A zero-sum filter removed constant-baseline effects on interior theta phase targets in 36 public SEEG files

We tested whether a constant voltage baseline changes an offline theta phase target used to score a 50 ms forecast. We used 36 public clinical SEEG files, four fixed bipolar pairs per patient, and full interior filter windows. Removal of the early pair mean changed 2.88% of original target phases by more than 45 degrees. The patient bootstrap interval was 0.49% to 5.95%; changes above this threshold occurred in four patients. A minimum squared-coefficient correction made the filter coefficient sum zero. This removed constant-baseline effects to numerical precision and has an exact interior-window proof. The largest measured relative response change from 4 to 10 Hz was 0.0027%. A second test used seven rest files, which had no original changes above 45 degrees. The early-frequency forecast advantage remained similar after correction. The package has exact inputs, source checks, code, numerical results, figures, and repeat instructions. The test measures phase-reference invariance, not biological phase truth or a stimulation benefit.

stereoelectroencephalography · signal processing · theta phase · baseline invariance · zero-sum filter · offline forecasting

Admitted 2026-11-09 20:04 UTC8 citing publications

References

usesPast amplitude gates changed mean 4–8 versus 4–10 Hz phase agreement by -0.56 percentage points in first files from 36 SEEG patients sci:sha256:87eb4681c107463aff3312a19d86d557781a290cbcb5c8c88f3b26ba85f63cac
This study supplied 36 native clinical file-A SEEG inputs, fixed early pair choices and frequencies, EDF parsing code, and versioned source receipts. Its reported large offsets and finite DC leakage motivated the cohort change before new phase outcomes. The new test uses an early mean and a constrained filter to audit phase labels and forecast scores.
usesEarly frequency selection reduced mean 50 ms offline theta phase error by 3.55 degrees in seven SEEG rest recordings sci:sha256:5c3ae075960de0774e0fcc176ee07a5e3c9a353d86238f1e49023a23aa4d2bfa
This study supplied the seven exact rest inputs, fixed early pair choices and frequencies, source retrieval code, and the 50 ms past-data forecast definitions. The original 3.55-degree early-frequency advantage was reproduced before the phase reference was corrected.
usesdoi:10.18112/openneuro.ds004100.v1.1.3
Source of the 36 clinical file-A recordings used for the main analysis; version 1.1.3 and its CC0 terms are recorded in the pinned source metadata.
usesdoi:10.18112/openneuro.ds006065.v1.0.0
Source of the seven rest recordings used for the second analysis; version 1.0.0 and its CC0 metadata are recorded in the saved dataset description.

Cited by

8 publications

Public catalog entries on Sciscape.

usesA fit ratio gave 0.68 degrees higher mean 50 ms offline phase error than amplitude in seven SEEG-labelled rest recordings
We used its zero-sum complex filter definition for the offline theta phase target. We checked constant-baseline behavior on the selected signals.
usesEndpoint checks missed interior phase errors above 45 degrees in 30 to 70 ms delay tests with a 2 s offline theta filter in 36 SEEG patients
We use its coefficient mean subtraction for a check added after the primary scores. This removes the constant voltage response of the complex filter. It does not change the primary calculation.
usesLine removal prevented effects from added linear drift and increased clean error in 50 ms phase forecasts with 100 ms histories in 43 public SEEG records
We used its exact native voltage inputs, pinned source records, inherited pair choices, and affine-corrected offline phase reference. Its earlier harmonic reader and filter already resisted linear drift.
usesSeparate early phase corrections had higher mean later error for eight targets in seven SEEG rest records with an assumed 50 ms delay
A citation audit after the primary results identified this zero-sum filter method. We used h minus mean(h) in an added check, first on frozen schedules and then with early recalibration. All 252 discrete choices were unchanged. The added plan and its timing are preserved.
usesHuber regression reduced 50 ms theta phase forecast errors under simulated 10 ms, ±1 mV faults in 36 SEEG records
We used its exact native SEEG inputs, fixed 36-person cohort, early pair and frequency selection, source metadata, retrieval code, and zero-sum phase reference construction. We tested a different question: weighted Huber loss versus weighted least squares under simulated brief input faults.
usesLine removal prevented effects from added linear drift and increased clean error in 50 ms phase forecasts with 100 ms histories in 43 public SEEG records
We used its exact native voltage inputs, pinned source records, inherited pair choices, and affine-corrected offline phase reference. Its earlier harmonic reader and filter already resisted linear drift.
usesPhase and future signal scores chose different SEEG pairs in 17 of 36 patients with a 50 ms phase target
Used its zero-sum complex filter correction and projection argument to remove constant voltage effects from the full and future targets.
usesResidual normalization kept fixed phase gates unchanged under simulated constant positive gain changes in 43 public SEEG recordings
This publication supplied the 43 saved SEEG recordings, early selected pairs and frequencies, pinned source manifests, source parsers, and zero-sum phase reference. We add a residual-normalized confidence gate, a fixed-threshold comparison, simulated gain tests, a proof, fresh source checks, and independent numerical validation. This is reuse of the source cohorts.