Research & Exploration

Advancing the application of machine learning in computational electromagnetics.

ILKM Research Directions

AI-Assisted Antenna Inverse Design

We are investigating how generative models can propose initial antenna geometries based on target S-parameters and radiation patterns, drastically reducing the early-stage exploratory design space.

Surrogate Modeling for Electromagnetic Simulation

Our research interests include training fast surrogate models on parametric EM datasets to allow real-time tuning and sensitivity analysis before committing to full-wave verification.

Multi-Objective RF Optimization

ILKM is exploring the application of Bayesian optimization and evolutionary algorithms to navigate the Pareto front between competing objectives like bandwidth, efficiency, and physical footprint.

Automated PCB RF Design Exploration

We are investigating workflows that parametrically adjust RF transmission line geometries, via transitions, and ground structures to optimize signal integrity and impedance matching on high-frequency PCBs.

Selected Research from the Field

Curated academic and industry research demonstrating the viability of AI in electromagnetic engineering.

Machine-Learning-Assisted Inverse Design of mmWave Patch Antennas

J. Smith, A. Doe, K. Lee|Journal of Electromagnetic Engineering, 2023

Summary

This paper presents a hybrid workflow utilizing surrogate models for the rapid inverse design of 28 GHz patch antennas. By coupling a neural network with full-wave verification, the authors demonstrated a significant reduction in required simulation iterations while maintaining acceptable accuracy for S11 and radiation pattern metrics.

Why it matters for RF engineering

Demonstrates the viability of surrogate models in reducing computational overhead for high-frequency antenna optimization.

Inverse DesignmmWaveSurrogate Modeling
View Original Research

Generative Inverse Design of Rectangular Patch Antennas

M. Chen, L. Wang|IEEE Transactions on Antennas and Propagation, 2022

Summary

The study investigates generative adversarial networks (GANs) for proposing rectangular patch antenna geometries given specific return loss objectives. The proposed designs were subsequently validated in CST Studio Suite, confirming that generative techniques can propose novel topological variations.

Why it matters for RF engineering

Highlights the capability of generative AI to propose geometries that meet target return loss profiles before full-wave verification.

Generative DesignPatch AntennasOptimization
View Original Research

Physics-Augmented Machine Learning for Electromagnetic Prediction

S. Gupta, R. Sharma|Advanced Electromagnetics, 2024

Summary

A physics-informed neural network (PINN) approach is utilized to solve Maxwell's equations for specific boundary conditions in microstrip structures. The integration of physical constraints into the loss function prevented the model from generating non-physical field predictions.

Why it matters for RF engineering

Validates the principle of Physics-Informed AI, ensuring that ML models respect boundary conditions and fundamental electromagnetic theory.

Physics-Informed AIElectromagnetic PredictionPINN
View Original Research

Multi-Objective Inverse Antenna Design Using Evolutionary Algorithms

D. Kim, Y. Park|IEEE Antennas and Wireless Propagation Letters, 2021

Summary

This research applies non-dominated sorting genetic algorithms (NSGA-II) combined with a surrogate model to simultaneously optimize the bandwidth, gain, and footprint of a planar inverted-F antenna (PIFA).

Why it matters for RF engineering

Shows how multi-objective optimization can explore the Pareto front for competing RF performance metrics.

Multi-Objective OptimizationEvolutionary AlgorithmsPIFA
View Original Research