Digital-twin-assisted autonomous vehicular communication networks require bounded delay, high packet delivery, and rapid service recovery under mobility, congestion, handovers, and network faults. Existing softwarized approaches remain limited by reactive control, fixed twin fidelity, redundant telemetry transfer, and incomplete coupling between prediction and programmable enforcement. This study presents TwinSoft, a closed-loop framework that converts time-aligned vehicular telemetry into proactive SDN/NFV control decisions. The framework applies semantic filtering to per-hop delay, queue occupancy, packet-loss, handover, and link-state measurements, retaining updates that are relevant to latency, reliability, or topology. Hierarchical MEC-local twins maintain high-resolution state around congested RSUs, unstable handover regions, overloaded edge resources, and fault-exposed service paths, while a cloud-level twin preserves cross-domain resource and service-chain constraints. Short-horizon simulations estimate congestion growth, latency violations, packet-loss exposure, handover instability, and fault propagation. A risk-bounded selector rejects actions that violate queue-stability, residual-risk, control-bandwidth, or VNF-capacity constraints. In the simulation environment, telemetry acquisition and forwarding control are represented using INT-like OpenFlow measurements, Ryu, and Open vSwitch, while NFV resource decisions are emulated through Python-based MEC modules. The results show that TwinSoft achieves 8.7 ms data-plane end-to-end packet latency under high-density mobility, maintains 96.1% PDR under large-scale faults, restricts control overhead to 11.3% at 2000 vehicles, and reduces recovery time to 37.8 ms compared with 95.5–138.7 ms for the evaluated baselines.
With the development of connected autonomous vehicles (CAVs), conventional communication methods are inadequate for transmitting the vast amount of real-time data collected by numerous onboard sensors. Semantic communication that exploits deep learning based joint source-channel coding (DeepJSCC) can achieve exciting compression rate (CR) and noise-resiliency performance for wireless image transmission tasks, especially in low signal-to-noise ratio (SNR) environments. However, existing DeepJSCC-based semantic communication methods are generally trained under a specific SNR, without taking into account the dynamic channel conditions of CAVs. To balance image reconstruction quality and transmission data volume, we propose an adaptive semantic communication method, that can adaptively adjust feature weights according to SNRs, thus compress the features by masks, and enhance the feature granularity through a multi-head attention module. To address the low SNR and high-speed mobility communication conditions of CAVs, a Doppler-fading channel model influenced by path loss and Doppler shift effects is built. Simulation results demonstrate that the proposed method exhibits better reconstruction quality under low SNR and low CRs compared to state-of-the-art methods.
The rapid emergence of Cooperative Intelligent Transport Systems (C-ITS) and autonomous driving has necessitated ultra-reliable low-latency communication (URLLC) within highly dynamic vehicular environments. However, the prevalence of severe Doppler shifts and Non-Line-of-Sight (NLoS) conditions poses significant challenges for traditional Low-Density Parity-Check (LDPC) decoding algorithms. This paper proposes two novel Deep Reinforcement Learning (DRL)-based decoding frameworks that adaptively optimize error-correction performance over volatile Vehicle-to-Vehicle (V2V) and Vehicle-to-Infrastructure (V2I) links. By leveraging the decision-making capabilities of DRL, the proposed algorithms—specifically the Bootstrap-Driven (B-D) DRL-Proposed Algorithm (2)—demonstrate a transformative improvement in both reliability and efficiency. Extensive simulations under Line-of-Sight (LoS) and NLoS scenarios reveal that the proposed frameworks significantly outperform the baseline DRL-WBF scheme. Most notably, B-D DRL-Proposed Algorithm (2) achieves a peak throughput exceeding 6 × 106 bps in LoS conditions and maintains a near-constant decoding latency of less than 0.005 ms, even at low Eb/No. Furthermore, it achieves a superior Bit Error Rate (BER) 10−5 in NLoS environments. These results underscore the potential of DRL to provide robust, real-time, and high-throughput decoding solutions for next-generation V2X communication standards. These results underscore the potential of DRL to provide robust, real-time, and high-throughput decoding solutions for next-generation V2X communication standards, thereby providing the deterministic low-latency foundation required to guarantee upper-layer statistical quality-of-service (QoS) in time-sensitive vehicular networks.
