An improved Wigner distribution based algorithm for signal identification

A new modified Wigner distribution, in which the cumbersome cross-terms can be totally eliminated, is combined with the wavelet packet decomposition to get a better time-frequency analysis algorithm. The non-stationary signal are first decomposed into approximation and detail signals by the wavelet packet analysis, and then each constituent signal is analyzed by the modified distribution to get a time-frequency representation data matrix. To simplify the computation for identification, the Karhunen-Loe`ve transform is used to find the principal eigenvector of the distribution data matrix. The principal eigenvectors of the approximation and detail signals construct the signal feature database for identification. In addition to numerical simulations, experiments of some underwater acoustic signal identification are conducted.