DML-Based Single-Snapshot DOA Estimation and Ghost Suppression for Sparse MIMO Radar
Abstract
This article presents a computationally efficient framework for direction-of-arrival (DOA) estimation and ghost target suppression in single-snapshot sparse multiple-input multiple-output (MIMO) radar. A look-up table (LUT)-accelerated deterministic maximum likelihood (DML) estimator is proposed, incorporating a signal-to-noise ratio (SNR)-adaptive scenario classification, a null-topology-driven iterative refinement, and a parabolic interpolation of the DML cost that jointly enable sub-grid DOA accuracy without repeated real-time matrix computation. Range-Doppler cells containing more than two dominant targets can be flagged by the normalized DML residual at no additional cost. To address multipath-induced ghost targets, a reconstruction-based one-dimensional (1-D) direction-of-departure (DOD) evaluation is proposed, which decomposes the two-dimensional DOD-DOA search into independent 1-D searches guided by the estimated DOA, and resolves grating lobe ambiguities through joint signal reconstruction. Monte Carlo simulations under a 77 GHz automotive MIMO radar show that the DOA estimator achieves a root-mean-square error (RMSE) of 0.14° at 10 dB SNR, outperforming subspace-based and parametric baselines at all tested SNRs and angular separations, while achieving accuracy competitive with a sparse-recovery baseline. The reconstruction-based validation rejects a grating lobe exceeding the true mainlobe and achieves a ghost suppression success rate increasing from 62.0% at 10 dB to 94.2% at 40 dB, while the false rejection rate for direct-path targets decreases from 19.0% at 10 dB to 3.0% at 40 dB. With a DOA estimation cost comparable to greedy sparse recovery, the framework significantly reduces computational complexity compared to convex sparse recovery, making it viable for real-time automotive ECU deployment.