Modulation Pattern Recognition of Non-cooperative Underwater Acoustic Communication Signals Based on LSTM Network

Modulation pattern recognition of non-cooperative underwater acoustic communication signals is an important part of underwater acoustic confrontation. Machine learning can be a good option to automatically realize modulation pattern recognition. Long-short-time memory(LSTM) network based on instantaneous features of communication signal is studied in this paper. It is shown that this method can achieve considerable classification results using the experimental data from four trials. The recognition accuracy can exceed 80%. Some simulations are also carried out in this paper which indicates that the recognition accuracy can exceed 90% with the signal-to-noise ratio higher than −10dB.