Internet of Vehicles (IoV) is a typical extension of Internet of Things (IoT). Specifically, Federated Learning (FL) is capable of alleviating the knowledge sharing and privacy protection problems of IoV, and further enhancing the driving experience and service quality. However, dueto the scanty data results of new environment and the high-cost of expert data-labeling, posing an imminent challenge of how to reinforcement the vehicles’ data. In this letter, a data reinforcement mechanism is proposed to utilize the vehicles’ unlabeled dataset sufficiently and enhance the vehicle collaboration ultimately. Particularly, the dataset distribution characteristics of vehicles’ datasets are calculated to measure the dataset similarity. Furthermore, the labeling models are distributed to vehicles to empower the unlabeled data with the assistance of dataset distribution characteristics. Experimental results show that the proposed data reinforcement mechanism is capable of labeling the unlabeled data accurately and improving the performance of FL.

Cooperative perception has emerged as a promising solution to overcome the limitations of individual sensing in autonomous driving. However, its performance in real-world scenarios is often inconsistent due to dynamic environments and fluctuating wireless communication quality. In particular, the quality of service (QoS) in both vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) links is frequently affected by physical constraints, while limited communication and computation resources further exacerbate system latency and degrade perception accuracy. To address these challenges, we propose a novel communication-sensing-computation integrated cooperative perception framework for connected vehicular networks. This framework considers V2V and V2I cooperation and introduces a unified optimization strategy that simultaneously handles resource allocation, fusion method selection, and cooperative agent selection. We formulate this as a mixed-integer nonlinear programming (MINLP) problem and decompose it into two subproblems, which are efficiently solved using a block coordinate descent method. To enable adaptive and scalable decision-making, we design a deep reinforcement learning (DRL)-based joint resource allocation scheme for V2I, and extend it with a federated learning (FL) variant for decentralized V2V scenarios. Extensive multi-factor simulations demonstrate that our method significantly improves perception accuracy, reduces latency, and enhances robustness under diverse network conditions. These results validate the effectiveness and practicality of our proposed framework in real-world cooperative driving environments.

This paper investigates a novel client selection for the volatile Federated Learning (FL) systems, where volatility means that the state of the client set, client datasets, and client training status will change over time. We study how to select clients dynamically to mitigate the volatility. Particularly, the volatile client selection problem is formulated as a classification problem, and we propose two new metric features. The Multi-Feature Volatile Client Selection (MFVCS) algorithm, which considers client training capacity, client-weighted data quality, and client historical selection entropy, is proposed to solve the volatile client selection problem. Moreover, we have developed an adaptive dynamic weighting algorithm that allows for dynamic updating of the weight for each feature. We propose a volatility ratio to measure client volatility. The experimental results indicate that the proposed algorithm demonstrates strong robustness and better performance under different volatility ratios of the client set. In particular, the proposed MFVCS algorithm improves the model accuracy at most by $\mathbf{9.2\%}, \mathbf{9.6\%}$ and $\mathbf{12.5\%}$ under 0.01 volatility ratio, 0.05 volatility ratio and 0.1 volatility ratio, respectively.

With the increasing number of Internet of Things devices and sensors in vehicular network, a huge amount of data is generated. Vehicle Edge Computing (VEC) utilises the computation resources at the edge of the network and can efficiently process this big data through computational offloading techniques. However, due to the neglect of communication network relationships among vehicles, current Vehicle-to-Vehicle (V2V) computation offloading schemes encounter challenges such as high communication latency, substantial communication overhead, and the wastage of computation resources. To address these challenges, we design a computation offloading mechanism based on Federated Graph Neural Network (GNN) for vehicular networks, that is, vehicular FedGNN (V-FedGNN). Firstly, our modeling approach considers features including vehicle speed, location, available resources, and wireless network, which are embedded in the graph structure. Secondly, we design weighted vehicular communication network topology and propose weighted total delay optimization problem. Finally, this paper proposes a prediction model based on Federated Learning (FL) and GNN to minimize the weighted computation offloading delays among vehicle nodes. Experimental results demonstrate that our proposed scheme achieves high offloading prediction accuracy, with an average value of 98.1% and achieves low offloading latency, correspondingly.

