本論文探討雜訊的性質，是否符合一般分類演算法的基本假設「正規分布」，並且從此筆雜訊中擷取的雜訊，是否符合獨立。若雜訊較符合正規分布，則我們使用的分類方法較有效能。我們可經由雜訊直方圖的觀察，探討其正規性。並從雜訊的直方圖中發現異常突起的特性，且此異常突起大比例均勻散佈於雜訊中。由其他實驗雜訊的比較，發現異常的雜訊可能為不同儀器設備所致，或是由於電極接收到其他微弱訊號。 在頻率域上的表現，發現頻譜上的異常突起，就是造成直方圖異常突起的原因。另外，我們發現雜訊在頻率域上的能量呈現穩定起伏，於是本論文提出一種新的濾波方式，使得雜訊更符合正規分布，獨立性也更好。另一方面，在真實的神經訊號分類中，利用雜訊在頻率域上能量的穩定起伏，對神經訊號做濾波，發現濾波後會使得分類程式對於 PCA 散佈圖有部分重疊的情形分類較佳。 This paper probes into property of noise and discusses whether its distribution is the 'normal distribution' which is generally the basic assumption of algorithm of classification. And getting noise from it, whether its distribution is independent and identically distributed. If the distribution of noise is normal, the classifications are relatively efficient. In addition, by observing the histogram of noise and its normalization model, we find abnormality, unformly scattering with large percentage near the center. By comparing with noise of other experiments, we conclude that abnormal noise in the histogram may be resulted from the instruments or just the other weak signals received by the electrode . When observing the representation of noise in the frequency domain, we find irregularity in the frequency domain which causes abnormality in the histogram in the time domain. Moreover, we find the power of noise shows steady wave in the frequency domain. Accordingly, we propose a kind of new filtering method, to make the distribution of noise conform to the normal distribution and make it more independent than the original one. On the other hand, in real nerve signals, the classification will be improved by a fillering based on the fact that the power of noise shows steady wave in the frequency domain. After filtering, the classification is better than that by PCA scattering plot.
