Logistics‐Grade UHF RFID Tag Design via Surrogate‐Guided Bayesian Optimisation

ABSTRACT

Logistics deployments demand Ultra‐High Frequency Radio‐Frequency Identification UHF RFID tags that are reliably matched inside the European Telecommunications Standards Institute (ETSI) 865–868 MHz corridor, with edge equalisation (–24.94 dB at 865/868 MHz; VSWR = 1.12), a deep centre at 866.5 MHz, flat in‐band response, and shallow out‐of‐band behaviour. We address this narrow‐band, standards‐aware objective with a data‐efficient pipeline that couples adaptive DOE (with near‐corridor refinements and replay) to a multitask DNN surrogate predicting the entire return‐loss spectrum plus auxiliary targets (resonance shift and minimum return loss). The surrogate then steers a Bayesian optimisation (BO) loop using EI with a near‐hit prescreener (a corridor‐aware feasibility gate) and a trust‐region‐like fast stop that halts exploration once the predicted resonance enters [863, 870] MHz. All methods operate on the same 13‐parameter box, frequency grid (0.1 MHz densification in‐band), material stack, and CST solver settings to ensure like‐for‐like comparisons against TRF, PSO and NMSA. On this basis, our approach yields the closest corridor‐compliant response: RL (865 MHz) = –26.41 dB (within 1.5 dB of the –24.94 dB target), RL (868 MHz) = –22.91 dB (≈2 dB shallow), and 2.7 MHz contiguous in‐band –15 dB bandwidth. To validate real‐world applicability, a physical prototype was fabricated and evaluated in an anechoic chamber, demonstrating a measured directivity of 5.01 dBi and maintaining topological robustness despite fixture‐induced detuning. Furthermore, environmental resilience was explicitly verified through a packaging‐induced detuning analysis on a 4 mm corrugated cardboard substrate. Compared against recent state‐of‐the‐art (SOTA) journal publications, the results demonstrate that our surrogate‐guided, corridor‐aware BO concentrates EM effort where equalised edges are decided, producing highly deployable tag responses under a reproducible and solver‐agnostic workflow.