Block-Sparse Learning Enabled Approach Towards Efficient Channel Estimation for Underwater Visible Light Communications

The sophisticated underwater environment makes it difficult to implement reliable and robust channel estimation for underwater visible light communication (UVLC). It has been observed that the UVLC channel has a block-sparse structure in the time domain, making it possible to use block-sparse recovery methods. To this end, this paper combines sparse learning theory to establish a block-sparse learning based underwater visible light channel estimation (SBL-UVCE) scheme. Specifically, classic unfold convex optimization based approximate message passing (AMP) algorithm into different network layers of a deep-unfolding neural network. Concurrently, in order to promote the network to better extract and utilize the block-sparse structure channel of UVLC, a Gaussian mixture denoiser based on minimum mean square error is introduced Simulation results have verified that the proposed scheme is significance better the existing compressed-sensing based schemes and the state-of-the-art sparse learning based scheme, especially under the circumstances of low signal-to-noise ratio (SNR) and insufficient number of observation pilots.