Common spectrum controls gave a 0.22 percent error decrease from additional pairs for linear forecasts of future theta terms in seven SEEG rest records

Extra electrodes can help a brain interface predict a signal before it applies stimulation. We tested a linear reader with one target pair and up to three extra SEEG pairs in seven public rest records from OpenNeuro ds006065. The target was the future part of a theta filter with a center 50 ms after the cutoff. It used stored samples from the cutoff to about 1.05 s later. We fitted the reader again with 39 controls per person. These controls kept each block auto and cross spectrum. Extra pairs reduced mean later error by 0.30 percent in recorded data and 0.22 percent in these controls. The paired difference was 0.08 percentage points, with a descriptive 95 percent person bootstrap interval from -0.20 to 0.35. Controls with separate phase rotations gave -0.17 percent mean gain. The small recorded gain did not give clear evidence for timing structure beyond the preserved spectra. This does not prove equal performance or a benefit in a live interface. The package has exact inputs, source checks, all fitted models, results, figures, and code.

SEEG · neural signal forecasting · theta · Fourier phase controls · ridge regression · spectrum preservation · brain-computer interfaces