Shared bipolar subtraction gave unequal linear prediction gains at a 4 ms lag in independent-contact models of two SEEG records

We tested shared bipolar subtraction with public SEEG spectra from two patients in OpenNeuro ds005670 version 1.0.0. Each patient supplied seven 20 second windows. We kept each recorded contact power spectrum and set contact cross spectra to zero. At a 4 ms lag, the shared-edge median absolute net linear gain was 8.03e-6 and 5.01e-6. A closed-form proof gives the condition for each gain to be positive. The exact time reversal contrast is zero. Disjoint channels have zero exact gains at this lag. In the public-data phase benchmark, output phase tests did not show excess flags. A separate exploratory test used 40 trials of three independent Gaussian AR contacts with coefficients 0.8, 0.8, and 0.2. Output phase controls flagged the time reversal contrast in 30 trials; contact phase controls flagged it in three. Thus a zero population contrast does not ensure a valid output phase test for contact independence. This result concerns a finite-lag linear score and the stated null procedures. It does not establish neural direction or a patient artifact rate. The package includes all 14 input arrays, a raw retrieval recipe, source locks, code, proof, edge results, control decisions, and figures. All input arrays matched fresh raw reconstruction. Eleven numerical result files matched on repeat runs.

stereoelectroencephalography · bipolar subtraction · linear prediction · time reversal · phase randomization · signal processing