UACC-GAN: A Stochastic Channel Simulator for Underwater Acoustic Communication

Abstract

Due to the high cost of sea trials and the variability of sea states, the duration of experiments is usually too short to fully verify underwater acoustic communication (UAC) performance in a real-ocean environment. Moreover, traditional UAC channel simulators also face issues of inaccurate environmental parameters or mismatched statistical models. To tackle these challenges, we propose UACC-GAN, a data-driven stochastic channel simulator for UAC, offering an innovative solution for channel data augmentation. UACC-GAN uses the generative adversarial network model to learn the latent space of the measured channel data set and then maps the random sampling points in this space into a new time-varying impulse response (TVIR). Our simulator is validated using the small-scale WATERMARK data set collected at sea. The results indicate that the generated TVIR has a realistic delay–Doppler spread and can reproduce time-varying delay path characteristics. The cumulative distribution of multiple 0-D properties also proves the realism of the entire distribution of the generated channel data set. In addition, by relying on the continuity of the latent space, UACC-GAN generates channel characteristics with random fluctuations, such as Doppler spectrum shape, delay energy distribution, and tap covariance, which contributes to more diverse communication testing conditions. Finally, we pass frequency-hopping spread spectrum and orthogonal frequency-division multiplexing communication signals through the generated and measured channels. The comparable results of simulated bit error rate (BER) and actual BER underline the value of the UACC-GAN simulator for communication system design and testing.