A five-pulse template gave mixed six-bit error changes at 86 selected SEEG readouts with known pulse timing

We tested whether a fixed mean response from five early pulses reduced later SEEG voltage code error. The test used processed public data from OpenNeuro ds004696, version 1.0.1. There were 86 selected bipolar readouts in eight people with epilepsy. Both codes used known pulse timing, separate before and after ranges, and the same calibration search. Five final pulses at each site supplied the test. The primary six-bit uniform comparison gave a mean person log ratio of -0.224 dB. Its descriptive 95% interval was -0.738 to +0.269 dB. Four of eight person log effects favored the template. Equal-person arithmetic mean error increased by 6.48%. Each float32 template added 3,932 model bytes. Total bytes increased by 53.2% for five six-bit test pulses. Secondary four-bit comparisons reduced error within both code families. They did not give lower total bytes or error than six-bit direct coding at the tested length. An independent reader checked 344 streams and all six-bit model fits. A second run gave 698 identical result and stream files. Three fresh complete source files reproduced the archived voltage exactly. The SCI package retains numerical inputs, source records, results, and three figures; GPL analysis and retrieval code is omitted, so the exact analysis implementation is unavailable here. The study does not test live preprocessing, device energy, or clinical benefit.

SEEG · neural recording · template subtraction · uniform quantization · signal compression · brain-computer interfaces · reproducibility