A fiber optic bundle is a collection of thousands of individual optical fibers packed tightly together so they function as a single unit, carrying light or images from one end to the other. Some bundles preserve the spatial arrangement of fibers so that an image entering one end exits the other end intact; others scramble the arrangement deliberately to spread light evenly for illumination or sensing. This seemingly simple technology sits at the heart of medical endoscopes, astronomical instruments, explosive-detection systems, and a growing number of computational imaging platforms that push well beyond what the raw optics can do alone.
Ordered Versus Unordered Bundles
The single most important distinction in fiber optic bundles is whether the individual fibers maintain the same relative positions at both ends. In an ordered bundle, there is a one-to-one correspondence between where a fiber sits at the input face and where it emerges at the output face. That means a pattern of light entering one end comes out the other end looking like the same pattern, which is what you need if you want to transmit an image. Ordered bundles are often called “coherent” bundles, though this has nothing to do with the optical coherence that laser physicists talk about; it just means the spatial layout is preserved.1RP Photonics Encyclopedia. Fiber Bundles
An unordered bundle, by contrast, deliberately shuffles the fiber positions so that there is a pseudo-random relationship between input and output. This sounds like a defect, but it is actually useful when the goal is uniform illumination rather than image transfer. Shining a lamp into one end of an unordered bundle produces a smooth, evenly distributed light field at the other end, which is exactly what you want for lighting a surgical field or illuminating a sample in a spectrometer.2RP Photonics Encyclopedia. Fiber Bundles
How Fiber Bundles Are Made
Manufacturing a bundle with tens of thousands of individual cores in perfect alignment is not trivial. One common technique is the stack-and-draw method: individual glass rods or preforms are stacked together in the desired arrangement, heated in a furnace, and drawn out into a thin, fused structure. The process is favorable for achieving extremely fine spatial resolution without introducing the kinds of defects that would ruin the image, such as individual cores breaking or shifting out of position during the draw.3PubMed. Study of microbubble defects in high-NA silicate-glass multi-core imaging fibers fabricated by the stack-and-draw method That said, microbubbles and other tiny imperfections can still form in the glass during fabrication, and understanding where those defects originate remains an active area of research.
The resulting bundles can range from a few hundred fibers for specialized sensing applications to tens of thousands of cores for high-resolution imaging. Individual core diameters in imaging bundles are typically just a few micrometers, and the overall bundle diameter may be as small as a millimeter or two, which is part of what makes them so useful for getting into tight spaces like airways, blood vessels, and needle biopsy channels.
The Honeycomb Problem
If you have ever looked through a fiber bundle imaging system and noticed a grid-like pattern overlaid on the scene, you have encountered the most annoying limitation of the technology. Because each fiber core is a discrete light channel separated by cladding material that does not transmit the image, every fiber bundle image comes with a honeycomb-like artifact stamped across it. The pattern arises from the inter-fiber spacing, and it obscures fine details of whatever you are trying to look at.4PubMed Central. Depixelation and enhancement of fiber bundle images by bundle rotation
Think of it this way: a bundle with 30,000 cores can only deliver 30,000 independent pixels of information, and those pixels are arranged in a fixed hexagonal grid with dead zones between them. That is a far cry from the millions of pixels in even a modest digital camera sensor. The honeycomb artifact does not just make images look strange; it can genuinely interfere with clinical or scientific interpretation. A pathologist trying to identify abnormal cells through a fiber bundle endomicroscope, for instance, may struggle to distinguish cellular features that fall in the gaps between cores.
Software Solutions and Deep Learning
Hardware improvements can shrink the core spacing somewhat, but the real breakthroughs in image quality over the past several years have come from computation. One straightforward approach is to physically rotate the bundle during acquisition, capturing slightly shifted versions of the same scene. Software then combines those shifted frames to fill in the gaps, producing a smoother image with finer detail than any single frame could provide.5PubMed Central. Depixelation and enhancement of fiber bundle images by bundle rotation
More recently, deep learning has taken this idea much further. Researchers have trained neural networks to learn the mapping between raw fiber bundle image sequences and their ground-truth high-resolution equivalents. A motion estimation network first aligns a sequence of raw frames, and then a 3D convolutional network reconstructs a single high-resolution image. Testing on resolution targets and tissue samples shows these methods can significantly improve spatial resolution compared to what the bundle delivers on its own.6PubMed Central. Fiber bundle imaging resolution enhancement using deep learning
A separate line of work tackles the honeycomb artifact and low signal-to-noise ratio simultaneously using a physics-informed neural network that alternates between enforcing fidelity to the measured data and applying learned image priors. The network essentially “regenerates” lost spatial information, and because it incorporates a physical model of how the bundle distorts the image, it avoids the kind of hallucinated details that purely data-driven networks sometimes produce.7Defence Technology. Model-based deep learning for fiber bundle infrared image restoration These computational approaches are changing the economics of fiber bundle imaging: instead of needing an expensive bundle with ever-finer core spacing, you can get comparable results from a cheaper bundle paired with smart software.
