Realization of an all-optical effective negative-mass oscillator for coherent quantum noise cancellation Nived Johny, Jonas Junker, Bernd Schulte, Dennis Wilken, Klemens Hammerer, Michèle Heurs Abstract We report the realization of an all-optical, tabletop effective-negative-mass oscillator (ENMO) scheme capable of canceling quantum noise when cascaded with an opto-mechanical sensor susceptible to (quantum) radiation pressure noise. Our coherent quantum noise cancellation (CQNC) scheme offers a broadband cancellation capability with a tunable, wavelength-flexible, and compact system. This is achieved through the implementation of an optical equivalent of an opto-mechanical interaction, facilitated by a down-conversion and a beam-splitting process Tsang and Caves 2010 . The intricate nature of the system and its multiple interacting components made characterizing the interdependent parameters with conventional methods ineffective, leading to the development of an in - situ characterization scheme. The obtained parameters meet the targets for CQNC set in previous studies Wimmer et al. 2014 . With our current realization, we project a broadband quantum noise reduction of 3.6 dB 3.6\text{\,}\mathrm{d}\mathrm{B} , corresponding to a 77 % 77\text{\,}\mathrm{\%} reduction in quantum back-action noise at the optimal frequency of maximum reduction, indicating the readiness of the ENMO for application. We discuss the prospects for new applications in quantum information and communication using the same platform. I Introduction An opto-mechanical interaction occurs when an optical light field reflects on a mechanical oscillator, a phenomenon that spans diverse scales Aspelmeyer et al. 2014 , from atoms in a light field Ritsch et al. 2013 ; Tebbenjohanns et al. 2021 ; Monteiro et al. 2020 to kilogram-scale mirrors like those used in LIGO and Virgo Capote et al. 2025 ; Ganapathy et al. 2023 ; The Virgo Collaboration et al. 2020 ; Abac et al. 2025 . These interactions have been employed in a wide range of applications, including fundamental science research Chen 2013 ; Safavi-Naeini et al. 2012 ; Murch et al. 2008 , inertial sensors Yver-Leduc et al. 2003 , gravitational wave detection Braginsky and Minakova 1964 ; Braginskiǐ and Manukin 1967a ; Caves 1980 ; Braginsky et al. 1997 ; Schnabel et al. 2010 ; McClelland et al. 2011 ; Danilishin and Khalili 2012 , and even for quantum computing Resch and Karpuzcu 2021 and quantum information processing Barnum et al. 1998 . However, as the sensitivities of these systems have improved due to recent advances, they have become increasingly limited by quantum noise Caves 1981 ; Caves 1980 . To further improve the sensitivity of these systems, it is essential to address the optical power-dependent trade-off between quantum back-action (QBA) noise at lower frequencies and shot noise (SN) at higher frequencies. The limit caused by this trade-off is known as the standard quantum limit (SQL) Caves 1981 ; Braginsky et al. 1992 ; Schnabel et al. 2010 . Several schemes have been developed to achieve sub-SQL sensitivities. While frequency-independent squeezing has been proven effective for narrowband sensitivity enhancement Abadie et al. 2011 ; Lough et al. 2021 ; Aasi et al. 2013 , schemes like variational readout Kimble et al. 2002 ; Kampel et al. 2017 or frequency-dependent squeezing using filter cavities Junker et al. 2022a ; McCuller et al. 2020 ; LIGO O4 Detector Collaboration et al. 2023 ; Virgo Collaboration et al. 2023 have been proven for broadband sub-SQL sensitivity. However, these approaches often require long filter cavities, which are impractical and costly to implement in a general quantum sensor, such as a small inertial sensor. Coherent quantum noise cancellation (CQNC) is an alternative scheme that enables broadband sub-SQL performance using a tabletop setup, and can be adapted for different quantum sensors Tsang and Caves 2010 ; Wimmer et al. 2014 . CQNC employs coherent feedforward control by introducing an anti-noise process to destructively interfere with the back-action noise. The opto-mechanical back-action process can be described as a down-conversion and beam-splitting process, which couples the optical field with the phononic excitation of the mechanical oscillator. The CQNC scheme employs an effective negative mass oscillator which mimics this process, but with an inverted sign. By cascading the ENMO and the opto-mechanical system (OMS), the back-action noise