Sparsity-driven minimum kernel risk-sensitive loss criterion-based adaptive line enhancer under non-gaussian noise

The tonal components within the noise radiated from surface or underwater vehicles are significant characteristics for passive sonar detection. Adaptive line enhancer (ALE) is widely utilized as a preprocessor in passive sonar systems to enhance the tonals, facilitating subsequent tonal detection. The conventional ALE (CALE), which relies on the second-order statistics, performs well under Gaussian noise conditions. However, the occurrence of impulses and outliers in the background noise, which arise from diverse natural events and human-made interferences, introduces non-Gaussian characteristics. This deviation from Gaussian noise results in diminished performance of CALE. To address this issue, this paper proposes an ALE based on the minimum kernel risk-sensitive loss (MKRSL) criterion and the frequency-domain sparsity of tonals. The KRSL serves as a metric for assessing the similarity between tonals and noise. By adopting the MKRSL criterion, the ALE effectively suppresses impulses and outliers. Additionally, incorporating a sparsity-based penalty into the adaptation process further improves the signal-to-noise ratio (SNR) gain by reducing weight noise. Both simulations and sea trial data processing results demonstrate the effectiveness of the proposed ALE under non-Gaussian noise conditions.