本論文主要是探討在分碼多工系統中,改進機率資料連結演算法之多使用者偵測器計算速度,以提升系統效能。雖然使用機率資料連結演算法之多使用者偵測器可以得到近乎最佳化多使用者偵測器的效能,但其計算複雜度相較於最小均方誤差多使用者偵測器仍然偏高。為了增快其收斂速度,本論文使用電腦模擬來探討不同的機率初始值設定在同步及非同步分碼多工系統中對機率資料連結演算法之多使用者偵測器的效能與收斂速度之影響。進而提出以最小均方誤差或解相關多使用者偵測器之輸出訊號,經由軟性決定機率作為機率資料連結演算法之多使用者偵測器之機率初始值的新方法。此一機率初始值之設定,可在系統維持相同的位元錯誤下,有效的降低機率資料連結演算法之多使用者偵測器計算量。最後我們將此一方法與使用固定機率初始值機率資料連結演算法之多使用者偵測器應用於直序分碼多工系統及緩慢衰落平坦通道下之多載波分碼多工系統中,以驗證其可行性與效能改進。
The main purpose of this thesis is to improve the performance of multi-user detector using probabilistic data association in CDMA systems by reducing its computational complexity. The performance of multi-user detector using probabilistic data association is close to optimal multi-user detector, but its computational complexity is still higher than the linear MMSE multi-user detector. To reduce its computational complexity, the different initial settings of probabilistic data association are used in this thesis to analyze the effects of convergence time of this multi-user detector in synchronous and asynchronous CDMA systems. A new method has been proposed in this thesis. We set the initial values of probabilistic data association by using the soft decision of the output signals in MMSE or decorrelating multi-user detectors. It can reduce the computational complexity while maintain the same bit error rate in the CDMA systems. Finally, the new proposed method and other methods of fixed initial values of probability are applied to both the DS-CDMA system and slowly fading flat channel MC-CDMA system to verify its performance.