使用線性估測方法的直接盲式MMSE等化器

Abstract

由於使用傳統的高階統計方法做盲式等化器會使收斂十分緩慢。近年來的研究多轉向Tong首所提出的二階方法，其中如Ding在西元2000年所提出的Ding演算法是屬於新式的外積分解法(OPDA)，它擁有比許多演算法更好的效能，但缺點是必需用到矩陣pseudo-inverse的運算，這會造成數值問題，且它也無法應用於時變頻道。基於Ding演算法，並藉Fan所提出的線性估測(LP)法，我們結合了這兩個演算法的優點，推導出改良的新演算法，我們稱之為LP-OPDA。LP-OPDA不需用到矩陣的pseudo-inverse，故可擁有比Ding演算法更優良的效能，且它可應用於時變頻道。LP-OPDA亦是屬於低運算量的演算法。由模擬中我們可以發現，LP-OPDA擁有比Ding演算法及其它許多演算法更好的效能。

The channel equalization using the high-order statistics methods has a slow convergence rate. In recent years, the second-order statistics (SOS) methods have become a popular research. One of the SOS methods, such as the Ding algorithm proposed by Ding in 2000 is an advanced type of outer-product decomposition algorithm (OPDA), has been shown to have better performance than many existing algorithms. But Ding algorithm needs the pseudo-inverse of the correlation matrix, thus the computation is not simple and could cause numerical problems. It is also not suit for tracking time-varying channels. By the use of linear prediction (LP) method proposed by Fan, we deduce a new algorithm based on Ding algorithm. We name the new algorithm as linear prediction based outer-product decomposition algorithm (LP-OPDA). LP-OPDA combines both the advantages of LP and Ding algorithm and has its new advantages. LP-OPDA does not need the pseudo-inverse operation, thus have superior performance over Ding algorithm. LP-OPDA is available for tracking time-varying channels and also computationally efficient. From the simulation results, we can see LP-OPDA has superior performance over Ding algorithm and many other existing algorithms in many ways.