A common phase output reduced the mean maximum error against two theta targets by 2.02 degrees in 50 ms offline forecasts from seven SEEG rest records

We tested a common phase output for two theta targets in seven public SEEG rest records from OpenNeuro ds006065 version 1.0.0. Each person had four fixed bipolar pairs. Separate ridge readers used only 250 past samples. They forecast target phases 50 ms after each cutoff. Both targets used centered filters and later voltage. The planned joint cosine output reduced the mean larger error against the two targets from 77.52 to 75.50 degrees. All seven people had a decrease. The mean decrease was 2.02 degrees, with a descriptive person interval of 1.13 to 2.96 degrees. A simple phase midpoint gave a lower mean error of 74.17 degrees. An early output gate gave 51.39 percent simultaneous 45 degree hits at 24.80 percent yield. The common output increased error against the broad target. The cohort and test block were used in earlier work. These results test agreement with defined signal targets. They do not establish a true neural phase or stimulation efficacy. The 18.56 MB package has code, exact inputs, all forecasts, source hashes, three figures, and reproduction instructions. A fresh public source check matched all input samples. A repeat analysis gave 23 byte-identical numerical products.

neural signal processing · theta phase · phase forecasting · SEEG · offline prediction · circular error · ridge regression · signal targets