User selection has became crucial for improving energy efficiency in communication of federated learning (FL) over wireless networks. However, centralized user selection causes additional system complexity. This study proposes a network intrinsic approach of distributed user selection that leverages the radio resource competition mechanism in random access. Taking the carrier sensing multiple access (CSMA) mechanism as an example of random access, we manipulate the contention window (CW) size to prioritize certain users for obtaining radio resources in each round of training. Training data bias is used as a target scenario for FL with user selection. Prioritization is based on the distance between the newly trained local model and the global model of the previous round. To avoid “excessive contribution” by certain users, a counting mechanism is used to ensure fairness. Simulations with various datasets demonstrate that the proposed method can rapidly achieve convergence similar to that of the centralized user selection approach.

With cutting-edge sensors and learning algorithms developed for vehicular perception, breakthrough advancements have been made in the perception-based end-to-end autonomous driving in recent years. However, the reliability of autonomous driving systems could be compromised by the vulnerability of perception module to occlusion. To address this issue, the integration of vehicle-to-vehicle communication enabled perception data sharing in the dynamic driving task has been proposed and has yielded notable results, as demonstrated by COOPERNAUT, a cooperative system based on distributed lidar perception. In this paper, we introduce ICOP, an end-to-end driving system based on multi-agent camera cooperative perception, to select sensor sharing nodes and to fuse intermediate image data features for learning a driving policy. In the ICOP system, each agent encodes image information into Bird’s Eye View (BEV) representations individually, and these representations are then transmitted as payloads of V2X (vehicle-to-everything) messages via wireless connection, thus enables capturing global spatial interactions among agents to form comprehensive BEV perception information used for final control decision-making. Supported by our designed mechanism of vehicle-to-vehicle communication and transformer block to achieve acceptable image sensory data size for transmission, the experiments suggest that the proposed cooperative perception driving system achieves better results than lidar-based systems in challenging driving situations compared to prior works.

Unlocking the Potential of Smart Cities: Our paper details a groundbreaking field test validating the direct camera-to-RSU connection in a C-V2X pedestrian warning scenario. By leveraging local AI computing within the Sony IMX500 smart camera, our approach eliminates the need for traditional computing servers, leading to significant improvements in pedestrian warning speed. Test results demonstrate a reduction in response time by up to approximately 1 second, showcasing the efficiency gains and transformative benefits of building smarter cities.

Platoon-based vehicular network in NR-V2X has been considered as a promising technology to assist reducing traffic congestion, saving vehicle fuel, and enhancing driving experience. Resource allocation is the basis for ensuring stable and safety vehicular networks. In this paper, we propose a Distributed Resource Allocation algorithm using Federated Multi agent Deep Reinforcement Learning (DRAFRL), which mathematically utilize the federated averaging (FedAvg) mechanism to reduce the variance between agents and achieve better transmission performance. The proposed algorithm consists of four steps: Firstly, each agent updates local model by deep deterministic policy gradient (DDPG) algorithm. Secondly, the agents upload local model parameters to the base station (BS) for federated aggregation. Thirdly, the BS performs weight aggregation using the FedAvg method and updates the global model. Finally, the BS distributes the optimized global model parameters to each agent. The simulation results show that the proposed algorithm outperforms other baseline algorithms while reducing the variance between agents by 93.5% and 99.1% compared with two baselines.

