AI-Enhanced 3-D Channel Modeling for Aerial Double-RIS Assisted-V2V Communications With UAV Fluctuation Compensation

Abstract

Intelligent vehicular networks require reliable high-capacity links in dynamic non-stationary environments. Conventional geometry-based stochastic models (GBSMs) struggle to accurately capture the complex propagation effects in aerial-assisted links, particularly those affected by unmanned aerial vehicle (UAV) attitude fluctuations. In this paper, we propose a three-dimensional (3D) artificial intelligence (AI)-enhanced channel model for aerial double reconfigurable intelligent surface (RIS)-assisted vehicle-to-vehicle (V2V) communications. The proposed model integrates UAV mobility, random attitude fluctuations, and double-RIS configurations. We derive the complex channel impulse response (CIR), spatial cross-correlation function (CCF), temporal autocorrelation function (ACF), and frequency correlation function (FCF). A lightweight multilayer perceptron (MLP) neural network is embedded to compensate for RIS perturbation errors induced by UAV dynamics. Furthermore, the twin delayed deep deterministic policy gradient (TD3) method is employed to jointly optimize UAV trajectories and RIS parameters. Simulation results demonstrate that the proposed model effectively captures spatial-temporal-frequency non-stationary characteristics, achieving superior channel correlation and communication capacity compared to conventional single-RIS models and other deep reinforcement learning-based algorithms.