Deep Unfolding for Sparse Distance Recovery in PMCW MIMO Automotive Radar

Phase-Modulated Continuous Wave (PMCW) radars have attracted significant attention due to advances in mm-wave technology, waveform design, and digital signal processing. The de facto technique for distance estimation in PMCW radar receivers is matched filtering with a bank of correlators. There is an inherent trade-off between support for multiple antennas (MIMO support), sequences with good correlation properties and the maximum achievable unambiguous range/velocity. An important challenge in PMCW MIMO radars is to design receivers that result in low sidelobe levels in the range domain, for a given choice of sequences. In this paper, we propose a novel range processing scheme by formulating an optimization problem with ℓ1-norm regularization that promotes sparse distance estimates. To solve this, we propose deep unfolded FISTA and ADMM algorithms for distance sidelobe suppression and restoration of the orthogonality of the transmitted codewords. We show that the proposed method achieves better dynamic range compared to the traditional matched filtering approach.