Underwater acoustic multipath sparse channel estimation via gridless relevance vector machine method

In the scenario of underwater acoustic sparse channel estimation with training sequences, grid points in the measuring matrix are caused by discretizing procedure. Estimating accuracy might not be guaranteed with state-of-the-art methods when multipath delays don't exactly locate on the grid points. In this paper, we construct a gridless measuring matrix for sparse channel estimation which contains a off-grid adjusting factor and further using Relevance Vector Machine algorithm to estimate this factor to estimate the offset. This paper first describes the approach and then testifies its estimating result in numerical experiments on two different underwater channels. The results demonstrate that this method outperforms conventional ones in estimating error and bit error rate and this is even more obvious when the grid gets coarser.