A construction method of reverberation suppression filter using an end-to-end network

Area of Science:

• Signal Processing
• Underwater Acoustics
• Machine Learning

Background:

• Reverberation is a major interference in active sonar systems, particularly in shallow waters, degrading target detection.
• Reverberation suppression techniques are crucial for enhancing underwater echo signal clarity and accuracy.

Purpose of the Study:

• To propose an end-to-end network, the Reverberation Suppression Network (RS-U-Net), for effective suppression of reverberation in underwater sonar echo signals.
• To improve the signal-to-reverberation ratio (SRR) and enhance target detection accuracy in shallow water environments.

Main Methods:

• An end-to-end network architecture, RS-U-Net, was developed using a one-dimensional convolutional network (1D-CNN) for feature extraction.
• The network utilizes sonar echo signal data as input and employs an auto-encoder for initializing weights to optimize training time and performance.

Main Results:

• The RS-U-Net demonstrated effective suppression of reverberation, improving the signal-to-reverberation ratio (SRR) of echo signals.
• Experimental validation showed an improvement in echo signal detection by approximately 10 dB, outperforming existing methods.
• The proposed method achieved approximately 2 dB better reverberation suppression compared to other leading techniques.

Conclusions:

• The RS-U-Net offers a significant advancement in reverberation suppression for active sonar systems in shallow waters.
• The network architecture and training strategy effectively enhance the performance and accuracy of underwater target detection.