Abstract
A multi-stage adaptive filter has been reported in a previous study as an advanced adaptive noise cancellation system for ultrasonic non-destructive testing applications. However, the multi-stage adaptive filter is limited by the slow convergence speed of its least-mean-squares (LMS) algorithm, as well as additional implementation processes and increased computational complexity. This work proposes to overcome these limitations by initialising the second-stage filter with coefficients obtained from the first stage and by using a step-size estimator for the first-stage filter. Furthermore, a sign LMS filter is proposed to replace the conventional LMS algorithm in the filter to reduce the computational complexity. Flaw echo signals masked by grain noise, obtained from a stainless steel block with a side-drilled hole and from two welded plates with a root crack and a slag inclusion, were used to evaluate the performance of the filters. The results showed improvements in convergence speed and signalto- noise ratio. The sign LMS algorithm was successfully implemented in the proposed multi-stage filter, offering a less computationally complex, alternative filter.