High-Speed Multi-Dimensional Optical Field Measurement via MMF-MCF Spatial-Temporal Mapping Architecture

Technical Summary: High-Speed Multi-Dimensional Optical Field Measurement via MMF-MCF Spatial-Temporal Mapping

Problem Statement
The simultaneous acquisition of high-dimensional optical information, specifically wavelength and state of polarization (SOP), is critical for applications in optical communications, material characterization, and biomedical imaging. Conventional methodologies relying on cascaded discrete components (e.g., diffraction gratings, rotating waveplates) suffer from large physical footprints and mechanical alignment constraints. While nanophotonic on-chip spectropolarimeters offer miniaturization, they often face a trade-off between ultra-high spectral resolution and fabrication scalability or cost. Furthermore, existing multimode fiber (MMF) speckle-based architectures, though cost-effective, are fundamentally limited by the frame rates and data-transfer latencies of camera-based image sensors (CCD/CMOS), precluding the observation of transient optical phenomena. There is a specific gap in achieving high-speed, single-shot demodulation of both wavelength and SOP from sparse signals.

Methodology
The authors propose a compact, high-dimensional optical field analyzer utilizing a discrete spatiotemporal sampling architecture based on the integration of multimode fiber (MMF) and multicore fiber (MCF).

• Physical Principle: Coherent light launched into an MMF generates a complex spatial speckle pattern at the output facet due to inter-modal interference. This pattern is deterministically reconfigured by variations in input wavelength or SOP.
• Spatial-Temporal Mapping: Instead of capturing the full 2D speckle pattern with a camera, the system employs a 7-core MCF to discretely sample the spatial intensity distribution at specific coordinates. These parallel spatial signals are converted into a serial temporal sequence using a fiber delay line array (with 0.2 m increments), creating a single-shot sparse temporal pulse sequence of seven distinct peaks.
• Signal Decoding: A residual multilayer perceptron (ResMLP) neural network is employed to decode the 1×7 discrete optical intensity pulse signals. The network architecture includes a projection layer mapping inputs to a 1024-dimensional latent manifold, followed by six stacked residual blocks for feature extraction, and a hierarchical decoding structure to output the wavelength and three Stokes parameters.
• Experimental Setup: The system uses a tunable laser (1549.64 nm) modulated by an electro-optic modulator (100 MHz repetition rate, 0.5 ns pulse width). A motorized polarization controller (MPC) modulates the input SOP. The MMF (5 m, 105/125 µm) is spliced to a 7-core MCF, which feeds into a delay line array and a high-speed photodetector connected to a digital storage oscilloscope.

Key Contributions

1. Architecture: Introduction of a spatiotemporal mapping framework that transforms 2D spatial speckle patterns into 1D temporal pulse sequences, enabling high-speed single-pixel detection without the latency of camera-based systems.
2. Algorithm: Implementation of a ResMLP network capable of simultaneously demodulating wavelength and full-Stokes polarization from highly downsampled signals (only 7 sampling points).
3. Sampling Analysis: Systematic investigation into the trade-off between spatial sampling density and reconstruction accuracy, identifying the minimum number of sampling points required for multi-parameter analysis.
4. Robustness Verification: Demonstration of isotropic fault tolerance, proving that optical field information is redundantly encoded across the fiber cross-section rather than localized in specific channels.

Results

• Performance Metrics: Under a standard full-sampling configuration (7 cores), the system achieves a wavelength mean absolute error (MAE) of 0.25 pm (with a spectral resolution of 2 pm) and a polarization resolution of 0.2015 (in normalized Stokes space, measured as mean Euclidean distance).
• Speed: The system operates at a 100 MHz repetition rate, significantly surpassing the kHz or Hz limits of camera-based spectropolarimeters.
• Sampling Density: Analysis reveals that 5 to 6 sampling points are the minimum required to balance measurement rate and accuracy. Reducing the core count below five leads to a rapid degradation in wavelength estimation accuracy, while polarization accuracy shows a linear correlation with the number of cores.
• Fault Tolerance: Experiments involving the iterative masking of single cores show that the system maintains stable performance regardless of which core is removed. Even under large dynamic range conditions (20 pm intervals, 40-step SOP variations), the polarization error remains stable between 0.20 and 0.24, confirming that no single core is critical.
• Comparison: Compared to existing schemes (e.g., Refs [13, 26]), this work achieves full-Stokes demodulation and picometer-level spectral resolution at 100 MHz, whereas camera-based approaches are limited by readout speeds.

Significance
The paper claims that this framework provides a viable solution for multiparameter decoupling under conditions of severe spatial downsampling (only 6.3% of the near-field speckle information is sampled). By validating that optical field information is redundantly encoded across the entire fiber cross-section, the work offers useful insights for designing next-generation, high-speed, and robust all-fiber analysis systems. The architecture effectively bridges the gap between high-resolution metrology and the need for real-time, high-speed data acquisition, relaxing the requirements for precise alignment and splicing angles in practical implementations.