Six-bit asinh encoding gave lower balanced error than tuned uniform encoding with one range in processed SEEG at 86 selected sites after five calibration pulses per site

We tested six-bit digital encoding in processed public SEEG from eight people with epilepsy. There were 86 selected bipolar readouts. The first five pulses at each site set the parameters. The final five pulses supplied the test data. Asinh encoding gave a 72.5% decrease in the geometric balanced error ratio relative to tuned uniform encoding with one range per site. All eight person log effects favored asinh. The arithmetic mean error decreased by 34.2%. A further uniform control used two ranges selected by known pulse timing. Its arithmetic mean error was 67.2% lower than asinh, although the mean log comparison favored asinh. This control was added after the primary scores were known. Thus, the main claim applies to the tested uniform encoder with one range. The work adds a public benchmark, fixed encoders, and saved six-bit streams. All eight complete source files passed fresh hash checks and reproduced the archived voltage exactly. The supporting archive retains voltage, provenance, results, three figures, and packed streams. GPL analysis and retrieval scripts are omitted, so the public package does not provide an executable reproduction path. The study does not test live recording, an analog converter, or clinical benefit.

SEEG · neural signal processing · asinh encoding · quantization · electrical stimulation · balanced error · brain-computer interface