Quantum-Enhanced Laser Phase Noise Filter

• Quantum-enhanced laser phase noise filters use nonclassical resources, such as squeezed vacuum, to reduce in-loop noise below the classical shot-noise limit.
• They integrate advanced passive cavity, comb filtering, and feedforward techniques to achieve up to 5 dB suppression across key frequency ranges.
• These filters are crucial for precision timing, atomic control, and quantum networking by preserving coherence in complex quantum systems.

A quantum-enhanced laser phase noise filter is a phase-noise suppression system in which the in-loop phase readout is improved beyond the classical shot-noise-limited reference by a nonclassical resource, most prominently squeezed vacuum injection. In the broader literature, however, the same phrase is also used more loosely for quantum-limited passive cavity filters, quantum-system-compatible active phase-noise cancellers, and quantum-channel-compatible fiber links. The strict and broad usages are both well represented: a true squeezed-vacuum-enhanced phase-noise feedback loop was experimentally demonstrated in 2025 (Li et al., 8 Jul 2025), whereas earlier and parallel work developed passive quantum-limited comb filtering (Schmeissner et al., 2014), cavity- and delay-line-based active suppression for AMO control (Chao et al., 2023, Denecker et al., 2024), and phase-coherent optical distribution for quantum networking (Johnson et al., 10 Sep 2025).

1. Concept and terminology

The literature distinguishes three notions that are often conflated. A quantum-enhanced phase-noise filter uses a nonclassical optical state to reduce the in-loop readout noise below the classical shot-noise limit, so that feedback suppresses laser phase noise beyond the classical stabilization floor. A quantum-limited phase-noise filter reaches or approaches the shot-noise-defined standard quantum limit without using nonclassical resources. A quantum-enabling or quantum-channel-compatible phase-noise filter is a classical stabilization architecture whose main significance is that it preserves coherence in a quantum experiment, such as STIRAP, Raman control, TF-QKD, or squeezed-light interferometry (Li et al., 8 Jul 2025, Schmeissner et al., 2014, Johnson et al., 10 Sep 2025).

This distinction is central because several prominent architectures are explicitly not quantum-enhanced in the strict metrological sense. The phase-coherent-fiber drift-correction system of 2025 is described as a “quantum-channel-compatible active phase stabilizer plus laser-frequency-drift compensation system,” not as a filter using quantum resources (Johnson et al., 10 Sep 2025). The PDH-feedforward family is similarly classical but “quantum-enabling,” being designed for precise and high-speed control of atomic and molecular quantum states rather than for beating a quantum readout limit (Chao et al., 2023). The same caution applies to cavity-transmission self-injection locking for trapped-ion experiments (Krinner et al., 2023).

Adjacent technologies further broaden the terminology. Quantum noise filter cavities in gravitational-wave detectors filter squeezed states rather than the source-laser phase itself, and their performance is dominated by optical loss and scatter (Vander-Hyde et al., 2014). Likewise, coherent-control schemes for squeezed-vacuum phase alignment suppress effective quadrature-angle noise rather than free-laser phase noise (Dooley et al., 2014). These systems are closely related in measurement theory, but they are not laser phase-noise filters in the narrow sense.

2. Measurement principle and the quantum enhancement mechanism

The main technical obstacle is that laser phase noise is not directly observable by simple power detection. The 2025 squeezed-vacuum implementation identifies three consequences of this fact. First, phase noise cannot be extracted from the laser output by the same direct scheme used for amplitude-noise stabilization. Second, heterodyne phase readout carries a $3$ dB quantum-noise penalty. Third, direct optical phase readout by a cavity does not measure a pure phase quadrature; it rotates the optical noise ellipse by a small angle, so the detected photocurrent contains both amplitude and phase contributions (Li et al., 8 Jul 2025).

In the direct-readout scheme, an over-coupled cavity rotates the quadratures so that the detected in-loop quadrature is

$$X_{\mathrm{in\text{-}loop}}=\sqrt{t}\,X_{\mathrm{in}}\cos\theta_1+\sqrt{t}\,Y_{\mathrm{in}}\sin\theta_1+\sqrt{r}\,X_s .$$

This relation shows why amplitude-noise contamination is a first-order design problem: the measured signal contains residual laser amplitude noise, converted laser phase noise, and squeezed-vacuum noise simultaneously. The corresponding in-loop noise floor can be written in relative-intensity-noise form as a sum of a residual amplitude term, a converted phase-noise term proportional to $\omega^2/\kappa_1^2$, and a quantum term set by the squeezed vacuum (Li et al., 8 Jul 2025). The architecture therefore requires both a phase-sensitive optical discriminator and a separate suppression stage for excess amplitude noise.

