Radar-Embedded SATCOM with Deep Neural Network Demodulation

In this work, the feasibility, design, and implementation of radar-embedded communications with satellite applications are investigated. We design a deep neural network (DNN) machine learning detector to demodulate SATCOM data. The performance result is compared with the detection method of using maximum likelihood estimation (MLE) to estimate the amplitude and phase of the radar signal, which is followed by a maximum likelihood detection (MLD) receiver. Pulsed radar and linear frequency modulation (LFM) waveforms are chosen to embed communications symbols. Quaternary phase-shift keying (QPSK) and eight phase-shift keying (8PSK) modulations are used for illustration. In this work, three DNN demodulators for radar-embedded communications are developed. One of the DNN detectors actually outperforms the MLD demodulator and is shown to be robust for pulsed radar-embedded communications. One of our goals is to embed satellite communications into LFM waveform, which is used in synthetic aperture radar (SAR). The DNN works well for LFM radar-embedded communications when the received LFM phase offset is removed a priori. However, the DNN symbol error rate (SER) performance suffers when the LFM phase offset is introduced for large RCR. Lastly, we perform laboratory transmission and reception tests: a) shielded cable and b) over-the-air (OTA) tests. It is shown that pulsed radar-embedded communication is feasible with both MLE-MLD and DNN detectors with reasonable SER performance.