Threshold-based extended minimum-sum algorithm for non-binary LDPC codes

Abstract

In this paper, we propose a new improved algorithm to reduce the decoding complexity of NB-LDPC codes especially when the number of iteration is large. The proposed algorithm is based on the extended min-sum (EMS) which truncates the message vectors and trellis-based EMS (T-EMS) which employs the trellis representation. Our modification of EMS introduces the thresholds to further truncate the incoming messages to the check node. The performance of our improved new algorithm is only 0.05dB (resp 0.2dB) to BP in the waterfall region over GF(8) (resp GF(128)), but the complexity will reduce to lower than half of the EMS when the threshold is chosen reasonable.