Quasi Frequency Diversity Processing of Ultrasonic Signals — A Review

Abstract

Signal processing techniques capable of improving signal to noise ratio in the presence of strong backscattered grain echoes interfering with the echo reflected from material defects are considered in this paper. A number of algorithms belonging to a class of Quasi Frequency Diversity (QFD) techniques are presented and compared. The best known method in this class is the Split Spectrum Processing (SSP), which has been proven successful in suppressing material noise. The SSP is based on the idea of expanding the received ultrasonic signal into a set of split signals corresponding to a number of narrow frequency bands. However, performance of the SSP is highly dependent upon number of split signals, their center frequencies and bandwidths, that may be difficult to estimate properly. Robust and adaptive QFD techniques are proposed as a solution to this problem. Two robust QFD techniques based on removal of narrow frequency bands from the ultrasonic signal spectrum are presented in the paper. The first one, referred to as Cut Spectrum Processing (CSP), does not require any spectral parameters at the price of an increased computational burden. The resultant CSP algorithm includes one single parameter, which can be easily tuned interactively by the operator. The second technique, referred to as Fragment Spectrum Processing (FSP) is based on an idea of fragmentation of the signal spectrum by using a set of irregular comb filters, i.e., filters having a frequency response consisting of a large number of alternating stop- and pass-bands. Efficiency is accomplished by using a set of the comb filters with virtually different (possibly orthogonal) frequency responses. The adaptive QFD technique presented here is based on the fact that the polarity thresholding SSP can be formulated as a multilayer perceptron artificial neural network operating in feedforward mode with binary neurons and bipolar input signals.