Deep-Learning-Based Pixelated Microwave Filter Design and Characterization using Electro-Optical Electric-Field Measurements
Researchers have developed a novel deep learning method to automate the design of pixelated microwave filters, overcoming the limitations of traditional iterative tuning processes. This approach, which combines convolutional neural networks with genetic algorithms, was experimentally validated using both S-parameter and electro-optical electric-field measurements. The synthesized low-pass filter showed strong agreement between simulation and experimental results, with a 7 GHz passband and over 20 dB suppression beyond 9.5 GHz. Notably, the electro-optical measurements provided unprecedented insights into the electric field patterns of AI-generated designs, revealing structures akin to coupled transmission-lines. AI
IMPACT This research demonstrates AI's capability to accelerate and innovate in specialized engineering fields, potentially leading to more efficient and novel component designs.