This paper investigates adaptive uplink power control for a low Earth orbit (LEO) satellite-assisted vehicular communication network, where a mobile satellite ground terminal reuses the same spectrum as nearby terrestrial base stations (BSs). Such satellite–terrestrial spectrum sharing can extend service coverage for vehicular and remote mobile users, but it may also produce harmful interference to terrestrial receivers. A spectral-efficiency (SE) maximization problem is first formulated under average transmit-power and average interference constraints. By applying the Lagrangian method, a closed-form interference-aware water-filling policy is derived, which adapts to both the desired LEO satellite channel and the strongest terrestrial interference channel. The framework is then extended to direct energy-efficiency (EE) maximization. The resulting fractional program is solved using Dinkelbach’s method, and the EE-optimal power-control policy is derived in closed form for each Dinkelbach iteration. In addition, a hybrid learning-and-optimization benchmark is developed, where an AI-assisted power predictor is combined with one-step Dinkelbach refinement to reduce online computational complexity while approaching the EE-optimal benchmark. Simulation results reveal the transition between interference-limited and power-limited regimes. They also show that the SE-optimal policy achieves higher throughput in the high-threshold region, whereas the EE-optimal policy significantly improves EE by avoiding excessive transmit power. These results demonstrate the importance of objective-aware power control for balancing throughput, energy consumption, and terrestrial interference protection in LEO satellite-assisted vehicular uplink communications.
This paper tackles forwarding unavailability prediction in crowdsourced Internet of Vehicles (IoV), where rational and independently operated vehicles contribute their resources to support packet forwarding and multi-hop connectivity. Although incentives can encourage participation, user-owned vehicles may become unavailable during their trips due to contextual factors related to trip conditions, resources, environment, or user preferences, causing packet drops, route breaks, higher delay, and reduced packet delivery. Existing works mainly focus on malicious forwarding behavior or network-level prediction, while overlooking context-driven forwarding unavailability caused by rational vehicle decisions. Moreover, blockchain-enabled solutions support transparent dissemination of behavior-related information, but usually keep the prediction process off-chain, limiting its verification in decentralized environments. To address these limitations, this paper proposes a blockchain-assisted context-aware framework that predicts a vehicle’s forwarding unavailability probability before route establishment in crowdsourced IoV using temporal, spatial, and weather-related context. Each vehicle uses a lightweight XGBoost model trained on its trip behavior data and deployed on-chain through a Forwarding Behavior Predictor smart contract, while the predicted probability and validation parameters are recorded and disseminated via a Prediction Registry smart contract. The prediction is then integrated into Forwarding Behavior-Aware QoS-OLSR to proactively avoid vehicles likely to become unavailable during relay selection and routing-cost computation. Simulation results show improved packet delivery, higher successful flow completion, and fewer router drops compared with the baseline under different contexts and prediction uncertainty.
Modern intelligent transportation systems rely fundamentally on vehicular networks capable of ultra-responsive communication, a capability brought into reach by 5G and beyond technologies for applications such as autonomous driving. A persistent obstacle within these fast-changing, mixed-technology networks is the dual and often competing requirement to slash both data transmission delays and power usage, a multi-faceted optimization that existing solutions, focused on singular objectives, fail to address adequately.Confronting this trade-off, we present SONG-PO-DRL, a new hybrid framework that fuses a swarm-optimized non-dominated sorting genetic algorithm with deep reinforcement learning to drive adaptive, multi-criteria decision-making. Evaluation with seven algorithms as baselines, using simulation implemented with OMNET++, Simu5G, Veins, and SUMO, shows that this approach surpasses current state-of-the-art methods in critical areas such as latency and competitively; energy consumption, cluster head longevity, congestion levels, and processing overhead. By providing a foundational design strategy, this framework enables the development of higher-performance, more stable, and scalable vehicular networks, thereby contributing directly to the evolution of dependable and energy-conscious future transportation infrastructures.
