Quadratic fits reduced mean SEEG shape error against a bounded power law plus constant floor between record ends at 20–100 Hz with 55–65 Hz removed

We tested whether a power law with a constant floor predicts SEEG spectral shape better than a quadratic model. SEEG means stereoelectroencephalography. We used the first and last 60 s from public OpenNeuro ds005624 version 1.0.0. The test included 24 participants, 34 records, and 4,163 named bipolar pair records. The primary band was 20 to 100 Hz, with 55 to 65 Hz removed. Each excerpt had its mean log10 power removed. Both models had two shape parameters. The floor exponent had bounds 0 to 8. The first minute supplied the fit. The last minute supplied the test. Mean participant error was 0.006702 for the floor and 0.006032 for the quadratic. The quadratic reduced error by 10.0 percent. The 95 percent participant bootstrap interval was 8.2 to 11.8 percent. All 24 participant errors were lower with the quadratic. The floor also improved on a straight fit. All seven fixed checks are reported. This secondary analysis reused known spectral inputs. The intervals assume independent participant units. The result limits the constant-floor description in this band. It does not identify a physical source. Code, exact inputs, source pins, all scores, and checks are preserved.

stereoelectroencephalography · SEEG · neural spectra · spectral shape · power spectrum · power law · constant floor · quadratic model · Welch method · model comparison · OpenNeuro