IEEE Transactions on Microwave Theory and Techniques · 1999 · 250 citations · 9 references
EngineeringElectrical EngineeringComputational ElectromagneticsMicroelectronicsMicrowave EngineeringDevice ModelingMicrowave FiltersMicrowave Integrated CircuitsMicrowave Network AnalysisMicrowave Semiconductor DevicesNeural NetworksLarge-signal Device ModelingMicrostrip Circuit AnalysisMicrostrip Circuit ModelingMicrowave Circuit ModelingMicrowave Circuit SimulationMicrowave ModelingNeural Network ModelingSmall-signal Device Modeling
The study introduces space‑mapping‑based neural network models for microwave circuits. The authors propose five techniques to build SM‑based neuromodels, including frequency‑sensitive neuromapping and Huber‑optimized training, and compare them to classical and state‑of‑the‑art methods using a microstrip bend and a high‑temperature superconducting filter as demonstrations. These models lower training cost, improve generalization, and simplify network topology versus classical approaches.
For the first time, we present modeling of microwave circuits using artificial neural networks (ANN's) based on space-mapping (SM) technology, SM-based neuromodels decrease the cost of training, improve generalization ability, and reduce the complexity of the ANN topology with respect to the classical neuromodeling approach. Five creative techniques are proposed to generate SM-based neuromodels. A frequency-sensitive neuromapping is applied to overcome the limitations of empirical models developed under quasi-static conditions, Huber optimization is used to train the ANN's. We contrast SM-based neuromodeling with the classical neuromodeling approach as well as with other state-of-the-art neuromodeling techniques. The SM-based neuromodeling techniques are illustrated by a microstrip bend and a high-temperature superconducting filter.
9