Signal preprocessing for ML with wavelets.

FerroWave helps convert noisy, non-stationary signals into denoised series, multi-scale features, and model-ready tensors.

Why this needs production signal processing

!

Raw time series often mix trend, noise, jumps, and regime changes in one channel.

!

Preprocessing steps written for notebooks can be difficult to reproduce in production.

!

ML teams need transparent features that can be audited and regenerated.

How FerroWave handles it

01

Separate scales

Wavelet and adaptive decompositions separate slow components, high-frequency detail, and localized events.

02

Denoise or summarize

Pipelines can denoise the signal, extract features, or produce scaleogram tensors for model input.

03

Keep parity

Reference-pinned behavior helps preserve the research result when moving to Rust deployment.

What makes this FerroWave-shaped

FerroWave includes denoising, DWT/MODWT/SWT/WPT/CWT, and adaptive decomposition workflows.

Feature extraction for AI/ML pipelines is explicit in the site headline.

Verification and reference pinning support the reliability story.

Common concerns

Is signal preprocessing the same as feature extraction?

They overlap. Preprocessing can denoise or decompose a signal; feature extraction converts that structure into model inputs.

Can FerroWave produce image-like inputs?

Yes. Scaleogram tensors for CNN-style workflows are part of the public ML positioning.