Underwater Communication Signals’ Modulation Recognition Based on Sparse Autoencoding Network

This paper studies the method of underwater communication signals’ recognition based on sparse autoencoding network. Firstly, the power spectrums of the received signals are classified using a sparse autoencoding network, getting modulation types of signals except of PSK. Then make the quartic spectrums of the signals whose recognition results are PSK in the first step. Finally, the classification of QPSK and 8PSK’s quartic spectrums is completed by using another sparse autoencoding network. Simulation experiments show that the sparse autoencoding network can automatically extract effective spectrum features from the received signals’ spectrum information. Compared with the traditional recognition method based on power spectrum feature extraction, the proposed algorithm reduces the feature extraction process based on domain knowledge, and the recognition performance is better than the traditional algorithm.