- 作者: 吳文榕
- 作者服務機構: 交通大學電信工程系
- 中文摘要: 盲目等化器是一種不需訓練訊號(training Sequence)而能自行收斂的等化器。在很多情況下,這種等化器特別有用,例如有些時候訓練訊號很難傳送或是會延遲一段很長的時閒;而使用盲目等化器最大的缺點在於其收斂速度緩慢,這是由於其所使用之LMS演算法所致。LMS使用隨機梯度(stochastic gradient)的搜尋方式,因此其方向不一定在最陡的方向,這種現象在盲目等化之演算法中特別明顯。在本論文中我們提出了一新的等化方式,能非常有效的改進收斂速度。在傳統的盲目等化作法上,主要是利用一非convex的成本函數(cost function)而將其期望(expectation)最小化,而我們所提出的做法是將此函數在某一區塊的資料的平均加以最小化,此種做法類似最小平方法(least square),但因我們所使用之成本函數為非convex,因此無法得到遞迴(recursive)的式子,因此我們將資料分成區塊(block),而在區塊內得到最佳值,我們稱這種等化方式為區塊更新(block updated),就像最小平方法的收斂速度遠高於LMS一樣,我們這種做法也將使盲目等化的收斂速度大為提高。
- 英文摘要: A blind equalizer can converge without using any a priori known training sequence. The maindrawback of using a blind equalizer is its slow convergence. This is due to the LMS type of algorithmemployed in the equalization The LMS algorithm uses a stochastic gradient, and the search is notnecessarily in the direction of the steepest descent. This is particularly true for the blind equalizer. Inthis paper, we propose a new scheme to accelerate the convergent speed. The conventional blind equalizationschemes are derived by minimizing the mean of some non-quadratic cost functions. Instead of doing so,we propose to minimize the time average of the cost functions. This is similar to the concept of the leastsquare method. However, since the cost function is non-quadratic, a recursive formula cannot be obtained.Thus, we use a sub-optimal approach. We first partition the inputs into blocks. The optimal solutionin a block is found by some iteration method. Its initial value is obtained from the optimal solutionsof previous blocks. We also investigate the relation of our algorithm and Bussgang's. Simulations showthat our algorithm significantly improves the convergent rate of blind equalizers.
- 中文關鍵字: blind equalization; LMS algorithm; blind equalizer; Bussgang's algorithm
- 英文關鍵字: --