摘要 直序式分碼多工 (Direct sequence-code division multiple access, DS-CDMA) 系統在蜂巢式行動電話系統被廣泛的研究。在此篇論文裡,特別討論DS-CDMA的系統在受到非高斯雜訊干擾時,因為信號在傳輸過程中受到脈衝雜訊 (impulse noise) 干擾的影響,導致在接收端的信號無法正常還原的問題,而通常脈衝雜訊為人為的電磁波干擾或大量自然界中所產生的雜訊,經由實驗證實此種脈衝雜訊為一非高斯雜訊分佈,所以我們進而利用在 [4] 中的一種高斯混合型的模型,是利用兩種變異數不同的高斯雜訊所組合而成的,來去模擬出信號在傳輸的過程中所受到的非高斯雜訊。 在這裡我們將在接收端的部分使用一些方法,來估測還原出原來的訊號,有利用遞迴的動作來壓抑雜訊的估測器,如線性解相關器least square (LS) 、M解相關估測器 (Minimax decorrelating detector) 或者是利用在統計的觀點,將受到雜訊干擾過於嚴重的資料視為離群值(outlier),找出並加以去除的方法,如最小平方中位數法least median of square (LMedS)、重複再加權最小平方法reweighted least square (RLS)、逐次平均去除法consecutive mean excision (CME) 、順向逐次平均去除法forward consecutive mean excision (FCME) 等,來解決脈衝雜訊影響系統的問題,並且進行系統的程式模擬來比較各種估測器的效能表現。 另外,我們將簡化各種方法的複雜度運算來得到新的方法:如改善LMedS和RLS過於複雜的問題以及找出CME 與 FCME的效能上限,並且利用M-estimator [4] 的分析方法,比較出各種方法的效能在理論上的差異性,以及算出各種的估測器運算複雜度,找出一個低複雜度高效能的方法。
Abstract Direct sequence-code division multiple access (DS-CDMA) has been extensively investigated in the cellular mobile telephone system. In this thesis, we discuss non-Gaussian noise effect to the DS-CDMA system. Because the signal is influenced by the impulse noise in the course of transmitting, the signal cannot return to original signal at receiver. The impulse noise is usually due to the human-made electromagnetic interference and a large number of natural noise as well. It is known through experiment measurements to be a non-Gaussian distribution. And then we can exploit a Gaussian mixture model in [4], it is composed of two kinds of Gaussian distribution which have different variance. So we can use the mixture model to attain the impulse noise which affects the transmitted signal. We will use some methods to estimate the original signal in the receiver. There are some estimators that use iteration to suppress the impulse noise, least square and minimax decorrelating detector, etc. In statistics, there are also some estimators, for instance, least median of square (LMedS), reweighted least square (RLS), consecutive mean excision (CME) and forward consecutive mean excision (FCME). Those methods let the signals be an outlier and cancel them when they are interfered seriously. All of those methods want to solve the impulse noise problem in the system and we will use simulation program to compare the performance of all estimators. We will also simplify the complexity operation of different methods to obtain the new method here. Such as, our proposed methods improve the overly complicated problem in LMedS and RLS. The performance upper bound of CME and FCME are found by our proposed method. Using the method of M-estimator [4], we analyze all methods and compare the performance of different methods analytically. In the end, we compute the complexity operation of all estimators. So we can find a low complexity and high performance method.