Nitrogen-vacancy (NV) centers in diamond have attracted attention as magnetic field sensors with high spatial resolution (Balasubramanian et al., 2008; Maze et al., 2008; Rittweger et al., 2009) and sensitivity (Barry et al., 2016; Chatzidrosos et al., 2017; Zhang et al., 2021). Their sensitivity has been demonstrated to be as low as sub-pT level, for AC fields (Wolf et al., 2015) and with hybrid magnetometers at the low-frequency regime (Xie et al., 2021) The range of their application includes, but is not limited to, single neuron-action potential detection (Barry et al., 2016), single protein spectroscopy (Lovchinsky et al., 2016), as well as in vivo thermometry (Kucsko et al., 2013). Advantages of NV-based magnetometers, compared to other magnetic field sensors, include their ability to operate in wide temperature and magnetic field ranges (Zheng et al., 2017). The ability to operate them also without the use of microwaves (Wickenbrock et al., 2016; Zheng et al., 2017; Zheng et al., 2020), has recently enabled a variety of new applications in environments where microwaves (MW) would be detrimental (Chatzidrosos et al., 2019).

The NV consists of a substitutional nitrogen and an adjacent vacant carbon site. It can appear in different orientations along the crystallographic axes of the diamond lattice. This enables vector measurements of magnetic fields (Clevenson et al., 2018). Vector magnetometry itself can be useful in magnetic navigation applications (Cochrane et al., 2016), magnetic anomaly detection, current and position sensing, and the measurement of biological magnetic fields (Barry et al., 2016). Vector measurements near a background field of ∼ 100 mT, where the NV’s ground state level anti-crossing (GSLAC) occurs, are of particular interest (Zheng et al., 2020). Some of the challenges of vector measurements near the GSLAC field include the necessity to precisely align the NV and the bias magnetic field axes for optimum sensitivity of the microwave-free method (Zheng et al., 2017) or the need to account for transversal field related nonlinearities of the NV gyromagnetic ratio when performing microwave spectroscopy.

One of the challenges NV magnetometers face is their low photon-collection efficiency. Approaches to increase the efficiency include, use of solid immersion lenses (Hadden et al., 2010; Siyushev et al., 2010; Sage et al., 2012), or employment of infrared absorption (Acosta et al., 2010; Dumeige et al., 2013; Jensen et al., 2014; Chatzidrosos et al., 2017; Dumeige et al., 2019). Photoluminescence (PL) for fiberized sensors is preferentially collected with the same fiber delivering the pump light but detected on the input side of the fiber (Patel et al., 2020). Despite considerable effort, even modern sensors typically just feature a PL-to-pump-light ratio of about 0.1% (Barry et al., 2016; Patel et al., 2020).

In this paper, we discuss the construction of a fiberized NV-based vector magnetic field sensors and compare it with another sensor of the same kind. One sensor is constructed in Helmholtz Institute Mainz, referred to as Mainz sensor in the following and the other one is constructed by Robert Bosch GmbH, referred to as Bosch sensor in the following. They both achieve sub nT / H z magnetic sensitivity. The sensor constructed in Mainz achieved a sensitivity of 0.45 nT/ H z , limited by intensity noise of the pump laser, with 11 pT/ H z photon-shot noise sensitivity and 0.5% PL-to-pump power ratio taking the fiber coupling efficiency into account. The sensor constructed by Bosch achieved a sensitivity of 0.34 nT/ H z , which was approximately one order of magnitude larger than the expected photon-shot noise limited sensitivity (Stürner et al., 2021). The PL-to-pump power ratio was 0.3% taking the fiber-coupling efficiency into account. The components used for the construction of both sensors are commercially available. They allow for a small sensor size (11 × 40 mm² for Mainz and 15 × 25 mm² for Bosch) with maximized PL-to-pump-light ratio, as well as robustness to movement, which also makes the sensors portable. All of this makes the sensors ideal for mapping magnetic fields and measuring in regions that are not easily accessible. The magnetometers are constructed in such a way that they allow close proximity of the sensors to magnetic field sources, thus allowing for high spatial resolution when mapping magnetic fields. After demonstrating the sensitivity of the sensors and explaining the principles of vector magnetometry with NV centers, we used the Mainz sensor to perform spatially resolved optically detected magnetic resonance (ODMR) measurements covering an area of ×20 30 mm² inside a Halbach-magnet array. The Halbach array itself provides a highly homogeneous magnetic field around 100 mT which makes it ideal for near-GSLAC magnetic field measurements and studies with NV centers. We present the analysis for the translation of the extracted frequency measurements into magnetic field. Details about the construction of the highly homogeneous Halbach array is subject of another publication (Wickenbrock et al., 2020).

The diamond sample used for the Mainz sensor is a 2.0 × 2.0 × 0.5 mm³ type Ib (100)-cut, high-pressure high-temperature (HPHT) grown sample, purchased from Element Six. The initial [N] concentration of the sample was specified as < 10 ppm. The sample was irradiated with 5 MeV electrons at a dose of 2 × 10¹⁹ cm⁻² and then annealed at 700°C for 8 h. The diamond sample used for the Bosch sensor is a 0.8 × 0.8 × 0.5 mm³ (111)-cut, 99.97% ¹²C enriched, HPHT grown diamond. The sample was irradiated with 2 MeV electrons at a dose of 2 × 10¹⁸ cm⁻² at room temperature and then annealed at 1,000°C for 2 h in vacuum. The [NV⁻] concentration was determined by electron spin resonance to be 0.4 ppm (Stürner et al., 2021).

Figure 1A shows the relevant energy levels of the negatively charged NV center, which we use for magnetometry. The ground and excited spin-triplet states of the NV are labeled ³A₂ and ³E, respectively [Figure 1A], the transition between them has a zero-phonon line at 637 nm, but can be exited by more energetic photons (of shorter wavelengths) due to phonon excitation in the diamond lattice. The lower and upper electronic singlet states are ¹E and ¹A₁, respectively. While optical transition rates are spin-independent, the probability of nonradiative intersystem crossing from ³E to the singlets is several times higher (Dumeige et al., 2013) for m _s = ±1 than that for m _s = 0. As a consequence, under continuous illumination with green pump light (532 nm), NV centers accumulate mostly in the ³A₂, m _s = 0 ground state sublevel and in the metastable ¹E singlet state. For metrology applications, the spins in the ³A₂ ground state can be coherently manipulated with microwave fields.

Figure 1B shows the experimental setup used for the measurements conducted in Mainz. To initialize the NV centers to the ground-triplet state, we use a 532 nm (green) laser (Laser Quantum, gem532). To reduce intensity noise from the laser source, the light intensity is stabilized using an acousto-optic modulator (AOM, ISOMET-1260C with an ISOMET 630C-350 driver) controlled through a proportional-integral-derivative controller (PID, SIM960), in a feedback loop. The green laser is normally stabilized to a laser power of 200 mW. The green laser is then coupled into a 2 m long, high-power multi-mode fiber cable (Thorlabs, MHP365L02).

The MW to manipulate the NV spins are generated with a MW function generator (FG; SRS, SG394). They are amplified with a 16 W amplifier (Amp; Mini-Circuits, ZHL-16W-43+) and passed through a circulator (Mini-Circuits, CS-3.000) before they are applied to the NV centers using a mm-sized wire loop. The other side of the wire is directly connected to ground. The radius of the wire used for the wire loop is 50 μm. The wire is then attached to the optical fiber allowing the two parts to move together. For the magnetic resonance measurements presented in this paper, the MW frequency was scanned between 3,800 MHz and 5,000 MHz. Information on the experimental setup used to conduct the measurements with the Bosch sensor can be found in Ref (Stürner et al., 2021). The data presented for the Bosch sensor are acquired at 20 mW of pump-light power and 0.06 mW of MW power.

A schematic of the Mainz fiberized magnetic field sensor head is shown in Figure 1C. The diamond is glued to a parabolic condenser lens, which itself is glued onto a 11 mm focal length lens and attached to a fiber collimator (Thorlabs, F220SMA-532) to which the high-power multi-mode fiber (Thorlabs, MHP365L02) is connected. The MW wire loop is attached to the other side of the diamond to provide the rapidly oscillating magnetic fields required for this magnetic field detection scheme. The light is delivered to the diamond via the parabolic condenser, lens, collimator and the high-power fiber. The same components collect the spin-state-dependent red PL of the NV ensemble. On the other side of the fibre the PL is filtered using a longpass dichroic filter (Thorlabs, DMLP605) which is also used to couple the incoming green light into the fiber. After the dichroic filter residual reflected green light is removed by a notch filter (Thorlabs, NF533-17) and the PL is focused onto a photodiode (PD; Thorlabs, PDA36a2). The detected signal with the PD is connected to a lock-in amplifier (LIA; SRS, SR830). With this setup we were able to achieve 0.5% PL-to-pump-light ratio, which is an order-of-magnitude improvement compared to other fiberized sensors (Patel et al., 2020). A photograph of the Bosch sensor can be seen in Figure 1D. The sensor head, containing a microwave resonator, a custom designed balanced photodetector, and the diamond, which was glued to the collimated output of a single-mode fiber, is located inside a custom designed Helmholtz coil. The Helmholtz coil was used to generate a magnetic bias field of 1.07 mT and the collimation of the laser beam was achieved with a gradient-index lens (GRIN). Further details on the setup used for the Bosch sensor can be found in Ref. (Stürner et al., 2021). The final sensor head, of the Mainz sensor, has a diameter of ≤11 mm and a height of 40 mm, in a configuration that allows for ≤300 μm average distance between sample and sensor, these dimensions refer to not just the NV-containing diamond but to the overall sensor-head footprint, including parabolic lens and fiber collimator. The 2 m long fiber with the attached MW wire and the fiberized sensor head for the Mainz sensor (shown in Figure 1C) can be moved independently of the other components. The footprint of the sensor head of the Bosch sensor was 15 × 25 mm², with these dimensions referring to the sensor-head as well. The smallest distance between the center of the diamond and the outer surface is roughly 2.4 mm, the sensor can move independently as well. To filter low-frequency systematic noise components, e.g. laser-power fluctuations, the MW-frequency ω _L is modulated and the detected PL is demodulated with the LIA. To produce the magnetic field maps presented here, the Mainz sensor assembly was mounted on a computer-controlled motorized 3D translation stage (Thorlabs, MTS25/M-Z8) in the center of the Halbach array (not shown in Figure 1).

Figure  2 depicts the response of the Bosch sensor for three different NV axes upon application of an AC magnetic field with varying frequencies and a constant amplitude of 0.7 μT. The three axes see different effective amplitudes due to the different angles of the NV axes with respect to the magnetic field vector as expected. The sensor response slightly decreases with increasing frequency of the test field independent of the axis. This decrease might stem from the fact, that with increasing frequency, the excitation field becomes under-sampled meaning that the full reconstruction of the sinusoidal excitation field is not possible anymore leading to a decrease in measured amplitude. This decrease is not observable for the Mainz sensor, whose bandwidth is currently limited by the integration constant of the lock-in amplifier used during data acquisition.

To estimate the magnetic sensitivity of the two sensors we follow another method. When the MW field is resonant with the ground-state m _s = 0 → m _s = ±1 transitions, population is transferred through the excited triplet state to the metastable singlet state, resulting in PL reduction. PL as a function of MW frequency is the ODMR signal. By focusing around a single of feature in this ODMR signal the sensitivity can be extracted. The magnetic resonance for the Mainz sensor features a 350 kHz linewidth and 1.6% contrast. The best sensitivity of the Bosch sensor was achieved with a linewidth of ≈92 kHz and a contrast of 0.6%. We modulate the MW frequency around a central frequency f _c , and record the first harmonic of the transmission signal with a LIA. This generates a demodulated signal, which together with the NV gyromagnetic ratio of | ^γ /_2π |∼ 28.024 GHz T⁻¹ can be used to translate PL fluctuations to effective magnetic field noise. The optimized parameters resulting in the best magnetic field sensitivity for the Mainz sensor were: a modulation frequency of f _mod = 13.6 kHz and modulation amplitude of f _amp = 260 kHz. The sensitivity of the Bosch sensor was optimized for three different NV axes as summarized in Table 1.

Figure  3 shows the magnetic-field-noise spectrum of the sensors. The blue trace of Figure 3 corresponds to magnetically sensitive data of the Mainz sensor, the orange to magnetically insensitive. The magnetically sensitive data are taken at a MW frequency resonant with the NV electronic sublevels, while magnetically insensitive data are taken off resonance. The purple and green trace correspond to magnetically sensitive spectrum of the Bosch sensor outside and inside a magnetic shield, respectively, finally the red trace corresponds to the insensitive plot. The magnetically insensitive spectrum can be obtained if f _c is selected to be far from the ODMR features. The peak at 50 Hz is attributed to magnetic field from the power line in the lab. The average sensitivity in the 60–90 Hz area is 0.45 nT/ H z and 0.34 nT/ H z , it is achieved for pump light power of 100 and 20 mW, for the Mainz and Bosch sensors respectively.

The noise traces for the Bosch sensor are based on continuous data series, that are recorded with a sampling rate of 1 kHz for 100 s. To calculated the amplitude spectral density (ASD), the data series was split in 100 consecutive intervals, each with a duration of 1 s. For each interval the ASD was calculated. The depicted is the average of the 100 ASDs.

We constructed two fiberized NV based magnetic field sensors. The sensors feature sub-nT/ H z magnetic field sensitivity with high PL-to-pump-light ratio. The components used to make the sensors are commercially available which makes their construction easy and reproducible. The design for the Mainz sensor was made in such a way as to allow free access to the front side of the diamond as well as for robustness, portability, and small size.

With one of the main challenges for NV magnetometry, especially fiberized, being the low photon-collection efficiency, which leads to poor photon-shot-noise limited sensitivity, the 0.5% PL-to-pump-light ratio collection efficiency, demonstrated in the Mainz sensor, can be used to achieve even higher sensitivity, if the effect of other noise sources, e.g. laser-intensity, thermal and magnetic noise, is minimized. This ratio could also be further optimized by implementing side collection on the diamond (Sage et al., 2012; Barry et al., 2020), as well as highly reflective coatings on the opposite (front) side of the diamond. These features were not implemented in this sensor in order to keep all of its components commercial and easily accessible.

The access to the front side of the diamond allows for close proximity to magnetic field sources, which in turn leads to high spatial resolution. The proximity is currently limited to about 300 μm by MW wire on top of the diamond. It could be optimized further by using a thinner diamond sample, a sample with shallow implanted NV centers or a thinner MW wire such as, for example, a capton-tape printed circuit board, all depending on the intended application, as these changes influence the sensitivity of the sensor. Such optimization would be especially beneficial for measurements of dipole fields or measurements that require high spatial resolution.

The robustness and portability of the sensors also make them attractive for mapping larger areas, or on-the-move measurements of magnetic fields. These features combined with the small size of the sensors suggest a use for measuring in hard-to-access places or small areas where maneuverability of the sensor would be required, e.g. endoscopic measurements inside the human body.

As a demonstration of the Mainz sensor we measured the magnetic field of a custom-designed Halbach array. For these measurements, the small fiberized diamond sensor showed a mm-scale spatial resolution, angular resolution of the magnetic field vector extracted in the measurements were ×9 10^–3 degrees and 18 × 10^–3 degrees for angles θ and ϕ, respectively. We note here, that the corresponding displacement due to such a rotation of the diamond sensor (considering a (0.5 × 0.5) mm² square cross section) could optically not be resolved.