High-Dynamic Range PMCW Radar Sensing Through Deep-Unfolded Successive Sparse Recovery

Phase-modulated continuous wave (PMCW) radar has gained significant interest due to highly flexible waveform designs and multiple-input–multiple-output (MIMO) scaling to achieve higher angular resolutions. However, imperfect code orthogonality and the presence of self-interference (SI) limit its applicability today due to insufficient dynamic range, when compared to frequency-modulated continuous wave (FMCW) radar. This work introduces a deep-unfolded successive network that aims at increasing the dynamic range in terms of detectable targets, i.e, detecting weaker targets in the presence of strong targets, after range-Doppler (RD) processing in code-division multiplexed PMCW radar. The successive network uses sparse recovery with group ℓ1 -regularization for sidelobe suppression. Through an ablation study, we substantiate that the proposed successive unrolled network outperforms the conventional unrolled network in terms of both magnitude and phase estimation accuracy. Moreover, we present how the proposed successive network robustly scales to large MIMO configurations (up to 32 transmit antennas), where the conventional methods tend to fail. Successive learned FISTA (L-FISTA) achieves a dynamic range of 99.5 dB for a 32×4 PMCW radar. Additionally, the methods are evaluated at various levels of sparsity, using range-Doppler maps (RD maps) in dense target scenarios. Finally, we compare the computational load of the presented methods using the floating-point operations (FLOPs) metric.