臺灣博碩士論文加值系統

在做生物訊號的辨識時，常常會遇到的一個問題就是自然界的雜訊，在雜訊的影響下，每次產生的生物訊號都會有所不同，以心跳為例，正常人每次產生的心跳訊號不會完全相同，但是我們不會因為這樣些微的差距，就判斷心跳是異常的，因此在做生物訊號的分類或辨識時，我們一定要把雜訊對它的影響一併考慮。擴散網路便提供了一套演算法，在式子中加入了雜訊因子，在學習的過程中因為一併考慮了雜訊對訊號的影響，所以當擴散網路習得一種訊號時，對雜訊的干擾將有一定程度的容忍力，此外擴散網路的另一個特點是考慮了時變項，這特色的優點是有機會達成即時的辨識，這對應用上來說是非常重要的一項優點。
不滿足於擴散網路的學習和辨識，我們更有興趣的是將擴散網路的理論實現成積體電路，它的好處在於攜帶的方便和運算的快速，試想如果今天有一種輕便的儀器能夠學習和辨識生物訊號，在醫療上將會是很大的幫助。在實現成積體電路的過程中，第一步是針對擴散網路中神經元的電路化著手，在這方面我們面臨的問題是如何從數學式子轉換成電路架構，以及如何從數值上的參數範圍轉換成電路上的值。當這些問題克服後，再來就是要探討神經元的電路系統在重建的過程中是不是符合數學上的運算。
最後讓這樣的電路實現成晶片系統，從晶片的量測上去探討真實電路和模擬上看到的誤差，以及解決的辦法，盼提出修正後的電路能夠達成擴散網路理論中訊號的重建。

To recognize biomedical signal, a problem that we often have to face is the noise in the real world. With the noise, biomedical signal will be different every time. For example, the heartbeat of one healthy person must be different a little every time, but we’ll not diagnose it is an abnormal heartbeat. So we have to consider the effect of the noise into the biomedical signal when we try to classify or recognize them. One kind of algorithm called Diffusion Network is suitable for this kind of condition. It includes the noise term in the equation. Once it learned the signal, the system can tollerance the noise interference in a certain extent. Another characteristic is the Diffusion Network considers the time-varying coefficient. The characteristic makes the system recognize the signal real time and it is a very strong point in the application.
Unsatisfying at the learning and recognition of the signal, we try to implement the Diffusion Network into VLSI technology, which will be portable and convenient. Thinking about it, if there is a portable instrument which can learn or recognize the biomedical signal, it is how helpful for the medical treatment. The first step is to make the neurons in the Diffusion Network be a ciruit system. The problem we face is how to transfrom these equations into circuit construction and what is the mapping of the parameter between the mathematical value and the value in the circuit. After overcome these questions, we try to reconstruct the signals with the circuit system, and then compare the result with the mathematical computation.
In the end, we implement the circuit into VLSI technology. From the chip testing, there are some errors between the chip and the simulation of the circuits. We discuss the reasons and try to modify them. After that, we hope the system can reconstruct the signals successfully.

Abstract...............................................ii
摘要...................................................iv
誌謝....................................................v
章節目錄...............................................vi
圖目錄.................................................ix
表目錄................................................xiv

第一章 內容介紹 1
1.1 研究動機與目標.................................1
1.2 研究結果貢獻...................................2
1.3 章節簡介.......................................3
第二章 相關文獻回顧 4
2.1 擴散網路模型...................................4
2.2 擴散網路學習理論...............................7
2.2.1 蒙地卡羅權重取樣法.............................7
2.2.2 蒙地卡羅期望值最大化學習法.....................9
2.3 擴散網路重建理論..............................12
2.4 擴散網路單一個神經元在超大型積體電路的實現....13
2.5 總結..........................................14
第三章 由數學模擬定出數值與電路間參數的對應關係 15
3.1 擬定擴散網路數學模擬時參數的範圍限制和初始值..15
3.1.1 數學模擬時參數的範圍限制......................15
3.1.2 數學模擬時參數的初始值........................18
3.2 用擴散網路學習人造訊號和生醫訊號..............20
3.2.1 學習不同頻率的弦波............................20
3.2.2 學習分支曲線..................................26
3.2.3 學習心電圖....................................29
3.2.4 學習螺旋訊號..................................32
3.2.5 學習希臘字母..................................35
3.3 擴散網路數學模擬的參數範圍和積體電路的對應關係37
3.4 總結..........................................41
第四章 擴散網路神經元的內部電路結構 42
4.1 電流乘法器電路架構............................42
4.2 電流傳送器電路架構............................49
4.3 可變電阻電路架構..............................53
4.4 電壓電流轉換器電路架構........................58
4.5 Sigmoid電路架構...............................62
4.6 補償電路架構和權重電流源電路架構..............66
4.7 電壓緩衝器電路架構............................70
4.8 雜訊產生器電路架構............................71
4.9 單一神經元的結構和外部控制訊號................78
4.10 總結..........................................80
第五章 利用擴散網路神經元電路系統模擬重建訊號 81
5.1 整個神經元網路的結構和外部控制訊號............81
5.2 由數學模擬取參數在電路上重建正弦波............85
5.3 由數學模擬取參數在電路上重建分支曲線..........94
5.4 由數學模擬取參數在電路上重建心電圖............96
5.5 由數學模擬取參數在電路上重建螺旋訊號..........99
5.6 由數學模擬取參數在電路上重建希臘字母.........101
5.7 總結.........................................103
第六章 實際積體電路上量測到擴散網路的特性 104
6.1 單一神經元內的電路區塊.......................104
6.1.1 可變電阻電路的量測...........................104
6.1.2 補償電路的量測...............................107
6.1.3 Sigmoid電路的量測............................108
6.1.4 電流乘法器、權重電流源和電流傳送器的量測.....115
6.2 總結.........................................120
第七章 結論 121
7.1 研究總結.....................................121
7.2 研究未來發展方向.............................122
參考文獻 124

