Sparse Analysis Model Based Dictionary Learning for Signal Declipping

Clipping is a common type of distortion in which the amplitude signal truncated if it exceeds certain threshold. Sparse representation has underpinned several algorithms developed recently for reconstruction original from clipped observations. However, these declipping are often built on synthesis model, where represented by dictionary weighted sparse coding coefficients. In contrast to works, we propose analysis-model-based (SAD) method, model formulated an analysis (i.e. transform) dictionary, and additional constraints characterizing clipping process. The updated using Analysis SimCO algorithm, recovered least-squares based method or projected gradient descent incorporating observable set. Numerical experiments speech music used demonstrate improved performance ratio (SDR) compared recent state-of-the-art methods including A-SPADE ConsDL.