DNN-based Signal Detection for Underwater OTFS Systems
Shumin Zhang, Yuzhi Zhang, Jiazheng Chang, Bin Wang, Weigang Bai
Abstract
Orthogonal time-frequency-space (OTFS) is a new two-dimensional (2D) modulation technique that offers reliable communications over time-frequency selective channels. In underwater acoustic (UWA) channel, the time selective fading and frequency selective fading are severe. The receiver has to recover the OTFS signal which is distorted by inter-symbol interference (ISI). The traditional UWA OTFS receiver performs explicitly channel estimation and equalization for the detection of received symbols, which require the prior knowledge of system. In this paper, a deep learning based signal detection method is proposed for UWA OTFS communication, where a deep neural network (DNN) can recover the received symbols after sufficient training. The proposed DNN OTFS detection method has been trained and tested under both simulation and experimental UWA channels. Numerical results demonstrate that the DNN based OTFS detection method performs lower bit error rate (BER) than classical detectors.