Deep Neural Network Detection for Pulsed Radar-Embedded M-PSK Communications
Christopher Y. Liu, Ric A. Romero · 2021
In this paper, we investigate the demodulation performance of radar-embedded communications, by utilizing deep neural network (DNN) machine learning to extract the communications symbols. We compare its symbol error rate (SER) performance with a detection method that uses a least squares estimator (LSE) to estimate the magnitude and phase of the radar signal and is coupled with maximum likelihood detection (MLD). We choose M-PSK modulation (specifically QPSK and 8-PSK) for illustration. When estimating and subtracting the radar amplitude prior to demodulation, SER improvements depend on collection time for both DNN machine learning and MLD demodulation methods. In this work, we develop three DNN demodulators for radar-embedded communications. The first two techniques perform very close to the traditionally optimal MLD detector while the third, which is designed for various radar-to-communications ratio (RCR), outperforms the MLD and is shown to be robust. Two interesting results to note when training the DNN: a) there is a near ideal communications signal-to-noise power ratio (C-SNR) and b) there is a near ideal radar-to-communications power ratio (RCR) in which to train the DNN for robust demodulation of radar-embedded communications that employ QPSK and 8-PSK.