The passive cavity-comb work of 2014 supplies the complementary quantum-limited picture. There the cavity acts as a second-order low-pass filter for comb CEO/CEP noise, with cutoff

$f_c=\frac{c}{F L},$

and becomes effective above about $100$ kHz because the measured cutoff is $f_c\approx 130$ kHz (Schmeissner et al., 2014). The balanced homodyne detector then measures relative phase noise down to the shot-noise floor, establishing the standard quantum limit as the reference. In that framework, the minimum resolvable relative phase noise is tied to

$$\langle\delta\phi^2\rangle_{\rm min}=\frac{8\hbar\omega_0}{P}=4S_{\rm SQL},$$

so passive cavity filtering is not itself quantum enhancement, but it can expose the quantum floor that a genuinely quantum-enhanced readout later surpasses (Schmeissner et al., 2014).

3. Direct experimental realizations

The first experimental demonstration of a quantum-enhanced laser phase noise filter was reported in 2025 (Li et al., 8 Jul 2025). It used a 1550 nm single-frequency fiber laser with 1 W output power, an AOM acting as the fast phase actuator with 200 kHz bandwidth, an OPO generating 1550 nm squeezed vacuum, and a 99:1 beam splitter that mixed the laser with the squeezed vacuum. The in-loop phase readout used an over-coupled cavity with linewidth 7.5 MHz; the out-of-loop monitor used an independent impedance-matched cavity with linewidth 6.8 MHz. The generated squeezed vacuum was 10.6 dB, the total optical efficiency was $88 \pm 0.8\%$, and the total phase fluctuation of the relevant phase-lock loops was $20 \pm 0.9\ \mathrm{mrad}$. Under these conditions, the system achieved 5 dB quantum-enhanced phase-noise suppression below the classical shot-noise-limited reference over 5 kHz to 60 kHz (Li et al., 8 Jul 2025).

A defining feature of that experiment was the use of an SHG stage as an excess-amplitude-noise suppressor before the phase-readout cavity. At about 70% conversion efficiency, the reflected 1550 nm fundamental wave was brought close to the shot-noise limit from the kHz to MHz range. The reported residual amplitude noise was about $-157\ \mathrm{dB/Hz}$, which lay 14.1 dB below the in-loop shot-noise level for 50 µW detected in-loop power. The classical shot-noise-limited reference was $$X_{\mathrm{in\text{-}loop}}=\sqrt{t}\,X_{\mathrm{in}}\cos\theta_1+\sqrt{t}\,Y_{\mathrm{in}}\sin\theta_1+\sqrt{r}\,X_s .$$0, and the squeezing-enhanced out-of-loop curve fell up to 5 dB below it (Li et al., 8 Jul 2025). This established experimentally that phase-noise suppression cannot be quantum-enhanced by amplitude squeezing alone unless the phase readout and amplitude contamination are treated jointly.

A different but closely related landmark is the broadband passive cavity filter for femtosecond comb CEO noise (Schmeissner et al., 2014). That system used a commercial mode-locked Ti:Sapphire oscillator with 25 fs pulse duration, 156 MHz repetition rate, 800 nm center wavelength, and 1 W average power, together with a passive bow-tie cavity of length 1.92 m, effective finesse $$X_{\mathrm{in\text{-}loop}}=\sqrt{t}\,X_{\mathrm{in}}\cos\theta_1+\sqrt{t}\,Y_{\mathrm{in}}\sin\theta_1+\sqrt{r}\,X_s .$$1, measured cutoff $$X_{\mathrm{in\text{-}loop}}=\sqrt{t}\,X_{\mathrm{in}}\cos\theta_1+\sqrt{t}\,Y_{\mathrm{in}}\sin\theta_1+\sqrt{r}\,X_s .$$2 kHz, transmitted bandwidth about 35 nm FWHM, and power transmission about 38%. The cavity reduced CEO phase noise sufficiently that the homodyne phase-noise signal fell into shot noise at about 5 MHz detection frequency, bringing the CEO noise to the quantum limit on microsecond timescales and improving homodyne pulse-timing sensitivity by up to 2 orders of magnitude (Schmeissner et al., 2014).

