Neuromodeling of microwave circuits exploiting space-mapping technology

What is it about?

For the first time, two powerful technologies for modeling RF and microwave circuits were combined in this work, namely: artificial neural networks and space mapping. In 1993 emerged methodological proposals for the effective utilization of artificial neural networks (ANN) in modeling of high-frequency circuits. The inherent properties of ANN made them very suitable to model linear and nonlinear RF and microwave circuits, in both frequency and time domains, following “black-box” approaches. On the other hand, in a totally separated manner, in 1994 emerged space mapping (SM) as an innovative computer-aided design (CAD) methodology for the automated and efficient optimization of RF and microwave circuits. SM intelligently exploits the accuracy of high fidelity but computationally expensive models (fine models), with the efficiency of computationally cheap but insufficiently accurate models (coarse models), yielding powerful algorithms for fast and accurate modeling and design optimization of high frequency circuits. For a number of years, these two CAD techniques evolved independently from each other, producing separated modeling and design optimization algorithms. It was until 1999 when Dr. Rayas-Sánchez et al. proposed combining artificial neural networks and space mapping for very efficient and highly accurate electromagnetics (EM)-based modeling of RF and microwave components.

Featured Image

Why is it important?

These neural space mapping approaches developed by Dr. Rayas-Sanchez can be essentially considered as knowledge-based neuronal techniques, however, neural space mapping significantly deviate from other ANN techniques that exploit prior knowledge, by focusing on approximating, through very simple neural networks, the mapping between the coarse and fine models mentioned above, making it very easy to implement in engineering practice.

Perspectives