---
title: Quantum-Enhanced Sensing Methods & Applications
url: https://www.emergentmind.com/topics/quantum-enhanced-sensing
type: topic
---

# Quantum-Enhanced Sensing Methods & Applications

Searching arXiv for the provided papers and closely related work on quantum-enhanced sensing.
Quantum-enhanced sensing denotes sensing protocols in which nonclassical resources reduce estimation error, lower detection thresholds, or improve search complexity beyond classical or shot-noise-limited baselines. In the literature surveyed here, the operative resources include spin squeezing, hyperentanglement, bosonic superpositions, interaction-based readout, hybrid-spin amplification and deamplification, criticality, non-Hermitian spectral structure, information scrambling, and quantum algorithms. The resulting tasks range from phase and field estimation in atomic and photonic interferometers to weak-target detection, broadband AC-signal search, and post-hoc learning of many properties of a classical field from a single quantum measurement record [2212.09124][1709.04513][1807.11934][2309.00177][2501.07625][2602.17591].

## 1. Metrological foundations and figures of merit

The canonical benchmark is the standard quantum limit (SQL), while the asymptotic aspirational benchmark is Heisenberg scaling. In the unitary pure-state setting, the Cramér–Rao bound is written as
\[
\delta\theta \ge \frac{1}{\sqrt{\nu F(\theta)}},
\]
with Fisher information \(F(\theta)\), repetition number \(\nu\), and quantum Fisher information (QFI) \(\mathcal F(\theta)\) obtained by optimizing over measurements; for pure states under unitary evolution, \(\mathcal F(\theta)=4\langle \Delta \hat H^2\rangle\) [1709.04513]. Mixed-state analyses also appear centrally in noisy-target and local-many-body sensing, where QFI is evaluated from the spectrum of the output density operator rather than directly from a variance formula [2004.09234][2010.09050].

In experimental quantum sensing, the abstract Fisher-information language is often converted into platform-specific gain metrics. In spin-squeezing-based interferometry on optical transitions, the Wineland parameter
\[
\xi^2 = \min_{\mathbf n_\perp} \frac{N\langle(\Delta \hat S_{\mathbf n_\perp})^2\rangle}{|\langle \hat{\mathbf S}\rangle|^2}
\]
is directly related to the SQL through
\[
\xi^2=\left(\frac{\Delta\phi}{\Delta\phi_{\rm SQL}}\right)^2,
\qquad
\Delta\phi_{\rm SQL}=\frac{1}{\sqrt N},
\]
so negative values in dB indicate sub-SQL operation [2303.10688]. In spin-1 condensate nonlinear interferometry, gain is reported as
\[
-20\log_{10}\!\left[\frac{\Delta\phi}{(\Delta\phi)_{\rm SQL}}\right],
\]
with both a two-mode SQL \((\Delta\phi)_{\rm SQL}=1/\sqrt N\) and a three-mode SQL \((\Delta\phi)_{\rm SQL}=1/(2\sqrt N)\) depending on the interferometric structure [2212.09124].

The same field also uses operational SNR criteria. In Fourier-domain interferometric sensing with \(N\)-photon probes, the detected coincidence probability in the small-modulation regime yields a PSD of the form
\[
S_N(f) \simeq \frac{1}{f_0^2} \left[ \frac{\lambda}{2N} + \frac{(\lambda V_N A_m)^2}{16}\,\delta(f-f_m) \right].
\]
This makes explicit that the spectral line at \(f_m\) is unchanged while the noise floor scales as \(1/N\), giving
\[
\mathrm{SNR}(N)=\frac{\lambda A_m^2}{8}\,N,
\qquad
\mathrm{SNR}(N)/\mathrm{SNR}(1)=N,
\]
and therefore a \(3\) dB improvement for the \(N=2\) versus \(N=1\) comparison [2602.16350]. This framing is important because quantum enhancement need not appear as signal amplification; it can appear as spectral-noise suppression.

