- Balija Santoshkumar, MSU
- Robust Zero-crossing Detection in Noisy Signals Using Topological Signal Processing
- 03/22/2021
- 2:00 PM - 3:00 PM
- Online (virtual meeting)
- Shelley Kandola (kandola2@msu.edu)
Detecting zero-crossings in noisy signals is a classical problem that has been researched for over
a century. Some of the prominent applications of zero-crossing detection are phase and frequency
determination in oscillatory systems, smooth switching operations in power systems, image processing and recognition, speech processing and reconstruction of audio signals, biometrics using human
iris, and bar code scanners. This work leverages Topological Signal Processing (TSP), more specifically persistent homology, to develop a simple but powerful zero crossing detection algorithm. The
algorithm utilizes zero-dimensional persistent homology to estimate the zero-crossings in a noisy
signal, and uses the resulting persistence diagram to find out the significant splits in data to bound
the zero crossings. We compare the accuracy and speed of our approach with available methods
in the literature using different types of noisy signals, as well as showing sensitivity studies that
consider the sampling frequency (SF), and Signal to Noise Ratio (SNR). Our results show that
TSP can be directly applied to unfiltered signals with moderate to high noise levels for finding
the zero-crossings. In contrast to many existing tools, our approach does not require any complex
analog circuitry, iterative solvers, or filtering.