A change from average reference to Cz gives a median alpha Riemannian distance of 11.29 in 88 ds004504 v1.0.9 records with a ridge fraction of 0.001 in all 19 channels

This study tested whether an EEG voltage reference change alone alters a Riemannian covariance distance. It used 88 processed records and 6,377 fixed windows from OpenNeuro ds004504 version 1.0.9. The primary test used the alpha band, 19 sensors, and a ridge fraction of 0.001. The ridge was 0.001 times the mean zero-sum contrast variance. Changing average reference to Cz gave a median full-channel distance of 11.29 for the same source data. The median distance between different average-reference records was 11.11. A method with 18 orthonormal zero-sum contrasts removed the reference effect. The largest primary reference-only contrast distance was 1.59e-12. A proof gives exact invariance under any common voltage subtraction with the same ordered sensors and estimator. All 88 records matched in the primary audit with both methods. The queries used identical source data. The package retains all 11 planned settings, two fresh source checks, independent numerical checks, source hashes, code, and reproduction instructions. These results describe a matrix method. They do not test clinical prediction or neural source accuracy.

electroencephalography · EEG reference · covariance distance · Riemannian geometry · zero-sum contrasts · signal processing