Radar and communication integrated systems design based on deep neural networks

Abstract

AbstractDue to the spectrum congestion in the future 6G networks and the dramatic increasing requirement for the equipment that can detect and communicate simultaneously, the design of bi‐static radar and communication integrated (RCI) system is receiving more and more attention. Orthogonal frequency division multiplexing (OFDM) is a commonly used waveform in bi‐static RCI system. There are two gaps in current research that need to be filled. First, high peak‐to‐average ratio (PAPR) of OFDM signal is less considered in the conventional bi‐static RCI system due to the difficulty in recovering in the receiver. Second, the target echo estimation depends on the reconstruction of the direct‐path communication (DPC) signal. To overcome these limitations, we firstly propose a deep learning (DL)‐based transceiver design for the bi‐static RCI systems. Specifically, we design a neural network (NN)‐based waveform designer (WDNN) to reduce the PAPR of the transmit waveform, an NN‐based signal separator (SSN) to directly separate the target echo and the DPC signal, and an NN‐based symbol recovery (SRN) to demodulate the symbols. By using the training data and channel realisations, these three networks are jointly trained offline to minimise the PAPR of the transmit waveform, the separation error of the target echo and DPC signals, and the bit error rate. Then, in the online deployment phase, the trained WDNN is used to modify the transmitted waveform, and the trained SSN and SRN are placed at the receiver to separate signal and demodulate symbol. Simulation results show that compared to the conventional method, the proposed scheme not only reduces the PAPR but also improves the estimation of target parameters and communication symbol.