## 2. Entanglement-assisted interferometry and interaction-based readout

A central route to quantum-enhanced sensing uses entangled many-body probe states together with readout protocols that preserve the metrological gain in the presence of realistic detection noise. In a spin-1 \({}^{87}\mathrm{Rb}\) Bose–Einstein condensate initialized in the polar state \(|F=1,m_F=0\rangle\), spin-exchange dynamics near \(q\approx |c_2|\) generate a spin-nematic squeezed vacuum governed by
\[
\hat H=\frac{c_2}{2N}\hat{\mathbf L}^2-q\hat N_0.
\]
The squeezing is quantified by
\[
\xi_\theta^2=-20\log_{10}\!\left(\frac{\Delta \hat L_z}{\sqrt N}\right),
\]
with best observed squeezing \(12.1\pm0.5\) dB, improving to \(14.0\pm0.5\) dB after subtraction of detection noise. The key readout innovation is an echo protocol: nonlinear splitting, a \(\pi/2\) state flip around \(\hat Q_0\) that exchanges squeezing and anti-squeezing axes, and nonlinear recombination. For a small-angle RF Rabi rotation, this yields \(15.6\pm0.5\) dB relative to the three-mode SQL, equivalent to \(21.6\pm0.5\) dB beyond the two-mode SQL for \(N=26400\) atoms. In a Ramsey interferometer it yields \(16.6\pm1.3\) dB over SQL and an extrapolated absolute near-resonant microwave-field sensitivity of \(103~\mathrm{pT}/\sqrt{\mathrm{Hz}}\) at a probe volume of \(18~\mu\mathrm m^3\) [2212.09124].

An optical-transition implementation on trapped-ion chains realizes a related but microscopically distinct form of collective enhancement. In 1D chains of up to \(51\) \({}^{40}\mathrm{Ca}^+\) ions, a power-law transverse-field Ising model operated in the \(B\gg J_0\) regime becomes an effective XX exchange model that reproduces the collective behavior associated with one-axis twisting (OAT). The dynamics preserve total transverse magnetization more effectively than a corresponding finite-range Ising evolution, suppress the growth of finite-\(k\) spin-wave excitations, generate a Wineland parameter of \(-3.9 \pm 0.3\) dB for \(N=12\), and reduce Ramsey measurement uncertainty by \(-3.2 \pm 0.5\) dB below the SQL for \(N=51\) [2303.10688]. The same experiment also observes non-Gaussian multi-headed cat states in the Husimi \(Q\)-distribution at longer times.

Hyperentanglement supplies a photonic realization in which multiple degrees of freedom contribute independently to the Fisher information. In the mirror-tilt protocol based on photon pairs entangled in polarization and spatial modes, the single-photon and two-photon encoding unitaries are
\[
\hat U^{(1)}(\theta)= e^{-i2k\theta\hat\sigma_z \hat x},
\qquad
\hat U^{(2)}(\theta)=\exp\!\left[-i2k\theta(\hat\sigma_{z_1}\hat x_1+\hat\sigma_{z_2}\hat x_2)\right].
\]
The corresponding QFI for a hyperentangled pair,
\[
\mathcal{F}^{(2)}(\theta)=32k^2\left(\sigma^2+\mathrm{Cov}[x_1,x_2]+2d^2\right),
\]
contains contributions from spatial spread, beam displacement, and spatial covariance. In the maximally correlated case this becomes
\[
\mathcal{F}^{(2)}_{\rm max}(\theta)=4\,\mathcal{F}^{(1)}(\theta),
\]
which is the two-photon Heisenberg enhancement. Experimentally, a simple binary polarization projection saturates the QFI in the small-angle regime and yields sub-shot-noise precision for large beam displacement \(d\approx 6\) mm [1709.04513].

