With clean endpoints, linear repair gave lower 70 to 110 Hz error than four-point cubic repair for simulated 3.17 ms SEEG gaps in 18 patients

We compared linear repair with four-point cubic repair in public SEEG baseline data. We used 735 channels from 18 patients in OpenNeuro ds006519, version 1.0.0. The input sample came from cited prior work. The primary mask removed 13 samples, or 3.17 ms, at 50 Hz. With no added endpoint error, mean patient median 70 to 110 Hz NMSE was 0.04520 for linear repair. It was 0.35889 for cubic repair. Linear repair gave lower error in all 18 patients. Cubic minus linear was 0.31369, with a descriptive paired patient interval of 0.27151 to 0.35512. An exact calculation gave 13.28 times more added filtered error for cubic repair under independent endpoint errors. One shared error per gap gave equal added channel error for both methods. A clean 90 Hz sine wave favored cubic repair. Thus, this smooth signal control did not predict the recorded SEEG result. Fresh source downloads matched all 18 saved excerpts. A pinned Git audit checked 320 metadata files and 35 signal pins. A separate analysis matched all 27 numerical files exactly. The package includes exact inputs, code, source records, proofs, tables, three figures, and repeat instructions. These simulated baseline gaps do not establish actual artifact removal, evoked signal recovery, or BCI task accuracy.

SEEG · gap repair · linear interpolation · cubic interpolation · signal processing · endpoint noise · brain-computer interfaces