As an important research field in time series processing, traffic prediction has a profound impact on people’s daily lives and social development. Conventional traffic prediction relies on complete observation data. However, data missing is common in cities due to equipment failure, network interruption, etc., which poses a huge obstacle to traffic prediction. In this paper, we design a novel Spatiotemporal Ego-graph Domain Adaptation framework (SEDA) to predict traffic state in data missing scenarios. Based on the multi-dimensional topological information of local network (ego-graph), isomorphic ego-graphs are aligned across the missing data in target domain and the external data in source domain to obtain alternative data. Furthermore, a Dual-branch Cross reCoupling method (DCC) is proposed to reconstruct missing features according to the alternative data. Experimental results on real public datasets with 10%-40% missing show that SEDA averagely outperforms both the state-of-the-art knowledge transfer-based prediction baselines and the incomplete data prediction baselines by more than 0.45% and 0.86%. Ablation experiments and visualization analysis further demonstrate the effectiveness of SEDA components.

As a promising distributed learning technology, federated learning (FL) is used in wireless communication to efficiently utilize distributed data. However, statistical heterogeneity is often ignored as a crucial factor affecting wireless federated learning (WFL) performance. Besides, free rider is common in real world. In this paper, we consider the statistical heterogeneity and free rides jointly with limited resources. We first define a new measurement considering the substitutability and wholeness of client, called contribution degree. Then we propose the Contribution Degree-based Client Selection (CDCS) algorithm to improve WFL performance. Experiments validate that the proposed algorithm improves the global model accuracy, achieves fast convergence and reduces total delay.

Traffic supply-demand mismatching has a severe impact on intelligent transportation systems. Fortunately, order dispatching is a promising option to mitigate the traffic supply-demand imbalance. Along this line, this article proposes the Multi-Driver Multi-Order Dispatching (MDMOD) method to make efficient order dispatching policy and enhance the experience of drivers and passengers. In the proposed MDMOD method, the Dynamic Multi-Objective Reward Learning (DMRL) algorithm is proposed to measure the driver-order-pair value, which illustrates the importance of a driver serving a specific order. A centralized matching algorithm is introduced to match all drivers and orders to maximize all driver-order-pair values. The multi-objective reward in the DMRL algorithm considers both immediate gains (i.e., pick-up distance) and future gains (i.e., the future traffic demand of order destination) to effectively improve the experience of drivers and passengers. Furthermore, by introducing the driver service level into the multi-objective reward, the “outstanding driver better reward” mechanism is realized to promote the ecological development of ride-sharing platforms. Notably, the Temporal-Graph Convolutional Network algorithm is proposed to predict the future traffic demand. Some virtual orders, which generated with the predicted future traffic demand, are dispatched to idle drivers to multiplex the traffic supply fully. A simulator is designed to test the performance of the proposed MDMOD method, experimental results demonstrate that the MDMOD method outperforms the state-of-the-art methods in terms of Average Driver Income and Order Response Rate.

Influenced by the era of the sharing economy and mobile payment, Dockless Bike-Sharing System (Dockless BSS) is expanding in many major cities. The mobility of users constantly leads to supply and demand imbalance, which seriously affects the total profit and customer satisfaction. In this paper, we propose the Spatio-Temporal Mixed Integer Program (STMIP) with Flow-graphed Community Discovery (FCD) approach to rebalancing the system. Different from existing studies that ignore the route of trucks and adopt a centralized rebalancing, our approach considers the spatio-temporal information of trucks and discovers station communities for truck-based rebalancing. First, we propose the FCD algorithm to detect station communities. Significantly, rebalancing communities decomposes the centralized system into a distributed multi-communities system. Then, by considering the routing and velocity of trucks, we design the STMIP model with the objective of maximizing total profit, to find a repositioning policy for each station community. We design a simulator built on real-world data from DiDi Chuxing to test the algorithm performance. The extensive experimental results demonstrate that our approach outperforms in terms of service level, profit, and complexity compared with the state-of-the-art approach.

