Operating a non-Hermitian atomic magnetometer with programmable digital electronics Authors: Tianyu Qian, Jipeng Xu, Ran Huang, Keyu Xia, Zhiqiang Xiong, Hongchang Zhao, Chao Wang, Yuchen Feng, Jintao Zheng, Jiajia Li, Hui Luo, Shiqiao Qin, Franco Nori, Hui Jing, Zhiguo Wang Published: 2026 ABSTRACT AbstractNon-Hermitian systems with exceptional points (EPs) have attracted much attention for their potential to increase sensitivity. However, EPs are difficult to generate and control accurately in most physical systems, as they are extremely sensitive to environmental disturbances. We propose a unique non-Hermitian system formed by hybridizing digital electronics with physical sensors. The programmability of digital electronics allows EPs to be created, manipulated, and applied easily and flexibly. More interestingly, the modular design of our system facilitates coupling with various physical sensors, increasing their linear response to an EP-enhanced sublinear response. By coupling digital electronics with an atomic magnetometer, we achieved a 4.27-fold sensitivity increase for weak magnetic field measurements. Our work opens a new frontier at the junction of three active fields, namely, atomic physics, electronics, and non-Hermitian devices. The programmability, scalability, and practicality of our approach could further increase the universality, complexity, and intelligence of non-Hermitian fields.Subject terms: Optics and photonics, Electrical and electronic engineering Introduction Sensor performance is mainly limited by loss during signal transmission, such as mirror reflection loss in optical cavities, friction loss in mechanical oscillators, and depolarization and decoherence (inherent loss) in atomic vapor cells. Traditional methods usually suppress these losses through the optimization of fabrication processes or the use of innovative structures, but overlook the potential for loss control. In recent years, sensing enhancement based on exceptional points (EPs) in non-Hermitian systems has shown the potential to increase measurement sensitivity1–7 The principle underlying the use of non-Hermitian systems to increase sensor sensitivity fundamentally differs from traditional approaches. When a non-Hermitian Hamiltonian traverses an EP in parameter space, the simultaneous degeneracy of both eigenvalues and eigenstates occurs, whereas in Hermitian systems, only eigenvalue degeneracy arises, with eigenstates remaining orthogonal8. This process is accompanied by a sublinear response enhancement scaling as ε1N (where ε denotes the perturbation strength and N is the system dimension)9–11. Thus, non-Hermitian systems can utilize loss to enhance the signal near the EPs12–16. Such enhancement mechanisms have been experimentally validated in optical17–23 and electronic platforms24–28, demonstrating performance improvements in displacement sensing25,26,29,30, thermal metrology31, inertial measurement13,32, and acceleration detection33. Moreover, non-Hermitian systems have also been used to study exceptional quantum effect34, loss-induced chirality inversion35, and spin-selective topological effects36. The implementation of non-Hermitian EP sensing in most systems often faces inherent limitations. The extreme sensitivity near EPs leads to fragile couplings, making gain-loss and phase control difficult to achieve simultaneously and prone to introducing excess noise37–41. In contrast to most physical samples, digitally synthesized non-Hermitian sensing systems exhibit increased tuning flexibility with reduced control complexity. However, no attempts have been made to integrate a physical non-Hermitian EP system with digital electronics to realize high-dimensional parameter control and ultrasensitive measurements. Furthermore, although EP sensitivity enhancement has been observed in electronic circuits33, optical resonators42, and magneto-optical media43, this phenomenon demands rigorous verification in additional physical architectures19,39,44–48. In this study, a method of coupling digital electronics with physical systems is proposed, and the establishment of resonant coupling between digital electronics and a physical sensor is explored, in which an EP-enhanced magnetometer is constructed. In this hybrid architecture, the atomic magnetometer (AM) unit enables ultrasensitive weak magnetic field detection, whereas the digital electronic (DE) unit enables real-time parametric control near EPs. Leveraging the high-precision and high-dimensional control capabilities of digital electronics, the combined system enables simultaneous manipulation of phase and gain while implementing relative noise suppression during measurements. The maximum response and sensitivity achieved by this prototype are 4.52 times and 4.27 times greater than those of a conventional AM, respectively. The modular architecture permits