Added word transforms reduced lossless 100 ms packet size in seven SEEG rest records and gave no gain in 35 stimulation records
We tested reversible word transforms in 84 two-second windows from 42 public SEEG records. The input had 23,110,160 exact source words. Each packet had a 32-byte header and no dependence on another packet. At 100 ms, the expanded codec used 26.963 bits per word in seven rest records. The four-mode baseline used 27.240 bits per word. The paired difference was -0.278 bits per word. The 95% person bootstrap interval was [-0.390, -0.174]. The added size reduction was 1.01%. The 35 stimulation records came from 21 people. Their mean difference was 0.000 bits per word. The 95th percentile host encode time in rest rose from 6.54 to 25.56 ms. All tested packets recovered every source bit. The package contains source windows, code, packet examples, figures, and source checks. The result applies to these source files and short windows. It does not establish implant energy use or neural decoding accuracy.
lossless compression · stereoelectroencephalography · SEEG · intracranial EEG · packet encoding · bit-plane transforms · word prediction · brain-computer interfaces
References
7- supersedesAdded word transforms reduced lossless 100 ms packet size in seven SEEG rest records and gave no gain in 35 stimulation records sci:sha256:87ea235fd19622db755c021c261bc63a0daa29afb3df5bac61761806578ad1a4
This version corrects the file audit after a download. The audit skips the supplemental metadata folder that sci adds. The scientific results are the same.
- usesIndependent 100 ms packets reduce data size and preserve source words in 42 public SEEG records sci:sha256:d68c6b03737397835b02cb1e307a69ed9a04735017100d2d9f7222331a753b80
The prior study supplied the 84 exact source windows, source metadata, selection rules, and four-mode packet baseline. We tested six added word transforms on these inputs. All 420 baseline window byte counts were reproduced.
- usesdoi:10.18112/openneuro.ds006519.v1.0.0
The selected stimulation records were from OpenNeuro ds006519 version 1.0.0.
- usesdoi:10.18112/openneuro.ds006065.v1.0.0
The selected rest records were from OpenNeuro ds006065 version 1.0.0.
- doi:10.1016/j.neuroimage.2026.121822
- doi:10.1038/s41467-025-59417-7
- supportsdoi:10.1109/TC.2008.131
The publication cites FPC as prior use of prediction and exact floating-point bit compression.
Cited by
2 publicationsPublic catalog entries on Sciscape.
- usesMean bits per source word were higher inside detected voltage runs for fixed lossless 100 ms packets in 21 ds006519 SEEG records
Uses the unchanged four-mode baseline code, packet format, and source channel map. Extends the size test into detected periodic voltage runs.
- usesChannel order reduced the mean size of independent 100 ms zlib packets for stored float32 SEEG in 35 stimulation records
This work supplied the unchanged ten-mode lossless codec. We reproduced its 100 ms packet sizes and used it as a secondary baseline for size and time.