Model-Driven Deep Learning for OTFS Detection With Phase Noise in V2X Communications

Orthogonal time frequency space (OTFS) is a promising multicarrier waveform, which effectively combats high Doppler effect in high mobility communications especially in Vehicle-to-Everything (V2X) scenarios. However, phase noise (PN) generated by oscillators is a key radio frequency impairment and inevitably presents in communication systems, resulting in severe performance deterioration. To this end, we design a model-driven deep learning for OTFS detection in the presence of unknown PN in V2X communication. In particular, we propose an iterative detection algorithm in the delay-Doppler domain to diminish the inter-carrier interference (ICI) caused by PN based on variational inference theory, which constitutes an approximate probabilistic inference technique associated with variational free energy minimization. Additionally, by inducing some trainable parameters, we further develop an unfolding approach to rapidly convergence and improve performance in deep learning manner. Simulation results demonstrate that the proposed detector reveals state-of-art performance comparing with other solutions in terms of bit error ratio (BER).