Correlation PCA

Correlation PCA — the same machinery applied after giving every feature equal sample variance: R = D⁻¹CD⁻¹, with D the diagonal of standard deviations. Spanish: PCA de correlación.

Ask this first: rescale one sensor channel by 100 without changing its information. Which PCA changes? Covariance PCA does — its first axis swings towards the inflated band. In correlation PCA the change cancels: the standardized coordinates are identical.

The price: correlation PCA can give a weak, noisy band as much initial variance as a strong band. Equal variance is a modelling decision, not a universal improvement.