Abstract
在這篇論文中，一種最佳雜訊消除過程是被使用於分析合成濾波器銀行系統中以便可以在子頻帶雜訊之下達成最佳訊號重建。利用等效方塊表示子頻帶訊號，我們使用狀態空間模式來處理這 M個頻帶附有子頻帶雜訊的濾波器銀行系統。綜合輸入訊號、分析濾波器和縮減器效力可表示成一種多變率狀態空間模式。這子頻帶雜訊可能是由編碼推演量化而產生或加上傳輸路徑上的雜訊所組成。我們考慮將子頻帶雜訊加入多變率狀態空間模式當做外來的雜訊，因此在 M個頻帶濾波器銀行系統中，子頻帶雜訊衰減問題變成了一種狀態估測過程在合成的多變率狀態空間模式中。因此這多變率卡門濾波平滑推演法能夠被推導出來，根據等效狀態空間模式以便能在子頻帶雜訊之下濾波器銀行系統中達成最佳訊號重建。基於最佳化估測理論，使用多變率卞門濾波過程以便成輸入訊號的最小變異數重建。有幾個數值範例也包括在這篇論文中。從模擬的結果中，我們可以發現在子頻帶雜訊濾波器銀行系統中有明顯地改進。除此以外，考慮如何改進頻譜在濾波器銀行中，因為有些訊號，如語音訊號等，需要有好的頻譜，我們利用分離每個頻道，獨立處理每個頻道，然後在在配合最佳化估測理論和卡門濾波器以得到較好的頻譜。而分離處理每個頻道的方法可得到較好的頻譜而性能差不多。In this paper, an optimal noise cancellation pr- ocedure isequipped in analysis/synthesis filter bank systems to achieveoptimal signal reconstruc- tion under subband noise. Based onan equivalent block representation of subband signals, a state-space model is introduced for the M-band filter bank systemswith subband noises. The composite effect of the input signals,analysis filter bank and decimators are represented by amultirate st- ate-space model, and the subband noises due toqu- antization from coding algorithm and/or corruption intransmission path is generally considered as an additive noisein the multirate state-space model. Hence the subband noisesattenuation problem in the M-band filter bank systems becomes astate estima- tion procedure in the resultant multirate state-s- pace model. Therefore the multirate Kalman filter- ing/smoothing algorithm can be derived according to a equivalentmultirate state-space model to achieve the optimal signalreconstruction in the analysis/ synthesis filter bank systemsunder subband noises. Based on the optimal estimation theory,the proposed multirate Kalman filtering procedure provides theminimum variance reconstruction of the input signals. For thebetter spectrum, we can use the band by band Kalman filters/smoothers and there are some discussion in this paper. A few ofnumerical examples are also included. From the simulationresults, we have found that the performance improvement ofsignal reconstruc- tion in noisy filter bank systems areremarkable.