Three DPSS tapers gave lower mean later Brier error than a Hann taper in eight reused SEEG records with fixed bipolar pairs and 500-sample buffers

We compared three DPSS tapers with one Hann taper for power features in public SEEG data. DPSS means discrete prolate spheroidal sequences. The time and half-bandwidth product was 2. Eight people had one task record each. All methods used the same eight bipolar pairs, 500-sample buffers, power bands, trial split, and fitting rule. Early trials set the models. Later trials tested rest versus a state before a known response. Mean later Brier error fell from 0.181243 to 0.169496. The paired change was -0.011747, or a 6.48% decrease. Seven of eight records had lower error. The descriptive 95% person bootstrap interval was [-0.018232, -0.005066]. A fixed penalty gave a similar result. A single first DPSS taper gave little change at that penalty. Known test signals showed less power variation with three tapers. They also showed changes in the frequency response at band edges. Thus the decoding gain cannot be assigned to noise reduction alone. The cohort and test rows were used before. The pair list was fixed from earlier training data. This result concerns offline task-state reading. It does not establish intention, online control, a new-session gain, or a stimulation effect. The package includes three figures, input waveforms, models, predictions, source checks, code, and repeat instructions.

stereoelectroencephalography · SEEG · multitaper spectral analysis · DPSS · brain-computer interface · Brier error · power features