A past voltage mean reduced average 70–110 Hz power error versus zero fill after simulated 50 ms gaps in SEEG post windows from 21 patients

We tested three gap fills on 35 public SEEG records from 21 patients in OpenNeuro ds006519 version 1.0.0. The fixed main test used known simulated 50 ms gaps in source post windows. The prior processing state was correct. The feature used a causal 70 to 110 Hz filter and a trailing 100 ms power mean. We scored the first 100 ms of observed signal after each gap. Patient weights were equal. The mean voltage from the preceding second reduced mean normalized RMS power error from 3.0135 to 1.3684 versus zero fill. The paired reduction was 1.6451, with a descriptive 95 percent patient bootstrap interval from 0.8997 to 2.4586. Error was lower in 17 of 21 patients. Holding the last voltage gave error 0.5572 as a secondary result. A constant-offset control and proof showed that past-voltage fills kept filtered error unchanged, while zero fill depended on voltage origin. All fills left error. Fresh source checks rebuilt all selected sample bytes. The full analysis reproduced exactly. The package preserves voltage, source pins, code, all scores, proof, figures, and reproduction instructions. This software benchmark does not test decoder accuracy, neural recovery, or a safe read delay after stimulation.

stereoelectroencephalography · SEEG · signal processing · band power · gap filling · causal filtering · brain-computer interface · simulated data gaps