🔬 optics

FLASH: Ultrafast beam quality characterization via spatial-to-temporal mapping

This paper introduces FLASH, a deep learning-enabled technique that converts 2D spatial beam profiles into 1D temporal signals using fiber arrays to achieve ultrafast 100 MHz beam quality characterization, overcoming the speed limitations of conventional cameras for real-time monitoring of complex nonlinear laser dynamics.

Original authors: J. Qiu, Y. Xiong, X. Hua, H. Wu, M. Tang

Published 2026-06-08

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: J. Qiu, Y. Xiong, X. Hua, H. Wu, M. Tang

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

Imagine you are trying to take a photo of a hummingbird's wings. The problem is that your camera is too slow; by the time it snaps the picture, the wings have already moved, and you just get a blurry mess. This is exactly the problem scientists face when trying to measure the quality of modern, super-fast laser beams.

Lasers today are so powerful and complex that their shape changes in the blink of an eye (nanoseconds). Traditional cameras are like slow-motion photographers; they can only take a few thousand pictures per second. But the laser is changing a million times faster than that. To see what's happening, you need a camera that is five orders of magnitude (100,000 times) faster.

Enter FLASH, a new invention by researchers at Huazhong University of Science and Technology. Think of FLASH not as a camera, but as a "spatial-to-temporal translator."

Here is how it works, using a simple analogy:

1. The "Fingerprint" Maker (The Multimode Fiber)

Imagine you have a complex, colorful pattern drawn on a piece of paper (this is the laser beam's shape). Instead of taking a photo of the whole paper, you shine the light into a special, thick glass rope called a Multimode Fiber.

Inside this rope, the light bounces around wildly, mixing and interfering with itself. When it comes out the other end, the original pattern has been scrambled into a unique, complex speckle of light—like a high-tech fingerprint. Crucially, this fingerprint changes instantly if the original pattern changes even slightly. The fiber has compressed the entire 2D picture into a single, messy, but information-rich "smear" of light.

2. The "Time-Travel" Line (The Multicore Fiber)

Now, we have this fingerprint, but we still can't take a picture of it fast enough. So, the researchers use a clever trick. They split the light coming out of the fiber into seven separate strands (like a 7-lane highway).

They then make each lane a slightly different length. Imagine running a race where seven runners start at the same time, but they have to run different distances.

Runner 1 runs 10 meters.
Runner 2 runs 10.2 meters.
Runner 3 runs 10.4 meters... and so on.

Because they run different distances, they arrive at the finish line one by one, not all at once. The researchers did this with light using a Multicore Fiber delay line. They turned the "side-by-side" information (spatial) into a "one-after-another" stream (temporal).

3. The "Super-Speed" Detective (The Single Detector)

Instead of a camera with millions of pixels, FLASH uses a single, incredibly fast light detector (like a super-fast eye). Because the seven strands of light arrive one after another, this single detector sees a rapid sequence of seven light pulses.

It's like listening to a drumbeat where the rhythm tells you the shape of the drum. The detector doesn't need to see the whole picture; it just needs to hear the "beat" of the seven pulses.

4. The "Brain" (Deep Learning)

The detector sends this rapid sequence of seven numbers to a computer brain (a Deep Learning model). This brain has been trained on millions of examples. It knows that "if the first pulse is bright and the last is dim, the laser shape was X." If the pattern changes, the brain instantly calculates the quality of the beam.

The Result

The paper claims this system can measure the laser beam's quality 100 million times per second (100 MHz).

Speed: It is 100,000 times faster than standard cameras.
Accuracy: It is incredibly precise, with an error rate of only 0.32%.

Why does this matter?

The authors say this acts like a "spatial oscilloscope." Just as an oscilloscope lets you see how a sound wave changes over time, FLASH lets you see how a laser beam's shape changes over time.

This allows scientists to finally "watch" ultra-fast events that were previously invisible, such as:

Spatio-temporal mode-locking: When different parts of a laser lock together in a complex dance.
Transient beam self-cleaning: When a messy laser beam suddenly organizes itself into a perfect shape in a split second.
Plasma-induced aberrations: When a laser hits a target and creates a plasma cloud that distorts the beam, requiring instant correction.

In short, FLASH turns a slow, blurry problem into a fast, clear stream of data, allowing us to finally keep up with the speed of modern lasers.

