Federated Learning mmWave Beamforming for V2X Communications with Imperfect CSI and Doppler Shift

초록

The proposed FL-mm-V2X approach addresses challenges in vehicle-to-everything (V2X) communications, leveraging mmWave beamforming to enhance spectrum efficiency and reduce interference in the presence of severe Doppler shift (DS) and imperfect channel state information (I-CSI). FL-mm-V2X combines federated learning (FL) with non-orthogonal multiple access (NOMA) for mmWave beamforming. Vehicle users conduct client training, and road side users (RSUs) collect gradients, optimizing power allocation through iterative updates. The optimization involves solving a non-convex problem with Lagrangian variables, adhering to Karush-Kuhn-Tucker conditions, utilizing a sub-gradient approach. The approach employs a convolutional neural network for DS estimation, evaluated against metrics such as bit error rate, DS estimation error, and signal-to-interference-plus-noise ratio (SINR). Comparative analysis includes SINR, the number of vehicle users, transmitted power, and complexity. Simulation results highlight the proposed approach's effectiveness in mitigating I-CSI and DS, demonstrating lower complexity compared to existing methods and confirming its suitability for dynamic performance in high-mobility mmWave channels.

키워드