應用於編碼-多輸入多輸出接收機之轉換式清單解映器

Abstract

最大對數清單解映器(max-log list demapper)已廣泛地應用於編碼-多輸入多輸出(coded multiple-input multiple-output，coded MIMO)接收機中，其原理主要是利用部分訊號向量-稱為候選清單(candidate list或簡稱為list)-來計算對數相似比(log-likelihood ratio)以降低複雜度。在文獻中，清單可以在訊號原始域(original domain)或是轉換域(transformation domain)來產生；前者是直接利用原始的訊號向量來產生清單，這樣的方式在遭遇不良(ill-conditioned)通道時，會有嚴重的效能衰退。而後者在產生清單之前，會先將訊號向量轉至轉換域，使得轉換之後有較佳狀況(better condition)的通道可以利用。在此論文中，我們針對coded MIMO接收機，設計以轉換為基礎的清單解映器。在第一部分，我們利用一種著名的轉換技術，稱為晶格優化(lattice reduction，LR)，來產生清單。透過LR的轉換，可得到較佳狀況的通道，以減緩在產生清單過程中的雜訊放大(noise enhancement)效應。在此部分，我們提出了兩種以LR為輔助的清單解映器，一種是針對迭代式(iterative)接收機而設計，另一種則是針對非迭代式。模擬結果顯示，和現有的解映器在相近複雜度下相比，我們所提出的清單解映器有較佳的效能，特別是使用小清單或是遭遇空間相關性高(不良)通道時。 接下來，我們研究了以奇異值分解(singular value decomposition，SVD)為輔助的清單產生方法。透過SVD，原始的通道矩陣將被轉換為一個沒有層級干擾(layer interference)的通道，因此可以簡化產生清單的程序，進而降低複雜度。在此部分，針對迭代式接收機，我們提出了兩種以SVD為輔助的清單解映器，第一種是以達到較佳效能為目標設計的，另一種則是考慮了實現相關的議題。藉由複雜度分析和效能模擬，和我們先前提出的以LR為輔助的清單解映器相比，以SVD為輔助的清單解映器具有較佳的效能-複雜度取捨(performance and complexity trade-offs)，特別是遭遇空間相關性高(不良)通道時。 最後，我們探討以搜尋較佳位元組(bit-tuple)的方式來產生清單。在文獻中，不管是在原始域或轉換域，清單皆是從符元的層級(symbol level)來產生。因為訊號(或符元)向量和位元組具有一對一對映(one-to-one mapping)的關係，我們可利用那些較有可能的位元組所對應的訊號向量來組成清單。在此部分，針對迭代式接收機，我們提出了兩種位元層級(bit level)的清單產生方法，第一種是以達到最佳效能為目標，第二種則是低複雜度、容易實現的方法。和其他符元層級的清單產生方法相比，前者雖然具有最佳的效能，卻因太過複雜而難以實現；而後者可透過增加清單大小的方式來平衡使用有缺陷、較為簡單的搜尋準則(search metric)，來達到較佳的效能。

The max-log list demapper has been widely employed in the implementations of a coded multiple-input–multiple-output (MIMO) receiver, where only a candidate list of sig-nal vectors is examined in the log-likelihood-ratio calculation to reduce complexity. In the literature, the list is generated either in the original domain or the transformation domain. In the former, the original signal vectors are used for the generation of the list, which, unfortunately, results in severe degradation in the performance if the channel is in ill-condition; while in the latter the signal vectors are transformed to the transformation domain before the list generation in a way that the transformed effective channel has a better condition. In this dissertation, we focus on the design of transformation-based list demappers for coded-MIMO receivers. In the first part, a famous transformation technique, called lattice reduction (LR), is utilized for the generation of list, where a better-conditioned channel matrix can be obtained and the adverse effect of noise enhancement can be mitigated during the list generation. Two LR-aided list demappers are proposed, one for iterative receivers and the other for non-iterative receivers. Simulation results show with similar complexity, the proposed demappers provide significant gains over existing demappers, particularly for the cases with a small list size and/or under spatially correlated channels. Next, the issue of list generation with the aid of singular value decomposition (SVD) is investigated. With SVD, the original channel matrix is transformed to one free of layer interference, and hence simplifies the list generation. Two SVD-aided list demappers for iterative receivers are proposed here, one is designed for better performance, while the other considers the issue of implementation. Compared to the proposed LR-aided list demapper, our results show that SVD-aided demappers have better performance and complexity tradeoffs, especially in the spatially correlated channels. Last, the list generation through the search of promising bit tuples is studied. In the literature, no matter which domain the list is generated, the search of list is performed at the symbol level. Since there is a relation of one-to-one mapping between a signal (symbol) vector and a bit tuple, a list can be generated in a way of collecting those signal vectors mapped from promising bit tuples. Two bit-level list generation schemes for iterative re-ceivers are proposed, one targets at better performance, while the other at low implementation complexity. Compared to other symbol-level methods, the former is shown to have the best performance but with a very high implementation complexity while the latter provides a good performance-complexity tradeoff by using a larger list size complemented by a simple search algorithm.