Early calibration reduced mean 70 to 110 Hz error by 2.31 points for causal fills of simulated 13-sample gaps in baseline SEEG from 20 patients

We tested a method to fill short gaps without a wait for future samples. The data were baseline SEEG from 20 patients and 1,671 native bipolar channels. We reused public OpenNeuro ds005169 version 1.0.0 data from an earlier sci study. The first 16 simulated gaps selected one slope setting per patient. The next 32 gaps tested that setting. Each gap removed 13 samples at 4096 Hz, or 3.17 ms. The reference was the recorded voltage through the same causal 70 to 110 Hz filter. Mean patient median RMS error fell from 52.06% of reference RMS with hold to 49.74% with the selected slope. The paired decrease was 2.31 percentage points. Its descriptive 95% patient bootstrap interval was 0.93 to 3.99 points. Fifteen patients improved and five did not. Delayed linear interpolation gave 25.70% error but needed up to 3.17 ms of future samples. The slope method increased filtered power more than hold did. All original source hashes and stored voltage values matched. Independent calculations and sample access checks passed. The package has data, code, tables, three figures, and source retrieval instructions. The test assumes exact gap times and clean preceding samples. It does not establish performance during stimulation, decoding accuracy, or clinical benefit.

stereo electroencephalography · gap filling · signal processing · causal filtering · high frequency · waveform error · sample and hold · linear interpolation