Machine Learning Doppler-Tolerant One-Bit Radar Detectors

Doppler-tolerant waveforms are some of the most common radar waveforms used in practice. However, their deterministic and repetitive nature impedes control of mutual interference when multiple radars operate in close proximity. Noise radar technology may address this problem but is not Doppler tolerant. In this study, we design a machine learning radar detector capable of Doppler-tolerant performance with noise waveforms. The detector is implemented as a feedforward multilayer neural network. We show that the detector may be trained to operate with one-bit data. Further, to evaluate the proposed detector's performance, we derive a closed-form expression of the receiver operating characteristic (ROC) for one-bit detection of a Swerling 1 target using the square-law detector under the assumption of low signal-to-noise ratio (SNR). Numerical results demonstrate that the proposed machine learning detector, when suitably trained, is capable of operating with Doppler tolerance over a wide range of Doppler shifts.