23.1 Ramsey frequency interrogation and clock feedback loops
23.1.1 Mapping a frequency detuning into population via free precession
Atomic clocks, optically pumped magnetometers, and atom-interferometer gravimeters all operate by converting an unknown physical parameter—such as a magnetic field \(B\) via the Zeeman shift \(\Delta = \gamma_B B\) or a local oscillator frequency offset \(\Delta = \omega_{\mathrm{LO}} - \omega_0\)—into a quantum phase \(\phi = \Delta\tau\) during a free-evolution interrogation interval \(\tau\). Using the Ramsey separated-oscillatory-field sequence introduced in Chapter 14, two \(\pi/2\) pulses separated by \(\tau\) and offset by relative phase \(\phi_{\mathrm{ref}}\) map the accumulated phase \(\Delta\tau\) into the excited-state readout probability
where \(W(\tau) = \exp[-(\tau/T_2)^p]\) is the coherence decay envelope.
23.1.2 Biasing to the maximum-slope quadrature point of a fringe
Near a fringe peak or trough (\(\Delta\tau + \phi_{\mathrm{ref}} = 0\) or \(\pi\)), the derivative \(\partial p/\partial\Delta\) vanishes to first order, making the population nearly insensitive to small frequency shifts \(\delta\Delta\). To achieve maximum linear sensitivity around \(\Delta \approx 0\), the reference phase of the second \(\pi/2\) pulse is biased to quadrature (\(\phi_{\mathrm{ref}} = -\pi/2\)), so that
At the quadrature operating point (\(\Delta \approx 0\)), the nominal probability is \(p_0 = 1/2\) and the population slope scales linearly with both the interrogation time \(\tau\) and the surviving coherence \(W(\tau)\).
23.1.3 Balancing longer phase accumulation against decoherence decay
Increasing the free-precession time \(\tau\) sharpens the Ramsey fringes (increasing the lever arm \(\tau\)), but waiting too long (\(\tau \gg T_2\)) collapses the envelope \(W(\tau) \to 0\) so the fringes flatten into a featureless \(p = 1/2\) line. Every quantum sensor therefore faces a fundamental trade-off between coherent phase-accumulation time \(\tau\) and decoherence time \(T_2\).
Visual animation · Section 23.1
Ramsey frequency interrogation and clock feedback loops
Interactive visual model for 23.1 Ramsey frequency interrogation and clock feedback loops (Mapping a frequency detuning into population via free precession · Biasing to the maximum-slope quadrature point of a fringe · Balancing longer phase accumulation against decoherence decay). Use Animate or Step to follow the physical state evolution and quantitative predictions.
23.2 Shot-noise error propagation and optimal interrogation time
23.2.1 Slope sensitivity and binomial projection noise at quadrature
In \(N\) independent probe measurements (either \(N\) uncorrelated atoms read out in parallel or \(N\) sequential shots), the number of excited-state counts \(n_1 \sim \mathrm{Binomial}(N, p)\) gives the empirical estimator \(\hat{p} = n_1/N\) with quantum projection noise (shot-noise) standard deviation
\[\sigma_{\hat{p}} = \sqrt{\frac{p(1-p)}{N}}.\]
By linear error propagation, the uncertainty \(\delta\Delta\) in the inferred frequency detuning is
At the quadrature operating point (\(\Delta \to 0\), \(p = 1/2\), \(\sqrt{p(1-p)} = 1/2\)), this yields the fundamental single-parameter Ramsey sensitivity formula
Fixed total campaign time \(T_{\mathrm{tot}}\) with zero dead time: When each shot takes duration \(\tau\) so that \(N = T_{\mathrm{tot}}/\tau\), substituting \(\sqrt{N} = \sqrt{T_{\mathrm{tot}}/\tau}\) gives
\[\delta\Delta(\tau) = \frac{e^{\tau/T_2}}{\sqrt{T_{\mathrm{tot}}}\,\sqrt{\tau}}.\]
