannot perform efficiently [21].

This study reviews V2X slicing using a proposed taxonomy that describes enablers, configurations, requirements, and ML/DL algorithms. The study is structured to target various schemes of comparison, including networking, the network slicing schemes in general, and a comparison between their use and limitations, followed by a comparison of the use and benefits of ML techniques in network slicing. Related survey papers are summarized in Table I.

Table I. Existing surveys

The surveys revised in this paper are classified into two groups based on the V2X perspective. The first group is the end-user perspective, and the second is the industry perspective. From an end-user perspective, such as in [22–24], security and safety are the two metrics of concern using ML algorithms. In [24], ML algorithms have been provided for network slicing in V2X. In [22], these networks are reviewed under safety consideration with the help of software-defined network (SDN) controllers.

The study by [23] investigates the ML algorithms under security measures under their application to the SDN controllers. The works in [25–29] take an industrial outlook and determine the control mechanisms under network requirements, including the eMBB, mMTC, and URLLC. To meet the URLLC slicing needs, [25,26] provides the use of mobile edge computing (MEC). These algorithms help decreasing the latency across the edge vehicles and help in lowering the capacity across the cellular BS. The work in [27] takes a perspective of motion planning in V2X instead of slicing. It uses the ML to the SDN in the context of autonomous vehicles. Bandwidth awareness is the work concern in [28] to provide vehicles with efficient spectral efficiency. In [29], Implementing 5G in vehicular networks has potential for increasing capacity, data rate, coverage, and device mobility. These studies’ limitations are that they focus on only one slicing area. There is no comprehensive review that discusses the V2X slicing in all directions. Thus, the contributions of this study are as follows:

• •To the best of our knowledge, this is the first survey that deals with applying ML algorithms in V2X slicing based on the network requirements from a dual perspective.
• •This paper discussed techniques for network slicing in conventional and machine learning approaches in V2X networks and the limitations identified by providing a proposed taxonomy.
• •Also, it explores and describes some challenges with possible solutions in applying these algorithms, like the availability of datasets and security issues in applying ML algorithms in V2X slicing.
• •This paper derived some open issues and future directions in V2X slicing based on ML algorithms like low computation process techniques, datasets collection, and controlling slicing.

The structure of this survey is organized as follows. Section II presents a taxonomy proposed for the V2X slicing based on enablers, requirements, configuration, and ML/DL algorithms used. Section III provides some challenges and solutions to applying ML in V2X slicing. Section IV discusses the open issues and future directions of V2X slicing. Finally, a summary of the review and the article is concluded in Section V.

II.V2X SLICING TAXONOMY

V2X slicing is a type of network slicing that is designed to support communication between vehicles and other entities in the transportation ecosystem, including pedestrians, infrastructure, and other vehicles. The taxonomy of V2X slicing typically includes the following categories:

• •Radio Access Network (RAN) Slicing: This type focuses on the radio interface between the vehicle and the infrastructure, including base stations and roadside units. RAN slicing enables the operator to allocate radio resources such as bandwidth, power, and frequency bands to different V2X use cases based on their specific requirements.
• •Core Network Slicing: (also called (Service/Application Slicing)) focuses on the higher layers of the V2X network stack, including the application layer where specific services or applications are hosted. This slicing type enables the operator to allocate resources such as compute power, memory, and storage to V2X use cases based on their specific requirements.
• •Context-aware Slicing: This is a more advanced form of slicing that considers the context of the V2X communication, such as the location of the vehicle, the type of service/application being used, and the level of security required. This slicing type enables the operator to allocate resources dynamically based on the changing needs of the V2X communication.

Overall, the taxonomy of V2X slicing provides a framework for creating and managing virtual networks that can be optimized for specific V2X use cases, enabling operators to provide more efficient, reliable, and secure V2X communication services.

This section elaborates on the V2X slicing taxonomy presented in Fig. 2, where a focus is gradually steered toward using ML algorithms for control and management. The approaches presented in taxonomy generally depend on the enablers techniques like SDN controller, network function virtualization (NFV), and cloud computing. The purpose of using ML is to enhance their performance and give them a competitive edge. Thus, ML/DL can play a role in enabling or assisting with network slicing technologies. By taking a holistic view of the network and using a dataset approach. Also, ML/DL can help determine actions to be taken within the network. The second element presented in the taxonomy is the two tiers of slicing, that is, inter-slicing and intra-slicing. Keeping these configurations under review allows taking decisions and actions based on ML. The third focus on the taxonomy is the underlying parameters, including latency, throughput, and bandwidth requirements for slice creation.

Fig. 2. V2x slicing taxonomy.

Thus, the presented taxonomy is an all-encompassing baseline for furthering the V2X slicing discussions. A one-stop review is provided to the readers in the context of V2X slicing and enabler while enabling them to apprehend the right enabler choice in a particular system design.

The inter- and intra-slicing configurations will be leveraged using the appropriate ML algorithm. The general requirements in V2X are the slicing based on the throughput, latency, and bandwidth requirements, which have been discussed in this review article.

