Machine Learning Optimization Techniques in Microwave Design

Advances in machine learning have transformed the methodology of microwave‐design optimisation, enabling rapid and robust solutions to challenges that once demanded extensive manual tuning and computational resources. Central to this transformation is the deployment of surrogate models, which approximate expensive electromagnetic simulations through approaches such as Gaussian processes, kriging, co‐kriging and fully connected regression networks. These surrogates reduce evaluation cost, permitting global search algorithms—genetic, evolutionary, particle swarm and Bayesian optimisation—to identify parameter sets that satisfy stringent performance criteria. Concurrently, deep learning techniques, including feedforward neural networks, recurrent architectures and deep neural networks, have been harnessed for forward and inverse modelling of components, translating geometrical parameters into S‐parameter responses and vice versa. Generative models, such as generative adversarial networks and conditional GANs, have further extended the paradigm by synthesising novel antenna geometries that meet desired frequency, bandwidth and quality‐factor targets without exhaustive manual redesign. Feature‐based frameworks integrate response sensitivities into trust‐region schemes, accelerating convergence to near‐global optima while maintaining computational efficiency. Together, these methods have demonstrated practical impact across filters, multi‐band and ultra‐wideband antennas, frequency‐selective surfaces and passive microwave circuits, offering a flexible toolkit that balances exploration and exploitation. The interconnection of surrogate modelling, evolutionary search and deep generative techniques underlines a maturing discipline poised to deliver compact, high‐performance microwave devices for communications, radar and sensing applications worldwide.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Research from all publishers

A surrogate model‐assisted evolutionary algorithm has been introduced to tackle the global optimisation of microwave filters. By combining Gaussian‐process surrogates with differential evolution and local Gaussian search, this framework balances exploration of the design landscape against surrogate accuracy. It achieves high‐quality filter responses comparable to full global searches, while drastically reducing the number of full‐wave simulations required.

An extensive survey of artificial neural networks in microwave computer-aided design highlights the role of feedforward, recurrent and deep neural networks in both forward and inverse modelling of active and passive components. The review outlines methods for neuro‐transfer‐function modelling, knowledge-based enhancements, and deep architectures, showcasing applications from filter tuning to modelling of VLSI interconnects and high-electron-mobility transistors.

Generative adversarial networks have recently been applied to microstrip antenna synthesis, producing geometry and electromagnetic characteristic data that closely match target specifications. Conditional GANs enable user‐defined central frequencies and substrate parameters, yielding antenna designs with targeted resonance, bandwidth and efficiency metrics. This approach dramatically shortens the design cycle by generating high-fidelity synthetic training data for subsequent optimisation stages.

Machine Learning Optimization Techniques in Microwave Design publication trend

The graph below shows the total number of articles in machine learning optimization techniques in microwave design across all publications each year (not limited to Nature Index journals).

Technical terms

Surrogate model: A data‐driven approximation of computationally expensive electromagnetic simulations, used to predict device performance from input parameters.

Generative adversarial network (GAN): A deep learning framework comprising competing generator and discriminator networks that produce synthetic data matching real examples.

Deep neural network (DNN): A multilayer feedforward neural network capable of learning complex nonlinear relationships in high-dimensional data.

Gaussian process: A probabilistic model used for regression, providing both predictions and associated uncertainty estimates for design variables.

Evolutionary algorithm: A population-based optimisation method inspired by natural selection, employing genetic operators to explore complex parameter spaces.

S-parameters: Scattering parameters that characterise how radio-frequency signals are transmitted and reflected by microwave components.

References

1. Microstrip Antenna Design Supported by Generative Adversarial Networks. AI (2024).
2. Global Optimization of Microwave Filters Based on a Surrogate Model-Assisted Evolutionary Algorithm. IEEE Transactions on Microwave Theory and Techniques (2017).
3. Artificial Neural Networks for Microwave Computer-Aided Design: The State of the Art. IEEE Transactions on Microwave Theory and Techniques (2022).