is reduced. While the initial proposal suggested an integration of positive and negative mass oscillators into a single system Tsang and Caves 2010 , later studies have shown that cascading both systems is possible and favorable Schweer et al. 2022 . This cascadability allows the ENMO to be a separate system from the OMS, providing flexibility in the choice and tuning of its operation parameters. A cascaded CQNC implementation using atomic ensembles has been demonstrated successfully Møller et al. 2017 . The wavelength of operation is limited to the atomic transition wavelengths, which are often different from the wavelengths of interest for quantum sensors. This limitation necessitates an additional step of two-color Einstein–Podolsky–Rosen (EPR) entanglement Brasil et al. 2022 . Further quantum - non - demolition (QND) measurements employing a cascaded reference - frame architecture have been realized Woolley and Clerk 2013 ; Caniard et al. 2007 ; Mercier De Lépinay et al. 2021 ; Hertzberg et al. 2010 ; Ockeloen-Korppi et al. 2016 ; Hammerer et al. 2010 ; Clerk et al. 2008 ; Møller et al. 2017 Here, we present an all-optical tabletop realization of an effective negative mass oscillator (ENMO) at a commonly used wavelength of 1064\text{\,}\mathrm{n}\mathrm{m} . The system parameters are precisely tunable, allowing for matching with a particular OMS. The Hamiltonian of an opto-mechanical interaction between an optical and a mechanical mode is recreated in our ENMO between two optical modes of two optical resonators. The coupling between these two optical modes is achieved through an optical 2-mode squeezing process and an independent beam-splitting process, enabling flexibility in the choice of wavelength and adaptability to different OMS configurations, including non-resonant operation. To overcome the challenge of operating a stable coupled optical resonator Gräf et al. 2013 ; Steinmeyer 2019 , we employed a polarization-coupled optical resonator topology with a type II parametric down-conversion. Due to the characterization challenge posed by the interdependent nature of system parameters, we developed a novel characterization method, which enabled the accurate determination of these parameters. Our new method enabled us to achieve the parameter values estimated in a previous publication Wimmer et al. 2014 , making the ENMO ready for cascading with an OMS (for instance, a membrane in the middle resonator with a Si 3 N 4 mechanical oscillator). From our measurements, we project that the ENMO could achieve a broadband cancellation of quantum back-action (QBA) noise of up to 3.6\text{\,}\mathrm{d}\mathrm{B} corresponding to a 77\text{\,}\mathrm{\%} reduction in quantum back-action noise at the frequency of maximum reduction when combined with the OMS. Furthermore, improvement in loss mitigation using state-of-the-art optical components shows great potential, enabling up to 11.9\text{\,}\mathrm{d}\mathrm{B} corresponding to an 81\text{\,}\mathrm{\%} QBA cancellation. Our realization of an ENMO will be further studied and modified for applications in quantum information, such as optical quantum memory and a single-photon source Yoshikawa et al. 2017 ; Yoshikawa et al. 2014 . Figure 1: a) The cascaded CQNC system with the ENMO in series to an OMS and its corresponding quantum noise ellipses at different frequencies are shown for each system. The tailored frequency-dependent inversely squeezed state (blue) destructively interferes with the ponderomotively squeezed state (orange), resulting in vacuum states (green). The scheme reveals the force signal (black arrow), which was masked by the additional phase quadrature noise due to ponderomotive squeezing. b) shows the schematic plots for the ideal susceptibilities of the oscillators. II Theoretical background Quantum back-action noise in an optomechanical system appears as an excess of phase - quadrature fluctuations in the light reflected from a mechanical oscillator. These unwanted phase fluctuations arise because the fluctuating radiation - pressure force exerted by the incident light displaces the oscillator. Since the radiation - pressure force is proportional to the optical power, it is linked to the amplitude quadrature of the incoming field. Consequently, the phase quadrature of the reflected light becomes correlated with the amplitude quadrature, generating a ponderomotive squeezing of the reflected field Braginsky and Minakova 1964 ; Braginskiǐ and