CycloTorch: A library for GPU accelerated cyclostationary analysis

Cyclostationary analysis of real signals has been a key component in condition monitoring of rotating machinery. However, the estimation of the cyclic spectral correlation or coherence can be very computationally expensive, which is why several computationally efficient algorithms have been proposed. Nevertheless, due to the lack of public code, not all of these algorithms have seen widespread adoption. Therefore, this Python library implements cyclic spectral correlation estimators and the frequency shifted filters to enable cyclostationary analysis of real signals. Furthermore, it employs PyTorch to enable GPU acceleration, which results in significant performance gains in comparison to purely CPU based implementations.