In this thesis, we investigate the blind equalization problem of both single-user and multi-user systems. We will present our contributions in five related topics. The first contribution is a blind equalizer designed for single-user systems by assuming that the user sequence is independent, identically distributed(i.i.d.) with nonzero 4th order cumulant. It is obtained by jointly diagonalizing a set of matrices which are computed from the received signal samples. This equalizer can be applied to multi-user systems for recovering one user sequence at a time....[ Read more ]

In this thesis, we investigate the blind equalization problem of both single-user and multi-user systems. We will present our contributions in five related topics. The first contribution is a blind equalizer designed for single-user systems by assuming that the user sequence is independent, identically distributed(i.i.d.) with nonzero 4th order cumulant. It is obtained by jointly diagonalizing a set of matrices which are computed from the received signal samples. This equalizer can be applied to multi-user systems for recovering one user sequence at a time.

The second contribution is a blind InterSymbol Interference(ISI) cancellation algorithm in multi-user systems. It is based on minimizing the error of three filter outputs with suitable time-shifts. The user sequences are assumed to be temporally white and mutually independent. In comparison with many existing algorithms, this algorithm is simpler and admits an adaptive implementation.

In order to reduce the sensitivity to the error in channel order estimate, our third contribution achieves the ISI cancellation by column anchoring. This algorithm does not need an exact channel order, and it can retain a pre-selected block column of the channel convolution matrix and force the remaining columns to zero. Simulation results show that the ISI cancellation method is applicable to the popular Globe System for Mobile(GSM).

As Blind Signal Separation(BSS) is required after ISI cancellation in multi-user systems, our fourth contribution, the algebraic BSS principle is established. Based on the principle, the smallest set of equations to date is obtained for the design of BSS algorithms.

Our last contribution is the general BSS algorithm with pipeline structure. It separates the source signals step by step. In each step, those source signals with nonzero given-order cumulants are extracted. Several simulation examples are presented to show the functions of this algorithm.