direct substitution of the AM unit with alternative sensing elements, such as an optical force sensor49, a microelectromechanical system (MEMS) gyroscope50, and a nitrogen-vacancy (NV) center magnetometer51, considerably increasing the applicability of the proposed system. The programmability of the DE unit would facilitate further research in which the hybrid system is combined with artificial intelligence (AI) algorithms, such as deep learning and evolutionary computation methods, to promote the application of non-Hermitian physics in various classical and quantum measurements. Results Hybrid non-Hermitian system The hybrid non-Hermitian system is shown in Fig. 1a. The AM unit is used as a sensor with inherent loss for detecting the magnetic field (Bz) by measuring the Larmor precession frequency of atoms (specific details are provided in the “Methods” section). The DE unit, which serves as an EP generator, regulates the loss, gain, and coupling of the system. Specifically, the atomic precession signal enters the computer, couples with the digital resonator, and is fed back to both the digital resonator and the atoms, completing non-Hermitian control. Hybrid EP-enhanced magnetometer. a Schematic of the hybrid EP-enhanced magnetometer, consisting of an atomic magnetometer (AM) unit and a digital electronic (DE) unit. The AM unit includes an atomic vapor cell, pump light, probe light, magnetic field coils, and detector. The DE unit is composed of a digitally simulated resonator and gain and phase shift modules. The gain module is implemented by a multiplier, the phase shift module is implemented by a shift register, and the signal coupling is performed by an addition operation. The atomic precession signal and the digital resonator signal are routed through the gain and phase shift modules and then coupled via an adder. The coupled signal is fed back to both the digital resonator and the atoms. b, c Simplified model of the hybrid non-Hermitian system and the response before and after coupling. After the AM unit couples with the DE unit, its linear response becomes an EP-enhanced response, with maximum enhancement near the EPs (as indicated by the stars); △ω is the eigenfrequency difference between atoms and the digital resonator. d Experimental results of the system response. The left figure illustrates the response of the AM and the EP-enhanced magnetometer to changes in the DC magnetic field ΔB. ω0 is the initial frequency of the system response when ΔB=0. The right figure shows a comparison of their responses to an AC-modulated magnetic field signal δB∙cosωt. EP response: The system response when the AM unit is coupled with the DE unit. Non-EP response: The system response when the AM unit is decoupled from the DE unit. This control strategy leverages the loss characteristics of the system, which were originally intended to be suppressed, representing a new approach to enhancing measurement capabilities. The hybrid system introduces gain feedback and energy exchange between subsystems, in addition to the inherent losses in the AM (Fig. 1b, c). As shown in Fig. 1c, after the introduction of the non-Hermitian control strategy, the response transforms from a simple linear response to an EP-enhanced response. Our platform enables in situ reconfiguration between the EP and non-EP regimes via dynamic toggling of the coupling. As the eigenvalue variation near the EP shifts from linear to sublinear, the system response is amplified. Figure 1d presents experimental measurement results for the non-Hermitian magnetometer before and after coupling. The coupled system demonstrates a significantly enhanced response for both DC and AC signals. The AM unit in the system can be described by the Bloch equations: Where Mx,y,z are the projections of the atomic magnetization vector M⃗ on the x,y and z axes, ωx,y,z=ΓBx,y,z is the atomic precession frequency about the x,y and z axes, and Γ is the atomic gyromagnetic ratio, T1 and T2 are the longitudinal and transverse relaxation times of the atom, respectively. We denote y1=M+=Mx−iMy=M⊥∙eiφ as the signal output of the atoms, andx1=Mzωx=MzΓBx as the signal input caused by the magnetic field Bx from the x-axis coil. Here, ωz=ω1 represents the eigenfrequency of atomic precession, 1/T2=γ1 indicates the loss, and ωy=0 signifies no input along the y-axis. Assuming Mz remains constant during the experiment, Eq. (1) can be rewritten as follows: The DE unit is a one-dimensional digital resonator simulated by a computer, which can be described by the following differential equation: here, x2 and y2 represent the input and output of DE unit, ω2 is the eigenfrequency, and γ2 is the loss. After applying approximation and order reduction to Eq. (3), we obtain: Next, as shown in Fig. 1a, we linearly add their outputs and use the result as the system input, thus obtaining