Physical Layer Security Enhancement in IRS-Assisted Interweave CIoV Networks: A Heterogeneous Multi-Agent Mamba RainbowDQN Method.

The Internet of Vehicles (IoV) relies on Vehicle-to-Everything (V2X) communications to enable cooperative perception among vehicles, infrastructures, and devices, where Vehicle-to-Infrastructure (V2I) links are crucial for reliable transmission. However, the openness of wireless channels exposes IoV to eavesdropping, threatening privacy and security. This paper investigates an Intelligent Reflecting Surface (IRS)-assisted interweave Cognitive IoV (CIoV) network to enhance physical layer security in V2I communications. A non-convex joint optimization problem involving spectrum allocation, transmit power for Vehicle Users (VUs), and IRS phase shifts is formulated. To address this challenge, a heterogeneous multi-agent (HMA) Mamba RainbowDQN algorithm is proposed, where homogeneous VUs and a heterogeneous secondary base station (SBS) act as distinct agents to simplify decision-making. Simulation results show that the proposed method significantly outperform benchmark schemes, achieving a 13.29% improvement in secrecy rate and a 54.2% reduction in secrecy outage probability (SOP). These results confirm the effectiveness of integrating IRS and deep reinforcement learning (DRL) for secure and efficient V2I communications in CIoV networks.