Objective To address the degradation of modulation features and recognition performance in underwater acoustic communication signals caused by random multipath propagation and noise interference, this study proposes a modulation recognition method incorporating denoising and feature enhancement.
Method The sparrow search algorithm (SSA) is employed to adaptively optimize the parameters of variational mode decomposition (VMD). The optimized VMD is then combined with wavelet thresholding to denoise and reconstruct underwater acoustic communication signals. Continuous wavelet transform (CWT) is subsequently applied to extract time-frequency features, which are fed into ShuffleNetV2 for initial six-class modulation classification. To resolve confusion between BPSK and QPSK, a second-harmonic feature is further extracted from the power spectrum of the squared signal. This feature is then used by a random forest classifier optimized using particle swarm optimization (PSO-RF) to provide supplementary discrimination between the two modulation schemes.
Results Simulation results show that recognition accuracy improves progressively from 75.60% for raw noisy signals classified using CWT and ShuffleNetV2 to 86.07% after SSA-optimized VMD and wavelet-threshold denoising, and further to 91.80% with the addition of the second-harmonic feature and PSO-RF-assisted discrimination. In offline experiments using field-collected data from Qiandao Lake, the overall recognition accuracy increases from 88.27% for the raw signals to 92.67% after denoising, and further to 95.60% after second-harmonic feature-assisted discrimination.
Conclusion The proposed method improves the discriminability of modulation features under random multipath propagation and noise interference, thereby enhancing overall modulation recognition performance. The second-harmonic feature provides complementary discriminative information for distinguishing between phase-modulated signals that are otherwise easily confused.
DownLoad: