In light of the ongoing transformation in the automotive industry, driven by the adoption of 5G and the proliferation of connected vehicles, network security has emerged as a critical concern. This is particularly true for the implementation of cutting-edge 5G services such as Network Slicing (NS), Software Defined Networking (SDN), and Multi-access Edge Computing (MEC). As these advanced services become more prevalent, they introduce new vulnerabilities that can be exploited by cyber attackers. Consequently, Network Intrusion Detection Systems (NIDSs) are pivotal in safeguarding vehicular networks against cyber threats. Still, their efficacy hinges on extensive data, which often contains sensitive and confidential information such as vehicle positions and owner’s behaviors, raising privacy concerns. To address this issue, we propose a Privacy-Preserving Self-Supervised Learning (SSL) based Intrusion Detection System for 5G-V2X networks. The majority of works in the literature relying on Federated Learning (FL) and often overlook data labeling on the end devices. Our methodology leverages SSL to pre-train NIDSs using unlabeled data. Post-training is then performed with a minimal amount of labeled data, which can be carefully crafted by an expert. This novel technique allows the training of NIDSs with huge datasets without compromising privacy, consequently enhancing the efficacy of cyber-attack protection. Our innovative SSL pre-training methodology has yielded remarkable results, demonstrating a substantial improvement of up to 9% in accuracy across a diverse range of training dataset sizes, including scenarios with as few as 200 data samples. Our approach highlights the potential to enhance automotive network security significantly, showcasing groundbreaking achievements that set a new standard in the field of automotive cybersecurity.

Security and privacy are crucial in V2X networks due to sensitive user information. Public Key Infrastructure (PKI) is widely used in C-ITS to ensure security and privacy. However, the practical implementation of PKI faces challenges in achieving seamless communication across diverse ITS projects worldwide. The absence of interoperability between PKI systems and unre-solved issues in existing PKI standards hinder global adoption. Despite available security standardizations using centralized PKI technology for V2X, a universally adopted PKI-based security architecture is necessary. Furthermore, the progress made in 5G V2X technology has demonstrated significant potential for revolutionizing V2X communication in the future. Enhancing the level of trust through integration of the 5G Core Network (5GC) into the PKI security mechanism can lead to more secure and efficient V2X communication. To address these challenges, we propose a blockchain-based architecture that integrates the 5GC network with the PKI infrastructure, aiming to enhance privacy and security in 5G V2X communication. Our solution is designed to be distributed and interoperable, aligned with existing ETSI ITS PKI standard. By utilizing Hyperledger Fabric (HLF) platform, a permissioned blockchain framework, we present the architecture and conduct a comprehensive security analysis to ensure compliance with security and privacy requirements of V2X communications.

Network slicing security in 5G and beyond 5G (B5G) networks is critical due to the wide range of supported services and applications. Existing literature focuses on reactive AI-based security that can detect and respond to threats after occurrence. In contrast, proactive security solutions, such as moving target defense (MTD), possess great promise. MTD involves constantly altering system configurations to increase uncertainty for attackers. Despite its potential, existing work that incorporates MTD often overlooks the intricate balance between enhancing security and maintaining network operational effi-ciency. This work proposes a novel approach to integrating Deep Reinforcement Learning (DRL) with MTD for network slicing security, our approach creates a moving target by dynamically reconfiguring IP addresses, complicating reconnaissance efforts, and thwarting potential attacks. Experimental results show that our solution achieves approximately 98 % effectiveness against Distributed Denial of Service (DDoS) attacks, demonstrating its efficacy in proactively mitigating threats.

