AI-Assisted Supermode Tracking and Low-Complexity Equalization in Strongly Coupled Multicore Fiber Transmission Systems

The world of high-speed internet relies on a simple but powerful idea: sending more data through the same thin strand of glass. For decades, engineers have pushed the limits of how much information a single fiber-optic cable can carry by packing more colors of light into it. However, there is a physical ceiling to this approach. To break through, scientists are turning to space itself, using a technique called space-division multiplexing. Instead of just one path for light, they are using fibers that contain multiple tiny channels, or cores, running side by side. In some of these advanced fibers, the cores are placed so close together that the light in one core naturally bleeds into its neighbors. This strong connection forces the light waves to travel together as a single, unified pattern known as a supermode. While this offers a way to multiply capacity, it introduces a new problem: the environment. Just as a long road might shift under the weight of a truck or the heat of the sun, these fibers change shape slightly due to temperature swings, bending, or twisting. These tiny shifts alter how the light waves mix together, scrambling the data unless the receiver can constantly relearn the new pattern.

A researcher at Abdullah Gül University has developed a new way to handle this scrambling, using artificial intelligence to keep the data stream clear without overwhelming the computer systems that process it. In their work, they tackled the challenge of maintaining order in these strongly coupled fibers, where the light's path is constantly drifting. They proposed a receiver that does not try to fix the mess after it happens with massive, slow calculations. Instead, it uses a compact, smart neural network to predict exactly how the light is mixing at any given moment. By learning the current shape of the signal's pattern, the system can instantly rotate the data back into its correct alignment before the heavy lifting of error correction begins. This approach allows the system to use much simpler filters to clean up the signal, making it possible to scale these high-capacity fibers to dozens of cores without the processing power needed growing out of control.

The researcher tested their idea through detailed computer simulations that mimicked real-world conditions over long distances. They modeled a fiber with six cores, sending data encoded in a complex format known as 16-QAM, and subjected the signal to the kind of random jiggling and stretching that happens in actual cables. They compared their new AI-assisted method against two older approaches. The first was a static method that assumed the light pattern never changed, which failed quickly as the fiber drifted. The second was a traditional method that used powerful, adaptive filters to fix the mixing, but this required a huge amount of computing power that grew exponentially as more cores were added. The simulations showed that the new AI method successfully tracked the changing patterns of the light. Over a distance of 1,500 kilometers, the AI-assisted receiver achieved a significantly lower error rate than the static method and approached the performance of the heavy-duty traditional system, while offering substantially reduced complexity.

What makes this finding particularly significant is how it changes the math behind the processing. In the traditional approach, adding more cores to the fiber meant the computer had to do a quadratic amount of extra work, meaning the effort doubled and doubled again with every new channel. The new method, by contrast, scales linearly. It treats the problem by first identifying the current pattern and then applying simple, independent corrections to each part of the signal. This means that as fiber systems grow to include twelve or twenty-four cores, the new method remains efficient while the old one would become too slow and expensive to run. The study confirms that by using a lightweight neural network to understand the physical behavior of the light, engineers can build receivers that are both robust against environmental changes and computationally lean.

The researcher did not just rely on theory; they built a complete simulation framework that included the generation of random data, the propagation of light through a fiber with realistic imperfections, and the final decoding of the signal. They found that the AI network could learn the relationship between the messy, mixed-up signals arriving at the receiver and the clean, original patterns they represented. Once the network identified this relationship, it could effectively untangle the signal, allowing the rest of the system to work with much shorter and simpler filters. This hybrid approach, which combines the physical understanding of how light travels with the pattern-recognition power of machine learning, offers a practical path forward. It suggests that future high-capacity networks can handle the inevitable drift of the physical world without needing to build impossibly large supercomputers into every receiver. The work demonstrates that for strongly coupled fibers, the key to unlocking massive capacity lies not in brute force, but in smart, adaptive tracking of the light's own shifting nature.