Theoretical work has also proposed intrinsically quantum-enhanced filtering inside driven quantum systems. In an optomechanical setting with intracavity Kerr nonlinearity and mechanical Duffing nonlinearity, the laser phase-noise coupling is reduced by a squeezing transformation, while a broadband squeezed-vacuum reservoir cancels the associated effective thermal-noise increase (Zeng et al., 2021). In that proposal, the Kerr nonlinearity restrains laser phase noise, the Duffing nonlinearity strengthens the effective optomechanical coupling, and the squeezed-vacuum environment supplies the genuinely quantum part of the enhancement (Zeng et al., 2021).

4. Classical and quantum-enabling architectures often grouped with the term

Much of the modern literature uses the language of phase-noise filtering for architectures that are classical in mechanism but indispensable in quantum experiments. These systems define the engineering background against which genuine quantum enhancement is evaluated.

Architecture	Core mechanism	Reported performance
PDH feedforward (Chao et al., 2023)	Residual PDH signal drives a delayed output EOM	more than 30 dB from $$X_{\mathrm{in\text{-}loop}}=\sqrt{t}\,X_{\mathrm{in}}\cos\theta_1+\sqrt{t}\,Y_{\mathrm{in}}\sin\theta_1+\sqrt{r}\,X_s .$$3 to $$X_{\mathrm{in\text{-}loop}}=\sqrt{t}\,X_{\mathrm{in}}\cos\theta_1+\sqrt{t}\,Y_{\mathrm{in}}\sin\theta_1+\sqrt{r}\,X_s .$$4; up to 42–43 dB near $$X_{\mathrm{in\text{-}loop}}=\sqrt{t}\,X_{\mathrm{in}}\cos\theta_1+\sqrt{t}\,Y_{\mathrm{in}}\sin\theta_1+\sqrt{r}\,X_s .$$5
Adaptive PDH feedforward (Chao et al., 2024)	Feedforward gain normalized by cavity transmission	robust $$X_{\mathrm{in\text{-}loop}}=\sqrt{t}\,X_{\mathrm{in}}\cos\theta_1+\sqrt{t}\,Y_{\mathrm{in}}\sin\theta_1+\sqrt{r}\,X_s .$$6 dB around $$X_{\mathrm{in\text{-}loop}}=\sqrt{t}\,X_{\mathrm{in}}\cos\theta_1+\sqrt{t}\,Y_{\mathrm{in}}\sin\theta_1+\sqrt{r}\,X_s .$$7; bandwidth up to 50 MHz
Delay-line feedback (Parniak et al., 2020)	50 m fiber-delay discriminator and EOM feedback	peak reduction more than 10 dB in 300 kHz-wide bands; below $$X_{\mathrm{in\text{-}loop}}=\sqrt{t}\,X_{\mathrm{in}}\cos\theta_1+\sqrt{t}\,Y_{\mathrm{in}}\sin\theta_1+\sqrt{r}\,X_s .$$8 at $$X_{\mathrm{in\text{-}loop}}=\sqrt{t}\,X_{\mathrm{in}}\cos\theta_1+\sqrt{t}\,Y_{\mathrm{in}}\sin\theta_1+\sqrt{r}\,X_s .$$9
Fully fiberized feed-forward phase-noise eater (Denecker et al., 2024)	Delayed MZI discriminator and downstream fiber EOM	measurement floor less than $\omega^2/\kappa_1^2$0; more than 20 dB from 1 to 10 MHz; up to 30 dB at 3 MHz
Cavity-transmission self-injection locking (Krinner et al., 2023)	Medium-finesse cavity transmission low-pass plus optical self-injection	20 to 30 dB improvement; noise floor between $\omega^2/\kappa_1^2$1 and $\omega^2/\kappa_1^2$2 from 100 kHz to $\omega^2/\kappa_1^2$3 MHz
Phase-coherent fiber plus drift correction (Johnson et al., 10 Sep 2025)	Round-trip fiber-noise cancellation and laser-drift correction	up to 47.5 dB phase-noise suppression; drift from 33.8(1) mHz/s to 0.05(12) mHz/s

