Sparse deconvolution of ultrasonic NDE traces-a preliminary study

The paper deals with the problem of deconvolving sparse ultrasonic NDE traces with time-varying pulses. A sparse dictionary learning algorithm is utilized to learn a time-varying pulse matrix. Each element of the matrix represents an individual pattern of possible local impulse responses. An ultrasonic signal is then decomposed into a sparse representation by the sparse Bayesian learning algorithm over the learned pulse matrix. The reflectivity sequence is finally estimated from the resulting sparse representation. The proposed method has been tested using NDE data taken from automotive thick film hybrid circuit boards.