xn=gn(κ1e−iθy1+κ2e−iφy2). Here e−iθ and e−iφ are phase-shift factors and κ1,κ2,g1, and g2 are gain factors, all of which can be tuned in the computer simulation. Finally, by defining Ψ=(y1y2)T, the system of equations can be written in a Schrödinger-like form, namely Ψ°=−iHeffΨ, where the effective Hamiltonian is given by: where Δω is the eigenfrequency difference between atoms and the digital resonator. Additionally, the Hamiltonian contains phase shift factors e−iθ and e−iφ and gain factors κ1, κ2, g1, and g2. Similarly, the eigenvalues contain these parameters, providing a diverse way to generate and control the hybrid EPs. Changing Bz alters the atomic eigenfrequency, and adjusting the phase and gain modules alters the gain and energy exchange of the hybrid system, which is crucial for controlling the hybrid EPs (The complete derivation process of the Hamiltonian is detailed in the Supplementary Materials). In summary, we coupled the AM unit with the DE unit to form a non-Hermitian system. Connected via signal transmission and feedback, the two units still maintain a degree of independence in line with modular design. Thus, the DE unit can also be easily coupled with other sensors. Moreover, the programmability of the digital circuit makes system control more flexible. Phase and gain control of the system To more thoroughly investigate the effects of gain and phase control, we regulated the phase shift φ and gain g2 to manipulate the hybrid system and EPs, with a schematic of the system setup shown in Fig. 2a. Notably, the signal frequencies of the atoms and the digital resonator can separately reflect changes in the eigenvalues of the system. As shown in Fig. 2b, unlike the purely physical system, the hybrid system can simultaneously acquire various signals, including the AM unit signal y1, the DE unit signal y2, and the total coupled signal y1+y2, which is beneficial for signal extraction and analysis. See Supplementary Fig. 1–4 for the complete signal processing procedure. Flexible control of the hybrid non-Hermitian magnetometer. a Control and signal extraction procedure for our hybrid non-Hermitian system. The system control consists of gain g2 and phase φ. The hybrid system can simultaneously acquire three signals: the atomic signal y1, the digital resonator signal y2, and the combined signal y1 + y2, through separate ports. b The Fourier spectra of the system output signal. The mutually independent data ports of the digital electronics can be used to separately extract signals from the magnetometer and the digital resonator or to extract the total signal. c Dependence of the real parts of the eigenvalues on the phase shift φ and the gain module g2. d Experimental results for φ and g2 control. The gray triangles represent the signal of the conventional AM, while the orange circles correspond to the magnetometer signal y1 in the EP-enhanced magnetometer, and the purple circles represent the signal of the digital resonator y2. The scatter points are experimental signals, and the curves are fits to data. The relationship between the voltage and the magnetic field is Bz=1.26nTmV−1∙U. Before controlling the phase φ, we normalize the gain modules by setting κ1=κ2=1; g1=g2=g0 and adjust the loss so that γ1=γ2=γ0, which makes the form of the Hamiltonian more concise and clear. When θ=π/2 and φ=π/2, the eigenvalues take the form λ1=i(g0−γ0)±(Δω2−g02), and the EPs occur at Δω=±g0. In contrast, when θ=π/2 and φ=−π/2, the eigenvalues become λ2=−iγ0±(Δω2+2ig0Δω), and the EPs occur at Δω=0. As shown in Fig. 2c, as φ varies from π/2 to −π/2, the real parts of the eigenvalues transition from a dual-EP state to a single-EP state (Changes in the imaginary parts of the eigenvalues are detailed in the Supplementary Materials.) In the dual-EP state (φ=π/2), the system response increases only when the system is in the PT-symmetry broken phase. In the PT-symmetric phase, the response is weaker than that of the conventional AM. Conversely, in the single-EP state (φ=−π/2), the system response is increased on both sides of the EP. In addition to phase φ control, control using gain coefficient g2 can be easily achieved here. As shown in Fig. 2c, taking the single-EP state (θ=π/2 and φ=−π/2) as an example, the system’s response at Δω=0 gradually decreases from infinity and becomes more linear as g2 increases from g0. As a result, the system transitions from an EP-enhanced response to a linear response, indicating gradual decoupling between the atoms and the digital resonator. In Fig. 2d, by adjusting the voltage U of the Bz coil, we changed the eigenfrequency difference Δω between the resonators, and measured the system response frequency under different conditions of φ and g2. As shown in the left and middle panels of Fig. 2d, when the phases are θ=π/2 and φ=π/2, a degeneracy exists in the PT-symmetric phase, where