From Gaussian to Non-Gaussian Quantum Sensing: Squeezing, Amplification, and Beyond

Quantum noise limits high-precision sensing, such as in gravitational wave detectors. Caves first proposed beating this shot-noise limit by injecting squeezed vacuum at the dark port of an interferometer. Frequency-dependent squeezing through a filter cavity now delivers broadband quantum noise reductions of up to 5. 2 dB at Hanford and 6. 1 dB at Livingston in the LIGO detectors' fourth observing run. Building on these advances, this thesis develops techniques for interferometric sensing under realistic loss and phase noise, spanning Gaussian state preparation, optomechanical readout and feedback control, and non-Gaussian state preparation. In Parts I and II, we work in the two-photon formalism, where states are Gaussian and dynamics are linear. In Part I, we present a waveguide squeezer as an alternative to the traditional bulk cavity-based source; it is single-pass, broadband, and compact, and its modularity opens a path to non-Gaussian state generation. We report the detection of squeezed light from waveguide sources at 1064 nm in the sub-MHz regime, and we phase-stabilize the squeezed vacuum with quantum noise locking. We also analyze the loss and phase-noise tolerance of Gaussian states. In Part II, we design a tabletop phase-sensitive optomechanical amplifier and analyze its performance in terms of cavity stability, control system, and noise budget. We then implement the locking scheme, including Pound-Drever-Hall locking of the cavity, and characterize the noise budget at the initial stage. We also propose using optomechanics for coherent feedback control, which could evade back-action and broaden the detection bandwidth of future gravitational wave detectors. In Part III, we turn to non-Gaussian states. Phase noise does not preserve Gaussianity, so the best input state under combined loss and phase noise need not be Gaussian. We numerically optimize the input state for displacement sensing over the (, _) landscape, integrating the Lindblad master equation in a truncated Fock basis and computing the quantum Fisher information. Within the Gaussian sub-family, the optimizer returns a displaced squeezed state once phase noise is appreciable, rather than pure squeezed vacuum. Without the Gaussian restriction, the optimized states sort roughly into four classes (Fock-like, cubic-phase-like, discrete-rotational, and squeezed), each dominant in a different region of the parameter space. In the experimentally relevant region of 1% loss and 100 mrad phase noise, cat states outperform squeezed vacuum even with balanced homodyne readout, an improvement implementable in the near term. We also show that the non-Gaussian advantage persists at higher photon budgets, well above N = 5. Together, these results show how state preparation, readout, and feedback choices trade off against realistic loss and phase noise, and identify the regimes where non-Gaussian states perform better.