Turbo multiuser detection for coded DS-CDMA systems: a Gibbs sampling approach

This paper deals with joint detection and decoding techniques for coded CDMA (code division multiple access) systems. A promising approach in this content consists of combining the results of a soft output multiuser detector (MUD) with single user soft-input soft-output (SISO) decoders in an iterative fashion (so called "turbo" principle). In a first part of the paper we describe the CDMA channel under the form of a probabilistic graphical model (also known as Bayesian, or belief, network) which provides a very generic and natural way of deriving turbo algorithms. The structure of the algorithm is then obtained by direct application of general probability propagation rules rather than by using the context dependent notions of intrinsic and extrinsic information. It turns out however that the obtained algorithm still requires soft output multiuser detection in a pseudo model where the symbols emitted by the user are a priori independent, which is not computationally feasible. The second part of the paper describes a simulation based MUD scheme which draws upon advances in Markov chain Monte Carlo methods. The performance of the overall turbo multiuser decoder is compared with that of a state-of-the-art algorithm with comparable computational cost.