I.J. Intelligent Systems and Applications, 2019, 2, 1-8 Published Online February 2019 in MECS (http://www.mecs-press.org/) DOI: 10.5815/ijisa.2019.02.01 Copyright © 2019 MECS I.J. Intelligent Systems and Applications, 2019, 2, 1-8 A Hybrid Model of 1-D Signal Adaptive Filter Based on the Complex Use of Huang Transform and Wavelet Analysis Sergii Babichev Jan Evangelista Purkyně University in Ustí nad Labem, Ustí nad Labem, Czech Republic E-mail: sergii.babichev@ujep.cz Oleksandr Mikhalyov National Metallurgical Academy of Ukraine, Dnipro, Ukraine E-mail: maillich2@gmail.com Received: 06 August 2018; Accepted: 25 November 2018; Published: 08 February 2019 Abstract—The paper presents the results of the research concerning the development of the hybrid model of 1-D signal adaptive filter based on the complex use of both the empirical mode decomposition and the wavelet analysis. Implementation of the proposed model involves three stages. Firstly, the initial signal is decomposed to the empirical modes by the Huang transform with allocation the components, which contain the noise. Then the wavelet filtering is performed to remove the noise component. The optimal parameters of the wavelet filter are determined based on the minimal value of ratio of Shannon entropy for the filtered data and the allocated noise component and these parameters are determined depending on type of the studied component of the signal. Finally, the signal is reconstructed with the use of the processed modes. The results of the simulation with the use of the test data have shown higher effectiveness of the proposed method in comparison with standard method of the signal denoising based on wavelet analysis. Index Terms—Denoising, Empirical mode decomposition, Huang transform, Wavelet analysis, Thresholding, Shannon entropy. I. INTRODUCTION 1-D and 2-D signals denoising are one of the current problems of modern informatics. Allocation of the noise component from the studied data allows us to improve the quality of the investigated signal that improves the effectiveness of the following stage of the data processing within the framework of the current problem. A lot of denoising techniques based on different methods of the signal processing exist nowadays. The use of Kalman [1] or Wiener [2] filters allows us to smooth the signal by using the extrapolation technique in the first case and by minimizing the mean square error between the estimated random process and the desired process in the second case. However, it should be noted, that these techniques are not effective in the case of processing of non- stationary and non-linear signals with local particularities. Implementation of these techniques in these cases promotes to the loss of the large amount of useful information. The modern techniques of the non-stationary and non- linear signals processing are based on decomposition of the investigated signals to the components with the following processing of these components in order to remove the noise. So, the research concerning the use of fast Fourie transforms (FFT) for estimation of the anisotropic relaxation of composites and nonwovens is presented in [3]. In paper [4] the authors implemented the time-frequency analysis of pressure pulsation signal based on FFT. The frequency spectrum including frequency-domain structure and approximate frequency- scope was obtained during the simulation process. However, it should be noted, that FFT is effective in the case of stationary signals processing and features extraction from the data. In the case of more complex non-stationary and non-linear signal processing the effectiveness of the FFT technique decreases. The wavelet analysis is the alternative and the logical continuation of the FFT [5, 6]. The approximation and detail coefficients are calculated during wavelet- decomposition process. The detail coefficients contain the information about the noise component in the most cases, thus these coefficients are processed to remove the noise component. Reconstruction of the denoised signal is performed with the use of both the approximation coefficients and the processed detail coefficients at levels of the wavelet decomposition from 1 to N. The wavelet analysis technique is widely used in different area of the scientific research [7–14]. The effectiveness of this technique implementation depends on the choice of the type of the used wavelet, level of the wavelet decomposition and determination of the thresholding coefficient value for detail coefficient processing. It should be noted that an effective technique for these parameters objective determining is absent nowadays.