Medical Endoscopy and Endomicroscopy
The single largest application area for fiber optic bundles is medicine. Flexible, miniaturized bundles can be threaded through the working channels of endoscopes, biopsy needles, and catheters to deliver high-resolution images from inside the body. Fibre-bundle endomicroscopy is currently the most widely used optical biopsy platform and has been approved for clinical use across a range of organ systems.8PubMed Central. Image computing for fibre-bundle endomicroscopy: A review
In practice, a clinician can slide a bundle-based probe down a bronchoscope to examine lung tissue at near-cellular resolution, or pass one through a needle during a pancreatic biopsy to see whether the tissue looks suspicious before committing to a full surgical procedure. The appeal is immediate, real-time visual feedback without needing to remove and process a tissue sample. The tradeoff, as discussed above, is the limited pixel count and honeycomb artifact, which is why the computational enhancement methods are so actively pursued in the medical imaging community.
Beyond imaging, fiber bundles also serve as light-delivery and light-collection pipelines in therapeutic settings. A fiber-coupled dosimeter system, for example, can monitor photodynamic therapy by collecting the faint phosphorescence signal from reactive oxygen species generated during treatment. One such system achieves more than 50 percent optical throughput with enough spectral resolution (around 7 nm) to detect the phosphorescence of singlet oxygen even against a strong luminescence background in living tissue.9Wiley Online Library. High optical-throughput spectroscopic singlet oxygen and photosensitizer luminescence dosimeter for monitoring of photodynamic therapy
Fiber Bundles in Astronomy
Astronomers face a different version of the same fundamental problem that endoscopists do: they need to collect light from many spatial positions simultaneously and deliver it to a detector. In their case, the detector is usually a spectrograph rather than a camera. The MaNGA survey on the Sloan 2.5-meter telescope at Apache Point Observatory uses bare-fiber integral field units, essentially custom fiber bundles, to capture spectra across the full face of nearby galaxies. Each bundle feeds light from a different spatial region of a galaxy into a spectrograph that covers wavelengths from 360 to 1030 nm, allowing astronomers to map how stars, gas, and dust vary from the center of a galaxy to its outskirts.10The Astronomical Journal. The MaNGA Integral Field Unit Fiber Feed System for the Sloan 2.5 M Telescope
The MaNGA survey is designed to observe about 10,000 galaxies in total, making it one of the largest integral-field spectroscopic surveys ever attempted. The engineering challenge is considerable: each fiber in the bundle must be precisely positioned at the telescope’s focal plane, and the output ends must be aligned along a slit so the spectrograph can read them out efficiently. The fact that bare-fiber bundles can be manufactured to the required tolerances at reasonable cost is a big part of what made this survey feasible.
Remote Sensing and Spectroscopy
Fiber bundles are also used to separate the operator from a hazardous sample. In one system designed for standoff detection of explosives, a fiber optic bundle carries laser excitation light to a distant target and collects the Raman-scattered light that comes back. The collected signal is then fed into a spectrometer. Spectra of explosives including TNT, RDX, HMX, and nitroguanidine were obtained in very short acquisition times and matched well with results from conventional laboratory instruments.11PubMed Central. AOTF Raman spectrometer for remote detection of explosives The bundle in this context is not imaging anything; it is simply acting as a flexible light pipe that lets the laser and detector sit at a safe distance from the material being analyzed.
Collection efficiency matters a lot in these applications because the Raman signal is inherently weak. Researchers have shown that it is possible to calculate the collection efficiency of a fiber optic bundle quite precisely if you know the angular acceptance characteristics of its individual fibers, which lets you design bundles optimized for a particular sensing geometry.12Elsevier / Optics Communications. Construction of fiber-optic bundle light-collection systems and calculations of collection efficiency Getting this right can mean the difference between a usable signal and noise.