Current flow-based Network Intrusion Detection Systems (NIDSs) have the drawback of detecting attacks only once the flow has ended, resulting in potential delays in attack detection and increasing the risk of damage due to the infiltration of a greater number of malicious packets. Moreover, the delay provides attackers with an extended period of presence within the network, enabling them to execute subsequent attacks. To overcome this drawback, this work addresses the issue of early flow classification in NIDSs that incorporates a Deep Learning (DL) model. This model leverages Recurrent Neural Networks (RNNs), including Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU), coupled with attention mechanisms. This strategic combination allows the system to harness the inherent sequential nature of packets within network flows, enhancing the efficiency of early flow classification. We conducted experiments on two up-to-date network intrusion datasets, namely CIC-IDS2017 and 5G-NIDD. Our findings demonstrate the effectiveness and accuracy of the proposed NIDS in classifying network flows. Additionally, our approach showcases its efficacy by promptly identifying and detecting attacks in their early stages without the need for flow termination. This results in a reduction in both the number of initial packets required for classification and the time needed for detection.

Being part of the 6G ecosystem vision, Connected and Automated Vehicles (CAVs) will enjoy sophisticated tailored services offering road safety and entertainment for users. As one of the 6G cornerstones, Network Slicing (NS) allows the creation of various customized 6G-V2X (Vehicle-to-Everything) use cases on the same physical infrastructure. However, 6G-NS advances can open up breaches to cyber-attacks aiming to break 6G-V2X Network slices to inflict maximum damage on CAVs and their users. Crossing borders, where CAVs leave their V2X-NS (V2X Network Slice) in the Home Mobile Network Operator (H-MNO) toward a similar V2X-NS in the Visited MNO (V-MNO), is an attractive opportunity to exploit by attackers. Detecting and mitigating attacks, in this case, becomes a priority, confronted by NS requirements and MNOs not ready to share their private data. To this end, this paper proposes a 3GPP-compliant privacy preservation collaborative learning scheme for 6G-NS security, focusing on V2X-NS cross-border areas. Our scheme leverages multi-process Federated Learning (FL) architecture to build efficient V2X-NS security-related models while preserving 6G V2X-NS isolation. In addition, it uses differential privacy-enabled stacking to build up attack detection knowledge at the V2X-NSs and MNOs levels while ensuring privacy preservation. We conducted an experimental study on the 5G-NIDD dataset, which is one of the most realistic publicly available 5G datasets. Our results demonstrate that multi-process FL with stacking can deliver high accuracy while ensuring isolation between 6G-V2X-NSs and privacy preservation between H-MNO and V-MNO.

Cyber security for 5G and Beyond (5GB) network slicing is drawing much attention due to the increase of complex and dangerous cyber-attacks that could target the critical components of network slicing, such as radio access and core network. This paper proposes a new cyber defense approach based on two-layer Federated Learning (FL) to protect 5GB network slicing from the most dangerous network attacks and a mean-field game to safeguard the FL-enabled defense system from poisoning attacks. Our proposed distributed defense systems cooperate, intending to detect internal and external attacks targeting the critical components of 5GB network slicing and detecting infected parts in the 5GB defense system. Our experimental results show that our cooperative defense systems exhibit high accuracy detection rates against network attacks, namely (distributed) denial of service and botnets while being robust against poisoning attacks and requiring a few overheads generated by defense systems. To the best of our knowledge, we are the first to propose lightweight and accurate cooperative defense systems based on two-layer FL and non-cooperative games to enhance security against attackers in 5GB network slicing.

As part of the 5G, Connected and Automated Vehicles (CAVs) will benefit from Network Slicing (NS) in several tailored 5G- Vehicle-to-Everything (V2X) services running on the same physical infrastructure. However, the use of 5G- NS may also increase the risk of cyber-attacks that could compromise 5G-V2X network slices (5G-V2X-NSs) and cause significant harm to CAV's passengers. This risk is particularly high at cross-borders, where CAVs move from their Home Mobile Network Operator (H-MNO) to a Visited MNO (V-MNO), with similar 5G-V2X-NSs in place. Therefore, deploying security services to neutralize 5G- V2X NS threats in this scenario is mandatory. However, if H-MNO and V-MNO act independently, deploying these security services could be inefficient and may result in increased memory, processing, and network resource consumption. Thus, MNOs should collaborate to orchestrate their security services to neutralize 5G-V2X NS attacks and optimize their costs efficiently. In this context, this paper proposes a novel approach to enhance the security of 5G-V2X NS at cross-borders using Reinforcement Learning (RL) based security orchestration. Specifically, we trained and deployed an RL agent interacting with both H-MNO and V-MNO. The RL agent efficiently deploys security services to effectively remove threats, optimize resource utilization, and minimize the impact on 5G-V2X-NSs. The performance results show that the RL-based security orchestration neutralizes threats with an average success rate of almost 100%. Additionally, resource consumption is minimal at less than 8 %, and the acceptable impact on 5G- V2X - NSs is negligible, averaging less than 12 %.

