Abstract—Belief-Propagation (BP) iterative decoding is considered in massive Multiple-Input Multiple-Output (MIMO) wireless communication systems. A problem with this BP decoding method that has yet to be solved is the evaluation of the residues in the Parallel Interference Canceller (PIC), which cannot be calculated directly. Furthermore, although there are various methods through which the residue effect may be accounted for, no comparative studies have thus far been reported. Hence, in this study, we consider the residue component as a random variable and construct BP decoders in which the residue effect is included into the likelihood of the PIC in different ways. We numerically compare the decoding performance among them. The results suggest that the decoder has high performance when it contains the residue effect in the variance of the likelihood.