Artificial intelligence for microwave circuit simulations
Loading...
Date
2017-06
Authors
Sam, Kah Seng
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
For the past decades, the Artificial Neural Networks (ANNs) have emerged as a popular
tool for analysing the various type of parameters in radio frequency (RF) and microwave
device modelling and designing. This type of analysis has been shown to be fast and
accurate, both theoretically and experimentally from the past work. In today high-speed
digital world delay lines and signal integrity plays an important role in microwave devices.
In this work, serpentine lines or meander lines will be investigated base on their
correlation between the propagation delay and the physical parameters of the microstrip
lines. Some of the critical parameters to be considered while designing the microstrip line
are serpentine line width, length, spacing, the number of bends and types of bends. These
parameters will be the input-target pairs of data to be fed to the neural network for training
session and validation purpose later on. The Momentum simulator in Advanced Design
System (ADS) will be used to simulate the meander lines as it is an electromagnetic solver,
to get the S-parameters and generating layout. The S-parameters generated will be used
to create the transient modelling which can determine the propagation delay in meander
lines. MATLAB is used in this research to create the desired neural network model for
propagation delay prediction. Approximation method is employed to find the best number
of hidden neurones which can optimise the performance of the neural network in term of
the training speed and the accuracy. Finally, both the ADS and ANN results for simulated
delay times of meander lines are compared to validate the performance and to justify the
exactitude of proposed analysis. The results indicate that the ANN achieves an accuracy
of above 0.995 (with a reference of 1.0) with a training time of less than a second.