A digital-signal-processing technique for ultrasonic signal modeling and classification

In this paper, we face the problem of constructing a robust feature space for automatic classification of signals fi om narrow-band in-air ultrasonic sensors. In consideration of the existing sensor bandwidth restrictions, the importance of selecting a suitable signal descriptor is highlighted. We assume that the characteristics of the ultrasonic sources which produce the signals are impressed in the shape of their echo envelopes. A technique based an orthonormal Laguerre polynomials is applied to the echo envelopes for constructing the feature space. Different methods for computing the Laguerre coefficients are discussed, and the properties of the resulting feature space are investigated. For the experimental verification of the method, a set of acoustic sources is synthesized by submitting a high-frequency piezoelectric transducer to varying levels of electrical damping. How some factors, i.e,, the signal-to-noise ratio (SNR) of the return signals, and the sampling rate to digitize them; affect the achievable recognition rate is discussed. High recognition rates are obtained in our experiments, in spite of the Bet that, by visual inspection, the shapes of the signals from the synthesized sources are very similar to one another.