Expand description
Multivariate Variational Mode Decomposition and the methods built on its solver.
- MVMD: Rehman and Aftab (2019), IEEE Transactions on Signal Processing. Ported from the reference at github.com/Dmocrito/mvmd (MIT; notice in THIRD_PARTY_NOTICES.md).
- VMD: Dragomiretskiy and Zosso (2014), the single-channel case, run on each channel.
- NA-MVMD: Zisou, Apostolidis and Hadjileontiadis (2021), MVMD with white-noise channels.
Structs§
- Mvmd
- Mvmd
Params - Defaults follow the reference implementation.
- NaMvmd
- Noise-assisted MVMD (Zisou, Apostolidis and Hadjileontiadis, ICASSP 2021). Each channel is scaled to unit SD (ddof = 1), white-noise channels are appended, MVMD runs on the whole set, the noise channels are dropped and the modes scaled back. The paper’s generalised cross-spectrum centroid for one snapshot is Σ_c Σ_f f|û_c|² / Σ_c Σ_f |û_c|², the update MVMD already uses.
- Noise
Params - White-noise channels added before MVMD. Defaults follow the reference
na_mvmd.m. - Vmd
- Variational Mode Decomposition of each channel on its own: MVMD with one channel, which is VMD (Rehman and Aftab 2019, section III; same grid, mirror and update as vmdpy). Mode k is the k-th lowest centre frequency of its channel, so modes are not aligned across channels.
Enums§
- Frequency
Init - Starting centre frequencies, in normalised frequency (cycles per sample).