Sparsity-Aware Channel Estimation for Underwater Acoustic Wireless Networks: A Generative Adversarial Network Enabled Approach

In order to effectively deal with the performance bottlenecks faced by underwater acoustic channel estimation, especially under harsh conditions with sophisticated background noise and insufficient spectrum resources. This paper proposed an UAC estimation method based on sparsity-aware Generative Adversarial Network in a compressed sensing framework. This method exploits the strong learning ability of the channel generator network (CGN) and establishes an explicit mapping relationship in the sample data distribution through adversarial training manner, thereby directly learning the sparse characteristics of the channel impulse response (CIR) of the UAC. Moreover, by introducing a regularized term to the traditional loss function of SA-GAN, a compound loss function is designed in this method, which aims to learn the characteristics of the channel. Simulation results show that, compared with the up-to-date UAC estimation methods, the proposed method significantly improves the channel estimation accuracy and spectral efficiency.