Manukin 1967b ; Corbitt et al. 2006 ; Danilishin et al. 2019 . In this process, the anti-squeezing noise projection masks the desired signal in the phase quadrature. The frequency dependence of the effect is governed by the mechanical displacement susceptibility of the oscillator to the applied radiation - pressure force. To perfectly cancel ponderomotive squeezing, the ENMO must provide an opposite, frequency - dependent interaction that generates an “inversely squeezed” state in the light entering the OMS (Fig. 1 a)). This inverse squeezing undoes the optomechanical squeezing, restoring a vacuum - like output and revealing the signal Kimble et al. 2002 . As shown in Fig. 1 b), the mechanical susceptibilities of the two oscillators must be matched in magnitude but opposite in phase. The theoretical basis for this cancellation scheme is presented by Tsang and Caves Tsang and Caves 2010 , while Refs. Wimmer et al. 2014 ; Schweer et al. 2022 discuss its benefits, limitations, and realistic case studies. Here, we summarize in a three-step process the theoretical background for our implementation of an all-optical ENMO where a high-finesse cavity, the ancilla cavity, is the analog of the mechanical oscillator. First, we mimic the radiation pressure interaction by reproducing its Hamiltonian in the optical domain. Second, we will match the susceptibilities of the mechanical oscillator in the OMS and the corresponding optical resonator of the ENMO with opposite phase. Third, we will couple these oscillators in a similar way by matching the readout optical cavities and the coupling strengths. The radiation pressure interaction Hamiltonian of an opto-mechanical cavity as shown in Fig. 2 a) is Schweer et al. 2022 ; Bernd Wolfgang Schulte 2023 \displaystyle\begin{split}H_{\mathrm{\scriptscriptstyle{rp}}}&=\frac{g_{\mathrm{om}}}{2}\left[(bc_{\mathrm{om}}+b^{\dagger}c^{\dagger}_{\mathrm{om}})+(bc^{\dagger}_{\mathrm{om}}+b^{\dagger}c_{\mathrm{om}})\right],\end{split} (1) where g_{\mathrm{om}} is the opto-mechanical coupling strength, c_{\mathrm{om}} ( c_{\mathrm{om}}^{\dagger} ) and b_{\mathrm{om}} ( b_{\mathrm{om}}^{\dagger} ) are the annihilation (creation) operators of the optical cavity mode and the mechanical phonon mode, respectively. The first term of the Hamiltonian denotes a two-mode squeezing process, and the second term denotes beam-splitting processes between the optical mode c and the mechanical mode b . To emulate the same interaction all-optically in a photon-photon mode, one can literally use a beamsplitter and a down-conversion process. As shown in Fig. 2 b), the mechanical oscillator is replaced with an optical oscillator with the power beam-splitter achieving the beam-splitting interaction. The meter cavity c (depicted in red) and the ancilla cavity a (depicted in purple). The 2-mode squeezing process is not represented here. Figure 2: a) Representation of a typical opto-mechanical cavity. Here, the optical resonator (meter cavity) field shown in red interacts with the mechanical oscillator mirror shown in purple. The output field from the opto-mechanical cavity has QBA noise imprinted (ponderomotive squeezing), which is detected using a homodyne detection scheme. b) Here, the opto-mechanical oscillator is replaced by a high-finesse optical resonator (ancilla cavity) shown in purple. The power beam-splitter couples the two oscillators through the beam-splitting process. The down-conversion process is not present here. c) In this realization, the high-finesse ancilla cavity is an orthogonal polarization mode instead of a spatially separated cavity. The opto-mechanical interaction is then mimicked by an optical-optical interaction by using a type II down-conversion crystal (down-conversion process) and a waveplate (beam-splitter process), and careful selection of the oscillator configuration. However, for ease of experimental realization and also to introduce the 2-mode squeezing process, a polarization-coupled cavity scheme is chosen instead of a power-coupled cavity as shown in Fig. 2 c). In this case, the waveplate acts as a beam-splitter coupling power between the two orthogonal polarization modes (see Appendix. A ). A type II nonlinear crystal is used to realize the 2-mode squeezing process between the two polarization modes. The susceptibilities of the mechanical oscillator \chi_{\mathrm{m}} and the ancilla cavity \chi_{\mathrm{a}} dictate the frequencies at which the above-mentioned interaction