Investigation of Spectral-Based Techniques for Classification of WidebandTransient Signals in Additive White Gaussian Noise

摘要

Spectral-based classification schemes designed to separate various wide bandtransient signals in added noise have been identified and their performances compared along with those obtained using a back-propagation neural network implementation. The spectral-based measures used include: the normalized cross-correlation coefficient; the modified normalized cross-correlation coefficient, and; the divergence and the Bhattacharyya distance. Noise was added to the signals to create signal to noise ratios of 0 dB to -20 dB. Results show that as noise levels increase, the modified normalized cross-correlation coefficient spectral measure remains the most robust scheme.