A fixed Rice code saved a mean 39.1 percent of bytes for rounded high gamma SEEG power in eight second sessions

We tested exact transport of rounded high gamma SEEG power features. The source was OpenNeuro ds005624, version 1.0.0. We used all eight people with two sessions. We reused the exact feature files from sci b8579edd-891d-40f5-a527-90ad5c9340ff. The new test measured packet bytes. Coding parameters used only 120 seconds from the first session. We rounded log10 power to steps of 0.001. This changes positive power by at most 0.116 percent. The primary code used a fixed center and Rice coding, with a literal escape. On 300 sampled seconds per person in the second session, ten-second packets saved a mean 39.08 percent of bytes against literal 16-bit integers. The descriptive 95 percent person bootstrap interval was 38.03 to 40.00 percent. All eight people had savings. The count includes models, packet fields, padding, and error checks. One-second packets saved 27.69 percent. A separate decoder recovered 1,488,000 integers with no mismatch. Fresh public signal clips matched the saved features for all eight people. A full rerun matched 53 computed files. The package preserves code, all packets, models, source files, tables, figures, and pinned retrieval instructions. Exact recovery applies to rounded features. This offline test does not measure task decoding, radio energy, or stimulation.

Rice coding · SEEG · high gamma power · feature transport · brain-computer interface · lossless coding · session transfer