Vehicular-to-Everything (V2X) communication standards ensure reliable and high-performance data exchange among vehicles, pedestrians, and the roadside infrastructure. 5G New Radio (NR) is a crucial technology that enables Vehicle-to-Network (V2N) and Vehicle-to-Infrastructure (V2I) communications. In the security context, applications and network services that rely on these communication interfaces are subject to external attack sources like radio jamming that target the same control and data frequencies used by them. This causes system and network performance degradation and even Denial of Service (DoS) events, which could lead to traffic accidents involving vehicles and/or Vulnerable Road Users (VRUs). Radio jamming attacks can adopt a smart behavior by changing the targeted center frequency, bandwidth, duration, or time between two consecutive attack bursts over time. Given the context above, we propose in this paper a Deep Learning (DL)-based approach to detect radio jamming attacks on V2I/V2N communication interfaces. Our DL model is trained using a dataset collected from our 5G-V2X testbed. Results show that our DL model outperforms traditional ML algorithms and provides a detection accuracy of up to 96%, a false positive rate of less than 3%, and a detection time decrease of 39% minimum.

As the automotive industry grows, modern vehicles will be connected to 5G networks, creating a new Vehicular-to-Everything (V2X) ecosystem. Network Slicing (NS) supports this 5G-V2X ecosystem by enabling network operators to flexibly provide dedicated logical networks addressing use case specific-requirements on top of a shared physical infrastructure. Despite its benefits, NS is highly vulnerable to privacy and security threats, which can put Connected and Automated Vehicles (CAVs) in dangerous situations. Deep Learning-based Intrusion Detection Systems (DL-based IDSs) have been proposed as the first defense line to detect and report these attacks. However, current DL-based IDSs are processing and memory-consuming, increasing security costs and jeopardizing 5G-V2X acceptance. To this end, this paper proposes a lightweight intrusion detection scheme for 5G-V2X sliced networks. Our scheme leverages DL and Knowledge Distillation (KD) for training in the cloud and offloading knowledge to slice-tailored lightweight DL models running on CAVs. Our results show that our scheme provides an optimal trade-off between detection accuracy and security overhead. Specifically, it can reduce security overhead in computation and memory complexity to more than 50% while keeping almost the same performance as heavy DL-based IDSs.

Deploying Connected and Automated Vehicles (CAVs) on top of 5G and Beyond networks (5GB) makes them vulnerable to increasing vectors of security and privacy attacks. In this context, a wide range of advanced machine/deep learning-based solutions have been designed to accurately detect security attacks. Specifically, supervised learning techniques have been widely applied to train attack detection models. However, the main limitation of such solutions is their inability to detect attacks different from those seen during the training phase, or new attacks, also called zero-day attacks. Moreover, training the detection model requires significant data collection and labeling, which increases the communication overhead, and raises privacy concerns. To address the aforementioned limits, we propose in this paper a novel detection mechanism that leverages the ability of the deep auto-encoder method to detect attacks relying only on the benign network traffic pattern. Using federated learning, the proposed intrusion detection system can be trained with large and diverse benign network traffic, while preserving the CAVs' privacy, and minimizing the communication overhead. The in-depth experiment on a recent network traffic dataset shows that the proposed system achieved a high detection rate while minimizing the false positive rate, and the detection delay.

