Coin-Sized Dual-Band Millimeter-Wave (mmWave) Antenna with Machine-Learning-Guided Impedance Prediction
Ahmed Jamal Abdullah Al-Gburi
This study suggests a coin-sized (10 × 8 × 0.64 mm3) millimetre-wave antenna that simultaneously resonates at 28 GHz and 38 GHz and is supported by a machine-learning surrogate for near-instant impedance evaluation. Realised on Rogers 6010 LM laminate (εr = 10.2), the radiator maintains |S11| ≤ -10 dB across 26.5-29.9 GHz and 37.2-39.7 GHz while providing peak gains of 3.8 dBi and 4.1 dBi in the lower and upper bands, respectively. A design-of-experiments sweep, comprising 330 full-wave simulations, generated the training corpus for a random-forest regression model. The surrogate predicts frequency-resolved |S11| with a mean-absolute error below 0.7 dB and coefficients of determination of 0.93 at 28 GHz and 0.84 at 38 GHz. The evaluation time is reduced from approximately 155 s per full-wave electromagnetic simulation to 0.1 s per surrogate query, enabling real-time design exploration. Eight-fold cross-validation confirms model stability, while feature-importance analysis identifies the geometric parameters most influential to dual-band matching. The learning-guided workflow therefore offers a fast and reliable alternative to exhaustive simulation, accelerating the optimisation of compact mmWave antennas for instrumentation, sensing, and future front-end modules.
