A conflict-free access method for parallel turbo decoder

Abstract

Parallel Turbo decoding encounters conflicts during parallel data accesses because of the sequential data interleaving, and the conflicts lead to a serious decrease of throughput. Traditional conflict-free memory mapping schemes often require a huge cost of hardware, such as the area of memory, which is also a bottleneck in practical implementations. In this work, DSATUR (Degree of Saturation) coloring algorithm is presented to solve the memory mapping problem and a memory subsystem with multiple memory banks is designed to achieve conflict-free data accesses during parallel Turbo decoding. To reduce the area of memory, an optimization for DSATUR coloring algorithm and an optimization for data addresses are proposed. The results of multiple experiments show that the conflicts of data access never occurs for using our memory mapping scheme. The two optimizations reduce 42% of the total area of memory compared with the original DSATUR algorithm without optimizations.