Modulation classification method based on multi-feature fusion for underwater acoustic communication signals

Feature extraction of communication signal is easily affected by complex underwater acoustic interferences such as high noise and multipath effect, which lead to low classification rate. To solve this problem, a modulation classification method based on multi-feature fusion for underwater acoustic communication signals is proposed. First, the multi-feature fusion method is designed to obtain signal features with stronger noise immunity, including wavelet time-frequency spectrum and two stronger spectral line features from square power spectrum and autocorrelation spectrum. Secondly, based on transfer learning theory, a lightweight network model is constructed to classify 2FSK and 4FSK signals using time-frequency features. Finally, combined with support vector machine optimized by particle swarm algorithm, an interclass recognition for non-frequency shift keying signals (including BPSK, QPSK, DSSS, OFDM signals) is designed using two stronger spectral line features. The results of simulation experiments demonstrate that the proposed method has good generalization abilities and noise immunity against underwater acoustic channels. The sea trial data verifies that the proposed method outperforms common neural network models in the classification rate.