1: \begin{abstract}

2: In practice, LDPC codes are decoded using message passing methods.

3: These methods offer good performance but tend to converge slowly

4: and sometimes fail to converge and to decode the desired

5: codewords correctly. Recently, tree-reweighted message passing methods have

6: been modified to improve the convergence speed at little or no

7: additional complexity cost. This paper extends this line of work and

8: proposes a new class of locally optimized reweighting strategies,

9: which are suitable for both regular and irregular LDPC codes. The

10: proposed decoding algorithm first splits the factor graph into

11: subgraphs and subsequently performs a local optimization of

12: reweighting parameters. Simulations show that the proposed decoding

13: algorithm significantly outperforms the standard message passing and

14: existing reweighting techniques.\\

15: 

16: \end{abstract}

17: