Abstract:  Portable nuclear magnetic resonance (NMR) spectrometers often suffer from external electromagnetic interference (EMI), leading to low signal-to-noise ratios (SNR) and reduced accuracy in material analysis. To address this problem, we propose a novel method that uses multiple reference coils to monitor environmental EMI noise and a multi-channel Transformer reconstruction (MCTR) network to predict and suppress noise in the main receiver coil. By capturing spatial electromagnetic patterns via parallel channels and modeling long-range dependencies, the MCTR network effectively estimates and removes noise from raw NMR signals in real time. Experiments on both simulated and real NMR data demonstrate that this approach significantly enhances noise suppression and maintains high signal fidelity, outperforming traditional denoising techniques. This method improves the robustness and performance of portable NMR systems in EMI-prone environments, supporting broader field applications.

Key words: Portable NMR, Multi-channel Parallel Reception, Reference Coils, Noise Suppression, Transformer