The traditional Direction-of-Arrival (DOA) estimation methods include Fast Fourier Transform beamforming, Multiple Signal Classification (MUSIC), Compressed Sensing or Orthogonal Matching Pursuit (CS/OMP), etc., which have varying performance characteristics in terms of computation speed, spectral resolution and sparse recovery. They have not yet achieved the optimal state of a good compromise between estimation accuracy, computation cost and real-time performance in complicated mmWave radar scenarios. DOA estimation at this stage in radar-based target identification, tracking, and recognitions; Its accuracy directly impacts the subsequent processing steps of these functions. To solve the above problem, this paper presents an improved lightweight multiple Signal Classification algorithm (L-MUSIC). In light of the trade-offs often suffered by methods such as FFT, MUSIC and CS/OOMP, L-MUSIC introduces the MUSIC subspace-processing chain into its basic realization structure and alleviates computational intensity problems caused by covariances, eigenvalues, full-angles spectra calculations via recursive covariance estimation; adaptive dimensional reduction techniques (such as SVD), rapid update method for subspaces, two-stage angle search strategies. Candidate-angle filtering and local spectral evaluation further reduce the load on a full-angular domain spectrum search. To suppress the sub-space degradation due to multiple paths propagation and coherent interference, and to reduce the performance drop caused by model discrepancy in the array-based system, we propose a novel strategy that adds diagonal-loading pre-processing and forward-backward spatial-smoothing post-processing techniques. Experiments show that L-MUSIC has a larger concentration of its primary lobe, an increased low-frequency null with respect to untargeted angles, stronger sidelobes and spurs peak suppression performance. The estimating error of the proposed method is lower than that for other comparative schemes at all tested SNR levels; Compared with these schemes’ identifying success rate increases further under high SNR scenarios. At Close-angle Dual-Target, Low-Snapshot, Coherent-Source Conditions, L-MUSIC remains very discriminative; also shows high Resolution Performance and good Stability at this time. Computational cost-wise, compared with traditional MUSIC and CS/OMP, L-MUSIC’s average execution time is relatively short, yet it still has excellent performance in arrays’ position deviations. These results indicate that L-MUSIC provides a lightweight high-resolution solution to the common trade-off among estimation performance, computational efficiency, and robustness in FFT, MUSIC, and CS/OMP methods, and therefore ofA feasible way to estimate the distance continuously based on Continuous-Doppler-Attenuation.fers an effective approach for real-time DOA estimation in complex scenarios.
