A noise gate avoided added repair error on four SEEG pulse maps with no injected fault and known simulated Gaussian output noise

We tested whether a fixed noise threshold can prevent unnecessary repair of brain recording outputs. We used measured pulse responses from four people in public OpenNeuro SEEG dataset ds004977 version 1.2.0. The test included 660 contacts, 50 shafts, and 1,002 sparse voltage difference outputs. Faults and independent Gaussian output noise were added in software after the differences were formed. The gate used the known noise standard deviation. It matched ordinary reconstruction in all 20,000 trials with no injected fault. With one fixed output fault per shaft, it reduced mean relative squared error by 81.95% at noise SD equal to 0.1 times early voltage RMS and by 6.14% at 0.5 times RMS. At the two highest primary noise levels, it rejected repair that would add error. A disclosed follow-up showed small error increases near the repair transition. An underestimated noise SD returned the cost of unnecessary repair. A fixed-output risk identity explains the rule but does not prove lower risk after output selection and gating. The package preserves the exact compact data, four figures, source versions and hashes, all 40,000 new trials, proof, code, and reproduction instructions. A full fresh replay matched all saved results. These tests do not establish device, neural decoding, stimulation control, or clinical performance.

SEEG · signal repair · Gaussian noise · output faults · voltage maps · brain-computer interfaces