The PDH-feedforward line is especially important because it explicitly reconstructs cavity filtering on a high-power output beam rather than using only the weak cavity transmission. In the 2023 implementation, the residual PDH error from a 1013 nm ECDL locked to a ULE cavity of linewidth $\omega^2/\kappa_1^2$4 kHz was applied through a downstream fiber EOM after a 20 m delay fiber, giving more than 30 dB suppression over $\omega^2/\kappa_1^2$5 to $\omega^2/\kappa_1^2$6 and up to 42–43 dB near $\omega^2/\kappa_1^2$7 (Chao et al., 2023). The 2024 adaptive version normalized the gain to cavity transmission and maintained over 40 dB attenuation for 24 hours despite 10% peak-to-peak transmission fluctuations and 1.5 °C laboratory-temperature variation (Chao et al., 2024).

Delay-line discriminators form a second major class. A 50 m fiber-loop interferometer used as a phase discriminator produced suppression exceeding 10 dB in 300 kHz-wide bands around MHz offsets and reached phase noise below $\omega^2/\kappa_1^2$8 at $\omega^2/\kappa_1^2$9 (Parniak et al., 2020). A later fully fiberized version with calibrated delay $f_c=\frac{c}{F L},$0 ns and a downstream fiber EOM achieved more than 20 dB suppression from 1 to 10 MHz, up to 30 dB at 3 MHz, and a measurement floor below $f_c=\frac{c}{F L},$1, then demonstrated improved Raman coherence with two such stabilized lasers (Denecker et al., 2024).

Cavity transmission self-injection locking constitutes a third route. In a 729 nm Fabry–Pérot laser diode system, a 10 cm medium-finesse cavity of linewidth about 140 kHz simultaneously defined the optical resonance and low-pass filtered phase noise beyond the cavity linewidth; the transmitted field was then fed back optically to self-injection-lock the diode. The resulting phase noise in the 100 kHz to $f_c=\frac{c}{F L},$2 MHz band was suppressed to between $f_c=\frac{c}{F L},$3 and $f_c=\frac{c}{F L},$4, representing a 20 to 30 dB improvement over a state-of-the-art PDH-stabilized ECDL (Krinner et al., 2023).

Finally, phase-coherent-fiber stabilization shows how a “filter-like” active phase suppressor can be integrated with slow laser-drift correction for networked quantum optics. In that architecture the stabilized fiber itself acts as an optical delay line for drift sensing, while the same AOM corrects both fast fiber noise and slow source-laser drift. The system reached Allan deviation coefficients $f_c=\frac{c}{F L},$5 for a 3.3 km field-deployed fiber and $f_c=\frac{c}{F L},$6 for a 71 km spool fiber, with frequency-drift reduction from 33.8(1) mHz/s to 0.05(12) mHz/s in absolute-reference mode (Johnson et al., 10 Sep 2025).

5. Applications in quantum metrology, control, and networking

In precision timing and comb metrology, the dominant limitation can be CEO phase noise rather than amplitude noise. The passive comb-cavity work showed that repetition-rate-related noise was more than 60 dB below CEO phase noise, that CEO noise above the lock resonance followed approximately $f_c=\frac{c}{F L},$7, and that cavity filtering reduced the measured relative phase-noise slope to $f_c=\frac{c}{F L},$8. Because the minimum resolvable timing jitter scales with $f_c=\frac{c}{F L},$9, passive cavity filtering improved homodyne pulse-timing sensitivity by up to 2 orders of magnitude (Schmeissner et al., 2014).

In AMO coherent control, the same spectral region is often the one that matters most physically. Feedforward cancellation of cavity-stabilized laser phase noise in ultracold RbCs molecular STIRAP improved the one-way transfer efficiency to 98.7(1)%, with the error per passage reduced by a factor of 4.5(5). The same experiment showed that the effective dephasing time increased from $100$0 to $100$1 after feedforward suppression of the MHz-scale servo bump (Maddox et al., 2024). In cold-atom Raman control, a phase-noise eater based on delayed interferometric measurement removed a 700 kHz servo bump by more than 20 dB, eliminating the associated collapse-and-revival pattern in Ramsey contrast (Denecker et al., 2024). For trapped-ion optical qubits, cavity-transmission self-injection locking reduced fast phase noise in the 100 kHz to $100$2 MHz range and was explicitly connected to avoiding incoherent spin flips and enabling two-qubit operations with error below $100$3 (Krinner et al., 2023).

