Blind Detection of Underwater Acoustic Communication Signals Based on Deep Learning

Abstract

Blind detection of underwater acoustic communication (UWAC) signals is challenging in non-cooperative reception scenarios. Difficulties include but not limited to complex underwater acoustic channels, diversity of signal categories, and data scarcity. To address these problems, we propose a novel blind detection method for UWAC signals based on deep learning (DL). First, an impulsive noise preprocessor and a signal denoising generative adversarial network are built to mitigate the noise in the received signals. Second, a convolutional neural network-based binary classification network is built to automatically extract features and distinguish between the UWAC signals and noise. Moreover, a transfer data model is presented to overcome the insufficient data issue in the target water regions. The results of simulation experiments and practical signal tests both demonstrate that the proposed method is robust to ambient noise with wide dynamic ranges and complex distributions. Our approach significantly outperforms conventional algorithms and existing DL-based algorithms at low signal-to-noise ratios, while requiring no prior information about the testing channel.

Keywords

• Underwater acoustic communication signals
• blind detection
• generative adversarial network
• convolutional neural network
• noise reduction
• transfer learning