Riemannian covariance estimation (riecovest)

This is a package for estimation of signal covariance matrices from noisy signal and noise-only data. The package is using pymanopt to perform optimization over Riemannian manifolds, using JAX to as a backend for automatic differentiation.

More info and complete API documentation

References

The code was developed as part of the paper J. Brunnström, M. Moonen, and F. Elvander, “Robust signal and noise covariance matrix estimation using Riemannian optimization,” presented at the European Signal Processing Conference (EUSIPCO), Sep. 2024. The examples folder contains code to replicate the results from the paper. If you use this software in your research, please cite the paper.

@inproceedings{brunnstromRobust2024,
  title = {Robust Signal and Noise Covariance Matrix Estimation Using {{Riemannian}} Optimization},
  booktitle = {European {{Signal Processing Conference}} ({{EUSIPCO}})},
  author = {Brunnstr{\"o}m, Jesper and Moonen, Marc and Elvander, Filip},
  year = {2024},
  month = sep,
  keywords = {covariance estimation,GEVD,low-rank,manifolds,MWF,quotient manifold,riemannian optimization,robust,SPD}
}

License

The software is distributed under the MIT license. See the LICENSE file for more information.

Installation

The package can be installed via pip from the PyPi repository:

pip install riecovest

Alternatively, clone this repository and install the package using:

pip install path/to/riecovest

To run the examples "reproduce_covest_t_distribution.py" and "reproduce_spatial_filtering.py" which reproduces the results from the aforementioned paper, a longer list of dependencies is needed. These can be installed with pip using:

pip install riecovest[examples]

The example "reproduce_spatial_filtering.py" also makes use of the MeshRIR dataset. It must be downloaded from the original source along with the dataset loader irutilities.py. The dataset loader must be placed in the same folder as the example script.

Acknowledgements

The software has been developed during a PhD project as part of the SOUNDS ETN at KU Leuven. The SOUNDS project has recieved funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No. 956369.