A fixed line and five sine terms increased gamma power restart error in 35 SEEG records with a complete 100 ms past buffer

We tested a filter state model that uses past voltage on public OpenNeuro ds006519 v1.0.0. The test used 35 SEEG records from 21 patients. All voltage samples were available. The model fit a line and five fixed sine terms to a 100 ms past buffer. The primary sine penalty was 0.01 times the buffer sample count. After replay, mean patient gamma power error was 20.92% of prior power. A state set from the first past sample gave 6.87%. The paired increase was 14.05 percentage points. The descriptive 95% person bootstrap interval was 11.12 to 17.65 points. All 21 patients had higher error with the model. Beta power error also increased. The other fixed penalties and the prestim check did not reverse the group result. The package contains exact voltage buffers, reference states, all scores, figures, code, and source hashes. This is a software restart test on a reused clinical sample. It does not test missing signal or live BCI control.

SEEG · filter state · software restart · gamma power · beta power · signal processing · brain-computer interface