CNN-based Automatic Modulation Classification Over Underwater Acoustic Channels

Automatic modulation classification (AMC) aims to recognize modulation schemes from received communication signals. Such a task is especially challenging over underwater acoustic (UWA) channels due to their harsh conditions including long multipath, high Doppler effect, and so on. In recent years, deep learning for AMC has attracted increasing attentions for its powerful feature-extraction capability. In this paper, we explore the feasibility and performance of a convolutional neural network (CNN)-based AMC method over UWA channels. Three transmission modes: single-carrier (SC), orthogonal frequency-division multiplexing (OFDM), direct sequence spread spectrum (DSSS), are employed. The modulation schemes include four coherent modulations: BPSK, QPSK, 8PSK and 16QAM, and two non-coherent modulations: BFSK, QFSK. In total, fourteen classes of communication signals are considered for classification. It showed recognition of single-carrier coherent signal is more difficult than others and to improve the classification accuracy, two hard example mining mechanisms were adopted. Numerical simulations showed the proposed scheme achieves decent recognition performance.