Date of Award

2019

Publication Type

Doctoral Thesis

Degree Name

Ph.D.

Department

Electrical and Computer Engineering

Keywords

Blind Source Separation, ECG signal detection, FECG signal extraction, Null space Transformation Matrix, Signal Processing, Speech signal separation

Supervisor

E. Abdel-Raheem

Rights

info:eu-repo/semantics/embargoedAccess

Abstract

Blind source separation/extraction (BSS/BSE) is a powerful signal processing method and has been applied extensively in many fields such as biomedical sciences and speech signal processing, to extract a set of unknown input sources from a set of observations. Different algorithms of BSS were proposed in the literature, that need more investigations, related to the extraction approach, computational complexity, convergence speed, type of domain (time or frequency), mixture properties, and extraction performances. This work presents a three new BSS/BSE algorithms based on computing new transformation matrices used to extract the unknown signals. Type of signals considered in this dissertation are speech, Gaussian, and ECG signals. The first algorithm, named as the BSE-parallel linear predictor filter (BSE-PLP), computes a transformation matrix from the the covariance matrix of the whitened data. Then, use the matrix as an input to linear predictor filters whose coefficients being the unknown sources. The algorithm has very fast convergence in two iterations. Simulation results, using speech, Gaussian, and ECG signals, show that the model is capable of extracting the unknown source signals and removing noise when the input signal to noise ratio is varied from -20 dB to 80 dB. The second algorithm, named as the BSE-idempotent transformation matrix (BSE-ITM), computes its transformation matrix in iterative form, with less computational complexity. The proposed method is tested using speech, Gaussian, and ECG signals. Simulation results show that the proposed algorithm significantly separate the source signals with better performance measures as compared with other approaches used in the dissertation. The third algorithm, named null space idempotent transformation matrix (NSITM) has been designed using the principle of null space of the ITM, to separate the unknown sources. Simulation results show that the method is successfully separating speech, Gaussian, and ECG signals from their mixture. The algorithm has been used also to estimate average FECG heart rate. Results indicated considerable improvement in estimating the peaks over other algorithms used in this work.

Share

COinS