查看文章

Ultrasonic testing signals are sometimes hard to interpret because of low signal-to-noise ratio (SNR) or overlapping echoes. The combination of Wiener filtering and autoregressive spectral extrapolation has proven to be capable of improving the SNR and time resolution. However, these signal processing techniques are not sufficiently robust to be used in industrial non-destructive testing applications. This is mostly due to arbitrary manner of selection of the signal processing parameters associated with these techniques. In this paper, a new approach, which eliminates the need for arbitrary assignment of some of the parameters is described. This new approach is more robust and suitable for practical applications, and is demonstrated by considering both simulated and experimental signals.