Background
AI/ML based network synthesis is a new field in RF/microwave network design. Previously, Deep Learning algorithms have been used to design dual band patch antennas which was not possible beforehand. In a recent effort, Deep learning was used to design matching networks for a power amplifier with operating frequency range of 30-100 GHz. With the use of Neural Network design, it is possible to find solutions which are not possible with traditional matching techniques as the set of possibilities for geometric shapes goes beyond our classic methods.
Project Goals
In this project we aim to identify and explore the relevant algorithms for network synthesis and deployment with common circuit solvers. We will synthesize matching networks with AI algorithms in python and implement the resulting circuit in ADS or AWR. The circuit will be fabricated and measured in the lab.
Project Stages
- literature review
- study of various algorithms
- study of deep learning based convolutional neural network EM based emulator
- simulation (ADS, AWR) of a diplexer at Ku frequencies
- circuit fabrication
- measurements : efficiency, noise, s-parameters
- characterization and comparison with conventional techniques
Project Duration : One semester with the option of extension to an additional semester
- Good programming capabilities (knowledge in Python – advantage)
- Optimization/Learning courses – advantage
RFIC/Analog circuit design courses – advantage
Instructors
Dan Fishler (dan.fishler@campus.technion.ac.il)