As part of Intelligent Transportation Systems (ITS), Electronic toll collection (ETC) is a type of toll collection system (TCS) which is getting more and more popular as it can not only help to finance the government’s road infrastructure but also it can play a crucial role in pollution reduction and congestion management. As most of the traditional ETC schemes (ETCS) require identifying their users, they enable location tracking. This violates user privacy and poses challenges regarding the compliance of such systems with privacy regulations such as the EU General Data Protection Regulation (GDPR). So far, several privacy-preserving ETC schemes have been proposed. To the best of our knowledge, this is the first survey that systematically reviews and compares various characteristics of these schemes, including components, technologies, security properties, privacy properties, and attacks on ETCS. This survey first categorizes the ETCS based on two technologies, GNSS and DSRC. Then under these categories, the schemes are classified based on whether they provide formal proof of security and support security analysis. We also demonstrate which schemes specifically are/are not resistant to collusion and physical attacks. Then, based on these classifications, several limitations and shortcomings in privacy-preserving ETCS are revealed. Finally, we identify several directions for future research.

Unlike previous mobile networks, 5G and beyond (B5G) networks are expected to be the key enabler of various vertical industries such as eHealth, intelligent transportation, and Industrial IoT verticals. To support that, B5G networks enable to sharing of common physical resources (radio, computation, network) among different tenants, thanks to network slicing concept and network softwarization technologies, including Software Defined Networking (SDN) and Network Function Virtualization (NFV). Therefore, new research challenges related to B5G networks have emerged, such as resources management and orchestration, service chaining, security, and QoS management. However, there is a lack of a realistic platform enabling researchers to design and validate their solutions effectively, since B5G networks are still in their early stages. In this paper, we first discuss the different methods for deploying realistic B5G platforms for the V2X vertical, including the key B5G technologies. Then, we describe DRIVE-B5G, a novel platform that serves as an end-to-end test-bed to emulate a vehicular network environment, allowing researchers to provide proof of concept, validate, and evaluate their research approaches.

Intrusion detection systems (IDS) have already demonstrated their effectiveness in detecting various attacks in cellular vehicle-to-everything (C-V2X) networks, especially when using machine learning (ML) techniques. However, it has been shown that generating ML-based models in a centralized way consumes a massive quantity of network resources, such as CPU/memory and bandwidth, which may represent a critical issue in such networks. To avoid this problem, the new concept of Federated Learning (FL) emerged to build ML-based models in a distributed and collaborative way. In such an approach, the set of nodes, e.g., vehicles or gNodeB, collaborate to create a global ML model trained across these multiple decentralized nodes; each one with its respective data samples that are not shared with any other nodes. In this way, FL enables, on the one hand, data privacy since sharing data with a central location is not always feasible and, on the other hand, network overhead reduction. This paper designs a new IDS for C-V2X networks based on FL. It leverages edge computing to not only build a prediction model in a distributed way, but also to enable low latency intrusion detection. Moreover, we build our FL-based IDS on top of well-know CIC-IDS2018 dataset, that includes the main network attacks. Noting that, we first perform a feature engineering on the dataset using the ANOVA method to consider only the most informative features. Simulation results show the efficiency of our system compared to the existing solutions in terms of attack detection accuracy while reducing the network resource consumption.

As a leading enabler of 5G, Network Slicing (NS) aims at creating multiple virtual networks on the same shared and programmable physical infrastructure. Integrated with 5GVehicle-to-Everything (V2X) technology, NS enables various isolated 5G-V2X networks with different requirements such as autonomous driving and platooning. This combination has generated new attack surfaces against Connected and Automated Vehicles (CAVs), leading them to road hazards and putting users’ lives in danger. More specifically, such attacks can either intra-slice targeting the internal service within each V2X Network Slice (V2X-NS) or inter-slice targeting the cross V2X-NSs and breaking the isolation between them. However, detecting such attacks is challenging, especially inter-slice V2X attacks where security mechanisms should maintain privacy preservation and NS isolation. To this end, this paper addresses detecting inter-slice V2X attacks. To do so, we leverage both Virtual Security as a Service (VSaS) concept and Deep learning (DL) together with Federated learning (FL) to deploy a set of DL-empowered security Virtual Network Functions (sVNFs) over V2X-NSs. Our privacy preservation scheme is hierarchical and supports FL-based collaborative learning. It also integrates a game-theory-based mechanism to motivate FL clients (CAVs) to provide high-quality DL local models. We train, validate, and test our scheme using a publicly available dataset. The results show our scheme’s accuracy and efficiency in detecting inter-slice V2X attacks.

