Blind system identification and source separation in power lines using finite alphabet signals

This dissertation presents a new method of solving the underdetermined case of blind source separation (BSS) in power lines. An algorithm was developed based on four synergistic approaches: a finite alphabet coding of source signals, a maximization of the system noise margin by a loading impedance factor, a recursive implementation of Self-Organizing Map, and data synchronization by zero-crossing detection. In a multiple-source-single-receiver system, which represents the most difficult case of the underdetermined BSS problem, the algorithm is capable of identifying the mixing parameters of the system and recovering the data from multiple sources without having any a priori knowledge of the system or demanding any training sequence. The algorithm begins by estimating the system mixing parameters. Once the parameters are identified, the source data of the multiple sources are recovered by using an ordinary classification technique, which in this work is a least-square technique.; To evaluate performance and robustness of the developed algorithm, the algorithm was implemented to solve an underdetermined blind source separation (uBSS) problem in powerline surveillance sensor network. Since a powerline network is a highly-disturbed noisy system suffering from unpredictable fatal attenuation, multipath fading, and unpredictable super-Gaussian noise, it is a good candidate for testing the algorithm. The bit-error-rate (BER) and the potential number of data sources of the test system were verified and presented in this dissertation.; This work presents a real-world experimental realization of the underdetermined BSS in the legacy powerlines for a simple but robust surveillance sensor network.