頁籤選單縮合
題 名 | Neural Least-Squares Design of FIR Digital Filters with Trigonometric Properties=類神經最小平方結合三角函數之有限脈衝響應數位濾波器設計 |
---|---|
作 者 | 蘇樂群; 周裕達; 陳福坤; | 書刊名 | 黃埔學報 |
卷 期 | 54(工程科學類) 2008.03[民97.03] |
頁 次 | 頁103-110 |
分類號 | 448.533 |
關鍵詞 | FIR濾波器; 硬體實現; 霍普菲爾類神經網路; 即時; FIR filter; Hardware implementation; Hopfield neural network; Real-time; |
語 文 | 英文(English) |
中文摘要 | 本論文提出一有效率利用類神經網路來實現最小平方FIR濾波器的設計,此方法是將濾波器設計的最佳化問題轉換成類神經網路之Lyapunov能量函數,使的濾波器設計可以應用霍普菲爾類神經網路來實現,當最佳化問題可以透過類神經網路的相關連接權重與偏壓參數計算後,再透過動態方程式就可以得到濾波器的最佳係數。此外,本論文更針對類神經網路參數所具有之特性,透過三角函數之化簡,使得參數的計算量可以大幅降低,提高濾波器設計之效能,此方法將可適合透過硬體實作來達到即時處理之效能。 |
英文摘要 | In this paper, an efficient structure with neural network implementation is developed for the least-squares design of FIR filters. The method is based on formulating an error function between the designed and desired frequency response for the optimization of filter design as a Lyapunov energy function by employing Hopfield neural network. Once these Hopfield parameters are obtained, the optimal filter coefficients can be achieved by the dynamic non-linear equations of the network. Furthermore, the expressions with closed-form for the parameters of Hopfield related are carried out by some useful properties. Therefore, the proposed technique can achieve the least computation required. It is shown that the design results compare favorably with those obtained by using previous neural-based technique. The proposed technique has the advantages of high computational efficiency and suitable for hardware implementation for the real-time processing purpose. |
本系統中英文摘要資訊取自各篇刊載內容。