III.CHALLENGES AND SOLUTIONS

Although some of the ML algorithms serve as a potential solution for those optimized V2X networks, there are various challenges associated with their use. Furthermore, intelligent network slicing in V2X networks is an emerging research area that aims to provide customized and efficient network services to support various V2X applications such as safety, traffic efficiency, and infotainment. Nevertheless, there are still some possible solutions available, illustrated in Table V.

Table V. Summary of the open research challenges and their possible solutions

In summary, many open research challenges exist in intelligent network slicing in V2X networks. Possible solutions include developing efficient algorithms, protocols, and mechanisms to address these challenges and providing customized and efficient network services to support various V2X applications.

IV.OPEN ISSUES AND FUTURE DIRECTIONS

Intelligent network slicing in V2X networks is an emerging research area that aims to provide customized and efficient network services to support various V2X applications such as safety, traffic efficiency, and infotainment. While significant progress has been made in this area, several open issues and future directions still need to be addressed. Also, there is a need to research hybrid machine learning methods dedicated to vehicular communication. The open research issues to be addressed are as follows:

• •Low computation process techniques: The computational load increases significantly with the large datasets produced from V2X networks serving under highly dynamic conditions. This demands the use of high-end processing systems and hardware [60] data conditions. Generally, ML and DL algorithms have been used in network slicing to achieve a variety of objectives, including optimizing resource allocation, predicting network performance, and detecting anomalies. Supervised learning algorithms are commonly used to train models that can predict network performance based on input parameters. Unsupervised learning algorithms can be used to detect anomalies in network traffic or behavior, and reinforcement learning can be used to learn how to allocate resources in a way that maximizes performance while minimizing cost. Overall, ML methods have proven to be a powerful tool for network slicing, allowing network operators to optimize resource allocation, detect anomalies, and predict network performance with greater accuracy and efficiency. Some researchers, such as [25,61], propose hybrid ML methods, on the contrary, that are less process intensive and are specially designed for the taxonomy of vehicular systems. These shall involve a mix of distributed learning and traditional deep nets.
• •Datasets Collection: As also stated in the challenges section, the availability of an all-inclusive dataset is low in V2X due to the dynamic conditions [62]. In such conditions, collecting all the datasets may render a huge size emerging from various network nodes and devices. Therefore, one way is to use a simulative approach for producing a dataset representative of the real environment [63]. Devising data collection mechanisms in V2X is an open research work, with few datasets available.
• •Scenario-based ML Algorithms: It is desired to use hybrid ML algorithms that enable the selection of the most relevant dataset for training purposes [64]. The hybrid models enable generalization using optimization algorithms, including game theory and ML [65]. As a result, scenario-based models have been produced that help reduces the computational load during the training and testing phases.
• •Multi-RAT management and controlling slicing: This is a hot topic in research in the V2X context. The network slicing methods also produce different RAT requirements that should be managed and controlled [66]. Multi-RAT management in V2X slicing is considered an open issue, especially when we look toward autonomous driving, intelligent networks, and ML.
• •Multi-dimensional network slicing: Current research in intelligent network slicing in V2X networks mainly focuses on single-dimensional slicings, such as slicing based on QoS requirements. However, multi-dimensional network slicing is needed to support the diverse and complex requirements of V2X applications, which consider various dimensions such as location, time, and context.
• •Energy-efficient network slicing: V2X networks operate in a resource-constrained environment, and energy efficiency is a critical concern. Developing energy-efficient network slicing mechanisms can reduce energy consumption, improve network efficiency, and prolong the network’s lifetime.
• •Interoperability and standardization: V2X networks involve multiple stakeholders, including vehicles, roadside units, and network operators. Interoperability and standardization are crucial for the efficient operation of V2X networks. Developing standard interfaces and protocols for intelligent network slicing in V2X networks can enable seamless communication and interoperability among stakeholders.
• •ML-based network slicing: ML has shown promising results in various domains, including network slicing. Developing ML-based network slicing mechanisms can provide more accurate and efficient network slicing decisions and improve network performance.

In summary, the open issues and future directions for intelligent network slicing in V2X networks include multi-dimensional network slicing, edge computing in network slicing, energy-efficient network slicing, interoperability and standardization, and ML-based network slicing. Addressing these issues can lead to developing more efficient and customized network services to support various V2X applications.

V.CONCLUSION

The study thoroughly reviewed the network slicing in the V2X networks through intelligent ML algorithms. The work entailed presenting different slicing methods in general, then presenting the limitations and introducing review techniques based on ML. These techniques were further presented in the form of V2X network slicing techniques. Their potential ability to provide lower latency and high throughput has been highlighted. It was further established that data-driven approaches yet limited these intelligent schemes. At the end of this paper, some challenges and open research issues were highlighted. In the future, these techniques can be tested at full potential once the 5G enabling techniques get fully functional and widespread V2X networks can be experienced. In addition, the existing ML algorithms presented in this study shall be analyzed for their merits and demerits across vehicular networks and areas where improvements can be made. The ultimate aim shall be to use hybrid ML and traditional resource-sharing models to make the slicing of V2X networks a reality. In future work, there are several areas where future work could focus on the use of ML/DL for network slicing in V2X. Dynamic slicing, edge computing, improved resource allocation, standardization, and privacy and security are the guarantors of V2X privacy.