Connected and Automated Vehicles (CAVs) represent one of the main verticals of 5G to provide road safety, road traffic efficiency, and user convenience. As a key enabler of 5G, Network Slicing (NS) aims to create Vehicle-to-Everything (V2X) network slices with different network requirements on a shared and programmable physical infrastructure. However, NS has generated new network threats that might target CAVs leading to road hazards. More specifically, such attacks may target either the inner functioning of each V2X-NS (intra-slice) or break the NS isolation. In this paper, we aim to deal with the raised question of how to detect intra-slice V2X attacks. To do so, we leverage both Virtual Security as a Service (VSaS) concept and deep learning (DL) to deploy a set of DL-empowered security Virtual Network Functions (sVNFs) within V2X-NSs. These sVNFs are in charge of detecting such attacks, thanks to a DL model that we also build in this work. The proposed DL model is trained, validated, and tested using a publicly available dataset. The results show the efficiency and accuracy of our scheme to detect intra-slice V2X attacks.

With the integration of fog networks and vehicular networks, Vehicular Fog Computing (VFC) is a promising paradigm to the efficient collection of data for improving safety, mobility, and driver experience during journeys. To this end, we exploit the Software-Defined Networking (SDN) paradigm to propose a fully-programmable, self-configurable, and context-aware data collection scheme for VFC. This scheme leverages a stochastic model to dynamically estimate the number of fog stations to be deployed. Our simulation results demonstrate that our proposed scheme provides lower latency and higher resiliency compared to classical data collection schemes.

Vehicular networks are vulnerable to a variety of internal attacks. Misbehavior Detection Systems (MDS) are preferred over the cryptography solutions to detect such attacks. However, the existing misbehavior detection systems are static and do not adapt to the context of vehicles. To this end, we exploit the Software-Defined Networking (SDN) paradigm to propose a context-aware MDS. Based on the context, our proposed system can tune security parameters to provide accurate detection with low false positives. Our system is Sybil attack-resistant and compliant with vehicular privacy standards. The simulation results show that, under different contexts, our system provides a high detection ratio and low false positives compared to a static MDS.

Protecting the location privacy is one of the main challenges in vehicular ad-hoc networks (VANETs). Although, standardization bodies, such as IEEE and ETSI, have adopted a pseudonym-based scheme as a solution for this problem, an efficient pseudonym changing and management is still an open issue. In this paper, we propose PRIVANET, a complete and efficient pseudonym changing and management framework. The PRIVANET has a hierarchical structure and considers the vehicular geographic area as a grid. Each cell of this grid contains one or many logical zones, called vehicular location privacy zones (VLPZs). These zones can easily be deployed over the widespread roadside infrastructures (RIs), such as gas stations, to provide a secure changing and management of pseudonyms. The proposed framework consists of different building blocks: 1) an effective VLPZ-based pseudonym changing strategy; 2) a reputation-based mechanism to motivate selfish vehicles to enter VLPZs; 3) an adapted user-centric privacy model; 4) a secure hybrid mechanism for the distribution of pseudonyms sets and CRLs; 5) a method to generate the IP and MAC addresses from the pseudonym; 6) a stochastic model to estimate the number of VLPZs required at a given cell; and 7) a mathematical model for an optimal placement of the VLPZs over RIs to reduce the transportation cost of vehicles in terms of time. An extensive simulation study using a realistic map and with real traffic mobility measurements is carried out to evaluate and validate the performance of the PRIVANET. The simulation results demonstrate the effectiveness of the proposed framework.