the system response is lower than that of a conventional AM; as a result, the sensor response is increased only in the PT-symmetry broken phase. When the phase is changed to φ=−π/2, the system transforms from a dual-EP state to a single-EP state, eliminating the degeneracy at the EP. Thus, the sensor response is increased on both sides of the EP, which is beneficial for practical measurements. We next investigated how regulating the gain coefficient g2 affects the system. Increasing g2/g0 from 1 to 1.5 reduces the system response but significantly increases the linearity of the response (Fig. 2d, middle and right panels). Therefore, during actual measurements, the sensitivity and linear response of the sensor can be balanced by adjusting g2 according to specific measurement requirements. The experimental results fit well with our theory, highlighting the advantages of the high level of control in our system. As shown in Fig. 2d, our effective control of phase and gain coefficient confirms that the flexibility of the DE unit makes the regulation of EPs in non-Hermitian systems easier. Also, the figure clearly shows the response enhancement. The slope of the EP-enhanced magnetometer response curve is significantly larger than that of the conventional atomic magnetometer. EP-enhancement weak magnetic field measurements To further verify the response and sensitivity enhancement of the system, we applied an AC modulation to Bz and performed measurements using the hybrid non-Hermitian system in both the EP and non-EP states. We measured the system performance under different gain modules g2 and input signal modulation amplitudes. In the experiments of this work, we use the noise-equivalent magnetic field (NEM) to quantify the sensitivity. The sensitivity is related to the signal-to-noise ratio (SNR) by NEM∝1/SNR. Therefore, as shown in Fig. 3, we first measure the response and noise enhancement factors of the system, and then divide the former by the latter to obtain the SNR improvement factor, which is also the sensitivity enhancement factor of the system. (The details are provided in the Supplementary Materials). EP-enhanced sensitivity of the hybrid magnetometer. Experimental results for the response, noise, and sensitivity under AC modulation δB∙cosωt. The data points and error bars represent the mean values and the uncertainty (σ), respectively, obtained from five independent repeated experiments. a Response, noise, and sensitivity enhancement results under different g2 conditions. The gray dashed line represents the measurement results of the AM unit without coupling to the DE unit, so the enhancement factor of it remains at 1. The orange, purple, and green data points represent the measurement results of the AM unit after coupling to the DE unit. When g2/g0 = 1, the system achieves the best response and sensitivity enhancement. b Response, noise, and sensitivity enhancement results under different input signal amplitudes. When the amplitude is 0.1 mV, the system achieves the best enhancement. We varied g2 to study how the sensitivity enhancement changes as the system shifts from an EP-enhanced response to a linear response. Adjusting the input amplitude alters both the input signal and noise, allowing us to examine the relationship between sensitivity enhancement and the input signal. When g2/g0 approaches 1, the response is increased by a maximum of 4.52 times. As g2 increases further, the response enhancement decreases because the increasing g2 disrupts the symmetry between the atoms and the digital resonator, moving the system away from the EP. Moreover, increasing g2 amplifies the digital resonator feedback signal, introducing more current noise. Consequently, the sensitivity enhancement, which is influenced by both the response and noise enhancements, decreases with increasing g2, eventually leading to a sensitivity for the hybrid system that is lower than that of the conventional AM. At g2/g0=1, the sensitivity of the EP-enhanced magnetometer is still increased by 4.27 times, confirming its practicality in magnetic measurement applications. Next, we increased the input signal amplitude with g2/g0=1 and evaluated system performance. As shown in Fig. 3b, as the amplitude increased, the response enhancement decreased, the noise enhancement increased, and the sensitivity enhancement decreased. Despite the presence of higher noise than in the previous experiments, the response was increased by a maximum of 4.46 times, and the sensitivity was increased by 3.03 times. This trend reflects the results of the previous experiment. When the input signal becomes stronger, the system response range widens, forcing the system to operate further from the EP, thus weakening the response enhancement. Additionally, a larger input signal carries more noise, which reduces the relative background noise