.png){kind=icon}
.png){kind=icon}
Return
Transfer Learning for Accelerating Microwave Filter Design
.png){kind=icon}
PRE
AI
DOI：10.1109/NEMO59437.2024.11395790.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This research introduces an innovative method for designing microwave filters through transfer learning with convolutional neural network (CNN) models. Beginning with training on large coarse datasets and transitioning to detailed fine datasets, this approach enhances design parameter (DP) values for optimizing microwave filter designs. Demonstrating its efficacy, the application of this method to a 5-pole hairpin microstrip filter is detailed, highlighting the CNN models' ability to adapt and efficiently extract critical parameters. The methodology delivers rapid, accurate, and reliable DP values, tailored to different center frequency and percentage bandwidth specifications of the same filter structure, while showcasing the potential for significant computational savings. This approach yields substantial advancements in the development of tunable filters, effectively merging speed with precision in practical applications.
Keywords:
Bandpass filter
convolutional neural network
machine learning
microwave filters
optimization
space mapping