Researchers have demonstrated a new way to capture a far wider range of sound without using more bits, challenging a nearly 90-year-old convention in digitisation and opening possibilities for healthcare, environmental sensing and AI.
What does a faint heartbeat have in common with a thunderous concert? Both present a challenge for digital technology: capturing the quietest details without losing the loudest sounds.
For nearly 90 years, engineers have managed this trade-off by balancing the loudness of sound a device can capture against the number of bits used to represent them.
"Reeves and Shannon established the ideas that made digital communication possible... What has changed is that modern computation allows us to ask whether the analogue signal must still be presented to the converter in the same way." Dr Ruiming Guo Department of Electrical and Electronic Engineering
Now, researchers at Imperial College London have demonstrated a radically different approach called USF-Audio. Rather than allowing loud signals to be clipped, their system deliberately folds them into a smaller range before digitisation. Although the resulting waveform can look or sound scrambled, its mathematical structure allows the original sound to be reconstructed.
The approach, known as Unlimited Sensing, changes what digital bits represent, rather than simply adding more of them.
In experiments using just six bits per sample, the prototype achieved at least 18 dB greater dynamic range than a conventional six-bit converter. A listening study involving 52 participants also found that reconstructed audio received higher quality ratings than conventional recordings at the same bit depth.
The implications extend beyond music. From digital stethoscopes and environmental sensors to future AI systems that interpret the physical world, the technology opens new possibilities for capturing information that conventional digitisation can lose.
Breaking a 90-year-old trade-off
The breakthrough builds on nearly 90 years of digitisation research, dating back to Imperial alumnus Dr Alec Reeves CBE, who invented pulse-code modulation in 1937, and Claude Shannon, whose work laid the mathematical foundations of modern digital communications.
Conventional analogue-to-digital converters face a fundamental trade-off: with a fixed number of bits, accommodating a wider range of signals comes at the expense of resolving finer details. Signals that exceed the converter's range are clipped, losing information that software can only attempt to repair afterwards.
The Imperial team's solution turns this conventional thinking on its head. Imagine the hands of a clock wrapping around after completing a revolution, rather than stopping. Unlimited Sensing uses a similar principle to represent large signal amplitudes within a smaller range. Crucially, the folds follow a mathematical structure, allowing the original signal to be recovered from the sequence of measurements.
Researchers Dr Ruiming Guo, Yuliang Zhu and Dr Ayush Bhandari combined mathematical recovery theory, scalable audio algorithms and a complete microphone-to-digital hardware implementation. Their custom converter was built using off-the-shelf electronic components, demonstrating how a mathematical idea can be translated into working technology.
Dr Ruiming Guo, lead author and Research Associate in Imperial's Department of Electrical and Electronic Engineering, said: "Reeves and Shannon established the ideas that made digital communication possible. Those ideas are not wrong. What has changed is that modern computation allows us to ask whether the analogue signal must still be presented to the converter in the same way."
USF-Audio hardware from Ayush Bhandari's research team
Can listeners hear the difference?
The researchers tested their technology using real acoustic recordings, including speech, music, heart sounds and animal calls.
The improvement of at least 18 dB in dynamic range corresponds to approximately an eightfold increase in the range of signal amplitudes the converter can accommodate, without increasing its bit depth. But could listeners actually hear the difference?
"Great music lives in its contrasts — from the faintest whisper to the loudest crescendo. Yet conventional digital recording forces compromises between capturing these extremes." Dr Ayush Bhandari Department of Electrical and Electronic Engineering
In controlled listening experiments, participants rated audio reconstructed using Unlimited Sensing more highly than conventional six-bit recordings and recordings affected by clipping.
The findings demonstrate that the benefits are not confined to mathematical theory or laboratory measurements. Changing how sound is represented digitally can produce improvements that people can hear.
Dr Ayush Bhandari, Associate Professor in Imperial's Department of Electrical and Electronic Engineering and co-author of the study, said: "Great music lives in its contrasts — from the faintest whisper to the loudest crescendo. Yet conventional digital recording forces compromises between capturing these extremes.
"We asked a radically different question: what if we changed what the bits represent, rather than simply adding more of them? That is the idea behind Unlimited Sensing — capturing a far wider range of sound without increasing digital resolution."
