Hundred times faster signal processing with new calculation technique

Publication in Nature Computational Science

A graphic summary of how the new technique works, translating the signal into a more informative representation. "In short, we look at the signal in a different light!"

By re-implementing an existing calculation technique for signal processing, Utrecht computer scientists have succeeded in speeding up the technique by a factor of one hundred, without loss of quality. This can enable considerable improvements in countless applications that work with signals or data flows from sensors, from MRI scanners to systems that predict earthquakes. The researchers, Lukas Arts and Egon van den Broek, are publishing their results today in Nature Computational Science.

“There are two commonly used techniques for signal processing,” Van den Broek explains. “One works very precisely, but is so slow that it is hardly useful. This is why the other technique is often used in practical applications: it is lightning fast, but loses a lot of quality during processing.” The researchers managed to speed up the slow technique by a factor of one hundred, so that it combines the best properties of both techniques: high speed as well as high precision.

Brain signals

The new calculation method can be used in all sorts of applications, says Van den Broek. “Take, for example, a brain-computer interface that allows a fully paralysed person to control a wheelchair. The wheelchair needs to respond immediately to steering signals from the brain, so you have to use the fast algorithm. Consequently, the signal quality is so low that such wheelchairs can only distinguish a few signals: left, right, accelerate and decelerate. With this new calculation technique, you can make the wheelchair recognise many more different signals without compromising on speed.”

Classical computer science

The researchers have made the slower signal processing technique a factor of one hundred faster by combining the underlying mathematics with the latest insights into hardware and software. “This is actually a perfect example of classical computer science,” says Van den Broek. “You take the mathematical basis and make it run on a device in the fastest possible way.” As a result, the new implementation is also a factor of one hundred more energy efficient.

Interactive cuddly toy

“The technology is ready to be used in all kinds of devices in our daily lives,” says Van den Broek. “We’ve made our new algorithm available open source, so anyone can get started right away.” In an interdisciplinary European project, Van den Broek and Arts are working on an interactive cuddly toy for children with autism, to support diagnosis and therapy. “A perfect application of the technology: the cuddly toy uses sensors to measure very precisely what is happening, and can then respond in real time. This makes it possible to better understand and facilitate social interaction between people, animals and robots.”


The fast continuous wavelet transformation (fCWT) for real-time, high-quality, noise-resistant time-frequency analysis
Lukas P. A. Arts and Egon L. van den Broek
Nature Computational Science, 27 January 2022, DOI 10.1038/s43588-021-00183-z