Deep Residual Learning for OTFS Channel Estimation with Arbitrary Noise
Xiaoqi Zhang, Weijie Yuan, Chang Liu
Abstract
Orthogonal time frequency space (OTFS) modu-lation has proved its capability of achieving significant error performance advantages over orthogonal frequency division mul-tiplexing (OFDM) in high-mobility scenarios. One challenge for OTFS channel estimation is that the performance of model-based estimators will drop dramatically in the scenarios with unknown and burst noise. In this paper, we model the channel estimation as a denoising problem and adopt a deep residual denoising network (DRDN) approach to implicitly learn the residual noise for recovering the channel matrix. Different from existing model-based channel estimators which only work well under white Gaussian noise, our proposed DRDN-based method is able to handle arbitrary noise, including both the correlated Gaussian noise and the non-Gaussian noise (e.g., t-distribution noise) cases. Finally, our simulations verify the effectiveness of the proposed OTFS channel estimation approach in arbitrary noise environments.