Performance analysis of optimal channel estimation and multiuser detection in a randomly-spread CDMA channel

This paper analyzes performance of optimal channel estimation and multiuser detection (MUD) in a block-fading code-division multiple-access (CDMA) channel on the assumptions of random spreading and large-system limit, by using the replica method developed in statistical mechanics. The authors find that the asymptotic spectral efficiency of the linear minimum mean-squared error (LMMSE) MUD which was proposed and analyzed by Evans and Tse in 2000 is indistinguishable from that of the optimal MUD for small system loads. Our results imply that performance of MUD scarcely improves even if one spends more computational cost than that of the LMMSE MUD, i. e., at most the cube of the number of users, on the above-described conditions.