Blind Equalization via Blind Source Separation Techniques

This paper addresses the application of blind source separation (BSS) algorithms to the problem of blind equalization (BE) in a frequency selective multiuser environment. Compared with other algorithms that exploit the structural properties of the channel or data matrix, our approaches are insensitive to channel order estimation, robust to ill-conditioned channels and able to choose the \best" equalizer delay(s). Many batch and adaptive BSS algorithms are brieey discussed for the application to BE, but we mainly focus on the application of two particular well-known batch approaches, i.e., JADE (Joint Approximate Diagonalization of Eigenmatrices) and ACM (Analytical Constant Modulus). Their respective underlying principles, assumptions, methodologies, computational complexity and performance are analyzed respectively. The relation between output noise power and the delay at which the equalizer recovers the signal is studied, from which an MMSE criterion is introduced to the JADE algorithm to reduce the noise enhancement eeect. Simulations under various scenarios demonstrate promising results for both JADE and ACM, which clearly outperform SOS-based BE algorithms such as the subspace and linear prediction techniques.