Wasserstein GAN Based Underwater Acoustic Channel Simulator

Mingzhang Zhou,Junfeng Wang,Haixin Sun
DOI: https://doi.org/10.1109/icspcc59353.2023.10400328
2023-01-01
Abstract:Underwater acoustic channel is a crucial part for implementation of underwater communications, and its measurement is expensive. This leads to the lack of samples for the training of deep neural network (DNN)-based underwater communication receivers. To solve this problem, A Wasserstein generative adversarial network (WGAN)-based underwater acoustic channel simulator is proposed in this paper for channel sample augmentation. The overlap between the distributions of the measured channel and generator output is analyzed. Then the CNN-based WGAN is constructed with the earth-mover's distance as part of the loss function. Tested with the measured channels in Wuyuanwan Bay, Xiamen, the proposed WGAN performs steadily in simulating the complete channel impulse responses and their single taps distributions. Moreover, a simple DNN-based OFDM channel estimator is built and completely trained with the generated channels. The simulation results show that the BER DNN-based channel estimator outperforms the LS and MMSE estimators.
What problem does this paper attempt to address?