Deep Learning Based Single Carrier Communications Over Time-Varying Underwater Acoustic Channel

Abstract

In recent years, deep learning (DL) techniques have shown great potential in wireless communications. Unlike DL-based receivers for time-invariant or slow time-varying channels, we propose a new DL-based receiver for single carrier communication in time-varying underwater acoustic (UWA) channels. Without the off-line training, the proposed receiver alternately works with online training and test modes for accommodating the time variability of UWA channels. Simulation results show a better detection performance achieved by the proposed DL-based receiver and with a considerable reduction in training overhead compared to the traditional channel-estimate (CE)-based decision feedback equalizer (DFE) in simulation scenarios with a measured sound speed profile. The proposed receiver has also been tested by using the data recorded in an experiment in the South China Sea at a communication range of 8 km. The performance of the receiver is evaluated for various training overheads and noise levels. Experimental results demonstrate that the proposed DL-based receiver can achieve error-free transmission for all 288 burst packets with lower training overhead compared to the traditional receiver with a CE-based DFE.

Keywords

• Channel equalization
• deep learning
• deep neural network
• DFE
• machine learning
• single carrier communication