🔬 optics

Comb-enabled spectral-domain image transport through perturbation-prone multimode fibers

This paper demonstrates a perturbation-resilient image transport system through multimode fibers by encoding spatial information into spectral signatures via dual-comb spectroscopy and reconstructing images using neural networks, achieving high-fidelity, real-time imaging even under strong fiber perturbations.

Original authors: Maohan Li, Zijian Wang, Bowen Sun, Zhuoren Wan, Xiangze Ma, Xiuxiu Zhang, Yuan Chen, Mei Yang, Qi Wen, Zhaoyang Wen, Ming Yan, Heping Zeng

Published 2026-06-30

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Maohan Li, Zijian Wang, Bowen Sun, Zhuoren Wan, Xiangze Ma, Xiuxiu Zhang, Yuan Chen, Mei Yang, Qi Wen, Zhaoyang Wen, Ming Yan, Heping Zeng

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Problem: The "Shaky Rope" of Light

Imagine you want to send a photograph through a long, flexible garden hose (a multimode fiber optic cable). Normally, light travels through this hose and comes out the other end as a clear picture.

However, if you bend, twist, or shake the hose while the light is traveling through it, the picture gets scrambled into a messy, confusing pattern of dots (called "speckle"). It's like trying to read a message written on a piece of paper that someone keeps crumpling and shaking. Every time the hose moves, the message changes completely, making it impossible to read without constantly re-calibrating your decoder.

The Solution: Turning Pictures into a "Musical Score"

The researchers in this paper found a clever way to stop the shaking from ruining the picture. Instead of trying to fix the messy dots at the end, they decided to change the language of the message before it even enters the hose.

The Translation (Image-to-Spectrum): They use a special device to translate the 2D picture into a "musical score" made of light. Imagine the picture isn't made of pixels, but of specific musical notes. Each part of the image corresponds to a specific "comb line" (a precise color or frequency of light).
The Journey: This "musical score" travels through the shaking hose. Here is the magic: even if the hose is twisted or bent, the pitch of the musical notes doesn't change. The hose might scramble the order of the notes slightly, but the notes themselves remain stable and recognizable. The information is carried by the colors of the light, not the shape of the light beam.
The Listening (Dual-Comb Detection): At the other end, they use a super-fast "ear" (a single detector) to listen to the light. Because they are using two slightly different "metronomes" (dual-comb lasers), the light beats create a radio-frequency signal. This allows them to read all the musical notes simultaneously and instantly, without needing a slow camera.

The Decoder: The "AI Translator"

Once they have the list of musical notes (the spectrum), they use a smart computer program (a neural network) to translate the notes back into the original picture.

Why it's smart: The AI is trained to understand that even if the "notes" are a little noisy or if some are missing (sparse data), it can still guess the full picture. It's like hearing a few notes of a song and instantly knowing the whole melody.
The Result: Even when the hose is being shaken violently, the AI can reconstruct the image with high clarity.

What They Achieved

Unshakeable: The system works perfectly even when the fiber is being bent and twisted. The image quality stays high (over 90% similarity to the original) regardless of the chaos in the hose.
Super Fast: They can take pictures incredibly quickly—up to 2.5 million frames per second. To put that in perspective, they can capture a digital mirror flipping on and off in the blink of an eye, tracking movements that are too fast for the human eye to see.
Efficient: They can send the picture even if they only capture a fraction of the data (compressive sensing), thanks to the AI filling in the gaps.

In Summary

The researchers built a system that stops fiber optic cables from being sensitive to movement. Instead of fighting the chaos of a shaking cable, they turned the picture into a stable "song" of light frequencies. A smart AI then listens to that song and instantly redraws the picture, allowing for high-speed, clear imaging through flexible cables, even when they are being jostled around.

Technical Summary: Comb-enabled Spectral-Domain Image Transport through Perturbation-Prone Multimode Fibers

Problem Statement
Multimode fibers (MMFs) offer a compact geometry and high modal capacity for transporting high-dimensional spatial information, making them attractive for applications such as endoscopy, remote inspection, and optical communications. However, their practical deployment is severely hindered by sensitivity to environmental perturbations. Fiber bending, twisting, and mechanical vibrations induce complex modal coupling, which rapidly alters the output speckle patterns. This instability invalidates conventional speckle-based calibrations and transmission-matrix approaches, which degrade quickly as the fiber state changes. While adaptive wavefront shaping and deep-learning methods have improved robustness, they often require extensive training data, specific feedback mechanisms, or remain limited to perturbations represented in the training set. Consequently, achieving robust, high-speed imaging through MMFs under unknown and time-varying perturbations remains a significant challenge.