occurs Wimmer et al. 2014 . Therefore, they are equalized, but with the opposite phase \chi_{\mathrm{m}}=-\chi_{\mathrm{a}} . This gives the following conditions Schweer et al. 2022 : \displaystyle\Delta_{\mathrm{a}} \displaystyle=-\omega_{\mathrm{m}}, (2a) \displaystyle\kappa_{\mathrm{a}} \displaystyle=\gamma_{\mathrm{m}} (2b) \displaystyle|\Delta_{\mathrm{a}}| \displaystyle\gg\kappa_{\mathrm{a}} (2c) where \Delta_{\mathrm{a}} and \kappa_{\mathrm{a}} are the detuning and linewidth of the ancilla cavity, respectively, and \omega_{m} and \gamma_{m} are the resonance frequency and linewidth of the mechanical oscillator, respectively. To ensure equal measurement strength of the ENMO and the OMS, the susceptibilities \chi_{c} and \chi_{\mathrm{om}} of the respective readout cavities must be matched. This gives the condition, \kappa_{\mathrm{c}}=\kappa_{\mathrm{om}}, (3) where \kappa_{\mathrm{om}} is the linewidth of the optomechanical meter cavity and \kappa_{\mathrm{c}} is the linewidth of the ENMO meter cavity. The coupling of the meter cavities to these oscillators are matched with the following condition with their single-photon coupling strengths \displaystyle g_{\mathrm{a}} \displaystyle=g\mathrm{\scriptscriptstyle{BS}}+g_{\mathrm{\scriptscriptstyle{DC}}}=g_{\mathrm{om}}, (4a) \displaystyle g\mathrm{\scriptscriptstyle{BS}} \displaystyle=g_{\mathrm{\scriptscriptstyle{DC}}}, (4b) where g_{\mathrm{a}} is the coupling strength of the ENMO, g_{\mathrm{\scriptscriptstyle{BS}}} is the beam-splitting coupling strength, and g\mathrm{\scriptscriptstyle{DC}} is the down-conversion coupling strength. With these matched, the interaction of the opto-mechanical cavity with the phononic mode is similar to the interaction of the meter cavity with the photonic mode of the ancilla cavity. The two systems are entangled. The inversely - squeezed state is generated when the pump field undergoes a two - mode squeezing interaction, producing a pair of orthogonally polarized, quantum - correlated fields. Within the ENMO the photons in the ancilla cavity are coherently exchanged with those in the meter cavity via a beam - splitting interaction; for perfect cancellation the two - mode - squeezing rate must equal the photon - exchange rate. Only photons exiting the meter cavity are available for detection. The relative phase between the two correlated photons determines the rotation of the squeezed quadrature; this phase acquires a frequency dependence because the ancilla cavity is detuned as in Eq. 2a . Likewise, the asymmetric overlap of the amplitude susceptibilities of the meter and ancilla cavities imparts a frequency - dependent modification of the squeezing magnitude. Consequently, the combined frequency - dependent squeezing factor and squeezing angle constitute the tailored inversely - squeezed state required for back - action evasion. The phase quadrature spectral density of the inversely squeezed state in an ideal CQNC scheme is Schweer et al. 2022 ; Bernd Wolfgang Schulte 2023 \displaystyle\begin{split}S(\omega)_{\mathrm{CQNC}}&=\frac{1}{2}+\frac{G_{\mathrm{a}}^{2}}{2}\left\lvert\chi_{\mathrm{m}}+\chi_{\mathrm{a}}\right\rvert^{2}\\ &\quad+\frac{G_{\mathrm{a}}\kappa_{\mathrm{a}}|\chi_{\mathrm{a}}|^{2}}{2}\left(\frac{\omega^{2}+\kappa_{\mathrm{a}}^{2}/4}{\Delta_{\mathrm{a}}^{2}}+1\right),\end{split} (5) where G_{\mathrm{a}}=g_{\mathrm{a}}^{2}\kappa_{\mathrm{a}}\lvert\chi_{\mathrm{a}}(\omega)\rvert^{2} is the frequency-dependent measurement strength of the ENMO. The corresponding OMS’s measurement strength is G_{\mathrm{om}}=g_{\mathrm{om}}^{2}\kappa_{\mathrm{om}}\lvert\chi_{\mathrm{om}}(\omega)\rvert^{2} . In an ideal CQNC scheme, with perfect matching, the second term vanishes because \chi_{\mathrm{a}}=-\chi_{\mathrm{m}} . The remaining terms are the shot noise of the incoming light field (first term) and the vacuum noise coupling due to the losses in the ancilla cavity (last term). These then determine the noise floor of the CQNC scheme. The study of Schweer et al. 2022 concludes that for the best performance, the EMNO should precede the quantum sensor. It studies the feasibility of realistic conditions and the effects of mismatches of parameters. It was shown that even with a realistic mismatch of the mechanical linewidth \gamma_{m} with the ancilla cavity optical linewidth \kappa_{\mathrm{a}} in Eq. ( 2b ), broadband sub-SQL performance is possible at off-resonance frequencies. Any mismatch between