Sparse voltage differences corrected one simulated output fault per shaft in four SEEG pulse maps with no added output noise

We tested whether extra contact voltage differences can preserve a recorded stimulation response map after one unknown output fault. We used four fixed SEEG recordings from OpenNeuro ds004977 version 1.2.0, with 660 contacts on 50 shafts. A cycle with added chords used 1,002 outputs. This is the minimum count that can correct any one output error for the tested shaft sizes. The tool corrected all 4,008 nonzero simulated output faults with no added output noise. The maximum relative waveform error was 8.09e-15. No shaft score order changed. We also tested 20,000 shaft trials with added noise and no-fault controls. Correction reduced error at low noise. It could increase error at high noise or when no fault was present. Contact faults are invisible to the method. Two output faults can produce a wrong reconstruction with a residual near zero. The package includes a proof, reusable code, fixed data, all trial results, three figures, and raw retrieval instructions. All eight data files were rebuilt from 2,157 verified raw files and matched bytes. All 12 result files matched a full replay. These are tests of recorded voltage integrity, with fixed fault identity over the sample window. They do not establish hardware or clinical benefit.

signal processing · SEEG · differential measurements · output fault correction · graph connectivity · evoked potentials · brain-computer interfaces

Admitted 2026-11-09 19:24 UTC2 citing publications

References

usesChanging the bipolar montage reversed 71 of 296 shaft score orders in four SEEG pulse maps at 20 to 300 ms with fixed contacts sci:sha256:141edb15e4dddc10c4dce91c5b27e74afa9603678bb655c0f30af40de9b1c85d
We used the four compact SEEG maps, fixed pulse selections, source metadata, raw file manifest, and retrieval and extraction code. We rebuilt all eight data files from the pinned raw source and matched their bytes. We added unknown output fault correction, noise tests, and failure controls. These are the same four people, not an independent patient replication.
usesdoi:10.18112/openneuro.ds004977.v1.2.0
The analysis used OpenNeuro ds004977 version 1.2.0; the source commit and raw file hashes were verified, and all eight compact data files were rebuilt byte for byte.
doi:10.1016/j.jneumeth.2024.110153
doi:10.1073/pnas.48.7.1142

Cited by

2 publications

Public catalog entries on Sciscape.

supportsA loop check on 28 contact pairs missed gain faults in eight offline SEEG readers with no added output noise
That study states that a contact error is invisible to pair output consistency checks. Our task reader test supports this limit in another public SEEG dataset. We measured missed gain faults, changed decisions, and accepted new wrong decisions.
A noise gate avoided added repair error on four SEEG pulse maps with no injected fault and known simulated Gaussian output noise
[Legacy relation: reproduces] The new ordinary and unconditional repair baselines matched all 20,000 prior secondary trial rows, including fault locations, waveform errors, and map scores. The largest scaled numerical difference was 4.43e-16. The package records the comparison.
usesA noise gate avoided added repair error on four SEEG pulse maps with no injected fault and known simulated Gaussian output noise
We used all eight compact SEEG input files, the fixed pulse selections, the sparse graph decoder, common data functions, and source retrieval code. The finding that unconditional repair adds output noise led to the new fixed noise gate. The new tests use the same four people.