Semi-blind Multiuser Detection in Zero-padded OFDM-CDMA Systems

In this paper, we propose a new semi-blind multiuser detector for zero-padded OFDM-CDMA systems. We show that the downlink transmission in a wireless zero-padded OFDM-CDMA system can be modeled as a linear mixture of the transmitted data symbols. Therefore, blind source separation (BSS) algorithms based on the natural gradient are good candidates to blindly recover those symbols. Before applying BSS, we propose to reduce the effect of the additive noise performing a sub-space projection. On the other hand, we propose to take advantage of any available, may be rough, channel estimation to initialize the blind algorithm. The benefits of initializing the BSS approach are twofold: we avoid the source scaling and arrangement problem and we achieve a quite faster convergence.