基於人工神經網路的太赫茲波導帶通濾波器設計

Abstract

本論文旨在研究太赫茲波導濾波器的設計，並採用電腦數值控制銑削技術製造出波導濾波器，最後通過量測結果以證明設計的可行性。本論文主要分為兩部分。第一部分是濾波器的基本理論。在本節中吾人分別以耦合矩陣理論與阻抗轉換器理論設計兩個工作於WR-3頻段(220 – 330 GHz)的非對稱電感性虹膜耦合的6階柴比雪夫波導濾波器。第一個濾波器的中心頻率為300 GHz，比例頻寬為13.3%；第二個則是基於IEEE 802.15.3d標準中的CH67 (278.64 – 304.56 GHz)所訂定。而第二部分則是人工智慧。在此節中吾人由訓練完成的人工神經網路來獲得所目標響應下的濾波器幾何尺寸。並且為了進一步提高其訓練速率，吾人先將濾波器的頻率響應做向量擬合以轉為對應的極點與殘值進而減少網路輸入個數。訓練結果表明濾波器尺寸的均方誤差接近電腦數值控製銑削的製作誤差。 量測結果表明300 GHz波導濾波器其在281 – 324 GHz的通帶內穿透損耗小於5 dB，反損耗優於13 dB。CH67濾波器在276 – 310 GHz的通帶內穿透損耗小於3 dB，反損耗優於17 dB。由於這兩個濾波器的性能並不完美，所以吾人作了後模擬以分析探討影響通帶內的穿透損耗的原因，以便在後續的設計中進行修正與改進。

The purpose of this thesis is to study terahertz waveguide filter designs. The designed waveguide filters are fabricated by the computer numerical controlled milling process and then measured to show the feasibility. The thesis is mainly divided in to two parts. The first part is the basic theory of the microwave filters. In this section, two asymmetric inductive iris coupled waveguide bandpass filters with 6-pole Chebyshev responses are designed by coupling matrix theory and impedance inverter theory, respectively. The first filter has a center frequency of 300 GHz and a fractional bandwidth of 13.3%; the second filter is designed to apply to Channel 67 of the IEEE 802.15.3d standard (278.64 – 304.56 GHz). The second part is the application of artificial intelligence to filter designs. In this section, an artificial neural network is used to provide the filter geometries using the desired frequency response. In order to improve the training speed, we first use the vector fitting technique to convert the frequency response into the corresponding poles and residues so that we can reduce the number of the network inputs. The results show that the mean squared error of the filter dimensions is close to the fabricating error of computer numerical controlled milling process. The measurement results show that the insertion loss of the 300 GHz filter is less than 5 dB and the return loss is better than 13 dB in the range 281 – 324 GHz. The insertion loss of the CH67 filter is less than 3 dB and the return loss is better than 17 dB in the range 276 – 310 GHz. Since the performances of these two filters are not perfect, I perform the post-simulation to analyze the causes of the insertion loss in the passband, in order to improve our future designs.