Making personal data anonymous is crucial to ensure the adoption of connected vehicles. One of the privacysensitive information is location, which once revealed can be used by adversaries to track drivers during their journey. Vehicular Location Privacy Zones (VLPZs) is a promising approach to ensure unlinkability. These logical zones can be easily deployed over roadside infrastructures (RIs) such as gas station or electric charging stations. However, the placement optimization problem of VLPZs is NP-hard and thus an efficient allocation of VLPZs to these RIs is needed to avoid their overload and the degradation of the QoS provided within theses RIs. This work considers the optimal placement of the VLPZs and proposes a geneticbased algorithm in a software defined vehicular network to ensure minimized trajectory cost of involved vehicles and hence less consumption of their pseudonyms. The analytical evaluation shows that the proposed approach is cost-efficient and ensures shorter response time.

The pseudonym-changing approach is the de-facto location privacy solution proposed by security standards to ensure that drivers are not tracked during their journey. Several Pseudonym Changing Strategies (PCSs) have been proposed to synchronize Pseudonym Changing Processes (PCPs) between connected vehicles. However, most of the existing strategies are static, rigid and do not adapt to the vehicles' context. In this paper, we exploit the Software Defined Network (SDN) paradigm to propose a context-aware pseudonym changing strategy (SDN-PCS) where SDN controllers orchestrate the dynamic update of the security parameters of the PCS. Simulation results demonstrate that SDN-PCS strategy outperforms typical static PCSs to perform efficient PCPs and protect the location privacy of vehicular network users.

The management of radio congestion in the control channel is one of the active research areas in Vehicular Ad-hoc Networks (VANETs). Many congestion control protocols have already been proposed to ensure an optimal management of the radio control channel. LIMERIC is a well-known congestion control protocol which was adopted by the current ETSI standardization process to be applied in the future deployment of VANETs. This protocol uses a mathematical equation to adjust the beaconing rate for each vehicle based on the measured channel load and a targeted channel load. However, efficiently managing all the available bandwidth using LIMERIC is yet to be achieved and still an open challenge. To address this issue, we propose a new approach that enhances LIMERIC protocol so that the available bandwidth would be used efficiently. Our aim is to bring the measured channel load as close as possible to the targeted level of channel load. Our method combines LIMERIC with a novel local density estimation approach called Segment based Local Density Estimation (SLDE). The performance evaluation shows that our approach uses the bandwidth efficiently and allows higher beaconing rate with a fair division of the available bandwidth.

The Internet of vehicles allows connecting vehicles to the Internet to make all data from vehicles available for applications aimed towards improving safety and comfort for passengers. Density is one of the most important sensed data to gather. This information is mainly obtained through periodic messages broadcast by the neighbouring vehicles. However, the availability of this information depends on the Internet. A low penetration rate of Internet of vehicles, or the loss of Internet connection, can significantly affect the accuracy of the sensed density. Moreover, the reception rate of the periodic messages seriously drops at short distances caused by the broadcast storm problem in high‐density scenarios. To address this problem, using inter‐vehicular communications, we propose a segment‐based approach for enhancing the accuracy of the local density estimation. This approach provides a highly accurate estimation with low overhead over the maximum vehicles transmission range to all the vehicles. The proposed approach is extensively evaluated analytically and by simulation. Performance evaluation results show that our approach SLDE allows about 3% of mean error ratio with low overhead over the maximum transmission range.

Protecting the location privacy is still one of the main challenges in Vehicular Ad-hoc Networks(VANETs). Although, standardization bodies such as IEEE and ETSI have adopted the pseudonymous scheme as a solution to this problem, an efficient pseudonym changing and management is still an open issue. In this paper, we propose a complete and efficient pseudonym management and changing scheme based on Vehicular Location Privacy Zone (VLPZ). We define VLPZ as a roadside infrastructure designed to pseudonyms management and changing. This scheme considers that the vehicular geographic area is partitioned as a grid, where each cell contains one or many VLPZs. The location privacy protection level provided by the scheme depends on the VLPZ capacity and the number of vehicles that are inside it at the same time. For this reason, we also propose a reputation mechanism to stimulate vehicles to enter to the VLPZ, and finally evaluate the performances of the proposed scheme using Veins Framework based on OMNet++ network simulator and SUMO mobility. Simulation results demonstrate the effectiveness of the proposed scheme.