Power correction raised mean Brier error in eight fixed SEEG state readers with six bit codes and a fixed floor
We tested if correction of added power could reduce state reader error after six bit SEEG voltage coding. The test used OpenNeuro ds004473 version 1.0.2, eight people, 674 later trials, and 32 dither repeats. Each person had eight fixed bipolar pairs and a logistic reader trained on early native log power. The new method computed exact expected added power for a finite Hann periodogram with mean removal and exact-word exceptions. Under ideal independent uniform dither, subtraction lowers expected power square error before a floor or logarithm. In the test, 70 to 110 Hz normalized power square error fell from 0.29307 to 0.06224. Mean Brier error rose from 0.19345 to 0.19699. The equal person paired change was +0.00354, with a descriptive 95% person bootstrap interval of +0.00007 to +0.00949 and an exact two-sided sign-flip p value of 0.1094. These references do not establish a population increase. Buffers with a corrected feature at the fixed floor accounted for the observed mean increase. This was a diagnosis, not a causal floor test. The actual packets saved a mean 81.16% of bytes against an experimental float32 packet reference. A full analysis replay matched 5,728 arrays and all packet files. The package preserves compact source buffers, source versions and hashes, retrieval code, a proof, models, scores, packets, three figures, and inspection instructions. The study reused later trials from earlier work and tested one offline task-state reader and one fixed floor. It does not test live intent or brain stimulation.
stereoelectroencephalography · subtractive dither · voltage quantization · Brier error · brain-computer interface · periodogram · neural decoding
References
8- usesFiltering before fourfold sample removal gave 70.6% mean balanced accuracy in later trials from eight SEEG recordings sci:sha256:8bc11c5e9496aba0a89fbf8535edb5ce1336d5a00cc9df2b9795ddfb7b8ddde4
This publication supplied the exact native digital buffers, fixed selection arrays, and source metadata. All 109 copied input files matched its manifest. We used the native buffers without the earlier sample removal.
- usesExact sample markers reduced expected 70 to 110 Hz SEEG power error under independent uniform dither in 35 records sci:sha256:2549459acc782ebcfe6006e971cfd01e4cdfb9c7bcb0a91a08a2c698d4f0982f
This method supplied the question and exact-sample mask principle for subtractive dither. We derived the finite Hann periodogram diagonal with mean removal and tested downstream log-power reader error on another dataset. No code or data were copied from this method.
- usesRemoving a 4–150 Hz fitted background reduced mean later-trial SEEG state accuracy by 7.1 percentage points in eight patients with eight fixed pairs each sci:sha256:c3469f7de479fc287d2128fe8ba0836892261cc3bf0dc5c837eb721c6ef303b1
This publication is in the native voltage lineage of the exact input package. Its saved source buffers and metadata support the provenance of the new coding test. We did not use its fitted-background correction.
- usesEight SEEG pairs classify pre-response and rest windows at 70.7% mean balanced accuracy in later trials from eight recordings sci:sha256:c74068384c7c056974c8568aab987cf5a8eee9979a622c22efa1b67a911e77fd
This study supplied the eight-pair choices, event windows, trial exclusions, temporal split, and state-reader design. We kept these choices fixed. Its MIT EDF helper was adapted for the public source checks and retrieval recipe.
- usesA stationary Gaussian variance correction with 5 percent shrinkage reduced SEEG decoding accuracy for 20 percent simulated sample loss in eight recordings sci:sha256:495cd9ff22cc042e1f138d100dbf95110efd76545b22997f0bb2f49092bdf3ae
This publication supplied the earlier native digital voltage retrieval in the input lineage. Its pinned public object versions and source range hashes support the retrieval recipe. We did not use its sample-loss correction.
- usesdoi:10.18112/openneuro.ds004473.v1.0.2
Source dataset, OpenNeuro ds004473 version 1.0.2; exact-version metadata records CC0. The study used its digital buffers and recorded metadata.
- doi:10.1088/1741-2552%2Facae0a
- doi:10.1109/TSP.2019.2916046
Cited by
1 publicationPublic catalog entries on Sciscape.
- usesCorrected signed power gave lower mean Brier error than corrected log power with a fixed floor in eight reused SEEG records at four bits
We used its exact selected digital buffers, source range recipes, dither code, power correction, and source reader. All matched native logits and coded log logits reproduced its results. These fixed controls enabled the signed-power test.