and degrades noise suppression (see Methods for details). As the input signal amplitude increases, the sensitivity enhancement eventually decreases below 1. In general, for the EP-enhanced magnetometer, the effects of response enhancement and sensitivity enhancement do exist. Moreover, the closer the system state is to the EPs and the smaller the input signal, the more obvious this effect is. Discussion Thanks to the hybrid of physical sensors and digital electronics, our system has many advantages unattainable for the structure based on pure physical systems. Our experimental results show that the DE unit effectively strengthens the response and sensitivity of the AM, which has considerable practical significance. This enhancement relies on the system working near the EP. Unlike purely physical non-Hermitian systems, which are often overly sensitive to the environment, the powerful digital programming of the DE unit enables the system to easily maintain the EPs. Another advantage of our system is its great programming capability. The digital resonator in the DE unit is fully implemented via computer software, so its physical properties, such as the frequency, loss, and order, can be easily controlled by programming. This not only makes system regulation flexible but also provides the necessary conditions for integrating AI algorithms. The programmability allows the system to overcome limitations on the number of components. Coupling multiple (even dozens of) physical sensors and digital units may thus be feasible. This creates opportunities for studying novel physical phenomena and mechanisms. The key advantage of our hybrid system is the complete separation of the sensing and DE control units, which is achieved through a plug-in modular design. The sampling frequency of the DE unit can easily reach the GHz level; thus, this design can be extended to several physical systems. For magnetic, optical, or mechanical sensing applications, as long as the sensing unit meets the physical characteristics of a dissipative resonator, it can serve as the sensing unit. Most sensing mechanisms can satisfy the condition of a dissipative resonator. For example, in mechanical sensors, elastic elements under damping exhibit characteristics of a dissipative resonator52. Similarly, the voltage or current response in RLC circuits53 and the optical modes within optical cavities21, when subjected to loss, can also be described via a dissipative resonator model. More importantly, the DE unit serves as an EP generator and is compatible with a linear physical sensor. This fundamentally transforms the physical properties of the sensing unit, enabling the system to shift from Hermitian to non-Hermitian and the emergence of EPs, which is essential for increasing measurement sensitivity. Therefore, this hybrid design could become a link between non-Hermitian theory research and precision measurement applications. In the future, it could find widespread use in both domains. Furthermore, although we have only demonstrated the sensitivity enhancement for magnetic field measurement by using 129Xe, we believe that using the DE unit to construct other non-Hermitian spin sensors is also feasible. For example, coupling other atoms (such as 3He, Cs, and Rb) with the DE unit can establish corresponding EP-enhanced magnetometers, which are expected to further improve the sensitivity of magnetic field measurement; coupling Xe isotopes (129Xe and 131Xe) with the DE unit can construct an EP-enhanced comagnetometer for improving the system’s rotation measurement sensitivity. Overall, thanks to the flexible advantages of the DE unit, our technical solution can meet diverse application requirements. We have to admit that our current system has limitations, such as the restricted response frequency of digital electronics and the non-linear effect of atomic magnetometers. But these don’t affect the current experimental results and will be our future research directions and priorities. In conclusion, we proposed a computer-based DE unit that can couple with physical systems to form a combined hybrid non-Hermitian and realize flexible phase and gain control. In experiments, we used this DE unit to construct an EP-enhanced magnetometer and performed highly sensitive magnetic field measurements. Compared with a conventional AM, the EP-enhanced magnetometer achieved increases of up to 4.52 times in response and 4.27 times in sensitivity. The addition of a suitable DE unit with physical sensors neither obscures the sensors’ original physical characteristics nor underutilizes their potential. Instead, it transforms the inherent loss, which has previously been regarded as a detrimental factor to be minimized, into an advantageous feature that amplifies the detection capabilities of such systems. This work could provide more possibilities and greater