Artificial neural network for crosstalk prediction in stripline transmission lines
Loading...
Date
2018-06
Authors
Kong, Chun Lei
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Crosstalk can cause serious electromagnetic interference problem and crosstalk
prediction in the early design stage is important. Several conventional modeling methods
such as RDSI and SPICE have previously presented to predict crosstalk in non-uniform
transmission lines and it needs large CPU memory consumption and long simulation time.
DOE is applied to efficiently select training data and reduce the number of EM
simulations in the Advanced Design System (ADS). Momentum EM Simulator is used to
extract S-parameters from coupled stripline with different design parameters and
generated an efficient dataset. Matlab Neural Network Toolbox is used to create neural
network models. Neural network models are trained to learn the characterization and
behavior of data for crosstalk estimation in stripline. Lastly, the neural model is validated
by comparing the simulated results and predicted results from ADS and ANN. The
performance evaluation shows that the crosstalk prediction in stripline achieved 99.9%
with training time of 0.2810s. In conclusion, this verified that the ANN is effective in the
stripline crosstalk prediction.