Fractional sampling in blind source separation for blind equalization, and related topics.

Format:

Book

Thesis/Dissertation

Author/Creator:

Zhang, Yinglu.

Contributor:

Kassam, Saleem A., 1949- advisor.

University of Pennsylvania.

Language:

English

Subjects (All):

Electrical engineering.

0544.

Penn dissertations--Electrical engineering.

Electrical engineering--Penn dissertations.

Local Subjects:

Penn dissertations--Electrical engineering.

Electrical engineering--Penn dissertations.

0544.

Physical Description:

239 pages

Contained In:

Dissertation Abstracts International 62-02B.

System Details:

Mode of access: World Wide Web.

text file

Summary:

Blind Source Separation (BSS) is the process of recovering a set of independent signals when only mixtures with unknown mixing coefficients are observed. Many important theories and applications have been investigated in BSS and snore generally in Independent Component Analysis (ICA). Blind equalization (BE) refers to the problem of determining the impulse response of the system or the input signal when the system is unknown and its input is inaccessible.

In digital communications, multiple independent sources may produce received signals from dispersive channels. Due to the presence of inter-symbol interference (ISI) in each channel and inter-user interference from adjacent channels, the received signals will be convolutive mixtures of source symbol sequences. Therefore, channel equalization is required to remove ISI, and source separation is necessary to separate each individual source signal from their mixtures. Blind equalization and blind source separation assume no training sequences, and little is assumed about the source signals and channel conditions. A novel technique for multichannel blind source separation and blind equalization using fractional sampling is proposed in this dissertation. Our method is also applicable to a single-input-single-output (SISO) channel.

Different BSS algorithms call be used to equalize a SISO channel. In this dissertation, we include performance comparisons between different BSS algorithms and other BE algorithms based on second-order cyclostationary statistics (SOCS). Mean-squared-error (MSE) of recovered signal is used as the performance measure of each algorithm. Our simulation results show that in general BSS combined with fractional sampling gives better performance at high SNR. It is also shown that the performance of the method using BSS and fractional sampling depends on the specific BSS algorithm chosen.

As an interesting application of BSS, we consider linear blind equalization for the global system for mobile communications (GSM) systems using source separation. The Gaussian minimum shift keying (GMSK) modulation employed in GSM systems is a nonlinear modulation technique, which necessitates the use of linear approximation models for the GMSK signal. In this dissertation, we consider two linear approximation models for the GMSK signal. The denotation technique and the denotation combined with fractional sampling method are discussed based on the two linear models. We introduce a novel equalizer structure that exploits both derotation and oversampling.

Robustness against deviations from nominal source probability density function (pdf) assumptions is very desirable in BSS algorithms. Little has bee done in the past on robustness issues in BSS. In this dissertation a new approach using ranks is proposed for robust BSS. Two different methods for evaluation of ranks are also introduced.

Notes:

Thesis (Ph.D. in Electrical Engineering) -- University of Pennsylvania, 2001.

Source: Dissertation Abstracts International, Volume: 62-02, Section: B, page: 1010.

Supervisor: Saleem A. Kassam.

Local Notes:

School code: 0175.

ISBN:

9780493134697

Access Restriction:

Restricted for use by site license.