Generative Model-Based Underwater Acoustic Channel Modeling with Partial Oceanic Environmental Parameters

In this paper, a generative model-based under-water acoustic (UWA) channel modeling method with partial oceanic environmental parameters is proposed. This method integrates the outputs produced by conventional channel models as priors to solving the dilemma posed by traditional UWA channel modeling tasks: difficulties obtaining comprehensive parameters from experimental observations due to the poor understanding of detailed underlying physics. Generative adversa rial networks (GANs) capture the distribution of UWA signals received within a period. Well-trained GANs can generate channel impulse responses (CIRs) that resemble real CIRs. The experimental results, supported by field data acquired from shallow water environments, indicate that the proposed method is able to model UWA channels and possesses the ability to generalize to small-scale temporal variations within a period. Specifically, the deep convolutional Wasser-stein generative adversarial network with a gradient penalty (DCWGAN-GP) significantly improves upon the performance of the traditional channel model, and the mean absolute error (MAE) of the CIRs is enhanced by 57.3 % under a SNR=10 dB.