Resource Allocation for IRS-Assisted V2I Anti-Jamming Communications in Interweave CIoV Networks: A Transformer-enhanced Multi-Agent DRL Method

This paper proposes a novel Intelligent Reflecting Surface (IRS)-assisted interweave Cognitive Internet of Vehicles (CIoV) network under malicious jamming attacks, where the IRS enhances communication performance by establishing additional links. In order to maximize the sum transmission rate of Vehicle-to-Infrastructure (V2I) links, we propose an optimization problem that jointly optimizes wireless resource allocation, such as spectrum and transmit power for Vehicle Users (VUs) and IRS phase shift. Because this problem is non-convex and complicated, we further propose a Heterogeneous Multi-agent Transformer-enhanced Dueling Double Deep Q-Network (HMA-TD3QN) based resource allocation method, where VUs and Secondary Base Station (SBS) act as distinct heterogeneous agents can independently perform resource allocation and phase shift optimization. The Transformer neural network architecture can better adapt to long sequence input states and extract relevant features from complex input states through the attention mechanism. Simulation results indicate that the proposed HMA-TD3QN method achieves improvements of 24.42%, 20.79%, and 22.25% over the basic HMA-DQN under three different jamming strategies, highlighting the effectiveness of IRS technology in enhancing the Quality of Service (QoS) and jamming resilience of CIoV network.