Design of a Neural Network Based Impulsive Noise Detector: with Application to Improve Iterative OFDM-based PLC Receivers

在電力線通訊系統中，訊號會受到脈衝雜訊(Impulsive Noise)的影響。我們可以透過將訊號轉到頻域上，一定程度地重建出雜訊。然而從重建的雜訊中將脈衝雜訊分離出來是一種非線性問題，使用一般的線性濾波器會導致系統的效能降低。本論文提出了一個利用類神經網路所設計的脈衝雜訊偵測器，用來偵測在正交分頻多工(Orthogonal Frequency-Division Multiplexing, OFDM)系統中的電力線收發機中的脈衝雜訊。類神經網路是一種非線性的機器學習模型，非常適合用來偵測脈衝雜訊。除此之外，類神經網路能夠針對傳輸環境不斷的調整神經模型，適合用在環境會隨時改變的通訊系統之中。本論文所提出的偵測器在一個遞迴式接收機之中使用，它含有兩個區塊，前脈衝雜訊處理(pre-IN mitigation)和後脈衝雜訊處理(post-IN mitigation)。前脈衝雜訊處理主要把較強的脈衝雜訊刪除，後脈衝雜訊處理則利用遞迴演算刪除剩餘的脈衝雜訊。本論文也提出一個目標產生器用來訓練類神經網路。電腦模擬結果顯示，利用類神經網路偵測器的電力線系統系統收發機能有效分離出脈衝雜訊與非脈衝雜訊，因此使系統效能大幅提升。The Impulsive noise (IN) in power-line communications (PLC) system affect the performance of system. We can transform the received signal into frequency domain, and hence the receiver can reconstruct the noise from received signal. However, finding the IN from the reconstructed noise is a non-linear problem, using a linear filter to handle this kind of problem would reduce the performance of the system. In this paper, we present a Neural Network based approach to detect IN in orthogonal frequency division multiplexing (OFDM) based baseband power-line communications system. Neural Network is a non-linear model and is suitable for the detection of IN; moreover, it is a self organizing system, which means that an Neural Network can adapt itself when the transmission environments change over time. The detection mechanism works in an iterative receiver that contains a pre-IN mitigation and a post-IN mitigation. The pre-IN mitigation is meant to null the stronger portion of IN, while the post-IN mitigation suppress the residual portion of IN by using an iterative process. We also present a target generator to train the Neural Network based IN detector in the training period. Simulation results show that the proposed Neural Network based IN detector can separate the IN from reconstructed noise, and hence improves PLC system performance