A frozen learned forecast reduced mean simulated zero-phase pulse error at 50 ms delay in seven SEEG rest records
We compared two fixed phase forecasts on public SEEG rest data from seven people with epilepsy. Each forecast selected its own simulated pulse times. The target was zero degrees, the assumed delay was 50 ms, and the minimum pulse interval was 0.5 s. Mean arrival error against a centered two-second phase reference was 69.76 degrees for a calibrated sinusoid and 57.60 degrees for a learned linear forecast. The mean decrease was 12.16 degrees. Mean rates were 1.695 and 1.703 pulses per second per pair. The learned forecast had lower person mean error in 7 of seven people. The package includes all selected voltage samples, unchanged early model weights, every pulse record, code, three figures, and source hash checks. The reference uses future voltage. This work reuses a previously scored rest block. It does not test a live device or a neural response to stimulation.
phase forecasting · intracranial EEG · SEEG · simulated stimulation · event replay · neural signal processing · offline evaluation
References
4- usesLearned filters reduced mean 50 ms offline theta phase error by 12.06 degrees against a calibrated sinusoid in later SEEG rest from seven people sci:sha256:3adc1388ee2f8a7376426deef4c35ed0f92b2d041c8f0962bbb58c3d1b7fb38a
We use unchanged 50 ms learned weights, full selected rest traces, the phase reference, and source provenance. We add event-selected pulse replay and common-schedule checks.
- usesDelay compensation reduced mean simulated pulse phase error in seven SEEG rest recordings with an assumed 50 ms delay sci:sha256:a530304ddb504b81ca4aec3a6dbb163fc0e113887ae4be98000a9863c9f01121
We use the wrapped positive phase crossing and minimum pulse interval rule from the sinusoid replay. We add learned crossings and fit a sinusoid offset for the 50 ms delay.
- usesdoi:10.18112/openneuro.ds006065.v1.0.0
Source data: selected rest SEEG records from OpenNeuro ds006065 version 1.0.0, which matches the version and records analyzed.
- doi:10.1038/s41467-025-59417-7
Cited by
3 publicationsPublic catalog entries on Sciscape.
- usesA 7 Hz phase input increased mean forecast error for later log response size after five calibration pulses at 86 SEEG readouts in eight patients
Used its simulated phase forecast study to define a separate test of response size after recorded pulses. Its voltage and models were not used.
- usesSimulated pulse schedules matched with and without 25-sample gaps at a 301-sample interval in seven SEEG rest records with fixed readers and delays
This publication supplied the 28 selected SEEG traces, frozen 50 ms readers, phase reference, crossing rule, source receipts, and no-gap events. We reproduced all 16,403 main event indices and reference phases. We added continuous replay in which each output causes a sensing gap, a finite-support interval proof, pulse-time measures, and rate costs.
- usesA 10 s candidate history reduced error from 50% retention at zero phase in offline tests of 36 SEEG patients
Used its prior native-sample forward phase-crossing design for simulated commands. The present test uses another dataset and frozen model bank.