適用於分頁配置儲存系統之低運算複雜度二維遞迴偵測設計

Abstract

被認為是下一代光儲存系統的分頁配置光儲存記憶體，由於其資料像素以極高密度記錄，一直以來存在著像素間嚴重互相干擾的問題。在描述其通道特性後，簡單門檻偵測被證明為不適用，而最佳化的最大可能性頁偵測也不具有實用性。在次佳的遞迴偵測策略中，本論文介紹並且比較了平行決定迴授等化與二維最大事後機率偵測。基於其極優越的偵測表現，二維最大事後機率偵測被決定採納。然而，在直接實作下，其運算複雜度仍顯過高。因此，本論文提出一套降低運算複雜度的設計方案，並證實在造成極少程度的效能損失下，其得以使二維最大事後機率偵測的運算複雜度顯著地降低。除了在演算法上作了改進，本論文也對此全平行化偵測策略的基本處理單元的架構提出了建議，同時也就一些其他實作議題進行討論。

Page-oriented optical memories, which are considered to be next-generation optical storages, have been suffered from severe inter-pixel interference (IPI) because of the dense packing of data pixels. After characterizing the channel, the simple threshold detection is demonstrated to be not suitable, while the optimal maximum-likelihood page detection (MLPD) is also impractical. Among the sub-optimal iterative detection schemes, parallel decision feedback equalization (PDFE) and 2D-MAP detection are introduced and compared. Owing to the superior performance, 2D-MAP detection is adopted. However, its computational complexity still appears too high for direct implementation. Therefore, a suite of complexity-reducing schemes are then proposed for 2D-MAP detection and demonstrated to be able to result in significant reduction on the amount of arithmetic operations, while at the cost of negligible performance degradation. In addition to improving the algorithm, the architecture of basic processing element for this fully-parallel detection scheme is also suggested, together with the discussion of some other implementation issues.