Applying the Whitening Transformation to Echo Signals for Reducing Pattern Noise in Ultrasonic Testing

Abstract

In the ultrasonic testing of industrial facilities, it is often necessary to deal with metals having an increased level of pattern noise formed by a multitude of pulses rescattered by the structural inhomogeneities of the material. Extracting a useful signal from the pattern noise background is a nontrivial task. The fact is that the spectra and spatial distribution of pattern noise pulses and of the echo signal due to a reflector are close to each other, thus making the common methods of reducing the level of interference inapplicable. Therefore, special methods have been developed to reduce the level of pattern noise. One of such methods is the whitening (decorrelation) transformation of echo signals, originally developed for radar. This technique consists in converting correlated pattern noise pulses into white noise, which is well suppressed by conventional coherent methods such as C-SAFT, for reconstructing reflector images. In numerical experiments, echoes with pattern noise were calculated using the method of finite differences in a time domain (FDTD) when forming domains according to the rule of constructing Voronoi diagrams. As an alternative, pattern noise was calculated taking account of multiple rescattering of the probing pulse by point reflectors. It is shown that pattern noise reduction is possible when the pattern-noise correlation matrix is known with an accuracy of at least $\approx$10%. In model experiments, the method of whitening transformation was successfully applied to eliminate reverberation noise.