對一個全平行高速低密度奇偶檢查碼解碼器(LDPC codes)而言,要達到高速(High Throughput)則必須將硬體內速度慢電路架構加以改進,所以我們提出機率性正規化最小和演算法(Probabilistic Normalized Min-Sum Algorithm),將查核節點(Check Node)內比較出第一小第二小的電路,改成只需比較出第一小與機率式的第二小的架構,利用此架構在遞迴(iteration)次數最多為7次的情況下,我們可以只犧牲0.05dB的位元錯誤率(bit error rate,BER),但卻可以得到最長路徑(Critical Path)變短與面積下降的好處,且我們根據最佳的正規化因子(Normalized factor)等於0.5和比較器電路改用查表的方式實作,可以實現出3bit硬體電路與4bit的位元錯誤率,然後再採用同時處理兩個字碼(codeword)的硬體架構和提早終止解碼機制(Early Termination)來實作。我們以TSMC 90nm 1P9M COMS製程來實作,APR(Automatic Place and Routing)後面積為7.97 mm^2,吞吐量(Throughput)為223.8(Gbps),吞吐量與面積的比率(TAR)為28.08(Gbps/mm^2),且與5bit的TPMP架構相比我們可以有62%的面積化簡與78.7%的最長路徑化簡。
In this thesis, we propose a Probabilistic Normalized Min-Sum Algorithm (PNMSA) for low-density parity-check (LDPC) decoders, where a probabilistic second minimum value is used in the check-node processing. Simulation results show that the proposed algorithm only introduces a minor performance degradation compared to the original normalized Min-Sum Algorithm. Based on the PNMSA, a fully-parallel decoder architecture is devised, where the check-node processing is implemented using several subunits and an efficient method is proposed to exchange messages between these subunits. With a carefully-chosen normalization factor, a satisfactory error-rate performance can be achieved using a lower number of quantization bits. In addition, look-up-table-based comparison with lower complexity is used to implement the check-node units. The proposed decoder was implemented using a 90-nm 1P9M CMOS process. Post-layout results show that the decoder occupies an area of 7.97 mm^2, achieves a throughput of 223.8-Gbps, and an energy efficiency of 14.9 pJ/bit.