Connected and autonomous vehicles require reliable collaborative learning for safety-critical perception tasks such as traffic sign recognition. However, in vehicular networks, raw driving data cannot be centralized because of privacy and bandwidth constraints, and conventional machine learning architectures struggle to accommodate the resulting scale and data heterogeneity. Federated learning (FL) addresses data privacy through local training, but most existing FL frameworks still rely on a central aggregation server, introducing a single point of failure and limiting system resilience in dynamic Internet of Vehicles (IoV) environments. To overcome this limitation, we propose a decentralized hierarchical blockchain-based FL framework in which vehicles train local models and submit updates to regional base stations (BSs). Each BS performs validation-based model selection using an independent validation dataset, while a two-tier blockchain, consisting of microblocks for regional selections and keyblocks for global consensus, coordinates decentralized aggregation without a central server. Evaluated on the German Traffic Sign Recognition Benchmark (GTSRB) under non-IID data distribution against FedAvg, Krum, and FLTrust, the proposed framework achieves 93.35% test accuracy, retains 92.78% accuracy under a 30% label-flipping attack, and maintains reliable operation with up to 50% base-station unavailability. The framework therefore offers a resilient decentralized learning solution for IoV applications that require reliable recognition performance and continued operation under adversarial behavior and partial infrastructure unavailability.
Efficient control of the wireless channel remains fundamental for the reliable operation of Cooperative Intelligent Transportation System (C-ITS), particularly as an increasing number of Facilities-layer services simultaneously exchange periodic and event-driven messages over a shared medium. While Decentralized Congestion Control (DCC) mechanisms regulate channel access in current European Telecommunications Standards Institute (ETSI) Intelligent Transportation System (ITS) standards, recent standardization efforts are moving towards broader Resource Management (RM) concepts that enable explicit coordination of channel usage across services. This paper introduces a modular RM framework that coordinates heterogeneous Facilities-layer services through explicit service budget allocation. Services declare their unconstrained transmission demand to the RM, which computes constrained service-specific budgets based on channel conditions and predefined prioritization policies. A lightweight RM interface is integrated into each service to translate these budgets into enforceable transmission allowances without modifying standardized service logic. The framework is implemented and experimentally evaluated on a real ITS-G5 testbed and integrated with the Cooperative Awareness Service (CAS) and Collective Perception Service (CPS). Results show that the proposed RM stabilizes channel load and significantly reduces packet loss and message latency under high station densities, while enabling services to adapt their transmission behavior in a controlled manner, rather than relying on uncontrolled packet drops.
Mobile edge computing (MEC) offers robust computational services, whereas vehicle computing delivers diverse sensing capabilities. A novel task interaction model is introduced, in which vehicles execute sensing tasks to collect data, and MEC servers perform analysis tasks to process the collected data. Within this task interaction model, MEC servers and vehicles collaborate to provide a comprehensive suite of services to users. We address the problem of resource allocation and task interaction in edge-vehicle collaborative computing (EVCC). A multitasking model is proposed in which each user has multiple sensing tasks that must be offloaded to distinct vehicles to minimize interference among vehicles. A three-way auction mechanism is introduced to maximize social welfare, involving three participants in the EVCC system: users, MEC servers, and vehicles. We demonstrate that the proposed mechanism is strategy-proof and outline how to design a strategy-proof mechanism for three participants, in which users, MEC servers, and vehicles can achieve maximum utility only by submitting truthful declarations. The proposed mechanism achieves individual rationality, consumer sovereignty, and budget balance. Furthermore, the approximate ratio of the proposed mechanism is analyzed. The experiment results show that the proposed mechanism can increase social welfare by an average of 14%, compared with Greedy-MMAPB-FA.
Cellular vehicle-to-everything (C-V2X) communications are becoming increasingly important for the development of intelligent transportation systems (ITS). Radio resource allocation (RRA) plays a critical role in ensuring reliable and efficient C-V2X communications. Both centralized and distributed RRA have been widely studied. Although centralized RRA has the potential to deliver superior performance, it also suffers from significant control signaling overhead. To address this challenge, we propose a fog-assisted system architecture with a proactive centralized scheduling workflow to mitigate the signaling overhead. Building on this architecture, we develop a priority-driven position-aware heuristic resource allocation (PPHRA) algorithm based on dynamic scheduling and spatial reuse for vehicle-to-vehicle (V2V) communications in the scenario of vehicle platooning. In addition, considering the half-duplex (HD) constraint inherent in C-V2X, we introduce a two-tier evaluation framework that assesses performance within a safety-critical distance and a broader communication distance, targeting ultra-high reliability with bounded timeliness in the former and high reliability with satisfactory timeliness in the latter. Simulation results show that the proposed algorithm achieves favorable reliability and timeliness performance, while ensuring efficient resource utilization and fairness under varying traffic densities, compared with the baseline greedy and semi-persistent scheduling schemes.
