Data-driven approaches for predictive modeling of N77/N78 band 5G microstrip patch array antenna using curve fitting...

Area of Science:

• Electrical Engineering
• Antenna Theory
• Machine Learning Applications

Background:

• The increasing demand for high-speed wireless communication necessitates efficient antennas for frequency bands like N77/N78.
• Designing antennas with desired characteristics, such as bandwidth and resonant frequency, can be complex and time-consuming.
• Traditional design methods often rely heavily on iterative simulations and physical prototyping.

Purpose of the Study:

• To design and validate a single-element and multi-element microstrip patch array antenna for N77/N78 band applications.
• To investigate the integration of machine learning (ML) algorithms for predicting key antenna parameters.
• To establish an effective antenna design methodology combining simulation, measurement, and ML prediction.

Main Methods:

• Design of microstrip patch array antennas on FR4 substrate.
• Electromagnetic simulation using CST Studio Suite and equivalent circuit modeling in Keysight ADS.
• Generation of 203 simulation-based data samples for ML model training.
• Training and evaluation of five supervised regression models (GB, ET, DT, RF, XGB) for parameter prediction.

Main Results:

• Good agreement was achieved between simulated and measured antenna characteristics.
• The Random Forest regression model demonstrated superior accuracy and lower error in predicting bandwidth and center resonant frequency.
• The developed ML models effectively estimated antenna parameters, reducing design iterations.

Conclusions:

• The proposed antenna design is suitable for N77/N78 band applications.
• The integration of full-wave simulation, measurement, equivalent circuit modeling, and ML-based prediction offers an effective and robust antenna design methodology.
• Machine learning significantly enhances the efficiency and accuracy of antenna design and parameter prediction.