Application Of Mutual Information Theory To Hilbert-Huang Transform

The massive experiments indicated the pronunciation signal is one kind of non-linearity, the non-steady signal. This causes based on the linearity, the steady traditional pronunciation signal processing method is unable in the performance to obtain further enhancement, therefore, studying on processing method of the advanced non-steady signal has the significant significance to the pronunciation signal processing research. The Hilbert-hang transformation is one new time-frequency analysis method which Chinese American N.E.Huang proposed, it is a major breakthrough on linear spectrum and a stable state based on Fourier transformation as the foundation and Hilbert-huang transformation has widespread application in earthquake wave analysis, speech recognition, mechanical breakdown diagnosis domains and so on. The idea of The Hilbert-huang transformation is that decomposes the complex multi-component signal to the simple single component signal, then carries on processing each single component signal. This paper which unifies the pronunciation signal the characteristic, proposes the corresponding improvement algorithm in view of the Hilbert-huang transformation existed question, carries on the extraction to characteristic of pronunciation signal by using the actual pronunciation signal data, and provides the foundational characteristic data for application based on the pronunciation characteristic.First of all, in the depth study on Hilbert Huang of the basic principles, the current signal analysis commonly used processing method are described in detail, classify, analyze and compare, summarize their advantages and disadvantages in signal processing. Second, we has studied the experience model decomposition algorithm with emphasis, analyzed and summarized its existence deficiency. This signal decomposition algorithm may obtain the quite reasonable decomposition to regarding the non-linearity, the non-steady signal. However, because this algorithm itself existed some questions, can expire appearorthogonality, the end effect, the false component production, and so on, in the course of decomposition certain signal these question existence can affect directly to the signal characteristic extraction, analysis, processing.This paper proposed one kind way based on the mutual information theory false component recognition methods, its basic thought is mutually, decomposes the EMD signal with primitive signal information content or the losing achievement judgment false component basis which each inherent modality function carries, thus improves the EMD decomposition algorithm the performance.Again, whether the precise envelope of signal curve or not, affect directly the signal decomposition result. Envelope is constructed, the extreme ends of the determination of the signal is extremely important step, if not properly identified at the extreme end will be generated at the endpoints end line effect, the direct result:a continued decomposition Will be gradual "pollution" within the data. This paper presents a BP network extension algorithm, the algorithm can generate for end effect suppression.Finally, in order to the simulation algorithm of the improved algorith m with the current comparison, we use Matlab 7.1 platform as toolkit of simulation and results show that the improved algorithm in this paper is i n the false components identification of aspects and components of the in hibitory aspects of end effect has been greatly improved.