Timely collection of sensor updates from ground Internet of Things (IoT) devices is critical for many monitoring and decision-making applications. Unmanned Aerial Vehicles (UAVs) provide flexible data collection capabilities, particularly in hard-to-reach environments; however, due to their limited flight endurance, dynamic IoT data generation times, and the need for coordinated multi-UAV operation, trajectory planning is a challenging optimization problem. Although recent research has investigated Age of Information (AoI)-aware trajectory design, most existing techniques fail to take into account dynamic data arrival, multi-UAV cooperation, and scalable optimization with numerous Ground Base Stations (GBSs) for effective data offloading. To address these limitations, this paper proposes ESTAMD, an Enhanced SFW (Superb Fairy-Wren)-Based Trajectory Planning Algorithm for AoI-aware multi-UAV IoT data collection. The proposed algorithm simultaneously minimizes Peak AoI, Average AoI, and total UAV flight distance while respecting each UAV’s maximum flight endurance and IoT devices’ data-generation timestamps. ESTAMD combines three significant enhancements-greedy-seeded initiation, adaptive hybrid mutation, and reflecting opposition exploration-to improve global exploration, maintain population variety, and avoid early stagnation in complex search fields. Extensive simulations across diverse IoT densities, UAV fleet sizes, and multiple GBS deployments show that ESTAMD consistently outperforms existing methods such as Heuristic, AoI-Optimal Trajectory Planning Combine Improved Ant Colony Optimization (AOTPACO), Improved Partheno-Genetic Algorithm with Enhancement Mechanisms (EIPGA), and classical SFW. The proposed approach achieves significantly lower AoI values, shorter UAV travel distances, and competitive computational efficiency, confirming its effectiveness, computational efficiency, and scalability for AoI-aware multi-UAV IoT data-collection scenarios of increasing problem size.
Transportation resilience is increasingly challenged by climate change and extreme weather, demanding robust and adaptive vehicular communication. Platooning, where autonomous vehicles travel closely in coordinated formations, can improve highway safety, traffic utility, and fuel efficiency, but its success hinges on reliable vehicle-to-everything (V2X) links. This paper proposes a visible light communication (VLC)-enabled V2X framework for platooning that integrates VLC and RF across infrastructure-to-vehicle (I2V) and vehicle-to-vehicle (V2V) links and employs Deep Q-Learning to optimize data rate under diverse weather conditions. We model four architectures: standalone radio frequency (RF), standalone VLC, hybrid RF-VLC (RF for I2V, VLC for V2V), and hybrid VLC-RF (VLC for I2V, RF for V2V), and evaluate their end-to-end performance using data rate and BER in clear air, fog, light rain, and heavy rain. Simulations show that standalone VLC achieves the highest peak rate in clear conditions (18 Mbps vs 10 Mbps for RF) but degrades sharply beyond 20 m and under heavy rain (near-zero by 30 m), while RF sustains 2 Mbps at 100 m. The hybrid RF-VLC setup provides a steady balance between high throughput and reliable performance across different distances and weather conditions. In clear air, VLC achieves a very low bit error rate (BER) of about 10-6, but this rises to nearly 0.05 during heavy rain. By combining the strong SNR efficiency of VLC with the resilience of RF, the hybrid system delivers superior BER across varying SNR levels. A Deep Q-Network (DQN) agent, modeled as a Markov Decision Process with switching and outage penalties, learns to select the best architecture in real time. This approach improves data rates by up to 60% compared to a baseline policy and continues to show strong gains as the platoon size increases. The agent's choices converge mainly to the RF-VLC mode, which matches the observed performance trends. Overall, these results demonstrate that DRL-enabled hybrid RF-VLC is a robust and high-throughput solution for platooning in diverse environments.
