Iterative multiuser detection with graphical modeling

Abstract

In this paper, we give a graphical interpretation of the multiuser detection problem for interference limited channels, using generalized inference techniques on graphs. One impediment to the implementation of multiuser detection techniques in practical systems (such as in DS-CDMA) has been the difficulty of dealing with the large number of random variables present in such systems. Graphical techniques, which have been previously applied to inference problems in many random variables, can give new insight into the handling of this problem. The main focus of our work is on asynchronous channels, which may be effectively modeled with a Markov chain model. A new implementation of the EM (expectation maximization) algorithm over factor graphs is introduced. Modifications to the EM algorithm to implement approximate inference with reduced complexity are discussed.