Modeliranje indeksa aktivnosti s statistikami višjih redov za vrednotenje redkih impulznih izvorov v konvolutivnih mešanicah : doktorska disertacija

slovenian

doktorska dela

2010

Univerza v Mariboru, Fakulteta za elektrotehniko računalništvo in informatiko

In this doctoral thesis, we address the evaluation of sparse impulse sources in convolutive signal mixtures. We develop statistical models of activity index, namely model of activity index average, variance and autocovariance sequence. The models can be used for the estimation of the system response lengths and the number of active impulse sources in the observed convolutive mixtures. We begin with an overview of the state of the art in the field of estimation of the number of sources and the length of system responses. Next, a model of convolutive mixtures of sparse impulse sources is introduced, which means a basis for all the derived activity index models. Next, a decomposition of surface EMG and the convolution kernel compensation (CKC) method are briefly presented. This is because activity index was introduced together with this decomposition method and it represents an essential part of CKC. The main part of this work is devoted to modelling the activity index. When doing this, we focus on the contributions of sources and noise separately. Developed models can be used for the estimation of system response lengths and the number of active impulse sources. For this purpose we developed two methods. The first one is based on the model of activity index variance and estimates the length of system responses and the number of sources by a reduction of search space. The second approach is based on the autocovariance sequence of activity index, which also estimates the length of the system responses. When the length of the system responses is estimated, the number of sources can be derived by using methods that rely on the system eigenvalues. The second approach has proved to be better. In the sequel, the possibilities for upgrading the activity index with higher-order statistics are presented. We address third- and fourth-order statistics and problems that appear when using them to extend the basic form of activity index. The developed models are verified on synthetic convolutive mixtures of sparse impulse sources.

07.08.2023