We present a framework for iterative multiuser joint decoding of code-division multiple-access (CDMA) signals, based on the factor-graph representation and on the sum-product algorithm. In this framework, known parallel and serial, hard and soft interference cancellation algorithms are derived in a unified way. The asymptotic performance of these algorithms in the limit of large code block length can be rigorously analyzed by using density evolution. We show that, for random spreading in the large-system limit, density evolution is considerably simplified. Moreover, by making a Gaussian approximation of the decoder soft output, we show that the behavior of iterative multiuser joint decoding is approximately characterized by the stable fixed points of a simple one-dimensional nonlinear dynamical system.
