基於多級樹狀結構之球狀解碼器實現

Abstract

在多重輸入輸出(MIMO)的通訊系統中，最大概似(ML)偵測具有相當突出的性能表現。但是最大概似偵測的主要缺點在於計算複雜度隨著天線數量與調變階數的遞增而指數增加。因此我們考慮以樹狀搜尋為基礎的 K-Best 演算法。K-Best 演算法的效能接近最大概似偵測，但計算複雜度卻比最大概似偵測低。本篇論文提出了多級 K-Best 演算法應用在多重輸入輸出系統。在所提出的方法中，主要是將高階星狀點分解成許多低階星狀點，並且應用有順序性地偵測低階星狀點的技巧來降低最大概似解錯失的機率。實驗結果發現多級 K-Best 演算法相較於傳統 K-Best 演算法不僅擁有較低的計算複雜度同時也能夠達到幾乎相同的效能。同時在本篇論文的最後也以目前數位積體電路技術來實現多級 K-Best 偵測器的架構。

In multiple-input multiple-output (MIMO) communication systems Maximum- Likelihood detection is the preferred detection method which achieves the optimal performance. However, the main drawback of Maximum-Likelihood detection is that it suffers from exponential computational complexity against the number of antennas and signal modulation methods. Thus we consider K-Best algorithm based on the breadth-first tree search. The K-Best algorithm provides near-ML performance while its computational complexity is reduced compared to the Maximum-Likelihood detection. This thesis proposes the multistage K-Best algorithm in MIMO systems. In the proposed method, we decompose higher order constellation into several lower order constellations and apply ordering of lower order constellation to reduce the probability of missing the Maximum-Likelihood solution. It can be seen that the multistage K-Best algorithm and conventional K-Best algorithm achieve almost identical PER performance at the same K value. However, the multistage K-Best algorithm is shown to achieve such performance with a significantly lower computational complexity compared to the conventional K-Best algorithm. At the same time the VLSI architecture for the implementation of the multistage K-Best algorithm based on 0.18-μm technology is presented.