Technical Summary: FLASH – Ultrafast Beam Quality Characterization via Spatial-to-Temporal Mapping

Problem Statement
Accurate, real-time monitoring of spatial beam quality is a prerequisite for intelligent optical field regulation and advanced laser applications, ranging from precision manufacturing to high-energy directed systems. However, modern high-power and multimode optical systems exhibit complex, nonlinear, and transient behaviors, such as spatio-temporal mode-locking and transient beam self-cleaning, where spatial profiles reorganize within nanosecond timeframes. Current characterization methods face a fundamental bottleneck: conventional two-dimensional image sensors (CCD/CMOS) are limited to kilohertz frame rates, which is orders of magnitude too slow to capture these ultrafast spatial dynamics. While computational reconstruction and deep learning methods have improved processing speeds, they remain constrained by the data acquisition rate of 2D cameras, leaving critical ultrafast spatial phenomena unresolved.

Methodology
To overcome the dimensional and temporal barriers of traditional imaging, the authors propose FLASH (Fiber-based Laser Assessment via Spatial-to-temporal High-speed-mapping). This non-imaging technique transforms two-dimensional spatial beam information into high-speed one-dimensional temporal pulse sequences. The system operates through three primary stages:

Spatial Encoding: A laser beam is coupled into a large-core Multimode Fiber (MMF). The incident light excites multiple guided modes that interfere, creating a complex speckle pattern at the distal end. This speckle acts as a high-dimensional, deterministic fingerprint encoding the input beam's spatial phase and amplitude variations.
Spatial-to-Temporal Serialization: The 2D speckle pattern is sampled by a 7-core Multicore Fiber (MCF). The cores are spliced to a fan-out device and connected to a fiber delay line array with precisely calculated length increments ( $\Delta L$ ). This converts the spatially distributed signals into a serial temporal sequence, ensuring pulses from different cores arrive at the detector sequentially without overlap.
Detection and Reconstruction: A single high-speed photodetector (20 GHz bandwidth) captures the serialized temporal pulse sequence. The resulting 7-dimensional intensity vector is processed by a deep Multi-Layer Perceptron (MLP) neural network. The network, trained on a large-scale dataset of simulated and experimental beam profiles with ground-truth $M^2$ labels, deciphers the serialized signals to predict the beam quality factor.

Key Contributions

Paradigm Shift: The paper introduces a spatial-to-temporal mapping architecture that bypasses the frame rate limitations of 2D sensors, enabling beam quality monitoring at speeds previously unattainable for spatial profiles.
Hardware Integration: The system integrates MMF-based diffractive encoding with MCF delay lines to compress rich spatial information into a format compatible with single-pixel, high-bandwidth detection.
Deep Learning Decoding: A specialized 11-layer MLP is developed to map the non-linear relationship between the serialized speckle fingerprints and the $M^2$ factor, achieving high precision despite the drastic reduction in spatial data points (from millions of pixels to 7 intensity values).

Results

Measurement Speed: The FLASH system achieves a measurement rate of 100 MHz, representing a five-order-of-magnitude improvement over standard camera-based methods.
Accuracy: The system demonstrates a minimal Mean Relative Error (MRE) of 0.32% across an $M^2$ factor range of 1.0 to 5.0. The Pearson correlation coefficient between predicted and ground-truth values exceeds 0.99.
Robustness: The model maintains high accuracy under varying modal complexities (up to 10 mixed modes) and environmental perturbations. The study notes that while increasing the number of sampling cores generally reduces error, a 7-core configuration offers an optimal balance between information richness and maximum repetition rate.
Performance Validation: Training and validation curves indicate strong generalization capabilities with no significant overfitting, even with inherent stochastic variations in the pulse sequences caused by system noise.

Significance
The authors posit that FLASH serves as a "transformative spatial oscilloscope," effectively bridging the gap between slow spatial imaging and the ultrafast demands of modern laser physics. By enabling nanosecond-level dynamic characterization, the technique unlocks the ability to observe hidden high-speed physical mechanisms, such as spatio-temporal mode-locking and transient beam self-cleaning. Furthermore, the system's ultra-high bandwidth and low latency provide a robust hardware foundation for real-time intelligent adaptive control, including the suppression of Transverse Mode Instability (TMI) in large-mode-area fiber lasers and the compensation of plasma-induced aberrations in precision manufacturing. The paper concludes that this spatial-to-temporal mapping paradigm offers a scalable diagnostic framework essential for the next generation of high-power and multimode laser systems.