News

Queen Mary University of London researchers develop intelligent communications system that dramatically improves data transmission when bandwidth is limited.

Anyone who has struggled through a poor phone call knows the frustration of missing half a conversation. Now, researchers at Queen Mary University of London, led by Paul Anthony Haigh, have developed an artificial intelligence system designed to solve an equivalent problem in the world of high-speed optical communications.

This groundbreaking work has the potential to address many real-world, sustainability, challenges industries face today. From reducing power consumption in data centres and national-scale internet links, to improving laser-based communication between aircraft, drones and spacecraft, and creating a more resilient communication in situations where bandwidth is limited, or conditions change rapidly.

Published in NPJ Wireless Technology, the research introduces a novel machine learning technique that helps optical receivers recover information that would otherwise be distorted or lost when communications channels become congested or bandwidth is restricted.

Rather than replacing traditional communications engineering with artificial intelligence, the researchers have combined the best of both worlds. Their system allows AI to make small, intelligent adjustments to an existing receiver design, preserving the reliability of established engineering while adding the adaptability of modern machine learning. The result is a communication system that becomes smarter only when it needs to be.

Experimental testing showed the approach significantly outperformed conventional receivers under challenging conditions, reducing communication errors by more than 40 per cent in severely bandwidth-limited scenarios. In some operating conditions, the technique reduced estimated bit error rates by two times. Just as importantly, when communication conditions were already favourable, the AI recognised it had nothing to add and automatically reverted to the conventional receiver, ensuring performance was never compromised, whilst saving energy that other AI-based communication systems might spend.

The breakthrough addresses one of the central challenges facing future communications. As demand for data continues to grow—from satellite internet and deep-space missions to airborne networks and next-generation wireless systems—engineers must find ways to transmit more information through limited bandwidth without dramatically increasing complexity or cost.

"Current AI approaches replace conventional communication systems with "black box" models that are computationally demanding and difficult to interpret. This research takes a different path and asks the AI to deviate from classical information theory if it improves the link performance. Instead of asking AI to do everything, it lets decades of communications engineering do most of the work, while AI makes small corrections only when they're needed. It's a partnership between human engineering knowledge and machine learning rather than a replacement for either." says Dr. Paul Anthony Haigh.

By allowing artificial intelligence to enhance rather than replace established communications theory, the Queen Mary team has demonstrated a practical route towards more resilient and efficient optical communication systems.

Although the technology remains at the research stage, it has the potential to support future satellite communications, laser links between aircraft and spacecraft, and faster, more reliable wireless networks capable of carrying increasing amounts of data without requiring entirely new hardware.

The research exemplifies Queen Mary University of London's strength in combining fundamental engineering with cutting-edge artificial intelligence to solve real-world challenges. By bringing together decades of communications theory with modern machine learning, the team has shown that innovation is often found not in replacing proven ideas, but in teaching them to adapt.

You can read the full paper via the following link.