intelligence for precision measurement and non-Hermitian research. The DE unit can potentially be used to upgrade traditional sensors to EP-enhanced sensors with higher responses and sensitivity. The DE unit is a flexible and precise experimental platform for studying various physical effects. In the future, we plan to further investigate higher-order EPs27,54, non-Hermitian topological phenomena7,53, and nonlinear effects2,55. Additionally, we will integrate AI algorithms into hybrid systems to further enhance their performance and explore their applications in the measurement of various physical quantities, such as magnetic fields, electric fields, sound, gravity, and displacement. Methods Atomic magnetometer (AM) unit The AM, constructed using high-temperature ⁸⁷Rb-¹²⁹Xe atoms, measures the external magnetic field strength via Larmor precession. The atoms are enclosed in a spherical glass vapor cell with a diameter of about 8 mm, containing 10 Torr of 129Xe, 300 Torr of buffer gas N2, and an excess of saturated Rb metal. During the experiment, the vapor cell is kept at 105 °C. The relationship between the precession frequency ωa of the atoms and the external magnetic field Bz is given by ωa = ΓBz, where Γ is the gyromagnetic ratio, which is approximately 2π∙11.86 mHz nT-1. The loss γ is approximately 0.0076 s⁻¹, and the eigenfrequency ω1 is approximately 161.1 Hz. The coil constant for the Bz magnetic field is approximately 1.26 nT mV‒1. To operate the system, the vapor cell is first heated to vaporize Rb, and then, the Rb atoms are polarized via a pump light source. The pump light has a power of 35.52 mW and a wavelength of 795 nm, corresponding to the D1 line of Rb atoms. The probe light has a power of 1.05 mW and a wavelength of 780 nm, which is detuned by 40 GHz from the D2 line of Rb atoms. The Xe atoms are also polarized via spin exchange optical pumping. Once the preparation is complete, the Xe atoms are excited into a magnetic resonance (a coherent state) via a magnetic field pulse produced by a pulse current through the coil. The precession frequency of Xe is measured by an embedded Rb magnetometer. Over time, the coherent state of the Xe atoms decays due to loss mechanisms, causing the signal to weaken and eventually disappear. Digital electronic (DE) unit We use the first-order Butterworth band-pass filter module in LabVIEW to implement the digital resonator. This module can readily simulate the signal and response of the resonator, and by changing the bandwidth and center frequency, we can correspondingly vary the loss and eigenfrequency of the DE unit. In the experiment, a 7856 R FPGA board is additionally used for sampling, with a sampling rate of 500 kHz. To ensure that the system state passes through the EP, the magnetometer and digital resonator must be symmetric, so the loss and frequency of the digital resonator are matched to those of the Xe atoms. Then, we compensate for phase shifts caused by signal transmission, setting θ=π/2 and φ=π/2, and adjust the gain modules to satisfy κ1=κ2=1; g1=g2=g0, which are easily achieved through programming. Phase and gain control Under the conditions κ1=κ2=1 and g1=g2=g0, when the phase θ and φ vary, the system’s Hamiltonian and eigenvalues change, leading to changes in the system’s behavior near the EP. When θ=π/2,φ=π/2, the system is in a dual-EP state, and the Hamiltonian becomes: When θ=π/2,φ=−π/2, the system is in a single-EP state, and the Hamiltonian becomes: In the dual-EP state, the degeneracy of the eigenvalues can cause a reduced response during measurement. Starting from the single-EP configuration, gain control can change the response characteristics of the system. When g2/g0 is no longer fixed at 1, the expression for the eigenvalues is as follows: By adjusting g2, we shift the system away from the EP, modifying the response zero-crossing slope and linearity. The response enhancement is related to the slope of the real part of the eigenvalue: According to this equation, when g2/g0>1 (or g2/g0<1), the response amplification decreases. When g2/g0 → +∞ (or g2/g0 → 0), the slope of the real part of the eigenvalue approaches 1. Therefore, as g2 increases (or decreases), the system response shifts from sublinear to linear, with linearity increasing gradually. Noise analysis To better understand the improved sensitivity, it is necessary to focus on the variation of system noise before and after coupling. As the response near the EP is enhanced, the input signal noise is amplified, but the background noise is not. The input signal can be split into a useful signal part SI and a noise part NI. Let the response enhancement of the original sensor be ga, and let the response enhancement of the EP-enhanced sensor be gad. The outputs of the original sensor Ya and the EP-enhanced sensor Yad are as follows: where Na and Nad are the