A redundant SEEG output reduced later power error from simulated constant gains in eight records with independent fractional noise

Four SEEG voltage differences form a closed ring. This constraint estimated relative output gains without a test signal. We tested 563 rings from eight public records in OpenNeuro ds004473 version 1.0.2. Gains were added after subtraction in software. With 1,000 calibration samples and 5 percent independent Gaussian noise, scaled total least squares gave 0.052 dB gain error. It reduced later 70 to 110 Hz power error from 3.18 to 0.32 dB. These values are means of eight participant medians. All three gains were within 5 percent in 95.96 percent of trials with equal participant weight. Common gain and contact gains before subtraction remain invisible. Timing shifts, correlated noise, and later drift need separate checks. This study tested software errors, not a physical circuit, a decoder, or stimulation.

SEEG · stereoelectroencephalography · total least squares · gain calibration · signal processing · intracranial EEG · noise · power estimation