Scientists at the Institute of Measurement Science have developed a new method for cleaning noisy signals

How can noise be reduced in measured signals without destroying information about the underlying dynamical system?

This important question is addressed in a new publication by Anna Krakovská and Jozef Jakubík from the Department of Theoretical Methods at the Institute of Measurement Science of the Slovak Academy of Sciences. The article, titled “Dynamics-aware noise reduction in time series,” was published in the prestigious journal Nonlinear Dynamics by Springer Nature.

As the authors show, conventional linear filtering, as well as recently popular denoising autoencoders, can strongly distort information about the underlying dynamical system, making it difficult to recover fundamental properties such as the number of degrees of freedom of the generating system.

The solution is to leverage the inherent dynamics of the process to distinguish the true signal from noise. A key contribution of the paper is the authors’ own simple denoising strategy within this class of methods, based on the fact that non-random components of the data can be predicted, whereas noise cannot. The proposed prediction-based method separates deterministic behaviour from noise so effectively that the noise reduction reaches up to an order of magnitude, even in signals corrupted by noise levels as high as 100%.

Congratulations to both authors on this contribution to nonlinear time series analysis.

Read the article here: https://link.springer.com/article/10.1007/s11071-026-12517-5