## 3. Bosonic, photonic, and spatially parallel implementations

Bosonic interferometers exploit large Hilbert spaces rather than large qubit registers. In a trapped-ion mechanical oscillator, Ramsey interferometry between \(|0\rangle\) and \(|n\rangle\) uses the motional superposition
\[
\lvert \Psi_n\rangle = \frac{1}{\sqrt{2}\left(\lvert 0\rangle + \lvert n\rangle\right),
\]
which acquires a relative phase \(n\phi\) under free evolution, producing
\[
P_0 = \frac{1}{2}\left[1-\cos(n\phi)\right].
\]
The ideal phase sensitivity is therefore \(1/n\), compared with the coherent-state benchmark \(1/\sqrt n\). Experimentally, number states were generated up to \(n=100\) and superpositions up to \(n=18\); the best observed sensing performance occurred at \(n=12\), with \(3.2(2)\) dB better sensitivity than an ideal classical interferometer with the same average occupation number, and frequency tracking of the axial mode of a single \({}^{9}\mathrm{Be}^+\) ion reached an Allan deviation \(2.6(2)\times 10^{-6}\) after about \(4\) s [1807.11934].

A single photonic or bosonic mode can also be used as a sensor via cat-state coherence. For the Hamiltonian \(H=\varepsilon a^\dagger a\), free evolution induces a phase-space rotation \(\theta=-\varepsilon T\). Using the amplitude cat state
\[
{\cal N} \big( |0\rangle + |\alpha_0\rangle \big),
\]
the protocol converts the rotation into qubit-readout fringes through a Ramsey sequence and a conditional parity gate. In the large-cat, small-angle regime,
\[
P_g \approx \frac{1-\cos(D\theta)}{2},
\qquad
D=|\alpha_0|^2,
\]
and the signal-to-noise ratio
\[
R = \frac{\left|\partial P_g/\partial \theta\right|}{\sqrt{P_g(1-P_g)}}
\]
satisfies \(R\simeq D\) at the optimal working point, i.e. Heisenberg-limit-like scaling in the ideal lossless case. With dissipation, the same work finds an optimum rather than indefinite improvement; for \(\kappa T = 0.02\), the maximal reported value is \(R_{\max}=36.25\) at \(D=100\) [2503.23531].

Optical platforms also show that quantum advantage can survive substantial loss and can be distributed across sensor arrays. In plasmonic surface-plasmon-resonance sensing, bright twin beams from four-wave mixing reduce the intensity-difference noise below the shot-noise level and improve refractive-index sensitivity from \(\Delta n_{\rm min}^{\rm classical}\approx 8.6\times10^{-10}\,\mathrm{RIU}/\sqrt{\mathrm{Hz}}\) to \(\Delta n_{\rm min}^{\rm quantum}\approx 5.5\times10^{-10}\,\mathrm{RIU}/\sqrt{\mathrm{Hz}}\), corresponding to a \(56\%\) enhancement; the same work reports \(9\) dB source squeezing, reduced to \(4\) dB after the sensor and other losses [1802.00410]. In a four-sensor quadrant plasmonic array illuminated by multi-spatial-mode twin beams, independent probe–conjugate spatial subregions permit simultaneous readout of local refractive-index changes with a quantum sensitivity enhancement in the range of \(22\%\) to \(24\%\) over the classical configuration [2311.01560]. In fiber interferometry, the spectral-domain viewpoint isolates a different operational mechanism: identical \(440\) Hz modulation peaks for single-photon and two-photon interference, but a \(3\) dB lower quantum noise floor, with the two-photon signal remaining resolvable when the single-photon signal is buried in the spectral background [2602.16350].