Methodology
The authors propose a fundamentally different strategy: encoding images into the spectral domain rather than relying on the spatial speckle field. The system utilizes a dual-comb architecture combined with image-to-spectrum encoding and neural-network-assisted reconstruction.

Image-to-Spectrum Encoding: Two optical frequency combs with slightly offset repetition rates ( $f_{r1}$ and $f_{r2}$ ) are generated. These combs are passed through an encoding unit consisting of a Virtually Imaged Phased Array (VIPA) and a diffraction grating. This setup spectrally resolves the comb lines and maps the one-dimensional comb spectrum into a two-dimensional spot array.
Spatial Modulation: A Digital Micromirror Device (DMD) selectively redirects specific spots from this array. By modulating the amplitudes of the corresponding comb lines, the DMD encodes 2D image information directly onto the optical spectrum.
Transmission: The encoded spectral information is transmitted through a 12-meter-long multimode fiber. Crucially, the spatial information is carried by the spectral signatures of the comb lines, which the authors posit are intrinsically more robust to modal perturbations than the spatial speckle pattern.
Dual-Comb Detection: Upon exiting the fiber, the two combs interfere on a single-pixel photodetector. This generates heterodyne beat notes in the radio-frequency (RF) domain. The RF spectrum is digitized and Fourier transformed, mapping the optical spectral information into the RF domain for massively parallel readout without moving parts.
Neural Network Reconstruction: The extracted RF comb-line peaks are arranged into a 1D tensor and fed into a neural network decoder. The architecture consists of a fully connected layer (FCL) for feature extraction, followed by an attention-enhanced UNet incorporating a Convolutional Block Attention Module (CBAM). This network performs compressive sensing reconstruction to recover the original image from the encoded spectra, even under sparse or noisy conditions.

Key Contributions

Perturbation-Resilient Transport: The paper demonstrates that encoding images into comb-line-resolved spectral signatures allows spatial information to be transported through MMFs with high fidelity, even under strong fiber perturbations (bending, twisting, vibration) that would destroy conventional speckle-based imaging.
High-Speed Single-Pixel Readout: By utilizing dual-comb heterodyne detection, the system achieves massively parallel spectral readout using a single photodetector, avoiding the bandwidth limitations of camera-based detection.
Compressive Recovery: The integration of a neural network decoder enables high-fidelity image reconstruction from sparse, noisy, and spectrally aliased measurements, relaxing the requirements for dense sampling and precise calibration.
Linear Approach: Unlike previous nonlinear frequency conversion methods for MMF imaging, this approach is linear, requiring less optical power and avoiding phase-matching constraints.

Results

Robustness: Under strong fiber perturbations, the system achieved Pearson correlation coefficients (PCC) exceeding 0.9 and maintained high Structural Similarity Index Measure (SSIM) and Peak Signal-to-Noise Ratio (PSNR) values. The reconstruction quality remained nearly identical between stable and perturbed fiber states, confirming that robustness stems from the spectral encoding rather than specific network training on perturbation states.
Complexity Handling: The method successfully reconstructed complex targets, including handwritten digits and dense Chinese characters, despite the limited spatial sampling density (effective resolution of approx. 11×8 spots).
Compressive Imaging: The system demonstrated effective reconstruction under both spatial-domain compression (overlapping spot patterns) and frequency-domain compression (intentional RF spectral aliasing), proving resilience to mixed and ambiguous encoding conditions.
High-Speed Dynamics: The system achieved frame rates up to 2.5 MHz (with a 2 µs integration time per frame). It successfully captured transient switching dynamics of the DMD, tracking target evolution in real-time.
Stability: The dual-comb readout showed excellent reproducibility over 150 measurements with minimal spectral deviation, and the system maintained performance across ten previously unseen fiber perturbation states.

Significance
The paper establishes spectral-domain image transport as a viable and powerful route for robust, real-time imaging through flexible multimode fibers. By shifting the information carrier from the perturbation-sensitive spatial speckle field to the perturbation-resilient spectral domain, the approach eliminates the need for repeated calibration and explicit field reconstruction. The authors claim this establishes a new tool for applications requiring flexible light delivery in complex environments, such as minimally invasive fiber endoscopy, remote sensing, and high-throughput optical monitoring. The combination of high power tolerance (due to the large mode area of MMFs), high-speed single-pixel detection, and perturbation resilience positions this platform as a promising foundation for future fiber-based photonic instruments and miniaturized imaging probes.