A low-memory intensive decoding architecture for double-binary convolutional turbo code

Memory accesses take a large part of the power consumption in the iterative decoding of double-binary convolutional turbo code (DB-CTC). To deal with this, a low-memory intensive decoding architecture is proposed for DB-CTC in this paper. The new scheme is based on an improved maximum a posteriori probability algorithm, where instead of storing all of the state metrics, only a part of these state metrics is stored in the state metrics cache (SMC), and the memory size of the SMC is thus reduced by 25%. Owing to a compare-select-recalculate processing (CSRP) module in the proposed decoding architecture, the unstored state metrics are recalculated by simple operations, while maintaining near optimal decoding performance.