| 摘要: | 在本論文中提出可改善傳統K-Best球面演算法的解碼性能的前瞻式架構。傳統的K-Best架構在進行解碼時, 在判斷存活路徑時僅就該層節點的PEDs進行判斷, 導致最大可能的解有一定的機率被排除在K個存活路徑之外, 容易造成錯誤的情況因而造成效能衰減。所提出的前瞻式架構有別於傳統K-Best架構的地方在於後者, 因此, 而前者也就是本論文所提出的前瞻式K-Best球面演算法在判斷時不僅以該層節點的PEDs進行判斷, 而是將該層節點延伸至下一層之子節點的PEDs拿來做判斷, 因為下一層子節點的PEDs所包含的兩層資訊遠多於只包含一層資訊的PEDs, 用來判斷將可提升存活路徑包含可能解的機率, 也就是說可以有效的提升效能。就模擬結果而言, 當K值皆為10時, 我們可以發現前瞻式架構的效能在錯誤率4×10-4左右的時候, 優於傳統K-Best演算法約4 分貝(dB)。在硬體的實作上, 採用管線式架構以達到高產出的目的, 同時以1-norm取代2-norm, 1’s complement取代2’s complement, 以及硬體共用等概念來降低硬體複雜度, 對於前瞻式架構中的前置排序處理模組來說, 此三種方法, 依據合成的結果可減少70%的複雜度。 A look-ahead algorithm that can improve the detection performance of the conventional K-Best sphere decoding algorithm is proposed in this thesis. In the conventional K-Best sphere decoder, which uses the partial Euclidean distance (PED) in the current layer to decide the K survival paths, the maximum likelihood (ML) solution may be expelled in the top layers and thus its performance is degraded. However, the proposed look-ahead technique uses not only the PEDs in the current layer, but also the PEDs of the best child node in the next layer. Because that the PEDs of survival nodes in two layers contain more information than the PEDs of the survival node in one single layer, the probability of reaching the ML solution can be increased. We can see that the one-layer look-ahead algorithm outperforms the conventional K-best algorithm about 4 dB when the BER is around 4×10-4. As to the hardware implementation, a pipeline architecture is employed to enhance the throughput. Moreover, we replace the 2-norm by 1-norm, 2’s complement by 1’s complement and also use common term extraction. With these techniques, the hardware of the pre-sorting process in the look-ahead unit can be reduced more than 70％. |