noise floors of the original and EP-enhanced sensors, respectively. Assuming that ga=1, we have: The response, noise, and sensitivity enhancement of the EP-enhanced sensor are compared with those of the original sensor. The response enhancement of the two sensors is compared as follows: The noise enhancement of the two sensors is compared as follows: The sensitivity enhancement of the two sensors is compared as follows: As gad increases, the sensitivity enhancement improves, approaching: This shows that a higher background noise Na of the original sensor makes the non-Hermitian system more advantageous. Thus, the main advantage of the EP-enhanced sensor is that it suppresses background noise. The noise sources of an atomic magnetometer can be divided into three main categories: quantum noise, environmental noise, and technical noise. Quantum noise includes spin-projection noise, photon shot noise, and AC-Stark–shift–associated noise. In general, these all belong to Na, i.e., the background noise. Environmental noise mainly consists of thermal noise and magnetic-field noise. In practical measurements, thermal noise is essentially independent of the signal strength to be measured and therefore belongs to Na, whereas magnetic-field noise is amplified together with the signal after it is input into the system and thus belongs to NI. Technical noise involves a wide variety of contributing factors, including laser, detector, electronic, and mechanical noise. These noises are distributed over all stages of the sensing process and are related to the hardware configuration. Consequently, part of them arises in the coupling stage of the system, and another part is introduced together with the input signal, contributing to Na and NI, respectively. Supplementary information Supplementary Information Peer Review file Acknowledgements The authors are grateful to Chengzhi Hou, Zhixian Liu, Ye Geng, and Wei Liu for fruitful discussions. H.J. is supported by the National Key R&D Program of China (Grant No. 2024YFE0102400), the National Natural Science Foundation of China (Grants nos. 11935006 and 12421005), and the Hunan Major Sci-Tech Program (Grant no. 2023ZJ1010). Z.W. is supported by the National Natural Science Foundation of China (Grant nos. U23B2066 and 12174446), the Hunan Provincial Natural Science Foundation of China for Distinguished Young Scholars (Grant no. 2023JJ10050), and the Innovation Research Foundation of National University of Defense Technology. J.X. is supported by the National Natural Science Foundation of China (Grant no. 62505360) and the National University of Defense Technology Foundation (Grant no. ZK25-40). R.H. is supported by the RIKEN Special Postdoctoral Researchers (SPDR) program. F. N. is supported in part by the Japan Science and Technology Agency (JST) [via the CREST Quantum Frontiers program Grant No. JPMJCR24I2, the Quantum Leap Flagship Program (Q-LEAP), and the Moonshot R&D Grant no. JPMJMS256E]. K.X. is supported by the National Natural Science Foundation of China (Grant Nos. 92365107 and U25A6008), the Innovation Program for Quantum Science and Technology (Grant no. 2021ZD0301400), and the Innovation Research Foundation of National University of Defense Technology (Grant no. 24-ZZCX-JDZ-11). Author contributions T.Q. and J.X. jointly contributed to the study as co-first authors. T.Q., J.X., R.H., K.X., F.N., H.J., and Z.W. conceived the idea. T.Q., Z.X., H.Z., C.W., J.L., J.Z., and Z.W. conducted the experiments and analyzed the data together with all the other authors. T.Q., J.X., R.H., H.L., S.Q., F.N., H.J., and Z.W. wrote the manuscript with contributions from all the other authors. Peer review Peer review information Nature Communications thanks Yong-Chun Liu and Lucas Rushton for their contribution to the peer review of this work. A peer review file is available. Data availability Source data have been deposited in the Figshare database and are publicly available for this paper (10.6084/m9.figshare.29433827). All other data that support the plots within this paper and other findings of this study are available from the corresponding authors upon request. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Tianyu Qian, Jipeng Xu. Contributor Information Hui Luo, Email: luohui.luo@163.com. Shiqiao Qin, Email: shiqiaoqincn@163.com. Hui Jing, Email: jinghui@hunnu.edu.cn. Zhiguo Wang, Email: maxborn@nudt.edu.cn. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-72864-0. References Associated Data Supplementary Materials Supplementary Information Peer Review file Data Availability Statement Source data have been deposited in the Figshare database and are publicly available for this paper (10.6084/m9.figshare.29433827). All other data that support the plots within this paper and other findings of this study are available from the corresponding authors upon request.