A novel Channel Estimation Method based on Deep Neural Network for OTFS system

Abstract

Orthogonal time-frequency space (OTFS) is a waveform technology designed in recent years, which can be applied to wireless communication scenarios with high Doppler extension. An accurate channel estimation result is critical in the OTFS system. Therefore, this paper focuses on channel estimation techniques based on the deep learning (DL) for the OTFS system. In our presented scheme, the delay-Doppler (DD) domain channel estimation problem is modeled as a recovery problem of sparse signal and then processed by orthogonal matching pursuit (OMP). Next, we present a five-layer deep neural network (DNN) to enhance the rough channel estimation result. Moreover, because our proposed DL-based channel estimation scheme is a model-driven paradigm, it has the advantages of a small scale of training data and a short training time. Simulation results prove that the presented DNN-based scheme obviously outperforms the traditional OMP algorithm, and the NMSE performance gain is about 5dB when the NMSE is 0.0012. In addition, we also show that the presented scheme has excellent robustness to channel mismatch and applies to different scenarios.