零填補之正交分頻多工系統的盲蔽式通道估計

Abstract

本論文提出一個經由盲蔽判別一個零填補之正交分頻多工系統通道以還原符元的方法。其判別方式是利用所接收訊號形成的自相關矩陣，計算出其通道係數乘積矩陣，再對一個由通道係數乘積矩陣所形成的厄米特矩陣作特徵分解。其通道係數向量，即為此厄米特矩陣經特徵分解後的最大特徵值所對應的特徵向量。所得的特徵向量與真實向量之間，存在了一個常數倍數的關係。而本文也提出了經由設計訓練符元的方式來得到真實的系統通道係數向量與計算而得的特徵向量之間的常數倍數。 為了減少通道雜訊對於估計通道準確度的影響，本文提出了利用觀察雜訊子空間與系統通道係數乘積子空間的相關程度來設計零填補之正交分頻多工系統的最佳週期性編碼序列。應用了最佳週期性編碼序列計算出通道之後，本篇論文以強制歸零等化器來還原符元。將實證結果與子空間方法比較後，首先，比起子空間法，本篇論文所提出的方法在計算上更為簡單。在低訊雜比與低傳送區塊的情況下，得到了較準確的通道估計與較低的錯誤率。

In this thesis, we proposed a blind channel identification scheme for zero-padding orthogonal frequency division multiplexing systems with periodic modulating sequence. The proposed method uses the block system model and exploits the channel matrix when some zeros are padded into the source block signal. When block signal is received, we can use its autocorrelation matrix to compute the products of channel coefficient. The channel impulse response vector can be identified, up to a scalar ambiguity, by computing the eigenvector associated with the maximal eigenvalue of a Hermitian matrix, which is formed by the products of channel coefficient. To remove the scalar ambiguity, we design the training symbols which are inserted in the data blocks. To minimize the decrease of channel estimation accuracy caused by channel noise, we design the optimal periodic modulating sequence for zero-padding orthogonal frequency division multiplexing by observing the orthogonality between the noise subspace and the channel product coefficient subspace. After applying the optimal periodic modulating sequence to identify the channel impulse response, zero-forcing equalizer is used to recover the symbol. Comparing to the subspace method, the proposed method is simpler in computation. In the low SNR regime and less data blocks, the proposed method has better performance in channel estimation accuracy and lower symbol error rate.