Minimizing \(e^{\tau/T_2}/\sqrt{\tau}\) sets \(\frac{\mathrm{d}}{\mathrm{d}\tau}(\tau^{1/2}e^{-\tau/T_2}) = 0\), which gives
\[\tau_{\mathrm{opt}}^{\mathrm{fixed}\,T_{\mathrm{tot}}} = \frac{T_2}{2}, \qquad \delta\Delta_{\min}^{\mathrm{fixed}\,T_{\mathrm{tot}}} = \sqrt{\frac{2e}{T_{\mathrm{tot}}\,T_2}}.\]
23.2.3 Fair resource accounting when comparing entangled and unentangled probes
For \(M\) unentangled probes, uncertainty scales at the Standard Quantum Limit (SQL) as \(1/\sqrt{M}\). Replacing the \(M\) independent probes with an \(M\)-qubit GHZ state \((|0\rangle^{\otimes M} + |1\rangle^{\otimes M})/\sqrt{2}\) multiplies the phase accumulation speed by \(M\) (\(\phi = M\Delta\tau\)), reaching Heisenberg scaling \(1/(M\tau)\) in the absence of noise. However, under independent local Markovian dephasing at rate \(1/T_2\) per qubit, the \(M\)-qubit GHZ coherence decays \(M\) times faster, \(W_{\mathrm{GHZ}}(\tau) = e^{-M\tau/T_2}\). Under a fixed total integration time \(T_{\mathrm{tot}}\), shortening \(\tau\) to \(T_2/(2M)\) restores the exact same minimum uncertainty \(\sqrt{2e/(M T_{\mathrm{tot}} T_2)}\) as \(M\) uncorrelated probes—highlighting why fair metrological comparisons must include the decoherence model and total time budget.
Visual animation · Section 23.2
Shot-noise error propagation and optimal interrogation time
Interactive visual model for 23.2 Shot-noise error propagation and optimal interrogation time (Slope sensitivity and binomial projection noise at quadrature · Optimal interrogation time under fixed shots versus fixed total time · Fair resource accounting when comparing entangled and unentangled probes). Use Animate or Step to follow the physical state evolution and quantitative predictions.
23.3 Systematic shifts, dead time, and stability metrics
23.3.1 Technical noise correlations, dead time, and Allan deviation
Every physical clock cycle includes a non-zero dead time \(\tau_d\) spent on state preparation, cooling, and fluorescence readout, so the cycle duration is \(T_c = \tau + \tau_d\) and \(N = T_{\mathrm{tot}}/(\tau + \tau_d)\). During \(\tau_d\), the local oscillator is not interrogated by the atoms, allowing high-frequency oscillator phase noise at harmonics of \(1/T_c\) to alias into the feedback loop (the Dick effect). Long-term clock stability is quantified by the overlapping Allan deviation \(\sigma_y(\tilde{\tau})\) as a function of averaging time \(\tilde{\tau}\): white frequency noise (projection noise) averages down as \(\sigma_y(\tilde{\tau}) \propto \tilde{\tau}^{-1/2}\), whereas flicker frequency floors and random-walk drifts cause \(\sigma_y(\tilde{\tau})\) to flatten and eventually rise at long averaging times.
23.3.2 Why optimal sensing times depend on the noise envelope and cost model
For non-Markovian or quasi-static Gaussian noise with stretched-exponential envelope \(W(\tau) = \exp[-(\tau/T_2)^p]\), the optimal interrogation time shifts to \(\tau_{\mathrm{opt}}^{\mathrm{fixed}\,N} = p^{-1/p}T_2\) for fixed \(N\) and \(\tau_{\mathrm{opt}}^{\mathrm{fixed}\,T_{\mathrm{tot}}} = (2p)^{-1/p}T_2\) for fixed total time \(T_{\mathrm{tot}}\) (for Gaussian dephasing \(p = 2\), \(\tau_{\mathrm{opt}}^{\mathrm{fixed}\,N} = T_2/\sqrt{2}\) and \(\tau_{\mathrm{opt}}^{\mathrm{fixed}\,T_{\mathrm{tot}}} = T_2/2\)). Including finite dead time \(\tau_d > 0\) shifts \(\tau_{\mathrm{opt}}^{\mathrm{fixed}\,T_{\mathrm{tot}}}\) continuously from \(T_2/2\) (when \(\tau_d \ll T_2\)) toward \(T_2\) (when \(\tau_d \gg T_2\)).
23.3.3 Bridging synthetic sensitivity curves to calibrated clock datasets
A precision measurement report pairs its statistical precision \(\delta\Delta_{\mathrm{stat}}\) (obtained from the Allan deviation at \(\tilde{\tau}\)) with a systematic error budget covering AC Stark shifts, second-order Zeeman shifts, blackbody radiation shifts, collisional shifts, and gravitational redshift calibration.
Visual animation · Section 23.3
Systematic shifts, dead time, and stability metrics
Interactive visual model for 23.3 Systematic shifts, dead time, and stability metrics (Technical noise correlations, dead time, and Allan deviation · Why optimal sensing times depend on the noise envelope and cost model · Bridging synthetic sensitivity curves to calibrated clock datasets). Use Animate or Step to follow the physical state evolution and quantitative predictions.
Explore the Apparatus
Adjust the interrogation time \(\tau\), the coherence time \(T_2\), the dead time \(\tau_d\), and the readout phase \(\phi_{\mathrm{ref}}\). Compare the quadrature slope \(\partial p/\partial\Delta\), the fixed-\(N\) frequency uncertainty \(\delta\Delta(\tau)\) (minimized at \(\tau = T_2\)), and the fixed-total-time uncertainty \(\delta\Delta(\tau)\) (minimized at \(\tau = T_2/2\) when \(\tau_d = 0\)).
Interactive model · Idealized predictions
Find a useful interrogation time
On a small screen, swipe the diagram to inspect its components. Moving marks illustrate the model; they are not measured trajectories.
Read the values as a table
How does a phase become a precise estimate of a field or frequency? The text and worked examples remain available without JavaScript. Animation starts only when requested; Step gives a motion-free way to follow the model.
Worked Check
Problem: A single-qubit Ramsey magnetometer has Markovian coherence time \(T_2 = 2.0\text{ ms}\) and zero dead time (\(\tau_d = 0\)). Over a total measurement window \(T_{\mathrm{tot}} = 10\text{ s}\), compare the angular-frequency sensitivity \(\delta\Delta\) achieved at the fixed-\(T_{\mathrm{tot}}\) optimum \(\tau_1 = T_2/2 = 1.0\text{ ms}\) against the sensitivity achieved at \(\tau_2 = T_2 = 2.0\text{ ms}\).
Solution:
1. At \(\tau_1 = T_2/2 = 1.0\text{ ms}\), the number of shots is \(N_1 = T_{\mathrm{tot}}/\tau_1 = 10\text{ s}/(1.0\text{ ms}) = 10{,}000\) and \(W(\tau_1) = e^{-1/2}\):
Operating at \(\tau_1 = T_2/2\) yields a \(14.2\%\) lower uncertainty than operating at \(\tau_2 = T_2\) for the same total integration time.
Summary
A Ramsey quantum sensor encodes an external magnetic field or clock frequency shift into a phase \(\phi = \gamma B\tau\), achieving maximum slope at the quadrature bias point \(\phi_0 = \pi/2\).
For \(N\) uncorrelated probes, quantum projection noise limits phase uncertainty to the Standard Quantum Limit \(\delta\phi_{\mathrm{SQL}} = 1/\sqrt{N}\), whereas \(N\)-partite GHZ or spin-squeezed states reach the Heisenberg scaling \(1/N\) in the absence of decoherence.
Under realistic dephasing at rate \(\gamma\), an \(N\)-qubit GHZ state decoheres \(N\) times faster than a single spin, requiring optimal interrogation times \(\tau \sim 1/(2\gamma)\).
Practice Problems
Problem 23.1: Optimal interrogation time with dead time \(\tau_d\)
When each shot requires interrogation time \(\tau\) plus dead time \(\tau_d\), the number of shots in total time \(T_{\mathrm{tot}}\) is \(N = T_{\mathrm{tot}}/(\tau + \tau_d)\). Show that for Markovian dephasing \(W(\tau) = e^{-\tau/T_2}\), minimizing \(\delta\Delta(\tau) \propto \sqrt{\tau + \tau_d}\,e^{\tau/T_2}/\tau\) gives the quadratic equation \(2\tau^2 + \tau(2\tau_d - T_2) - 2\tau_d T_2 = 0\), and find \(\tau_{\mathrm{opt}}\) when \(\tau_d = T_2\).
Clearing denominators in \(\frac{1}{2(\tau + \tau_d)} + \frac{1}{T_2} - \frac{1}{\tau} = 0\) gives \(\tau T_2 + 2\tau(\tau + \tau_d) - 2T_2(\tau + \tau_d) = 0\), or \(2\tau^2 + \tau(2\tau_d - T_2) - 2\tau_d T_2 = 0\). For \(\tau_d = T_2\), this becomes \(2\tau^2 + T_2\tau - 2T_2^2 = 0\), which has positive root \(\tau_{\mathrm{opt}} = \frac{-1 + \sqrt{17}}{4}T_2 \approx 0.781\,T_2\).
Problem 23.2: Optimal interrogation time under Gaussian inhomogeneous dephasing
Suppose the Ramsey fringe envelope is Gaussian, \(W(\tau) = \exp[-(\tau/T_2^*)^2]\). Find the interrogation time \(\tau_{\mathrm{opt}}\) that minimizes \(\delta\Delta(\tau) = [\sqrt{N}\,\tau\,W(\tau)]^{-1}\) when the number of shots \(N\) is fixed.
Hint
Maximize \(\tau\exp[-(\tau/T_2^*)^2]\) with respect to \(\tau\).
Problem 23.3: Unambiguous frequency capture range of a single Ramsey fringe
At quadrature (\(\phi_{\mathrm{ref}} = -\pi/2\)), the Ramsey signal \(p(\Delta) = \frac{1}{2}[1 + W(\tau)\sin(\Delta\tau)]\) is monotonic in \(\Delta\) only on the central half-fringe \(\Delta\tau \in [-\pi/2, +\pi/2]\). For \(\tau = 5.0\text{ ms}\), what is the maximum unambiguous linear frequency offset \(|\Delta f_{\max}| = |\Delta_{\max}|/(2\pi)\) before the fringe wraps around?
Hint
Set \(2\pi|\Delta f_{\max}|\tau = \pi/2\).
Solution
From \(2\pi|\Delta f_{\max}|\tau = \pi/2\), we find \(|\Delta f_{\max}| = 1/(4\tau) = 1/(20\text{ ms}) = 50\text{ Hz}\).
Key Takeaways
Biasing the second Ramsey pulse by \(\phi_{\mathrm{ref}} = -\pi/2\) places the readout at quadrature (\(p_0 = 1/2\)), maximizing the linear frequency slope \(\partial p/\partial\Delta = \frac{1}{2}\tau W(\tau)\).
With binomial projection noise \(\sigma_{\hat{p}} = 1/(2\sqrt{N})\), the frequency uncertainty is \(\delta\Delta(\tau) = [\sqrt{N}\,\tau\,W(\tau)]^{-1}\).
For Markovian dephasing \(W(\tau) = e^{-\tau/T_2}\), \(\delta\Delta(\tau)\) is minimized at \(\tau_{\mathrm{opt}} = T_2\) for fixed shot count \(N\), and at \(\tau_{\mathrm{opt}} = T_2/2\) for fixed total integration time \(T_{\mathrm{tot}} = N\tau\) with zero dead time.
Practical quantum sensors must balance interrogation time against dead time (the Dick effect), non-Markovian noise envelopes, and long-term systematic drifts quantified by the Allan deviation.
Sources and Further Reading
C. L. Degen, F. Reinhard, and P. Cappellaro, Quantum sensing, Reviews of Modern Physics 89, 035002 (2017).