All-fiber microendoscopic polarization sensing at single-photon level aided by deep-learning

TL;DR

All-fiber microendoscopic polarization sensing at single-photon level aided by deep-learning addresses the challenge of real-time, full-polarization characterization under extremely low-light and constrained-space conditions. The approach leverages intermodal interference in a short few-mode fiber, sparse sampling by a fiber array, and a neural-network calibrator to map detector counts to polarization states. It achieves single-photon sensitivity, high throughput, and month-long stability, validated on tissue, a USAF birefringent target, diatoms, and fast liquid-crystal transitions. This creates a compact, robust polarimetric tool for microendoscopy and materials science with potential for low-energy sensing and real-time metrology.

Abstract

The polarization of light conveys crucial information about the spatial ordering and optical properties of a specimen. However, precise polarization measurement in challenging conditions, including constrained spaces, low light levels, and high-speed scenarios, remains a severe challenge. Addressing this problem, we introduce a real-time polarization measurement method that is accurate down to a single-photon level and provides complete information about the polarization state. Free of moving components, the polarization sensor utilizes a short rigid piece of few-mode fiber followed by a fiber array and a detector array. The calibration of the sensor relies on a neural network yielding unprecedented accuracy across all polarization states, including partially polarized light. We validate the approach by visualizing the polarization structure of biological specimens and the liquid crystal polymer sample (birefringent USAF test). Our method offers an efficient and reliable solution for real-time polarization sensing and microendoscopy under low-light conditions.

Figures (8)

• Figure 1: A visual representation of the all-fiber polarization sensor. Polarized light emitted from a specimen is collected by a short piece of a few-mode fiber. The intermodal interference generates a granular speckle pattern at the output. Several isolated samples of this pattern impinge a fiber array and propagate to a corresponding number of single-photon detectors. The recorded detections are electronically processed using deep learning methods to characterize the incident polarization state.
• Figure 2: (a) Polarization error quantified by infidelity for varying numbers of active detectors. Variability bars represent uncertainties arising from different subset combinations. The achieved infidelities underscore the sufficiency of sparse sampling of the speckle pattern. (b) The relation between polarization infidelity, the collective number of detected photons across all detectors, and the measurement repetition rate. The colored area indicates the confidence interval of the experimental test set.
• Figure 3: Long-term stability evaluation of the all-fiber polarization sensor employing a 12 mm long few-mode fiber encased in a ceramic ferrule. The system performance was evaluated periodically with a 24-hour repetition interval. The average fidelities (blue dots) are accompanied by a two-parameter linear fit (black line) with a $4.7$$\times$$10^{-4}~\text{day}^{-1}$ negative slope, capturing the slow, gradual decline.
• Figure 4: Visualization of dense connective tissue: (left) an intensity image, (middle) a scan using the all-fiber polarization sensor, (right) an image using a stand-alone polarization microscope. The resulting Bloch parameters of the all-fiber scan are represented as an RGB false-colored image. The three highlighted pixels characterize the purple and green segments in the polarization structure alongside a reference background polarization. Their respective Bloch parameters are $\text{A}=(0.34, -0.43, 0.82),$$\text{B}=(-0.90, 0.05, 0.06),$ and $\text{C}=(-0.04, -0.18, 0.98).$
• Figure 5: Spatial distribution of a diatom birefringent structure measured by the polarimetric all-fiber sensor and characterized using three Bloch parameters. The red color corresponds to changes induced by an anisotropic diatom passing in front of the fiber tip. For comparison, the blue line represents the same environment without the diatom. The modulation visible in all three Bloch parameters indicates the properties of a polarization-affecting element moving in front of the sensor, allowing further study of the specimen.