Application of wavelet packet transform to signal recognition

A novel approach called neural network recognition approach based on wavelet packet transform and resemblance coefficient (NN-WPTRC) is proposed to recognize radar emitter signals with different intra-pulse modulations and plenty of noise. First of all, wavelet packet transform (WPT) is introduced to extract features from radar emitter signals. The principles of WPT and feature extraction algorithm of radar emitter signals are described in detail. Because the dimension of feature vector obtained from WPT is too high, a novel feature selection approach called resemblance coefficient (RC) method is presented subsequently. Definition and properties of RC are discussed and RC feature selection algorithm is introduced. Thirdly, neural network classifiers are designed to fulfill automatic recognition of radar emitter signals. Finally, 9 typical radar emitter signals are chosen to make simulation experiment to verify the effectiveness and feasibility of the proposed approach. 16 valid features are extracted from each of 9 radar emitter signals using WPT and the most important 2 features are selected from 16-dimension feature vector using resemblance coefficient feature selection approach. Experimental results show that NN-WPTRC has good capability of noise suppression and accurate recognition rate is up to 98.31%, which is much higher than that of sequential feature selection based on distance criterion function.