Tapered Bundles for Magnification and Coupling
Not all fiber bundles are straight cylinders. A tapered bundle has a larger face on one end and a smaller face on the other, so light entering the big end exits the small end at reduced size, or vice versa. This built-in magnification or demagnification is useful for coupling an image to a detector that is a different size than the source. A taper with a roughly 2:1 demagnification ratio, for example, can shrink an image from a large sensor area onto a smaller detector face. Compared to doing the same job with lenses, the fiber-optic taper is considerably shorter, lighter, and more mechanically stable, and it transmits more than an order of magnitude more light.13Elsevier / ScienceDirect. Imaging Characteristics of Rigid Coherent Fiber Optic Tapers
That last point deserves emphasis. The dramatically higher light throughput of a taper versus a lens system means that downstream amplifiers (such as microchannel plates in night-vision or scientific detectors) can be run at lower gain, which reduces electronic noise and extends the life of the amplifier. Tapers found early adoption in military night-vision devices and X-ray imaging systems, and they remain common wherever a compact, high-efficiency optical coupling is needed.
Crosstalk Between Cores
When individual fiber cores in a bundle sit very close together, light can leak from one core into its neighbors. This crosstalk degrades image contrast and blurs fine details, and it gets worse at longer wavelengths where light is less tightly confined to each core. In ordered imaging bundles designed for green-wavelength fluorescence imaging, researchers found that crosstalk between cores of the same structural class becomes a significant concern when the system is pushed toward longer wavelengths. If future bundles are designed for red or near-infrared imaging, managing this inter-core leakage will be a key design constraint.14Optics Letters. Core crosstalk in ordered imaging fiber bundles
This is one reason that simply shrinking core spacing to improve resolution is not a free lunch. Packing cores more tightly increases the potential for crosstalk, especially at wavelengths the bundle was not specifically optimized for. Bundle designers have to balance core size, spacing, cladding thickness, and the target wavelength range, and there is no single geometry that works perfectly for all applications.
Lensless Endoscopy and Wavefront Shaping
Perhaps the most futuristic use of fiber bundles treats the bundle not as a passive image pipe but as a remote phased array. Instead of simply relaying whatever pattern of light falls on the input face, the system uses wavefront shaping to control how light propagates through the bundle. A spatial light modulator at the proximal end shapes the incoming wavefield so that, after passing through the bundle, it focuses to a tight spot at the distal end. By scanning that focus point, you can build up an image without any lens at the far end of the bundle at all.
One implementation combines digital optical phase conjugation with a fast 2D galvo scanner and an adaptive lens to achieve video-rate imaging at 20 frames per second in two dimensions, with subcellular resolution demonstrated in three dimensions at 1 Hz.15PubMed. Video-rate lensless endoscope with self-calibration using wavefront shaping Removing the distal optics means the probe tip can be even smaller and more flexible than a conventional bundle endoscope, which opens the door to imaging in spaces that are currently too tight for any existing instrument. The challenge is that the wavefront correction has to be recalibrated when the bundle bends, since bending changes how light propagates through the cores. Self-calibration algorithms are making this more practical, but the technology is still largely in the laboratory.
What makes this approach so compelling is that it sidesteps the traditional resolution limit of fiber bundles entirely. Instead of being capped at one pixel per core, wavefront shaping can reconstruct images with resolution set by the optical diffraction limit of the whole bundle aperture. In principle, that means a bundle with 10,000 cores could deliver far more than 10,000 resolved points, though in practice the gains depend on how well the wavefront correction holds up under real-world conditions like motion and temperature change.
Choosing Between a Bundle and a Single Fiber
Fiber bundles and single optical fibers get used for very different things, and it is worth understanding when each one makes sense. A single fiber is ideal when you need to deliver light from point A to point B with minimal loss: think telecommunications, laser delivery for surgery, or coupling a light source to a spectrometer input. A bundle is what you reach for when you need spatial information, either transmitting an image or collecting light from multiple points simultaneously.
The tradeoff is cost, complexity, and flexibility. A single fiber is cheap and easy to route through tight bends. A bundle is more expensive, stiffer (especially if the fibers are fused), and introduces the artifacts and crosstalk discussed earlier. For pure light delivery or collection without spatial structure, an unordered bundle can be a good middle ground: it is more flexible than a rigid coherent bundle, collects light over a larger area than a single fiber, and does not need the precise alignment that imaging demands. Each category has a sweet spot, and many real-world instruments actually use both: a single fiber to deliver laser excitation and a bundle to collect the returning signal from across a sample.