Bayesian multi-user detection based on a state-space model of the DS-CDMA system

In this paper, we propose a new multiuser detector based on a symbol rate state-space model of the CDMA system. This state-space model presents a non-Gaussian state noise characteristic, Therefore, the Kalman filtering approach looses its MMSE optimality. As a solution, we propose to apply the Weighted Sum of Gaussian (WSG) approximation. We demonstrate that the WSG is propagated during iterations through a NKF structure. At each iteration, we inject only one gaussian term having the mean and the covariance matrix of the last obtained MMSE estimate of the state vector. The proposed structure improves the performance compared to the linear structures such as: the MMSE, the Kalman filter and the feedback structure such as DFE.