Frame-Based Time-Scale Filters for Underwater Acoustic Noise Reduction

Noise reduction for underwater acoustic signals has attracted considerable attention over the last few decades. Among the numerous techniques, wavelet soft-thresholding (STH) has been considered as one of the most effective noise reduction approaches, as it achieves near complete success in minimizing the mean-squared-error (MSE) and eliminating oscillations caused by noise. However, a limitation with STH is its preference towards lower frequency bands, which may cause distortions in the high frequency bands. Few previous research efforts have reported on the reduction of such frequency distortions. By introducing the time-scale filters (TSF), we present a novel technique for underwater noise reduction that improves the standard STH in reducing distortions in the joint time-frequency (TF) space. TSF is an advanced noise reduction algorithm which utilizes the signal's time-scale (TS) support region. It provides smooth reconstructions in both time and frequency spaces. We demonstrate the noise reduction results for two typical underwater noise sources: the snapping shrimp sound and the rainfall sound. We also introduce a TF distortion measurement as a criterion that compares the TF distributions of the denoised signal and the clean signal. For a signal-to-noise ratio (SNR) from -10 to 20 dB, the noise reduction results obtained using TSF have an average of 42.1% lower TF distortion than STH for the snapping shrimp noise, and a 23.3% lower TF distortion for the rainfall noise.