g_{\mathrm{\scriptscriptstyle{BS}}} and g_{\mathrm{\scriptscriptstyle{DC}}} drastically reduces the sensitivity at frequencies above the resonance. Therefore, precise control of g_{\mathrm{\scriptscriptstyle{BS}}} and g_{\mathrm{\scriptscriptstyle{DC}}} in Eq. ( 4b ) is important. Additionally, even for a parameter-mismatched system, a narrow-band enhancement of sensitivity other than at resonance is possible with a specific combination of \kappa_{c} , g_{\mathrm{\scriptscriptstyle{BS}}} and g_{\mathrm{\scriptscriptstyle{DC}}} Schweer et al. 2022 . III Experiment Figure 3: Schematic of the optical realization of the ENMO. It consists of a polarization-coupled cavity system with a type II PPKTP crystal. The cavity is 1.52\text{\,}\mathrm{m} long and has an FSR of 197.4\text{\,}\mathrm{M}\mathrm{H}\mathrm{z} . The input/output coupler has a reflectivity of 97\text{\,}\mathrm{\%} for p-polarization and reflectivity of 99.995\text{\,}\mathrm{\%} for s-polarization. The cavity is pumped with a single-pass 532\text{\,}\mathrm{n}\mathrm{m} light field in s-polarization. The dashed lines represent the vacuum states in p-polarization that are squeezed by the setup. Homodyne detection with a local oscillator in p-polarization detects the inversely squeezed states in the p-polarized light. Experimentally realizing the constraints given by Eqs. 2 , 3 and 4 is challenging. Most considered topologies Steinmeyer 2019 failed for the following reasons: 1. Realization of the beam-splitting process necessitates a coupled-cavity system. Stable operation of the system is difficult owing to the coupling; in particular, locking the lengths of the sub - cavities is challenging because of their interaction. 2. Each sub-cavity requires different line-widths to satisfy Eq. ( 2b ) and Eq. ( 3 ) while keeping the losses to a minimum. To avoid decoherence due to losses, the topology and the components have to be chosen carefully. 3. The coupling strengths g_{\mathrm{\scriptscriptstyle{BS}}} and g_{\mathrm{\scriptscriptstyle{DC}}} have to be precisely and independently tunable to satisfy Eq. ( 4a ) and Eq. ( 4b ). Satisfying Eq. ( 4a ) is crucial to match the ENMO to the OMS and satisfying Eq. ( 4b ) for high-frequency QBA cancellation Schweer et al. 2022 . Furthermore, easy modification of the cavity detuning \Delta_{\mathrm{a}} is necessary to facilitate adaptation to a different mechanical oscillator resonance frequency (refer to Eq. ( 2a )). To tackle the above challenges, firstly, we chose a polarization-based coupled cavity system as shown in Fig. 3 , where the two orthogonal polarization modes (p-pol and s-pol) correspond to the two modes of the ENMO. P-pol mode is the meter cavity mode. S-pol mode is the ancilla cavity mode, which mimics a mechanical oscillator. Locking the length of the two sub-cavities is then simplified to locking the same physical mirrors by the Pound-Drever-Hall technique in reflection using a counter-propagating beam (not shown in Fig. 3 ). Further, the relative detuning, and thus the relative cavity length, between the two polarizations could be tuned by changing the relative retardation in a birefringent bulk in the cavity. The birefringent nature of the type II nonlinear periodically poled potassium titanyl phosphate (PPKTP) crystal used for the down-conversion process is exploited for this by changing its temperature Junker et al. 2022a ; Junker et al. 2022b ; Junker et al. 2025 . The crystal’s polarization basis must be perfectly matched with that of the cavity to avoid the crystal undesirably introducing beamsplitter-like mode-coupling. This is important for the precise control of the coupling strength g_{\mathrm{\scriptscriptstyle{BS}}} . Appendix B.3 contains a more detailed explanation and techniques developed for precise execution. The setup is enclosed in a box to reduce acoustic, thermal, and air fluctuations, thus allowing for sufficient stability. Secondly, to achieve different line-widths \kappa_{\mathrm{a}} and \kappa_{c} for the two cavity modes while keeping losses minimal, an input/output coupler with different reflectivities for the two polarizations was chosen, while all other mirrors are high-reflective (HR) mirrors for both polarizations. In this way, the dominant part of the p-pol cavity decay is detected, thereby increasing the escape efficiency while fulfilling the requirements set by Eqs. 2b , 3 . The s-pol cavity is essentially realized with HR mirrors, with its dominant useful decay channel