AN EFFICIENT SPARSE SIGNAL DECOMPOSITION TECHNIQUE FOR ULTRASONIC SIGNAL ANALYSIS USING ENVELOPE AND INSTANTANEOUS PHASE

Sparse signal decomposition techniques have been widely used in recent years due to their efficiency in ultrasonic signal analysis. These techniques iteratively decompose ultrasonic signal in terms of model echoes (e.g., Gaussian echo, chirplet echo, etc.) that characterize local signal structures. The decomposed echoes (or the parameters) are then used for subsequent analysis, for example, for feature extraction and system identification. The first critical step in these decomposition techniques is the partitioning (i.e., windowing) of the ultrasonic data for identification of dominant signal features. This step has a great implication on the subsequent step that involves parameter estimation based on the assumed echo model or finding the best matched echo from a predefined dictionary of echoes. Therefore, a robust windowing technique that successively partitions ultrasonic data into dominant echo components is highly desirable. In this study, we obtain envelope and instantaneous phase via analytic signal representation to guide ultrasonic data partitioning. This type of partitioning is also meant to serve the initial guessing operation prior to the parameter estimation. Envelope and instantaneous phase provide important clues for local changes in the ultrasonic signal. The local maxima of the smooth envelope along with the changes in the instantaneous phase provide meaningful boundaries for echo structures. These boundaries are expected to provide an accurate data partition for the subsequent echo estimation. We present results that demonstrate the proposed echo windowing technique embedded with a model-based echo estimation is significantly faster than the Time-Frequency (TF) based echo localization techniques and provide meaningful echo localization and parameter estimation results. In particular, we test the algorithm in ultrasonic flaw detection using backscattered echoes from a steel sample that contains flaw echoes buried in the clutter (SNR is about 0 dB), and for the analysis of closely spaced reverberation echoes measured from a multi-layer test specimen.