Gate-fidelity theory clarifies why these frequency bands recur. For one-photon Rabi oscillations, the average gate error is

$100$4

so the driven qubit acts as an effective spectral filter that is maximally sensitive near $100$5 (Jiang et al., 2022). This is why servo bumps near the Rabi frequency are particularly harmful, and why changing the Rabi frequency can itself be a practical phase-noise filtering strategy at the control-protocol level (Jiang et al., 2022).

In quantum networking, the role of active phase-noise filtering is less about single-laser linewidth and more about preserving distributed coherence. The phase-coherent-fiber drift-correction system was designed to distribute nearly monochromatic photons that remain ultra-stable in both frequency and phase over fiber links. Using the TF-QKD scheme of Liu et al. and an integration time of $100$6, the authors estimated that replacing uncompensated fibers by the stabilized PCF links reduced the differential phase error from $100$7 to $100$8, and the channel-induced QBER term $100$9 from about $f_c\approx 130$0 to about $f_c\approx 130$1, corresponding to a ratio of about 73.7 (Johnson et al., 10 Sep 2025).

6. Limitations, losses, and conceptual boundaries

The main practical limit in true quantum-enhanced filtering is not the actuator but the in-loop measurement chain. In the squeezed-vacuum experiment, the gap between generated 10.6 dB squeezing and observed 5 dB enhancement was traced to three classes of degradation: 1.9 dB from optical loss, 0.6 dB from phase fluctuations, and 2.2 dB from noise cross-coupling. The reported cross-coupling at 8 kHz was $f_c\approx 130$2 electronic noise, $f_c\approx 130$3 residual excess amplitude noise, and $f_c\approx 130$4 in-loop frequency noise, giving a total of $f_c\approx 130$5 (Li et al., 8 Jul 2025). These figures define the central lesson of the field: quantum enhancement becomes visible only after the classical amplitude-, electronic-, and residual-phase-noise terms are pushed below the squeezed-vacuum-limited quantum floor.

Bandwidth extension introduces a different set of limits. Passive cavities filter only above their linewidth and inevitably trade transmission, dispersion, and bandwidth; the Ti:sapphire comb cavity, for example, became useful only above about 100 kHz, transmitted about 35 nm FWHM out of a 45 nm FWHM spectrum, and passed only about 38% of the power (Schmeissner et al., 2014). Feedforward systems are intrinsically sensitive to gain and delay mismatch. In the PDH-feedforward analysis, achieving attenuation better than 40 dB required the gain to satisfy $f_c\approx 130$6, while maintaining better than $f_c\approx 130$7 dB attenuation under cavity-transmission drift required transmission variation smaller than 1% (Chao et al., 2023). In the fully fiberized delay-line instrument, achieving $f_c\approx 130$8 dB suppression at 10 MHz demanded delay error below 0.5 ns, about 10 cm of cable (Denecker et al., 2024).

Fiber-based systems face additional physical floors. In the MHz-delay-line feedback experiment, thermoconductive and thermorefractive fiber noise dominated the detection floor up to around 2 MHz and contributed about 70% of the detection noise in the in-loop setup at 1.5 MHz (Parniak et al., 2020). In long phase-coherent links, environmental acoustic, seismic, and thermal perturbations, Rayleigh backscatter, and temperature-induced drift of local short fibers all contaminate the phase signal. The PCF work therefore required double integration in self-referenced drift correction to decorrelate true source drift from ambient-temperature-induced path-length drift (Johnson et al., 10 Sep 2025).

A final boundary concerns what phase-noise filtering is not. Some work uses laser phase noise as a resource rather than a nuisance: in laser-phase-noise QRNGs, spectral filtering is used to extract the flat randomness-rich portion of the detected interference spectrum, not to suppress source phase noise (Yang et al., 2023). Other work develops quantum-enhanced phase sensors rather than laser phase-noise filters: the integrated lithium-niobate phase sensor generated $f_c\approx 130$9 squeezing with 26.2 mW optical power and improved phase-measurement SNR by $$\langle\delta\phi^2\rangle_{\rm min}=\frac{8\hbar\omega_0}{P}=4S_{\rm SQL},$$0, but it did not itself stabilize a laser source (Stokowski et al., 2022). These distinctions are important because they locate the quantum-enhanced laser phase noise filter at the intersection of three separate agendas: nonclassical readout below the shot-noise limit, broadband or low-loss classical actuation, and application-specific coherence preservation in quantum technologies.