Premium
A novel trust‐region–based algorithm with flexible Jacobian updates for expedited optimization of high‐frequency structures
Author(s) -
PietrenkoDabrowska Anna
Publication year - 2019
Publication title -
international journal of numerical modelling: electronic networks, devices and fields
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.249
H-Index - 30
eISSN - 1099-1204
pISSN - 0894-3370
DOI - 10.1002/jnm.2613
Subject(s) - jacobian matrix and determinant , sensitivity (control systems) , computer science , wideband , process (computing) , trust region , scheme (mathematics) , engineering design process , algorithm , antenna (radio) , mathematical optimization , computer engineering , electronic engineering , engineering , mathematics , telecommunications , mechanical engineering , mathematical analysis , computer security , radius , operating system
Simulation‐driven design closure is mandatory in the design of contemporary high‐frequency components. It aims at improving the selected performance figures through adjustment of the structure's geometry (and/or material) parameters. The computational cost of this process when employing numerical optimization is often prohibitively high, which is a strong motivation for the development of more efficient methods. This is especially important in the case of complex and multiparameter structures. In the paper, an expedited trust‐region–based algorithm for electromagnetic (EM)‐driven design optimization of high‐frequency structures is proposed. The presented technique involves a flexible sensitivity update scheme depending on the relative design changes with respect to the trust‐region size, as well as a direction of the design relocation and its alignment with the coordinate system axes. This allows for performing finite‐differentiation–based sensitivity updates less frequently and, consequently, brings considerable computational savings. Numerical results obtained for an ultra‐wideband antenna and a microwave coupler demonstrate that the proposed algorithm outperforms the reference procedure in terms of the number of EM simulations necessary to arrive at the optimized solution (around 50%). At the same time, the design quality loss is minor.