Enhanced Security Asynchronous Federated Learning (ESAFL) is proposed for Internet-of-Vehicles scenarios characterized by high mobility, intermittent connectivity, and security-sensitive model updates. Unlike existing IoV federated learning approaches that handle authentication and asynchronous aggregation separately, ESAFL introduces a closed-loop framework that integrates rapid ECC-based initial and handover authentication, pre-aggregation update admission gating, two-stage asynchronous aggregation across RSU and MBS layers with staleness-aware weighting, and post-aggregation policy feedback for revocation and re-authentication. In this way, security events directly influence whether an update is admitted, how much it contributes under delay, and how subsequent cross-domain access is regulated. Simulations and prototype measurements show that ESAFL completes handover authentication in 8.26 ms and reaches the target accuracy within 150-400 rounds across four mobility scenarios, reducing the required training rounds by 10.0%-40.0% relative to the baselines. Under unauthorized or protocol-violating upload attempts, ESAFL preserves stable convergence and rejects most invalid submissions before they affect model fusion. These results demonstrate that ESAFL enables timely and trustworthy model refresh for safety-and QoS-sensitive IoV services under frequent handovers and unstable connectivity.
Unmanned Aerial Vehicles (UAVs) are increasingly used across fields such as precision agriculture, surveillance, and search and rescue. As missions become more complex and larger in scale, it is vital to have robust path planning strategies that account for communication constraints. Even though extensive research exists on UAV path planning, communication is often treated as a secondary or even optional factor, which makes many autonomous solutions less practical. This paper provides a systematic review of recent progress in communication-aware UAV path planning, with particular emphasis on artificial intelligence and hybrid frameworks. Using the PRISMA 2020 protocol, peer-reviewed studies from the last five years are examined, with a focus on methods that treat communication as a central design element. The gathered methods are grouped into a single taxonomy with five areas: swarm intelligence, reinforcement learning, graph and sampling, optimization, and potential field techniques. This analysis shows a clear trend toward learning-based, distributed, and hybrid setups, moving away from simple graph-based and centralized planners. Additionally, the review examines how the reliability of air-to-air and air-to-ground communication, network topology, and control structures affect planning in multi-UAV systems. A comparison is presented for communication-aware path planning across architectural patterns, protocol choices, environmental considerations, algorithmic families, and key design issues such as computational complexity, scalability, and simulation-to-real deployment readiness. A discussion focused on applications shows that missions involving surveillance, delivery, and swarms require optimizing trajectories and communication resources together. This study establishes a comprehensive taxonomy of performance metrics, providing a benchmark for evaluating the fundamental trade-offs between mobility, energy sustainability, and connectivity. The paper concludes by outlining current challenges and future research directions, including the development of scalable hybrid learning optimization models and resilient, communication-aware autonomous UAV systems.
With the rapid growth of the Internet of Things and advances in unmanned aerial vehicle technologies, UAV-aided communication systems have shown remarkable solutions for disaster management and emergency communication. However, traditional deep-reinforcement learning methods face challenges in handling non-convex decision-making optimization, multi-objective optimization, and mixed action space, which are critical to guarantee stable convergence of UAV deployment in an mmWave CF-mMIMO network. This work effectively proposed hybrid methods that combine proximal policy optimization (PPO) and soft actor-critic optimization (SAC) algorithms to jointly optimize UAV placement, user association, and access point coordination while explicitly accounting for traffic demand and user mobility. The hybrid algorithm integrates the regularized SAC entropy objective for efficient exploration in continuous action space with the clipped surrogate PPO objective for stable updates in discrete action spaces within the centralized training. By using joint optimization, the framework maximizes energy efficiency, spectral efficiency, and fairness while minimizing overall energy consumption. The simulation results demonstrate a tremendous improvement in system performance including energy consumption reduction to approximately 2.15 kj, energy efficiency gains of +1.5 bps/Hz/W at 30dBm, overwhelming downlink spectral efficiency of 9.94 bps/Hz at 50% with fairness index of 0.95 and sum rate gain of 13%, 33%, 52%, 85% better over SAC, PPO, SCA and heuristic baselines. These findings demonstrate the effectiveness of hybrid DRL-based 3D UAV deployment in mmWave cell-free massive MIMO systems.