Bibliography

[1] J. R. Movellan, P. Mineiro, and R. J. Williams, "A Monte Carlo EM approach for partially observable diffusion processes: Theory and applications to neural networks," Neural Computation, vol. 14, no. 7, pp. 1507-1544, 2002.
[2] H. Chen and A. F. Murray, "Continuous restricted Boltzmann machine with an implementable training algorithm," IEE Proceedings-Vision Image and Signal Processing, vol. 150, no. 3, pp. 153-158, 2003.
[3] H. Chen, P. C. D. Fleury, and A. F. Murray, "Continuous valued probabilistic behavior in a VLSI generative model," IEEE Transactions on Neural Networks, vol. 17, no. 3, pp. 755-770, 2006.
[4] Y.-S. Hsu, "Biomedical Signal Recognition Using Diffusion Networks." Master's Thesis of National Tsing Hua University,Taiwan, 2007.
[5] M.-R. Chu, "Design of the Neuron Circuit for the Diffusion Network in VLSI." Master's Thesis of National Tsing Hua University,Taiwan, 2006.
[6] S. A. Wartenberg and J. R. Hauser, "Substrate voltage and accumulation-mode MOS varactor capacitance," Electron Devices, IEEE Transactions on, vol. 52, no. 7, pp. 1563-1567, 2005.
[7] K. Tanno, O. Ishizuka, and Z. Tang, "Four-quadrant CMOS current-mode multiplier independent of device parameters," IEEE Transactions on Circuits and Systems Ii-Analog and Digital Signal Processing, vol. 47, no. 5, pp. 473-477, 2000.
[8] C Toumazou, FJ Lidgey, and D Haigh, Analogue IC design: the current-mode approach , U.K.: Peter Peregrinus, 1990.
[9] S. F. Al-Sarawi, "A novel topology for grounded-to-floating resistor conversion in CMOS technology," Microelectronics Journal, vol. 33, no. 12, pp. 1059-1069, 2002.
[10] G. Cauwenberghs, "An analog VLSI recurrent neural network learning a continuous-time trajectory," IEEE Transactions on Neural Networks, vol. 7, no. 2, pp. 346-361, 1996.
[11] F. Diotalevi, M. Valle, G. M. Bo, E. Biglieri, and D. D. Caviglia, "Analog CMOS current mode neural primitives,", 2 ed 2000, pp. 717-720.
[12] Shih-Chii Liu, Jorg Kramer, Giacomo Indiveri, Tobias Delbriick, and RodneyDouglas, Analog VLSI: Circuits and Principles ,MA: MIT Press, 2002.
[13] PR Gray and RG Meyer, Analysis and Design of Analog Integrated Circuits, 4th ed ,New York: Wiley, 1990.
[14] G. Cauwenberghs, "Delta-sigma cellular automata for analog VLSI random vector generation," Circuits and Systems II: Analog and Digital Signal Processing, IEEE Transactions on, vol. 46, no. 3, pp. 240-250, 1999.
[15] T. Serrano-Gotarredona and B. Linares-Barranco, "Log-domain implementation of complex dynamics reaction-diffusion neural networks," Neural Networks, IEEE Transactions on, vol. 14, no. 5, pp. 1337-1355, 2003.