n this paper we propose an extension of the space alternating generalized expectation maximization (SAGE) algorithm that accepts as input and provides as output soft information. This so-called soft-input soft-output (SISO) SAGE algorithm is applied to joint successive interference cancellation and channel estimation and combined with SISO single-user decoders in an iterative CDMA receiver. Monte Carlo simulations in flat Rayleigh fading channel show that this receiver is robust against channel estimation errors and can support a high system load.