Study On Signal Automation Identification System Of L04028A-044 Loudspeaker

Signal is the carrier of information, is a bridge to transmit and exchange information. Signal detection is an important way to access to information. Wavelet analysis, neural network and Support Vector Machine(SVM) are effective signal detection methods.This paper introduces the theoretical basis of wavelet analysis. A method is presented to extract frequency band energy feature by using wavelet package decomposition and reconstruction technique. The experimental result shows the validity of this method in L04028 A-044 loudspeaker signal processing. That application of wavelet package analysis of signal pre-processing provides good sample vector for signal decetion using in neural network and SVM.The loudspeaker's band energy are extracted based on the theory of wavelet packet, which provide input vectors for neural network fault diagnosis, while the loudspeaker's categories are as output vectors. Then establish the BP network model of L04028A-044 loudspeaker's defects recognition. Compared different numbers of implied layer notes that influence on the network performance to specify the number of implied layer note. Determined the learning rate and transfer functions by empirical. By comparing the errors caused by different goals of network to determine the training goal of BP neural network. By comparing the error curves caused by different training functions of the network to determine the training function of neural network model. Training, simulating and testing the BP neural network by using the normalized eigenvectors. Identifying the network model and classification accuracy by analysing the results of the loudspeaker realization.SVM is a relatively new theory in the field of signal detection. This paper presents the basic theory of SVM, the method and steps of pattern recognition. It describes the specific process of using SVM to classify L04028A-044 loudspeakers, which including of making a choice of kernel function and its parameters. Compared the results caused by neural network and SVM shows that neural network is more suitable for identification of L04028A-044 loudspeaker.