Direction-of-arrival (DOA) estimation in shallow-water underwater acoustics is challenged by coherent multipath and impulsive disturbances, which jointly cause covariance rank deficiency and outlier-driven subspace distortion. This paper proposes an embedded robust covariance-construction mechanism for coherent-plus-impulsive DOA estimation. The mechanism is implemented as mixture-correntropy-weighted simplified spatial smoothing (SS–MCC), in which snapshot reliability is enforced during subarray covariance accumulation rather than after decorrelation. A two-kernel residual-based weighting rule suppresses strongly contaminated snapshots while retaining moderately perturbed but informative snapshots. Under a controlled narrowband uniform linear array benchmark with fully coherent two-arrival multipath and Bernoulli–Gaussian impulsive noise, SS–MCC yields more stable DOA behavior than MUSIC, SS-MUSIC, and FLOM-MUSIC, especially in low-SNR, high-impulsiveness, and near-threshold regimes, although absolute strict recovery remains limited in the hardest cases. All-trial strict correct-two-peak statistics and ablation results show that the gain mainly comes from embedded covariance cleaning rather than post-processing or parameter tuning. A measured-noise-injected benchmark using NOAA–Navy SanctSound FK01 underwater recordings further confirms the same qualitative robustness trend under real noise waveforms, while remaining a semi-realistic noise-injection check rather than measured-array sea-trial validation. A simplified DOA- assisted MVDR benchmark indicates that improved covariance robustness can also support more favorable beamforming-oriented trends. The results provide controlled benchmark evidence that reliability-aware covariance construction can stabilize subspace extraction under joint coherent multipath and impulsive contamination; validation under wideband propagation, model mismatch, partial coherence, and measured array data remains future work.

Aiming at the problem that sparse channel estimation in underwater acoustic communication is susceptible to complex multipath propagation, non-Gaussian impulsive noise, and channel time variations, this paper proposes a bidirectional proportionate recursive maximum correntropy criterion algorithm, referred to as Bi-PRMCC. By introducing a bidirectional filtering structure into the proportionate recursive maximum correntropy criterion (PRMCC) framework, the proposed algorithm jointly exploits the information from forward and backward data sequences, thereby improving the estimation accuracy and block-based channel variation tracking capability for sparse underwater acoustic channels. Meanwhile, the maximum correntropy criterion enhances the robustness of the algorithm against non-Gaussian impulsive noise and outlier error samples, while the proportionate update mechanism improves its identification capability for dominant taps in sparse channels. To verify the effectiveness of the proposed algorithm, short-range sparse underwater acoustic channels and long-range complex multipath underwater acoustic channels are constructed based on the Bellhop ray-tracing model. Simulation experiments are then conducted under three typical non-Gaussian noise environments, namely Cauchy noise, $\alpha$-stable distribution noise, and Middleton noise. The experimental results show that, compared with recursive least squares (RLS), bidirectional recursive least squares (Bi-RLS), proportionate recursive least squares (PRLS), recursive maximum correntropy criterion (RMCC), and PRMCC, Bi-PRMCC achieves a lower steady-state normalized mean square deviation (NMSD) under different non-Gaussian noise conditions, indicating stronger robustness against impulsive noise. Under different signal-to-noise ratio conditions, the proposed algorithm still maintains superior steady-state estimation performance. In addition, in the channel abrupt-change tracking experiment, Bi-PRMCC can rapidly reconverge after channel variations occur, demonstrating favorable reconvergence capability under abrupt channel variations. The ablation study further verifies the stable performance gain brought by the bidirectional structure to PRMCC. Overall, the proposed Bi-PRMCC algorithm exhibits high estimation accuracy, robustness, and reconvergence capability under complex non-Gaussian noise and abrupt channel variation conditions.