1: \begin{abstract}

2: \footnotetext[1]{Contributed equally to this work.\\Supported in part by NSF Grant

3: No.~CCR-0514859 and EVERGROW, IP 1935 of the EU Sixth Framework.}

4: The canonical problem of solving a system of linear equations

5: arises in numerous contexts in information theory, communication

6: theory, and related fields. In this contribution, we develop a

7: solution based upon Gaussian belief propagation (GaBP) that does

8: not involve direct matrix inversion. The iterative nature of our

9: approach allows for a distributed message-passing implementation

10: of the solution algorithm. We also address some properties of the

11: GaBP solver, including convergence, exactness, its max-product

12: version and relation to classical solution methods. The

13: application example of decorrelation in CDMA is used to

14: demonstrate the faster convergence rate of the proposed solver in

15: comparison to conventional linear-algebraic iterative solution

16: methods.

17: \end{abstract}

18: