Projection Noise-Limited Readout

• Projection noise-limited readout is defined by irreducible quantum fluctuations, setting a measurement floor that scales as 1/√N in ensembles.
• Experiments across atomic interferometers, NV ensembles, superconducting circuits, and TPCs demonstrate how reducing technical noise unveils fundamental projection noise limits.
• Advanced techniques such as squeezed-readout and quantum non-demolition mapping enable performance improvements by isolating intrinsic fluctuations from classical noise sources.

Searching arXiv for recent and foundational papers relevant to projection-noise-limited readout across atomic, solid-state, superconducting, and detector contexts. arXiv search query: "projection noise limited readout atom interferometer NV ensemble superconducting TPC" Projection noise-limited readout denotes a measurement regime in which the dominant residual uncertainty is no longer set by technical readout noise, but by the irreducible fluctuations associated with the projection process or with the intrinsic projection-limited formation of the measured signal. In the literature summarized here, the expression has two distinct usages. In quantum metrology, it refers to the binomial fluctuations that arise when populations of independent two-level systems are measured projectively, yielding the familiar $1/\sqrt{N}$ scaling with ensemble size. In detector instrumentation for time projection chambers, it denotes operation in which electronics noise has been reduced sufficiently that intrinsic Landau fluctuations and diffusion dominate localization and calorimetric uncertainty rather than the front-end electronics (Döring et al., 2010, Dwyer et al., 2018).

1. Statistical basis of projection-noise-limited readout

For an ensemble of $N$ identical, independent spin-$\tfrac12$ particles prepared with probability $p$ in the “up” state and $1-p$ in the “down” state, a projective measurement of the collective spin operator $S_z=\sum_i s_z^{(i)}$ yields binomial statistics. The mean and variance are

$$\langle S_z\rangle = N\left(p-\tfrac12\right), \qquad \Delta S_z^2 = N\,p(1-p).$$

At the maximally uncertain point $p=\tfrac12$, one obtains $\Delta S_z=\sqrt{N}/2$. In population-readout language, the uncertainty of the measured fraction is

$$\sigma_p=\sqrt{\frac{p(1-p)}{N}} \simeq \frac{1}{2\sqrt{N}}$$

at $N$0, and the corresponding phase uncertainty in a near-unit-contrast Ramsey readout is $N$1 (Döring et al., 2010, Maier et al., 15 Sep 2025).

Equivalent expressions appear in solid-state spin counting. If $N$2 denotes the number of “up” outcomes, then the population difference $N$3 has variance

$N$4

and the normalized spin-polarization variance becomes

$N$5

at $N$6. This $N$7 variance law, or equivalently the $N$8 sensitivity law, is identified as the hallmark of the Standard Quantum Limit for $N$9 independent spins (Bechelli et al., 12 Jun 2026).

For higher-spin thermal ensembles, the same structure persists in modified form. For spins of length $\tfrac12$0, the thermal projection noise is

$\tfrac12$1

with normalized noise per spin

$\tfrac12$2

Projection-noise-limited readout, in this strict sense, means that photon shot noise, amplifier noise, drift, and other classical contributions have been suppressed below these intrinsic fluctuations, so that the measured variance follows the population statistics of the ensemble itself (Maier et al., 15 Sep 2025).

2. Ramsey interferometry with coherent atoms

A direct experimental realization was reported for a Ramsey-type interferometer using freely propagating coherent atoms. An atom-laser pulse initially in $\tfrac12$3 traverses two spatially separated Raman light sheets, each implementing a $\tfrac12$4-pulse on the two-level system $\tfrac12$5. After free evolution for a time $\tfrac12$6, the second $\tfrac12$7-pulse converts the accumulated phase into a population difference. The transfer probability is

$\tfrac12$8

where $\tfrac12$9 is the fringe contrast (Döring et al., 2010).

The apparatus uses an electro-optic modulator in an inherently stable Sagnac interferometer to generate the two-photon Raman coupling. Because the two counter-propagating beams experience identical fiber and thermal drifts, common-mode noise cancels and the two-photon Rabi frequency remains passively stable. State populations are measured by shot-noise-limited absorption imaging after Stern–Gerlach separation of $p$0 and $p$1, and the resulting $p$2 and $p$3 are used to reconstruct $p$4 and the Ramsey fringe (Döring et al., 2010).

The relevant noise decomposition explicitly separates quantum projection noise from technical terms. At $p$5, the single-shot signal-to-noise ratio is

$p$6

Photon shot noise in the imaging process contributes a variance $p$7, and Raman-laser intensity fluctuations were measured at $p$8, corresponding to $p$9, which lies below the projection-noise level for $1-p$0. The measured fringe contrast is typically $1-p$1–$1-p$2, and for $1-p$3 the experimental $1-p$4 follows the projection-noise line $1-p$5 with no excess noise observed above the fundamental limit. The lowest measured $1-p$6 matches the $1-p$7 expectation for $1-p$8 (Döring et al., 2010).

This realization is important because it isolates a genuine fundamental floor. It also fixes the meaning of “beyond projection noise”: reduced quantum uncertainty, such as spin-squeezed states, is required to surpass the $1-p$9 phase sensitivity of uncorrelated atoms (Döring et al., 2010).

3. Mesoscopic NV ensembles: QND access to intrinsic spin fluctuations

In solid-state spin ensembles, projection-noise-limited readout has historically been obstructed by photon shot noise substantially larger than the intrinsic spin fluctuations. A direct approach was demonstrated in a mesoscopic ensemble of nitrogen-vacancy centers in diamond by combining high magnetic fields, stabilization of the $S_z=\sum_i s_z^{(i)}$0N nuclear spin bath, and repetitive nuclear-assisted readout. The protocol performs a quantum non-demolition mapping of the nuclear spin population onto the NV electron spin, followed by optical fluorescence readout that leaves the nuclear spins nearly undisturbed (Maier et al., 15 Sep 2025).

Operationally, a narrowband microwave $S_z=\sum_i s_z^{(i)}$1-pulse selectively flips the NV electron spin only for one nuclear-spin manifold, a short $S_z=\sum_i s_z^{(i)}$2 532 nm laser pulse reads out and repolarizes the electron spin, and the subtraction of two successive fluorescence buckets removes slow drifts. Repeating the map-plus-readout cycle $S_z=\sum_i s_z^{(i)}$3 times yields a photon-count difference $S_z=\sum_i s_z^{(i)}$4 with Skellam statistics. The total variance obeys

$S_z=\sum_i s_z^{(i)}$5

where $S_z=\sum_i s_z^{(i)}$6 is photon shot noise, $S_z=\sum_i s_z^{(i)}$7 is the total number of photons collected per cycle, $S_z=\sum_i s_z^{(i)}$8 is the optical readout contrast, and $S_z=\sum_i s_z^{(i)}$9 is the spin-projection noise in spin units. In the low-$$\langle S_z\rangle = N\left(p-\tfrac12\right), \qquad \Delta S_z^2 = N\,p(1-p).$$0 regime, the readout is photon-shot-noise dominated; in the large-$$\langle S_z\rangle = N\left(p-\tfrac12\right), \qquad \Delta S_z^2 = N\,p(1-p).$$1 regime, the normalized noise saturates at the projection-noise floor $$\langle S_z\rangle = N\left(p-\tfrac12\right), \qquad \Delta S_z^2 = N\,p(1-p).$$2 (Maier et al., 15 Sep 2025).

The demonstrated operating point is highly specific. At $$\langle S_z\rangle = N\left(p-\tfrac12\right), \qquad \Delta S_z^2 = N\,p(1-p).$$3, the $$\langle S_z\rangle = N\left(p-\tfrac12\right), \qquad \Delta S_z^2 = N\,p(1-p).$$4N nuclear $$\langle S_z\rangle = N\left(p-\tfrac12\right), \qquad \Delta S_z^2 = N\,p(1-p).$$5 is $$\langle S_z\rangle = N\left(p-\tfrac12\right), \qquad \Delta S_z^2 = N\,p(1-p).$$6, enabling up to $$\langle S_z\rangle = N\left(p-\tfrac12\right), \qquad \Delta S_z^2 = N\,p(1-p).$$7 repeats. The confocal volume contains $$\langle S_z\rangle = N\left(p-\tfrac12\right), \qquad \Delta S_z^2 = N\,p(1-p).$$8 NV centers, the typical count rate is $$\langle S_z\rangle = N\left(p-\tfrac12\right), \qquad \Delta S_z^2 = N\,p(1-p).$$9 per cycle, and the crossover to the projection-noise plateau occurs near $p=\tfrac12$0, where $p=\tfrac12$1. The achieved plateau is reported as $p=\tfrac12$2 below the equivalent thermal projection-noise level, and fitting of the finite-$p=\tfrac12$3 model gives $p=\tfrac12$4 and $p=\tfrac12$5 under readout (Maier et al., 15 Sep 2025).

Once this regime is reached, the readout accesses the ensemble’s intrinsic fluctuation structure rather than the detector’s photon statistics. The same platform was then used to distinguish broadened spin-noise distributions produced by spatially correlated noisy microwave driving from the thermal width associated with uncorrelated phonon-induced $p=\tfrac12$6 noise, and to reconstruct signatures of correlated spin states through an XY8-based AC-field spectroscopy protocol. A plausible implication is that projection-noise-limited readout is not only a sensitivity milestone but also a prerequisite for directly observing higher-order collective statistics in mesoscopic solid-state systems (Maier et al., 15 Sep 2025).

4. Room-temperature spin counting and dispersive NV–cQED readout

A separate room-temperature route uses optical spin-to-charge conversion combined with polarization-selective excitation to isolate one crystallographic NV sub-ensemble with high contrast. In that implementation, green $p=\tfrac12$7 laser polarization is aligned to maximize overlap with the optical dipoles of the chosen NV orientation and suppress the off-axis families, raising the native ODMR contrast from $p=\tfrac12$8 to $p=\tfrac12$9. Spin-to-charge conversion then exploits the much larger ionization cross-section of the $\Delta S_z=\sqrt{N}/2$0 excited state relative to $\Delta S_z=\sqrt{N}/2$1, so that the final orange-readout pulse excites only $\Delta S_z=\sqrt{N}/2$2 and yields spin-dependent photoluminescence. The resulting single-shot contrast is $\Delta S_z=\sqrt{N}/2$3, corresponding to a $\Delta S_z=\sqrt{N}/2$4 improvement over standard green readout (Bechelli et al., 12 Jun 2026).

The noise analysis explicitly separates projection noise from classical readout noise. If the ensemble is prepared with probability $\Delta S_z=\sqrt{N}/2$5 in the “spin-up” state, the measured photon-count variance is

$\Delta S_z=\sqrt{N}/2$6

Subtracting the pole variances isolates the projection term,

$\Delta S_z=\sqrt{N}/2$7

For the normalized spin signal, the total noise is

$\Delta S_z=\sqrt{N}/2$8

where the first term is the irreducible projection noise and the second term is classical readout noise. Using diamond nanopillars with apex diameters from $\Delta S_z=\sqrt{N}/2$9 to $$\sigma_p=\sqrt{\frac{p(1-p)}{N}} \simeq \frac{1}{2\sqrt{N}}$$0, the extracted $$\sigma_p=\sqrt{\frac{p(1-p)}{N}} \simeq \frac{1}{2\sqrt{N}}$$1 fits $$\sigma_p=\sqrt{\frac{p(1-p)}{N}} \simeq \frac{1}{2\sqrt{N}}$$2, with best-fit ensemble size increasing up to $$\sigma_p=\sqrt{\frac{p(1-p)}{N}} \simeq \frac{1}{2\sqrt{N}}$$3. For $$\sigma_p=\sqrt{\frac{p(1-p)}{N}} \simeq \frac{1}{2\sqrt{N}}$$4, the projected spin-projection noise is $$\sigma_p=\sqrt{\frac{p(1-p)}{N}} \simeq \frac{1}{2\sqrt{N}}$$5 in normalized units, while the classical contribution remains within an order of magnitude of the SQL, with $$\sigma_p=\sqrt{\frac{p(1-p)}{N}} \simeq \frac{1}{2\sqrt{N}}$$6 at optimum readout time $$\sigma_p=\sqrt{\frac{p(1-p)}{N}} \simeq \frac{1}{2\sqrt{N}}$$7 (Bechelli et al., 12 Jun 2026).

A conceptually different solid-state strategy is dispersive microwave-cavity readout of an NV ensemble. In the off-resonant regime $$\sigma_p=\sqrt{\frac{p(1-p)}{N}} \simeq \frac{1}{2\sqrt{N}}$$8, the driven Tavis–Cummings model reduces to the dispersive Hamiltonian

$$\sigma_p=\sqrt{\frac{p(1-p)}{N}} \simeq \frac{1}{2\sqrt{N}}$$9

Because $N$00, the cavity field acquires a spin-dependent phase shift without inducing real spin flips. The reported analysis indicates near-unity inverse readout fidelity and femtotesla-level sensitivity using a commercially available diamond NV ensemble, with a sensitivity scaling that improves as $N$01 with increasing number of spins $N$02 (Wang et al., 24 May 2026).

Together, these room-temperature optical and cavity-mediated approaches show that projection-noise-limited or near-projection-noise-limited solid-state readout is not restricted to cryogenic or ultracold platforms. One route achieves direct spin counting from nanoscale ensembles of up to 43 spins without cryogenic cooling or high bias magnetic fields; the other seeks the SQL through dispersive collective coupling and cavity-phase readout (Bechelli et al., 12 Jun 2026, Wang et al., 24 May 2026).

5. Vacuum-fluctuation limits and squeezed readout in superconducting circuits

In superconducting qubit measurement, the relevant fundamental floor is often phrased not as spin projection noise but as the vacuum fluctuations of the coherent microwave field used for dispersive readout. For a coherent-state input, the measured quadratures satisfy

$N$03

and the power signal-to-noise ratio of a dispersive projective measurement is

$N$04

in the maximal-rotation limit. This sets the baseline against which squeezed-readout enhancement is defined (Liu et al., 2020).

An unbalanced two-mode squeezed interferometer formed from two Josephson Parametric Converters modifies that noise floor. The first JPC generates two-mode squeezed vacuum, one arm interacts dispersively with the transmon–cavity system, and the second JPC recombines the modes. When the pump phases are tuned for destructive interference of noise, the measured quadrature variance becomes

$N$05

so that

$N$06

In practice, photon loss and finite interferometer gain cap the effective squeezing, but the measured projective-readout performance showed a $N$07 improvement in power SNR relative to coherent-light readout at an optimal $N$08 ($N$09) (Liu et al., 2020).

The same experiment revealed a distinction between projective strength and quantum efficiency. At the bias point maximizing projective-readout SNR, the inferred quantum efficiency was $N$10, whereas tuning the SU(1,1) interferometer to be “as unprojective as possible” yielded $N$11. The reported interpretation is that the unprojective point minimizes internal which-path information leakage and hence reduces dephasing internal to the interferometer (Liu et al., 2020).

This result addresses a recurring misconception. Lowering the apparent readout noise floor is not identical to reaching the fundamental information-theoretic optimum of the measurement. In this setting, maximal projective discrimination and maximal efficiency occur at different operating points, even though both are governed by quantum-limited amplifier physics (Liu et al., 2020).

6. Time projection chambers: projection noise as a detector limit

In liquid argon time projection chambers, “projection noise-limited” has a different, non-quantum meaning. Conventional 2D projective readout uses several planes of parallel wires, each recording one coordinate versus drift time. Ambiguities arise when more than one track or shower projects onto the same set of wires at similar drift times, creating combinatorial ghost points in high-occupancy environments. The same geometry also produces projection-induced noise because each wire channel integrates charge from a large area and carries large capacitance (Dwyer et al., 2018).

Pixelated readout changes this limit. The demonstrated LArPix system used a $N$12 pad spacing, corresponding to $N$13 pads/$N$14, chosen to obtain $N$15 per MIP pad. Each pad has capacitance of order $N$16–$N$17, dominating the total input capacitance. The cryogenic heat-flux constraint is $N$18, so at $N$19 spacing the per-channel power must remain below $N$20. The LArPix ASIC, implemented in $N$21 CMOS with 32 independent channels per chip, provides charge-sensitive amplification, self-triggered digitization, and multiplexed readout from $N$22 to $N$23, while operating below $N$24 per channel (Dwyer et al., 2018).

Its measured electronic noise is sufficiently low that the remaining uncertainty becomes intrinsic to the charge cloud and ionization process rather than to the front end. Unconnected inputs showed $N$25 ENC at $N$26 and $N$27 ENC at $N$28; with pads connected, typical ENC remained below $N$29. For a $N$30-pitch MIP signal of about $N$31, the signal-to-noise ratio is approximately 40 at room temperature and 55 in cold operation. In the paper’s terminology, this yields “projection noise-limited performance”: with pixel ENC $N$32 and $N$33, the dominant uncertainty in 3D point localization and energy deposition becomes the intrinsic Landau fluctuations and diffusion rather than electronics noise (Dwyer et al., 2018).

The same design logic motivates newer TPC charge-readout architectures. GAMPix, developed for the GammaTPC gamma-ray instrument concept, combines coarse and fine scale instrumented electrodes to address “the twin problems of loss of measured charge after diffusion, and high readout power.” According to its abstract, it “enables low noise and ultra low power charge readout at the spatial scale limited by diffusion in a time projection chamber,” with possible applications including future DUNE modules (Shutt et al., 2024).

This detector usage is conceptually distinct from quantum projection noise. In quantum readout, the floor is set by the statistics of projective population measurement in a finite ensemble. In TPC instrumentation, the floor is set by intrinsic charge transport and energy-loss fluctuations once the electronics chain has been pushed below them. The shared phrase therefore identifies an endpoint of readout optimization, but the underlying noise source depends on the measurement modality (Dwyer et al., 2018, Shutt et al., 2024).