Recognition method of non-cooperation underwater communication signals using hierarchical semantic encoder transformer network

In order to solve the problem of automatic detection and modulation recognition of underwater acoustic communication signals under non-cooperative conditions, this paper proposes a non-cooperative underwater acoustic communication signal detection and modulation recognition method based on a deep learning network. This deep learning network is based on a hierarchical transformer structure and mask cyclic shifted window attention mechanism, which does not require recursion. This deep learning network overcomes the limitations of underwater acoustic sample size and computational resources. It has the ability to autonomously locate sound events, extract and identify signal features without prior signal information. The experimental results of identifying 2PSK, 2FSK, 4FSK, 4PSK, DSSS, and OFDM signals simulated has precisions with 92.4%. The recognition rate of three types of underwater acoustic communication signals, namely 2FSK, 4FSK, and 2PSK signals, measured on the lake is 96.1%. Therefore, the experimental results show the effectiveness of this method.