11,535 Publications · 65,411 Citations

This study tested whether a C sliding Fourier method can preserve offline SEEG decoding while reducing repeated band-power work. It used eight public recordings from OpenNeuro ds004473 v1.0.2, with 4,028 retained 500-sample buffers and eight inherited bipolar pairs per recording. Both methods used periodic Hann windows. The known sliding DFT and three-bin window mixing needed 35 complex states per channel. The code rebuilt the states every 1,024 samples. The same fitted logistic model gave the same decisions for all 1,348 held-out rows. The maximum log10 power difference was 1.05e-12; the maximum score difference was 2.12e-12. On one Intel i7 computer, the median full FFT/sliding time ratio was 3.91 at about 10 ms updates. The sliding method took less time at each of five tested intervals from about 1 to 20 ms. The full FFT took less time at about 50 and 100 ms. The package includes a packet API, exact inputs, source range receipts, all predictions, raw times, figures, proof, and repeat instructions. The joins between buffers are artificial. The labels use known future response times. The test does not show live intention decoding, stimulation benefit, or lower energy use. The contribution is a reproducible SEEG tool and a finite performance test of an established method.