Noise Analysis and Properties of Local Discriminant Basis Algorithm

K. Hazaveh and K. Raahemifar (Canada)


Local discriminant basis, Best basis, Time-frequency analysis, Non-stationary signal analysis, Wavelet packet, Local trigonometric transform, Gradient decent algorithm.


Local discriminant (LDB) basis algorithm is a supervised scheme for feature extraction and non-stationary signal classification. Due to its fast computational time, )log( nnO , and excellent time-frequency localization, it is a promising method for non-stationary signal analysis. An optimized version of local discriminant basis has been recently proposed that emphasizes certain regions of interest in different classes. From an engineering perspective some of the most important issues of such an algorithm are noise analysis and the effects of variable number of training signals, optimized LDB vectors and LDB features fed into classifier. In this paper a noise analysis is performed to study the behavior of a synthetic classification problem by using both original and optimized versions of local discriminant basis algorithm. Each experiment is performed with different SNR, different number of training signals and different number of optimized and classified LDB vectors to study these essential properties together. The results reveal interesting non-linear response to noise level and other parameters studied.

Important Links:

Go Back