Modulation Recognition of Underwater Acoustic Communication Signals Based on Deep Heterogeneous Network

Abstract

The complicated underwater environment and its variable characteristics bring certain challenges to the modulation recognition of acoustic communication signals. Traditional methods for the task are usually based on manually extracted features, which require sufficient prior knowledge and artificial cost. In this paper, considering the rapid development of machine learning technology, the original underwater acoustic signal data are compressed by PCA technique, so as to reduce the data dimension and suppress the noise interference. On the basis, a deep heterogeneous network combining hybrid dilated convolutional networks and Long-Short Term Memory network is built to automatically capture the hidden features of data series to achieve the modulation recognition of 4 underwater acoustic communication signals recognition, including OOK, 2FSK, 2PSK and QPSK. Under different SNRs, the simulation experimental results show that the proposed network is valid and robust to identify 4 modulation modes. In the actual experiment, recognition accuracy of 91.171% confirms the effectiveness of the proposed network for modulation classification.