Application of generalized adaptive wavelet neural network based on resemblance coefficient to signal recognition

A novel approach called generalized adaptive wavelet neural network based on resemblance coefficient (GAWNN-RC) is proposed to extract the most discriminatory information from radar emitter signals and to recognize different radar emitter signals with different intra-pulse modulation laws In this work. Because a weighted linear combination of dilated and translated replica of a mother wavelet can approximate a radar emitter signal in frequency domain and the weights, dilation and translation parameters can be estimated adaptively to minimize an approximation error in a least mean square sense, generalized adaptive wavelets (GAW) can be used to extract the most important features of radar emitter signals. Aiming at the characteristics of radar emitter signals, GAW is introduced to extract features for the first time. But there are three problems in GAW representation: (1) it is difficult to obtain the globally optimal solution using gradient descent algorithm (GDA) because GDA is very sensitive to initial values and drops into sub-optimum easily; (2) there are many difficulties in classifier design because the dimension of feature vector is too large; (3) multiple solutions exist in GAW representation, which is not considered in existing methods. So quantum evolutionary algorithm (QEA) and resemblance coefficient method are proposed to solve the problems effectively. Moreover, QEA and neural network (NN) are combined to design classifiers. Finally, 6 typical radar emitter signals are chosen to make the experiment of feature extraction and recognition. Experimental results show that high accurate recognition rate can be achieved and unknown radar emitter signals are also distinguished accurately, which indicates that the introduced approach is feasible and effective.