being the beam-splitting process, coupling it to the p-pol cavity. The different reflectivity for p-pol 97\text{\,}\mathrm{\%} and s-pol 99.995\text{\,}\mathrm{\%} is achieved by using an HR coated mirror for 4\text{\,}\mathrm{\SIUnitSymbolDegree}8\text{\,}\mathrm{\SIUnitSymbolDegree} angle of incidence (AOI) in a 45\text{\,}\mathrm{\SIUnitSymbolDegree} AOI configuration. This creates the unorthodox shape of the cavity as shown in Fig. 3 . Furthermore, the crystal and the waveplate in the cavity are anti-reflective (AR) coated for 1064\text{\,}\mathrm{n}\mathrm{m} wavelength. Further improvement of our system performance could be achieved by using the latest technology of AR-coatings (< 0.01\text{\,}\mathrm{\%} Las ) as it becomes available to us, because this is our dominant loss channel. The losses due to the HR mirrors are negligible. Lastly, the choice of a waveplate instead of an actual beam-splitter facilitates the tunability of g_{\mathrm{\scriptscriptstyle{BS}}} . In previous schemes, the beam-splitter was realized by a conventional power splitter and, accordingly, g_{\mathrm{\scriptscriptstyle{BS}}} could not be tuned Wimmer et al. 2014 ; Tsang and Caves 2010 . The beamsplitter coupling strength for a waveplate is given by g_{\mathrm{\scriptscriptstyle{BS}}}=\sin(2\delta)\,\theta\,\textit{FSR} Steinmeyer 2019 , where \theta is the relative phase shift between the fast and slow axis of the waveplate, \delta is the angle of the waveplate optical axis to the s-pol field and FSR is the free spectral range of the cavity. In previous attempts of polarization-coupled cavities, precise control of g_{\mathrm{\scriptscriptstyle{BS}}} was difficult due to the use of a typical half- or quarter-wave plate where \theta=\pi,~\frac{\pi}{2} is large and fixed Steinmeyer 2019 . Instead, in our case, an Ehringhaus compensator Burri 1953 was used, where we have \theta\approx 0 . Tilting the Ehringhaus compensator gives a degree of freedom to change \theta , thereby changing the sensitivity of g_{\mathrm{\scriptscriptstyle{BS}}} tuning under rotation of \delta . The noise variance measurements were taken at the first FSR due to excess power noise of our laser at baseband. We used a 532\text{\,}\mathrm{n}\mathrm{m} pump field in a single-pass configuration with a pump power of less than 140\text{\,}\mathrm{m}\mathrm{W} . The 1064\text{\,}\mathrm{n}\mathrm{m} coupled cavity has a p-pol linewidth measured to be \kappa_{c}\approx 2\pi\times 1\text{\,}\mathrm{M}\mathrm{H}\mathrm{z} , and the s-pol linewidth was inferred from the measurement fits to be \kappa_{a}\approx 2\pi\times 160\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} . Via an HR mirror, a seed beam is coupled in to weakly displace the vacuum-squeezed states to a bright squeezed state. The p-pol squeezed state is detected using a balanced homodyne detection scheme with an 11\text{\,}\mathrm{m}\mathrm{W} p-pol local oscillator. The AC part of the homodyne detection signal is sent to a spectrum analyzer to measure the noise variance, and the DC part of the signal is sent to an oscilloscope. The DC part depends on the interference between the seed field and the local oscillator field, from which the detection angle can be extracted and controlled. A full schematic of the system, including the locking schemes, is explained in Appendix B.1 . IV Characterization Figure 4: a) Measured variance (normalized to the shot noise, with dark noise subtracted) of the output field as a function of frequency for the ancilla cavity detuning \Delta_{\mathrm{a}} of -465\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} (light blue), -710\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} (light red) and -1050\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} (light green). For each detuning, the variance is shown for two approximately orthogonal detection-locked quadratures. The corresponding fits are shown as dashed lines in matching dark colors. The squeezing measurements correspond to an ENMO for a Si 3 N 4 membrane opto-mechanical system as compared in Table. 