This letter proposes a new travel time estimation model based on graph neural network (GraphTTE) to improve the accuracy of travel time estimation. We design a Multi-layer Spatiotemporal Graph frame (MSG), which consists of static network and dynamic networks, to fully consider the influence of traffic temporal characteristics and road network topological characteristics on travel time. Moreover, we design an Attention Graph Nodes Impact Index algorithm (AGNII) to score the impact of each node on travel time. In particular, the dynamic networks utilize the graph convolution network and gate recurrent unit to obtain the traffic characteristics, the static network utilizes graph convolution network to obtain the road basic attributes. We combine the real paths sequence with the impact score of nodes to extract the subgraph with a great impact on the trajectory. After the graph representation learning and deep residual network, the estimated time is obtained. A simulator was designed to train and test our model in Chengdu and Xi'an datasets, the results show that the mean absolute percent error (MAPE) is 12.58% and 14.01%, which is 1.54% and 1.78% lower than the baselines.

In this paper, a fast deployment strategy of unmanned aerial vehicles (UAVs) served as base stations (BSs) in an object region is investigated. To be specific, it solves a problem of how to find proper BSs position for multi-UAV as quickly as possible, and it also achieves the goal of maximizing the sum of downlink rates in a communication network. For this purpose, we design a geographical position information learning (GPI-Learning) algorithm to learn the GPI relationship between users and UAVs. This approach consumes less time by avoiding calculation of exact channels and fills a gap existed in the scenario of setting multi-UAV rapidly to serve multi-user. Without loss of generality, we apply GPI-Learning in different scenarios, such as changes in user number or area size. As for different area size, simulation reveals that a proper size is adequate to any smaller size on condition that the smaller size is included in training set. Numerical results witness the good performance of our proposed algorithm.

In this paper, an unmanned aerial vehicle (UAV) flies as a base station (BS) to provide wireless communication service. We propose two algorithms for designing the trajectory of the UAV and analyze the impact of different training approaches on transferring to new environments. When the UAV is used to track users that move along some specific paths, we propose a proximal policy optimization (PPO) -based algorithm to maximize the instantaneous sum rate (MSR-PPO). The UAV is modeled as a deep reinforcement learning (DRL) agent to learn how to move by interacting with the environment. When the UAV serves users along unknown paths for emergencies, we propose a random training proximal policy optimization (RT-PPO) algorithm which can transfer the pre-trained model to new tasks to achieve quick deployment. Unlike classical DRL algorithms that the agent is trained on the same task to learn its actions, RT-PPO randomizes the features of tasks to get the ability to transfer to new tasks. Numerical results reveal that MSR-PPO achieves a remarkable improvement and RT-PPO shows an effective generalization performance.

Jamming attack is a serious threat to wireless communications. Reactive jammer is one of the most power-efficient jammers, which can listen for activities on channel and jam receivers when it detects communications. But to the best of our knowledge, there is no existing research of defense scheme focused on sustaining the multi-user and multi-antenna communications under reactive jamming attack. In this paper, we propose iterative estimation algorithm and interference cancellation algorithm to maintain the multiuser and multi-antenna orthogonal frequency-division multiplexing (OFDM) communications under reactive jamming attack. The iterative estimation algorithm means inserting some pilots in frames of the transmitted signals to estimate the channel matrixes. Furthermore, in order to remove interference, the received signals are projected onto the orthogonal subspace of the jamming signals by using the interference cancellation algorithm. Finally, the simulation results prove that the multi-user and multi-antenna OFDM communications are nearly throttled by jamming attack and our defense mechanisms can effectively turn it into operational scenario with considerable performance under reactive jamming attack.

In this paper, we study the degrees of freedom (DoF) characterization of cache-aided 8-user multi-hop line network. The leftmost user 1 wishes to send messages to the remaining users and the messages are relayed by user 2, 3,..., 7. Each user is equipped with a local cache and we assume that the messages received by userkcan also be cached. Thus, interference signals from the messages transmitted by userk+ 1,k+ 2, ... and 8 can be eliminated since userkhas already cached the contents of messages in previous data transmission. With this cache-aided scheme, the multi-hop topology can be transformed into a special partially connected interference channel (IC). Then we derived the DoF outer bound of the network information-theoretically. To deal with the challenge of this partial connected IC model, where the number of interferences increases with the indexk, we proposed a DoF outbound achievable transmission method by assigning equidifferent signal space allocation.