Matrix reordering techniques for memory conflict reduction for pipelined QC-LDPC decoder

Abstract

Layered Decoding (LD) algorithm is widely applied in high throughput QC-LDPC decoders. Among all check node update algorithms in LD, Turbo-Decoding Message-Passing (TDMP) is adopted by many proposals. A-posteriori memory access conflict under pipelined TDMP decoder incurs serious throughput decline. In this paper, several matrix reordering techniques are introduced to reduce the conflict occurrences without incurring the performance loss, which includes Row Exchange method, element Sequence Reordering method, and a conflict detector with pipeline stall insertion. They are integrated in a joint recursive deep-first searching procedure. Test results show that the efficiency improvement reaches up to 60% compared to non-optimized scenarios for 802.11n and 802.16e standards.