Lightweight Single-Hidden-Layer MLP-Driven EMI Cancellation in SERF-OPM Arrays

Optically pumped magnetometers (OPMs) operating in the spin-exchange relaxation-free (SERF) regime are capable of femtotesla-level sensitivity for industrial magnetic field mapping. However, in complex electromagnetic environments, electromagnetic interference (EMI) degrades SERF-OPM array measurements, undermining spatial resolution and limiting deployment in harsh industrial settings. A lightweight multilayer perceptron (MLP) driven EMI cancellation technique is proposed, in which EMI is estimated via a nonlinear mapping from reference-sensor outputs to primary-sensor EMI. This mapping is implemented as a single-hidden-layer MLP with rectified linear unit activation, and a Gram-matrix condition-number criterion is designed to improve robustness against layout changes. Clean target signals are reconstructed by subtracting the MLP-predicted EMI from primary-sensor readings. Experimental validation on a 32-channel sensor array under both multiple-input multiple-output (MIMO) demonstrated that the MLP-based cancellation method achieves an average signal-to-noise ratio (SNR) improvement of 34.95 ± 8.06 dB in MIMO configurations. Results indicate that the proposed approach offers a favorable efficiency–performance balance relative to the other evaluated models in our benchmarking setting, occupying a Pareto-efficient region in the ΔSNR–latency plane. The framework enables reliable, high-sensitivity industrial OPM array measurements under adverse electromagnetic conditions.