1 b) Tomographical reconstruction of the quantum squeezed ellipse over frequency for \Delta_{\mathrm{a}} of -710\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} . ENMO Value/2 \pi OMS Value/2 \pi \Delta_{c} 0\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \pm 3\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \Delta^{\prime}_{\mathrm{om}} 0 MHz \Delta_{\mathrm{a}} -465\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \pm 3\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \omega^{\prime}_{m} \approx 417\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} -710\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \pm 3\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \approx 660\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} -1050\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \pm 3\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \approx 1065\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \kappa_{\mathrm{a}} 160\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \pm 20\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \gamma^{\prime}_{m} \approx 1\text{\,}\mathrm{H}\mathrm{z} \kappa_{c} 980\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \pm 50\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \kappa^{\prime}_{om} \approx 1\text{\,}\mathrm{M}\mathrm{H}\mathrm{z} g_{BS} 175\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \pm 5\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} - - g_{DC} 175\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \pm 5\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} - - g_{BS}+g_{DC} 350\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} \pm 10\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} g^{\prime}_{om} 350\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} Table 1: Shows the values extracted from the fit corresponding to Fig 4 . The parameters of the ENMO and its corresponding OMS equivalent are shown. This was chosen corresponding to a membrane in the middle opto-mechanical cavity with a Si 3 N 4 mechanical oscillator, which is available in our lab. Methods to individually characterize g_{\mathrm{\scriptscriptstyle{BS}}} and g_{\mathrm{\scriptscriptstyle{DC}}} do exist for a simpler system Steinmeyer 2019 . However, the system here with a coupled cavity of different linewidths ( \kappa_{c} and \kappa_{\mathrm{a}} ) and detuning ( \Delta_{c} and \Delta_{\mathrm{a}} ) for the two optical modes makes those methods impractical to implement. Furthermore, setting the experiment in different configurations to measure each system parameter is time-consuming and inaccurate. Therefore, we developed an in situ measurement and fitting routine to quickly characterize the coupling strengths ( g_{\mathrm{\scriptscriptstyle{BS}}} and g_{\mathrm{\scriptscriptstyle{DC}}} ), cavity line-widths ( \kappa_{c} and \kappa_{\mathrm{a}} ), and detunings ( \Delta_{c} and \Delta_{\mathrm{a}} ). This is helpful during applications when the ENMO is to be operated alongside a particular OMS. Fig. 4 a) shows measurement trace (light) pairs of 3 different detunings \Delta_{\mathrm{a}} and their corresponding fits (dashed dark) of the variance over frequencies. Each color pair corresponds to locked detection angles, which are approximately 90\text{\,}\mathrm{\SIUnitSymbolDegree} apart. The fit model is obtained by setting \chi_{m}=0 in Eq. 5 and implementing losses and rotation as in Appendix A.2 . These measurements were taken with a Keysight N9020A spectrum analyzer (resolution bandwidth (RBW) 100\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} ; video bandwidth (VBW) 12\text{\,}\mathrm{H}\mathrm{z} ; sweep time 1.01\text{\,}\mathrm{s} ; average 10). Because the measurements are taken at the first FSR, the measurement with the lowest \Delta_{\mathrm{a}} measurement (blue) is much more prone to effects due to \mathrm{FSR}_{s}\neq\mathrm{FSR}_{p} , and thus deviates from the fits at low frequencies. For baseband detection, this effect does not occur. For the case of \Delta_{\mathrm{a}}= -710\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} , a quantum tomography reconstruction of the squeezed ellipse at different frequencies is also shown in Fig. 4 b) Junker et al. 2022a . The squeezing factor and angles change over frequency, thus creating a tailored inversely squeezed state that can coherently and destructively interfere with a ponderomotively squeezed state from an OMS. Table 1 summarizes the results of the measurement fits. The ENMO was operated considering future operations alongside a membrane in the middle cavity with a Si 3 N 4 (silicon nitride) membrane, whose parameters are also shown. Here, we opted for a resonant OMS; therefore, the p-pol cavity is locked on resonance. The measurements are taken for 3 different s-pol cavity detunings \Delta_{\mathrm{a}} where our Si 3 N 4 