Statistical Analysis and Computer Generation of Spatially Correlated Acoustic Noise

Abstract

In ultrasonic NDE, simulation studies can play an important role in complimenting experimental validation of techniques under development. The utility of such simulations depends, in part, on the degree to which the simulated defect and noise signals are representative of the measured signals. In this paper, we describe an approach for generating simulated acoustic noise with a spatial correlation coefficient distribution and maximum extreme value (MEV) distribution which matches those distributions for measured acoustic noise. The procedure for generating noise signals is outlined for a line scan and for a raster scan. The basic approach forces the correlation of neighboring signals to the desired correlation by creating each signal as the sum of appropriately scaled neighboring signals plus a new random signal. For the line scan where each interior position has only two neighbors, this process is done sequentially without iteration. For the raster scan where each interior point has four nearest neighbors, iteration is required to simultaneously achieve the desired correlations with row and column neighbors. The MEV distribution is controlled in an outer iterative loop with the shape and position of the distribution dictated by spectral content of the noise signals and by controlling the signal energy, respectively. Results are shown which demonstrate the effectiveness of the approach. With this approach, a limited number of measured signals can be used to establish the correlation coefficient and MEV distributions which drive the computer generation of a large number of simulated acoustic noise signals.