臺灣博碩士論文加值系統

傳統上，一些模糊類神經網路系統都使用倒傳遞演算法(Back propagation, BP)調整其系統參數，但是不良的初始值往往會導致演算法收斂到局部最佳解（Local Minimum）。故，好的初始值是很重要的。而在模糊類神經網路系統結構的設計上，模糊規則數量會直接影響到系統的效能，在系統學習期間，若該數量保持不變，則可能導致模糊規則數不適當之情形產生。
為了解決BP法收斂到局部最佳解以及模糊類神經網路系統的結構設計問題，我們提出一種新型自我組織模糊類神經網路系統，它可根據輸入資訊自動建立系統結構以及設計初始值，再用BP訓練系統參數以達最佳效果。
最後，我們將提出的演算法，應用在非線性時變通道決策回授等化器上，並和傳統模糊類神經網路系統架構下的等化器做比較。從模擬結果可看出其有效降低位元錯誤率(Bit Error Rate, BER)及均方誤差收斂值(Mean Square Error, MSE)。

Conventionally, a fuzzy neural network (FNN) system may adopt back-propagation (BP) learning algorithm to adjust parameters. An improper initial value in BP may lead to local minimum. Therefore, initial value selection is very important for BP. Furthermore, on structure design of FNN, the fuzzy rule numbers may affect the performance.
Specially, if the number of fuzzy rules keeps unchanged during learning, which is prone to bring a deficiency or redundancy of fuzzy rules.
To overcome the local minimum problem and structure design of a FNN system, we propose a novel self-organizing fuzzy neural network (SOFNN) system, which establishes the structure and obtain the initial values of the system automatically. Then the BP is used to optimize the parameters of a FNN system.
Finally, the proposed algorithm is applied to decision feedback equalizer (DFE) of nonlinear time-varying channels for comparison with the conventional FNN. The simulation results show that SOFFN-based DFE can reduce both the BER and MSE effectively.

摘 要 i
Abstract ii
目 錄 iv
圖 目 錄 vi
表 目 錄 viii
第一章 緒論 1
1-1 前言 1
1-2 調適性濾波器 2
1-3 調適性等化器 4
1-4 本篇論文組織 6
第二章 模糊類神經網路 7
2-1 類神經網路 7
2-2 決策回授等化器 9
2-3 模糊系統 11
2-4 模糊類神經網路 14
2-5 模糊類神經網路等化器 17
第三章 學習演算法 20
3-1 競爭式學習演算法 20
3-2 自我組織學習演算法 26
3-2-1 結構學習演算法 27
3-2-2 參數學習演算法 30
第四章 模擬結果與分析 36
4-1 非時變通道的模擬結果 36
4-1-1 收斂特性-均方誤差收斂值： 39
4-1-2 位元錯誤率： 40
4-1-3 複雜度： 42
4-2 時變通道的模擬結果 43
4-2-1 收斂特性-均方誤差收斂值： 47
4-2-2 位元錯誤率： 48
4-2-3 複雜度： 49
第五章 結論 50
參考文獻 51

[1]S. Siu, C.-L. Ho and C.-M Lee, “TSK-based Decision Feedback Equalizer Using an Evolutionary Algorithm Applied to QAM Communication Systems,” IEEE Trans. Circuits Syst. II-Express Briefs, Vol.52, no.9, pp.596-600, Sept. 2005.

[2]C.-H. Lee and Y.-C. Lin, “An adaptive neuro-fuzzy filter design via periodic fuzzy neural network,” Signal Process., Vol.85, no.2, pp.401-411, Feb. 2005.

[3]H. Song, C. Wang, Y. He, S. Ma and J. Zuo, “Decision feedback equalizer based on non-singleton fuzzy regular neural networks,” Journal of Systems Engineering and Electronics, Vol.17, no.4, pp.896-900, Dec. 2006.

[4]S. Haykin, Adaptive filter theory, 3rd Edition, Prentice-Hall, Englewood Cliffs, NJ, 1996.

[5]S. Siu, G.J. Gibson and C.F.N. Cowan. “Decision feedback equalization using neural network structures and performance comparison with standard architecture,” IEEE proceedings, Vol.137, Part I, no.4, pp.221-225, Aug. 1990.

[6]蘇木春, 張孝德, “機器學習：類神經網路、模糊系統以及基因演算法則,” 二版, 全華科技圖書股份有限公司, 台北市, 民國95年

[7]L. Xu, A. Krzyzak and E. Oja, “Rival Penalized Competitive Learning for Clustering Analysis, RBF Net, and Curve Detection,” IEEE Trans. on Neural Networks, Vol. 4, no. 4, pp. 636-649, Jul. 1993.

[8]Y. M. Cheung, “On rival penalization controlled competitive learning for clustering with automatic cluster number selection,” IEEE Trans. on Knowledge and Data Engineering, Vol.17, no.11, pp.1583-1588, Nov. 2005.

[9]J. Ma and T. Wang, “A cost-function approach to rival penalized competitive learning (RPCL),” IEEE Trans. Syst., Man., Cybern.—part B, Vol. 36, no.4, pp. 722-737, Aug. 2006.

[10]R.-C. Lin, W.-D. Weng and C.T. Hsueh. “Design of an SCRFNN-based nonlinear channel equaliser,” IEE Proc.-Commun., Vol. 152, no.6, pp. 771-779, Dec. 2005.

[11]S. S. Yang, S. Siu, and C. L. Ho, “Analysis of the initial values in split-complex backpropagation algorithm,” IEEE Trans. on Neural Netw., Vol.19, no.9, pp. 1564-1573, Sept. 2008.

[12]Q. Liang and J. M. Mendel, “Equalization of nonlinear time-varying channels using type 2 fuzzy adaptive filters,” IEEE Trans. Fuzzy Syst., Vol.8, no.8, pp. 551-563, Oct. 2000.

[13]S. Siu, S. S. Yang, C. M. Lee, and C. L. Ho, ”Improving the Back-Propagation Algorithm Using Evolutionary Strategy,” IEEE Trans. Circuits Syst. II-Express Briefs, Vol.54, no.2, pp. 171-175, Feb. 2007.