Optimization in surrogate model building for RF circuit blocks.
L. De Tommasi (University of Antwerp), D. Gorissen (Ghent University, Belgium), J. Croon (NXP Semiconductors, Eindhoven, The Netherlands) and T. Dhaene (Ghent University, Belgium)
Surrogate models have become a cost effective alternative for replacing complex highly accurate computer simulations when exploring the design space, performing what-if analysis, optimization, sensitivity analysis. Crucial aspects of global surrogate modeling are model type selection [1] (rational functions, radial basis functions, support vector machines, kriging, artificial neural networks, splines…), adaptive sampling [2] (error-based, density-based and gradient based methods) and adaptive modeling [3] (model-parameters optimization via genetic algorithm, pattern search, gradient descent, particle swarm optimization, direct optimization…).
In this talk the most recent results concerning the surrogate modeling of simple RF circuit blocks are presented. Different optimization algorithms are compared on different model types and adaptive sampling strategies.
[1] D. Gorissen, L. De Tommasi, J. Croon, T. Dhaene, “Automatic Model Type Selection with Heterogeneous Evolution: An application to RF circuit block modeling”, submitted to IEEE WCCI 2008.
[2] D. Gorissen, L. De Tommasi, J. Croon, T. Dhaene, “Compact modeling of RF circuit blocks via Kriging surrogates”, submitted to MIKON 2008.
[3] K. Crombecq, D. Gorissen, L. De Tommasi, J. Croon, T. Dhaene, “A comparison of sequential design methods for RF circuit block modelling”, submitted to ICCS 2008.