1: \begin{abstract}

2: The task of CDMA multiuser detection is to simultaneously estimate 

3: binary symbols of $K$ synchronous users from the received $N$ base-band 

4: CDMA signals. Mathematically, this can be formulated as an inference 

5: problem on a complete bipartite graph. 

6: In the research on graphically represented statistical models, 

7: it is known that the belief propagation (BP) can 

8: exactly perform the inference in a polynomial time scale 

9: of the system size when the graph is free from cycles 

10: in spite that the necessary computation for general graphs

11: exponentially explodes in the worst case \cite{Pearl}. 

12: In addition, recent several researches revealed that 

13: BP can also serve as an excellent approximation algorithm 

14: even if the graph has cycles as far as they are relatively 

15: long \cite{MacKay,Kaba,Yedidia}. However, as there exit many short cycles 

16: in a complete bipartite graph, one might suspect that the 

17: BP would not provide a good performance when employed 

18: for the multiuser detection. 

19: 

20: The purpose of this paper is to make an objection to such suspicion. 

21: More specifically, we will show that appropriate 

22: employment of the central limit theorem and the law 

23: of large numbers to BP, which is

24: one of the standard techniques in statistical mechanics, 

25: makes it possible to 

26: develop a novel multiuser detection algorithm 

27: the convergence property of which is considerably 

28: better than that of the conventional multistage 

29: detection \cite{multistage} without increasing the computational 

30: cost significantly. Furthermore, we will also provide 

31: a scheme to analyse the dynamics of 

32: the proposed algorithm, which can be naturally linked to the 

33: equilibrium analysis recently presented by Tanaka in \cite{tanaka}. 

34: \end{abstract}

35: