Design of a Triple-Band Microstrip Patch Antenna Using Machine Learning-Assisted Prediction for Sub-6 GHz Applications

Abstract The increasing demand for compact multiband antennas in 5G/6G communications, cognitive radio, and other modern wireless communication systems requires efficient and intelligent antenna design approaches. Conventional antenna optimization methods based on full-wave electromagnetic simulations are time-consuming and computationally expensive due to repeated design iterations. Machine learning (ML)-assisted methods provide a faster prediction of optimal design/geometry parameters for efficient multiband anttena for targeted Sub-6 GHz applications. Multiband microstrip patch antennas are attractive for compact sub-6~GHz radios, but placing several resonances with full-wave simulation alone is slow. Published ML based studies mostly address one or two bands. They rarely test whether the simulated antenna meets a matching requirement at the required centre frequencies, or whether the training data are themselves consistent. This paper presents an ML-assisted workflow for a compact dual-slot, inset-fed microstrip antenna with a defected ground structure on FR-4. It targets the 2.4~GHz ISM, 3.5~GHz 5G n78 and 5.8~GHz ISM bands. A Latin Hypercube plan of 805 geometries over five parameters was simulated in CST Studio Suite, giving 630 usable samples with 15 outputs each. A first campaign was discarded after a consistency check showed unrelated outputs for identical geometries. Random forest (RF), XGBoost, and a neural network (ANN) MODELS are trained and compared, with and without electrical-length features. On 95 held-out designs, XGBoost with engineered features reached the highest mean $R^2$ across all fifteen outputs (0.72). This compares with 0.70 for the raw ANN, 0.68 for raw XGBoost and 0.67 for RF. The surrogates also separated passing from failing designs well, with an AUC from 0.93 for the raw ANN to 0.99 for the engineered-feature ANN. Even so, the difference between XGBoost and the neural network was within statistical noise. A surrogate-guided search followed by 40 verification simulations produced a design that meets $S_{11}\le-10$~dB at all three target frequencies ($-11.4$, $-21.5$ and $-11.6$~dB), where the baseline meets none. Six of eight tolerance variants still met all three targets. Gain, efficiency, VSWR, coverage of the standard bands and radiation patterns are reported for the final design. The design is narrow-band with modest gain and efficiency, and all results are from simulation.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-13
DOI
https://doi.org/10.21203/rs.3.rs-11352777/v1
Primary Topic
Antenna Design and Analysis
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Design of a Triple-Band Microstrip Patch Antenna Using Machine Learning-Assisted Prediction for Sub-6 GHz Applications

Srinu Sesham, Clement Temaneh Nyah, Pauly T. B. Halweendo
4 links
Research Square
Antenna Design and Analysis
preprint

Design of a Triple-Band Microstrip Patch Antenna Using Machine Learning-Assisted Prediction for Sub-6 GHz Applications

Srinu Sesham, Clement Temaneh Nyah, Pauly T. B. Halweendo
preprint en

Abstract

Abstract The increasing demand for compact multiband antennas in 5G/6G communications, cognitive radio, and other modern wireless communication systems requires efficient and intelligent antenna design approaches. Conventional antenna optimization methods based on full-wave electromagnetic simulations are time-consuming and computationally expensive due to repeated design iterations. Machine learning (ML)-assisted methods provide a faster prediction of optimal design/geometry parameters for efficient multiband anttena for targeted Sub-6 GHz applications. Multiband microstrip patch antennas are attractive for compact sub-6~GHz radios, but placing several resonances with full-wave simulation alone is slow. Published ML based studies mostly address one or two bands. They rarely test whether the simulated antenna meets a matching requirement at the required centre frequencies, or whether the training data are themselves consistent. This paper presents an ML-assisted workflow for a compact dual-slot, inset-fed microstrip antenna with a defected ground structure on FR-4. It targets the 2.4~GHz ISM, 3.5~GHz 5G n78 and 5.8~GHz ISM bands. A Latin Hypercube plan of 805 geometries over five parameters was simulated in CST Studio Suite, giving 630 usable samples with 15 outputs each. A first campaign was discarded after a consistency check showed unrelated outputs for identical geometries. Random forest (RF), XGBoost, and a neural network (ANN) MODELS are trained and compared, with and without electrical-length features. On 95 held-out designs, XGBoost with engineered features reached the highest mean $R^2$ across all fifteen outputs (0.72). This compares with 0.70 for the raw ANN, 0.68 for raw XGBoost and 0.67 for RF. The surrogates also separated passing from failing designs well, with an AUC from 0.93 for the raw ANN to 0.99 for the engineered-feature ANN. Even so, the difference between XGBoost and the neural network was within statistical noise. A surrogate-guided search followed by 40 verification simulations produced a design that meets $S_{11}\le-10$~dB at all three target frequencies ($-11.4$, $-21.5$ and $-11.6$~dB), where the baseline meets none. Six of eight tolerance variants still met all three targets. Gain, efficiency, VSWR, coverage of the standard bands and radiation patterns are reported for the final design. The design is narrow-band with modest gain and efficiency, and all results are from simulation.

Research Square
University of Namibia (NA)
Industry, innovation and infrastructure
Antenna Design and Analysis
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon