Characterizing Timing Parameters in Commercial SERF-OPM Multichannel Systems for Biomagnetic Field Sensing

TL;DR

These findings provide the first vendor-comprehensive benchmarks for timing parameters in commercial SERF-OPM systems, and highlight the trade-offs between bandwidth, delay, and calibration strategies, and underscore the need for rigorous timing characterization to ensure waveform fidelity.

Abstract

Spin-Exchange Relaxation-Free Optically Pumped Magnetometers (SERF-OPMs) are increasingly used in multichannel biomagnetic sensing, yet their timing performance remains poorly characterized. This contribution presents the first cross-platform study of time delay, group delay, intra-channel variability, and settling time across four commercial SERF-OPM systems: Neuro-1 and QZFM Gen.2 (QuSpin Inc.), and HEDscan and FLv2 (FieldLine Inc.). Measurements were performed inside a magnetically shielded room (BMSR-2.1, PTB, Berlin) using an already introduced test bench. Results show frequency-dependent delays of 1-10 ms, intra-channel spreads up to +-1 ms, group delays between 1-15 ms, and settling times of 2-55 ms. Clear differences in manufacturer strategies were observed: QuSpin minimizes intra-channel variability through digital delay equalization, whereas FieldLine employs per-sensor calibration to optimize bandwidth and phase matching. In all systems, the time delay deviation between channels is in the sub-millisecond range in the 20-140 Hz band, which is sufficient for magnetoencephalography source localization. However, longer settling times in some platforms limit performance for rapid stimulation protocols. These findings provide the first vendor-comprehensive benchmarks for timing parameters in commercial SERF-OPM systems. They highlight the trade-offs between bandwidth, delay, and calibration strategies, and underscore the need for rigorous timing characterization to ensure waveform fidelity. The results are directly relevant to applications such as stimulation-evoked responses, brain-computer interfaces, and closed-loop neuromodulation.

Figures (7)

• Figure 1: Simulated effects on a single-channel recorded SQUID-MCG signal: a) original waveform, b) introduced constant time delay of 15.6 ms, c) introduced frequency-dependent group delay rate of 7 ms per Hz, and d) introduced a settling time of $\tau$ = 98 ms, combined with an operating point loss triggered by a large field distortion event at the second heartbeat.
• Figure 2: a) 21 magnetometer heads of HEDscan OPM-MEG System (FieldLine Inc.) are mounted on the DALAC top using a CRP board with precision CNC-milled cut-outs to securely hold each housing. The unit cables are well-fixed to prevent any static B-field change caused by cable movement. b) Scheme of Time-delay evolution exemplified by Neuro-1 MCS: An excitation current applied to the coils generates the magnetic field $B_{\text{exc}}$. Eddy currents in the µ-metal shielding cause a global phase delay, while each sensor introduces an additional, sensor-specific delay. These are corrected in the data acquisition system during post-processing. The plot depicts the signals, with the Y-axis representing amplitude and the X-axis showing the sine argument in radians. The final time delay $t_{\text{d}}$ is subsequently determined.
• Figure 3: Illustration of settling $t_{\text{set}}$ in response to a magnetic field step change (DC-Change). The dashed red line represents the applied current (normalized to 100 %), and the solid blue line shows the response of the device under test, recorded with $f_{\text{s}}$ = 8 kHz. The step onset is defined at $t=0$ ms, corresponding to the moment when the applied current reaches 50 % of its final steady-state value. Settling time $t_{\text{set}}$ is measured from this point to the time at which the DuT output remains within a predefined error band, e.g., $\pm$5 % around the steady-state value, as indicated by the horizontal back dotted lines. The settling behavior reflects the combined effects of intrinsic time delay, slew rate limitations, and the sensor’s dynamic response characteristics.
• Figure 4: Neuro-1 results: a) Normalized time signal, overlap of 19 Y-channels with an applied test field of $f$ = 0.49 Hz. The inset highlights the zero-crossing region for visualizing relative time delay variation between the sensors. b) Time delays in dependency of frequency over sensitive axes (X, Y, and Z). Data points represent measured delays to the applied test field of $B_{\text{exc}}$ = 400 pT; dashed lines indicate typical time delay characteristic; black lines show calculated group delay. Notably, the X-axis exhibits significant deviations in the low-frequency range; therefore, no typical time delay or group delay is calculated. c) Tukey-style boxplots illustrate the phase delay variation, defined as the interquartile range (IQR) relative to the median, across MEG-relevant frequency bands and axes (X, Y, Z). The median of X-channels shows a systematic phase offset relative to Y-, Z-channels, additionally annotated above each frequency group. Phase variation is mostly within red dashed lines at $\pm 2.5^{\circ}$.
• Figure 5: HEDscan, FLv2 and QZFM Gen.2 results, all for an applied test field of $B_{\text{exc}}$ = 400 pT. Data points represent measured delays to the applied test field; black lines show calculated group delay: a) Time delays of HEDscan in dependency of frequency and X-, Y-, and Z-axis. Notably, Y- and Z-axes are characterized in dual mode; X-axis has a narrow bandwidth, corresponding to much larger time delays and significant deviations in the low-frequency range; therefore, no typical group delay is calculated. b) Tukey-style boxplots illustrate the phase delay variation of HEDscan in dual mode (Y, Z), defined as the inter-quartile range (IQR) relative to the median, across different frequencies. Phase variation is mostly within red dashed lines at $\pm 2.5^{\circ}$. c) Time delays of FLv2, single-axis sensors, for two operational modes (Open Loop and Closed Loop), including measured -3dB-bandwidth range for both modes. d) Time delays of QZFM Gen.2, dual axis, open loop mode, with a large spread of supported -3dB bandwidth.