## 4. Noise-robust transduction, amplification, and sensing in noisy backgrounds

Several recent protocols shift the emphasis from state preparation alone to preprocessing of the signal before final detection. In a warm-vapor hybrid sensor containing isotopically enriched \(^{87}\mathrm{Rb}\) and polarized \(^{129}\mathrm{Xe}\), rapid spin-exchange collisions generate coupled alkali and noble-gas dynamics with effective coupling \(J\approx 30\) Hz. Because the alkali response is broad (\(\Gamma_a\approx 30\) kHz) and the noble-gas response is narrow (\(\Gamma_b\approx 7\) mHz), their interference produces a Fano resonance
\[
F(\epsilon)=\mathscr{A}(\epsilon)\frac{(q+\epsilon)^2}{1+\epsilon^2}+\mathscr{B}(\epsilon).
\]
Near resonance, the effective-field amplification factor
\[
\eta=\frac{\gamma_b\lambda M_z^b}{2\widetilde{\Gamma}_b}
\]
is essentially equal to the Fano parameter, \(\eta\simeq q\); experimentally \(q\approx 524\), giving pre-amplification of more than two orders of magnitude. The same platform demonstrates magnetic-field measurement about \(54\) dB below the photon-shot-noise level of the \(^{87}\mathrm{Rb}\) magnetometer, a magnetic sensitivity about \(3.5~\mathrm{fT}/\sqrt{\mathrm{Hz}}\), and magnetic-background suppression of about \(24\) dB in deamplification mode [2309.00177]. This directly addresses the practical regime in which detector noise exceeds the intrinsic sensor sensitivity.

A complementary generalization appears in the bridge between quantum-enhanced sensing and quantum illumination. In the weak-target limit, a target of reflectivity \(\eta\ll 1\) is modeled by
\[
\hat{B}_{ab}(\eta,\varphi)\approx \exp[\eta(\hat{a}^\dagger \hat{b}e^{i\varphi}-\hat{a}\hat{b}^\dagger e^{-i\varphi})],
\]
and the resulting QFI at \(\varphi=\pi/2\) matches the QFI of phase sensing in a \(50{:}50\) interferometer. Under a fixed total input energy constraint, optimized \(N\)-photon entangled states outperform both TMSV and separable coherent states, and the paper connects the target sensitivity to a practical photon-number-difference measurement
\[
\hat M=\hat n_d-\hat n_e
\]
and to an SNR that functions as an effective error exponent for target detection. A striking reported result is that increasing thermal noise can improve both target sensitivity and SNR for optimized entangled states in the weak-target regime [2004.09234]. This does not state that thermal noise is generically beneficial; it states that in the specific model considered, the correlated structure of the entangled probe can turn thermal contributions into a net advantage.

A more expansive sensing architecture is Quantum Signal Learning (QSL), which treats a classical field as inducing an unknown displacement distribution rather than a single scalar parameter. In Bell QSL, a TMSV resource, passive optics, and static homodyne Bell measurements generate effective samples
\[
\zeta=\alpha+Z,
\qquad
Z\sim\mathcal N_{\mathbb C}(0,e^{-2r}),
\]
so both quadratures are estimated simultaneously with noise suppressed below the vacuum level. The protocol is explicitly designed for post-hoc estimation of many functionals of the displacement distribution from the same dataset, and the sample complexity for a property kernel \(\psi\) is bounded by
\[
N \ge 8\,\mathcal K_r(\psi)^2\,\epsilon^{-2}\log\!\left(\frac4\delta\right).
\]
The same work proves worst-case exponential separations from entanglement-free strategies for matched filtering and other classical-field tasks, and practical speedups over homodyne and heterodyne baselines [2602.17591].

## 5. Criticality, localization, driven many-body dynamics, and non-Hermitian sensors

Quantum criticality has long been identified as a metrological resource, but the recent literature broadens this notion beyond equilibrium ground states. In the Aubry–André–Harper lattice, the single-particle Hamiltonian
\[
\hat{H} = - \sum_i \left(\hat{c}^\dagger_i \hat{c}_{i+1} + h.c.\right) + V \sum_i \cos(2\pi i \omega)\,\hat{c}^\dagger_i \hat{c}_i
\]
has an energy-independent localization transition at \(V_c=2\). Near the transition, the QFI reaches essentially Heisenberg scaling, with \(F_Q^*(V=V^*) \sim L^{2.01(1)}\) in the single-particle case and \(F_Q^*(U=0) \sim L^{1.98(2)}\) at half filling; even with \(U=1.2\), the reported scaling is \(L^{1.95(4)}\). Experimentally meaningful observables do not always saturate this bound but can still be super-SQL: \(F_{cdw}^* \sim L^{1.54(4)}\) for the charge-density-wave imbalance, while dynamical quench protocols yield \(F_{H_2}\sim Lt^{3.92(1)}\approx Lt^4\) and \(F_{cdw}\sim Lt^{6.003(7)}\approx Lt^6\) in the transient regime [2305.02315].

When only part of a many-body system is accessible, equilibrium critical metrology degrades. For a block of \(L\) spins in the ground state of a driven \(XY\) chain, the local QFI scales only as \(aL^\eta\) with fitted exponents \(\eta \approx 1.8\), \(1.76\), and \(1.74\) for \(\gamma=0.1\), \(0.5\), and \(1\), respectively. Periodic driving alters this picture by replacing ground-state sensing with local steady-state sensing governed by Floquet resonances. The steady-state QFI satisfies \(F_Q^{ss}\sim L^2\) at \(\omega=2\), while for lower frequencies \(\omega<2\) the reported scaling is super-Heisenberg, \(F_Q^{ss}\sim L^\eta\) with \(\eta>2\). The enhancement is traced to closure of the Floquet quasienergy gap, the driven analogue of equilibrium gap closing at a critical point [2010.09050].

A non-equilibrium realization on superconducting hardware uses Stark–Wannier localization in a \(9\)-qubit chain with exchange coupling \(J=-8\,\mathrm{MHz}\). The unknown parameter is the linear gradient field \(h\). Projective computational-basis measurements at multiple evolution times are combined through a Bayesian estimator,
\[
f(h|\{n_j\})=\frac{f(\{n_j\}|h)f(h)}{f(\{n_j\})},
\qquad
h_{\rm est}=\arg\max_h f(h|\{n_j\}),
\]
and the resulting reciprocal variance scales as \(1/\delta^2 h\sim t^\beta\) with \(\beta\simeq 1.94\) for \(K=5\) time points and \(\beta\simeq 2.02\) for \(K=7\), i.e. near-Heisenberg-limited precision in time. The extended phase outperforms the localized phase, and the transition is identified numerically near \(h_c\approx -6\,\mathrm{MHz}\) for single excitation and \(h_c\approx -5\,\mathrm{MHz}\) for double excitation [2508.14409].

Non-Hermitian and non-unitary platforms add another route to enhanced susceptibility. In a photonic non-unitary quantum walk, transient-time dynamics already exhibit critical sensitivity associated with topological point-gap and line-gap closings. Experimentally, the maximum CFI scales as \(F^C_{\rm max}\sim N^b\) with \(b=2.116\) near the point-gap transition and \(b=2.707\) near the line-gap transition, whereas away from criticality the exponents drop to \(1.360\) and \(1.397\), respectively [2506.16133]. A time-modulated non-Hermitian qubit shows a related but more explicitly dynamical effect: eigenvalue-based sensing using the splitting \(\Delta_E(t)\) reaches a \(9.21\)-fold improvement over a conventional Hermitian sensor, while eigenstate-based sensing using time-dependent instantaneous-eigenstate populations reaches up to \(50\)-fold enhancement and can exhibit divergent susceptibility even when the trajectory is not close to an exceptional point [2503.16217].

## 6. Information scrambling, quantum computation, and generalized sensing tasks

A recent conceptual shift is the use of dynamical complexity itself as the sensing resource. Butterfly metrology prepares a coherent superposition of an unscrambled branch and a scrambled branch by applying forward evolution \(U=e^{-iHt}\), a local butterfly operator \(V\), backward evolution \(U^\dagger\), phase encoding \(e^{-i\phi S_z}\), and then a second forward evolution. In the local-control version, the butterfly state has the schematic form
\[
\frac{1}{\sqrt{2}\left(\mathbbm{1}+ i V(t)\right)\ket{0},
\qquad
V(t)=U^\dagger V U,
\]
and the small-\(\phi\) sensitivity is
\[
\eta^{-1}_{\phi=0} = \frac{N}{2} - \sum_{S_z} S_z\,P(S_z).
\]
For a fully scrambled, Haar-random state this gives \(\eta_{\phi=0}=2/N\), i.e. Heisenberg scaling up to normalization conventions. The same work makes the dependence on scrambling explicit through local OTOCs and operator space-time volumes, so the metrological gain is tied directly to operator spreading [2411.12794]. An experimental implementation on a superconducting processor with \(N=6,8,10\) qubits observes increasing phase-response slopes \(-2.97\), \(-3.93\), and \(-4.83\), OTOC decay, GME concurrence peaking around \(56\) ns for \(N=6\), and sensitivity beyond the SQL with normalized scaling linear in \(N\), consistent with a factor-of-two of the Heisenberg limit [2601.22503].

Quantum computation can also change the complexity class of a sensing task. For AC sensing of a weak oscillating field
\[
B(t)=B\cos(\omega t+\phi),
\]
with unknown amplitude, frequency, and phase, quantum search sensing partitions the frequency band into
\[
N\sim \left\lceil \frac{|\Delta\omega|}{n_S B_{\min}} \right\rceil
\]
bins, digitizes the analog response into a sensing oracle \(\mathcal O_{\rm AC}\), and applies Grover-style amplitude amplification. The resulting sensing time satisfies
\[
\tau_{\rm QSS} =\widetilde O\!\left( \frac{1}{n_S B_{\min} \sqrt{\left\lceil \frac{|\Delta\omega|}{n_S B_{\min}\right\rceil} } \right),
\]
while the Grover–Heisenberg lower bound is
\[
\tau=\Omega\!\left( \frac{1}{n_S B_{\min} \sqrt{\left\lceil \frac{|\Delta\omega|}{n_S B_{\min}\right\rceil} } \right).
\]
The same work proves that protocols based only on QFI, finite-lifetime quantum memory, or classical modulation obey strictly weaker bounds for this broadband detection problem [2501.07625].

Generalized target detection can likewise be recast as quantum-enhanced sensing. The phase-sensing/target-sensing equivalence established through QFI shows that weak-reflectivity target detection can be analyzed with metrological tools and then connected to practical discrimination through SNR and error exponents [2004.09234]. QSL extends this logic further by replacing single-parameter estimation with property learning over an unknown displacement distribution, making informational completeness and post-hoc reuse of data central rather than incidental [2602.17591].

Across these disparate formulations, several recurrent misconceptions are corrected by the recent literature. Quantum enhancement does not always mean a larger signal peak; in Fourier-domain interferometry it can mean a lower spectral noise floor [2602.16350]. It does not always require ultra-low-noise final detectors; interaction-based readout and hybrid-medium amplification can move the burden of sensitivity enhancement upstream of the detector [2212.09124][2309.00177]. It is not restricted to entangled probe states in the narrow sense; scrambling-based protocols, Groverized search, and QSL derive their gain from operator growth, coherent query complexity, or informationally complete sub-vacuum-noise datasets [2411.12794][2501.07625][2602.17591]. At the same time, the literature is equally clear that realized gains remain contingent on loss, decoherence, control infidelity, finite-time operation, and measurement nonoptimality, which is why many experimentally observed advantages remain below their ideal asymptotic limits [1807.11934][2303.10688][2508.14409][2010.09050].

Source: https://www.emergentmind.com/topics/quantum-enhanced-sensing