Improving Sonar Reconstruction with Sparse Reconstruction Techniques

Abstract

In many sonar applications, traditional processing can fail to provide adequate information to accomplish the respective goals, because the processed information is either strictly bandlimited to the transmit band, distorted (matched filtering places higher emphasis on larger frequency components), or subject to large error due to noise. By using sparse reconstruction techniques on systems using bandpass chirp transmit signals, we can extrapolate additional bandwidth beyond the transmit signal bandwidth without directly amplifying the noise. This approach could help significantly in areas where more than locational information is desired, such as in target classification. We approach this problem with a basic one-dimensional sparse reconstruction algorithm, extend this to blind deconvolution, and then extend this again by addressing a spatially-varying reconstruction problem.