I.А. Kirianov - Engineer, GSKB Almaz-Antey (Moscow). E-mail: cucutnog@bk.ru

Low-density parity-check (LDPC) codes have the best ability to error correction. The basic algorithm to decoding LDPC codes is «Belief propagation». It is powerful tool to decoding LDPC codes. But this algorithm has a high complexity since one uses hyperbolic tangent and arctangent function. Algorithm «Min-sum» allows reduce decoding complexity. Exchange between variable and check nodes goes without uses hyperbolic tangent and arctangent function. Payback for this simplification is degradation of error correction performance. It is about 0,7 dB. This loss can be reduced. Check messages of «Min-sum» is always greater than check massages of «Belief propagation». This is a reason for this loss. There are two ways to minimize this loss. First, it is algorithm «Min-sum normalized». Check messages of «Min-sum normalized» are computed by dividing check messages of «Min-sum» by some constant. Loss of this algorithm is about 0,1 dB. Second, it is algorithm «Min-sum offset». Check messages of «Min-sum offset» are computed by subtraction from check messages of «Min-sum» some constant. Loss of this algorithm is about 0,05 dB.