membrane ( 500\text{\,}\mathrm{\SIUnitSymbolMicro m} or 1000\text{\,}\mathrm{\SIUnitSymbolMicro m} side length) has its resonances. The value \kappa_{\mathrm{a}}=2\pi\times 160\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} is dominated by the losses in the cavity caused by the non-perfect anti-reflective coatings of the PPKTP crystal and the Ehringhaus compensator. The g_{\mathrm{\scriptscriptstyle{BS}}}=2\pi\times 175\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} was finely tuned to be equal to g_{\mathrm{\scriptscriptstyle{DC}}} using a motorized waveplate rotation stage. This corresponds to an opto-mechanical system with g_{\mathrm{om}}=2\pi\times 350\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} . The values obtained are consistent with the targets set in the past study Wimmer et al. 2014 . The system’s losses can be characterized by operating it as a single-mode squeezer. This is done by setting the waveplate to couple maximally between the s-pol and p-pol modes (i.e., in the analogy of a coupled cavity realization using power beam-splitter, the reflectivity of it is set to zero), thus effectively doubling the cavity length due to the high finesse of the s-pol cavity. In this configuration, the system could be considered as a simple single-mode squeezer to check for unaccounted losses Vahlbruch et al. 2016 . V Discussion and Outlook Figure 5: a) Simulated variance of the noise in the phase quadrature due to ponderomotive squeezing (orange line) and the potential broadband cancellation with the current ENMO (green line), normalized to shot noise. These are shown for the OMS and ENMO parameters given in Tab. 1 . The green-shaded area shows the reduction in noise. The projected noise with a potential future ENMO with improvement of \kappa_{\mathrm{a}}= 10\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} is shown in purple. The corresponding dashed lines show the canceled noise. b) shows the percentage of QBA noise that could be canceled with the current and future projected ENMO realization. Combining the measured ENMO with a simulated OMS, the noise reduction due to CQNC can be projected as in Fig. 5 . Fig. 5 a) shows in orange the phase quadrature noise increase over frequency due to quantum back action noise for an opto-mechanical system with the parameters shown in Table 1 . The green curve represents the reduced noise due to CQNC dictated by Eq. 5 with the currently achieved ENMO parameters. The ENMO suppresses noise for all off-resonance frequencies even with a mismatch of \gamma_{\mathrm{m}} and \kappa_{\mathrm{a}} . The corresponding force spectral density could also be calculated as in Schweer et al. 2022 . The purple curve represents the noise that could potentially be achieved with a future improved ENMO with reduced ancilla cavity linewidth \kappa_{\mathrm{a}}=2\pi\times 10\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} . The dotted lines represent the amount of noise that is reduced by the current (and future) ENMO over frequency. Fig. 5 b) shows the percentage of QBA noise canceled over frequency. With the current ENMO, a maximum noise reduction of 3.6\text{\,}\mathrm{d}\mathrm{B} corresponding to 77\text{\,}\mathrm{\%} could be achieved at frequencies approximately 0.67\times\omega_{m} . With \kappa_{\mathrm{a}}=2\pi\times 10\text{\,}\mathrm{k}\mathrm{H}\mathrm{z} , a maximum noise reduction of 11.9\text{\,}\mathrm{d}\mathrm{B} corresponding to 81\text{\,}\mathrm{\%} could be achieved at frequencies approximately 0.94\times\omega_{m} . The easy tunability of the parameters, such as \Delta_{\mathrm{a}} , g_{\mathrm{\scriptscriptstyle{BS}}} and g_{\mathrm{\scriptscriptstyle{DC}}} , makes the system easy to adapt for different OMS. g_{\mathrm{\scriptscriptstyle{BS}}} can be arbitrarily increased up to the cavity’s FSR. \Delta_{\mathrm{a}} could be increased desirably; however, for a constant g_{\mathrm{\scriptscriptstyle{DC}}} and g_{\mathrm{\scriptscriptstyle{BS}}} , the optical parametric oscillation threshold increases with higher detuning, increasing the demand on pump power. Simply increasing the pump power for higher g_{\mathrm{\scriptscriptstyle{DC}}} leads to the optical parametric oscillation, setting a limit to the maximum achievable g_{\mathrm{\scriptscriptstyle{DC}}} . Contrarily, keeping the pump power constant instead, increasing the detuning decreases the squeezing and anti-squeezing values. With the current system, a tuning range of 0