To decrease memory access of the decoder for double binary convolutional turbo code (DB CTC), an iterative decoding scheme is proposed. Instead of accessing all of the backward state metrics from the state metric cache (SMC), a part of them is computed by the recalculation unit (RU) in the forward direction. By analysis and simulations, both the amount of memory access and the size of SMC are reduced by about 45%. Moreover, combined with the scaling technique, the proposed scheme gets decoding performance near to that of the well-known Log-MAP algorithm.

Serious BER performance degradation due to finite numeric precision in VLSI implementation of Log-MAP and MAX-Log-MAP decoders where forward state metric is calculated using reverse tracing method, is analyzed, and two methods are proposed to overcome this problem, the loser storing method for MAX-Log-MAP and the periodic storing method for Log-MAP and MAX-Log-MAP. Both methods can reduce memory storage size effectively by half, but with additional circuit overhead. Our VLSI implementation examples show that, compared with original method, both methods give about 15% improvement in area and power consumption with identical BER performance.