Adaptive sparse representations of ultrasonic signals for acoustic microimaging

Acoustic microimaging (AMI) is a common nondestructive tool for failure analysis of microelectronic packages. Accurate estimation of the reflected ultrasonicechoes is essential for detection and location of defects inside the microelectronic packages. In this paper, an advanced AMI technique based on adaptive sparse representations is proposed to estimate the ultrasonicechoes and recover the reflectivity function. An adapted overcomplete dictionary capable of concise expression of ultrasonic signals is first learned by the focal underdetermined system solver-based column normalized dictionary learning algorithm. The ultrasonic A-scans generated by an AMI system are then decomposed into adaptive sparse representations over the learned dictionary using a sparse basis selection algorithm. Echo selection and echo estimation are further performed from the resulting adaptive sparse representations. The proposed technique offers a solution to the blind source separation problem for restoration of the reflectivity function and can separate closely spaced overlapping echoes beyond the resolution of the AMI system. Experimental verifications are carried out using both synthetic and measured data. The results show the proposed technique produces high resolution and accurate estimates for ultrasonicechoes.