A deep neural network based tuning technique of lossy microwave coupled resonator filters. Issue 9 (13th May 2019)

• Record Type: Journal Article
• Authors: Sun, Jia‐Jing; Sun, Sheng; Yu, Xi; Chen, Yongpin P.; Hu, Jun
• Is Part Of: Microwave and optical technology letters. Volume 61:Issue 9(2019:Sep.)
• Page Start: 2169
• Page End: 2173
• Publication Date: 2019-05-13
• Subjects: computer‐aided tuning (CAT) -- coupling matrix -- machine learning -- substrate integrated waveguide -- synthesis; Microwaves -- Periodicals; Optics -- Periodicals; 621
• DOI: 10.1002/mop.31866
• Languages: English
• ISSNs: 0895-2477

Abstract

Abstract: A novel tuning process of microwave filter is proposed based on deep neural network. Convolutional neural networks (CNNs) with several layers of convolution and nonlinear units are introduced to extract the coupling matrix of the simulated responses during the design procedure of lossy microwave filters. Then the proposed neural network model has been used to guide the tuning process of substrate integrated waveguide (SIW) filters. Compared to the traditional tuning method, this method can extract the coupling matrix with high accuracy as well as providing directions for the tuning rapidly. As an example, a fourth‐order bandpass filter centered at 20.45 GHz with 1.72‐dB insertion loss and 21.49‐dB return loss is finally designed using the proposed model. Numerical example validates the effectiveness of the proposed method.