We propose a distributed framework for traffic congestion estimation in 5G NR-V2X sidelink (Mode 2) networks that integrates rate-distortion-based compression with consensus filtering. The resulting Compressed Consensus Estimator (CCE) enables vehicles to exchange compact, quantized estimates within existing SCI/PSSCH fields, preserving full 3GPP compliance without requiring extra spectrum. Analytical results establish mean-square error bounds, convergence under quantization and packet loss, and large-deviation reliability. The framework is validated through both system-level simulations (WiLabV2X + SUMO) and a real vehicular dataset, showing that CCE achieves estimation accuracy close to the centralized Cram & eacute;r-Rao bound, converges within one second, and remains robust under dense traffic and high packet loss. Compared with uncompressed consensus, adaptive gossip, and observer-based baselines, CCE consistently yields lower estimation error and faster convergence, while meeting NR-V2X latency and scalability requirements, enabling practical ITS deployment.
The integration of electric vehicles as distributed energy resources through vehicle-to-grid technology promises decentralized coordination, yet existing peer-to-peer energy trading protocols critically overlook how vehicle-to-vehicle communication impairments degrade market efficiency and grid stability. This paper presents Cross-Layer Energy-Aware Trading (CLEAT), a novel reputation-based mechanism that jointly optimizes geographic routing and double-auction clearing under realistic wireless constraints. We formulate the social welfare maximization problem with explicit packet delivery ratio and latency constraints, then propose a distributed algorithm achieving order N-log-N complexity suitable for real-time operation. The unified reputation metric incentivizes cooperation in both packet relay and energy transaction fulfillment through exponential moving average updates. Extensive Python simulations with ten-seed statistical validation demonstrate that CLEAT achieves ninety percent of centralized ideal welfare while reducing grid peak-to-average ratio by twenty-three percent compared to communication-agnostic baselines. Under thirty percent packet loss, welfare degradation remains below four percent, and the mechanism isolates adversarial vehicles within three to five market rounds. These results establish communication-aware co-design as essential for practical vehicle-to-grid deployment, with direct implications for IEEE 2030.1 interoperability standards and future cellular-vehicle-to-everything infrastructure planning.
The rapid proliferation of data-intensive applications in the Internet of Vehicles (IoV) imposes stringent requirements on bandwidth and latency. However, conventional IoV routing protocols primarily rely on fixed spectrum allocation strategies, which renders them inadequate for highly dynamic, high-throughput networking environments. Cognitive Radio-enabled IoV (CIoV) addresses spectrum scarcity by allowing opportunistic access to idle licensed bands. In contrast, most existing CIoV routing methods overlook urban intersection dynamics, where traffic signal cycles induce severe network partitions, frequent link breakages, and intermittent connectivity. While intersection-aware strategies consider mobility variations, they often assume static or fully available spectrum, ignoring dynamic primary user (PU) activity. This separation of mobility and spectrum considerations limits routing performance. To address both challenges, we propose a unified delay-aware routing framework for CIoV intersections that jointly models traffic light phases, intersection-specific mobility, and spectrum dynamics within a routing decision process. By bringing these factors into a single framework, this study provides a more systematic way to support reliable and low-delay data forwarding in urban CIoV scenarios. A dual-mode strategy adapts to signal phases, combining joint relay and channel selection with an intersection-aware fallback mechanism. A composite cost function balances transmission delay, waiting time, and spectrum availability. Extensive evaluations using a joint SUMO-NS-3 mobility-network simulation framework show that TiDAR consistently outperforms two representative routing schemes under varying transmission distances, road segment lengths, vehicle densities, PU activity durations, packet sizes, traffic light switching times, and directionally asymmetric vehicle arrivals. On average, TiDAR reduces end-to-end delay by 36.4% and improves throughput by 62.8%, demonstrating that the coordinated integration of traffic-signal awareness, relay availability, and spectrum prediction can substantially improve routing robustness in signalized urban CIoV environments.
