Supply‐Chain‐Ready 915 MHz Radio‐Frequency Identification Meander Dipole

ABSTRACT

This paper presents a physics‐aware, spec‐driven machine‐learning (ML) optimization framework for synthesizing printable ultra‐high frequency (UHF) radio‐frequency identification (RFID) meandered dipoles targeting the heavily regulated 902–928 MHz operational corridor. Evaluated under a strictly capped 300‐solve computational budget within a full‐wave 3D EM environment, the proposed methodology—guided by a PCA‐hybrid surrogate and dynamic out‐of‐distribution (OOD) penalties—is comprehensively benchmarked against three solver‐integrated baselines: trust‐region frameworks (TRF), covariance matrix adaptation evolution strategy (CMA‐ES) and Classic Powell (CP). High‐fidelity free‐space results reveal a fundamental topological bifurcation. Unconstrained stochastic heuristics (CMA‐ES and CP) deceptively achieve ≈ 100% in‐band coverage through aggressive geometric over‐coupling, inducing massive out‐of‐band spectral leakage (≥ 22.71 MHz overflow) that violates regulatory emission masks. Crucially, when subjected to the dielectric shock of standard corrugated cardboard packaging ( ε _r = 2.0), these unconstrained topologies catastrophically overfit, suffering downward resonant drifts of up to 216.8 MHz into unregulated cellular bands. Conversely, the proposed Surrogate‐TR utilizes its OOD penalty to strictly enforce a zero‐leakage emission mask. By intentionally halting free‐space bandwidth expansion at an optimal Pareto boundary (≈ 70% coverage), the algorithm engineers a deliberate spectral ‘safety buffer’. Under cardboard dielectric loading, this buffer acts as a physical shock absorber, elegantly absorbing a 72.0 MHz detuning shift while maintaining a pristine −18.02 dB impedance match without structural deformation. Furthermore, the surrogate‐optimized geometry delivers an exceptional free‐space directivity of 1.86 dBi and ≈ 92.0% radiation efficiency, yielding a massive theoretical forward‐link read range of ≈ 26.1 m. To guarantee byte‐for‐byte reproducibility, all converged 11‐dimensional geometric vectors and spectra are explicitly released, providing a robust, zero‐leakage baseline for supply‐chain antenna engineering.