A simulated SEEG ripple trigger retained held-out F1 at 12 bits and 1000 Hz in 20 patients

We simulated an ideal analog-to-digital converter (ADC) for a causal hippocampal ripple trigger using 90 seconds of interictal NREM sleep SEEG from each of 20 patients, with 74 bipolar pairs and 12,309 held-out author-marked ripples. For each monopolar contact, we applied causal anti-alias filtering, decimated to 1000 or 500 Hz, clipped at ±5 mV, quantized to 12, 10, 8, or 6 bits, and then formed bipolar signals. Against the 2000 Hz float reference (median F1 0.45), 12-bit quantization met the pre-set criterion for retained accuracy at 2000, 1000, and 500 Hz. The mean held-out F1 change at 1000 Hz and 12 bits was −0.001 (95% bootstrap CI −0.010 to 0.010); at 500 Hz it was −0.016 (−0.024 to −0.008). Ten-bit quantization failed the pre-set criterion at every rate; 8- and 6-bit quantization reduced F1 by about 0.19 and 0.24. At 1000 Hz and 12 bits, the simulated stream was 12 kbit/s per contact with no detectable F1 loss. This was an ideal ADC simulation without noise or dither. The recorded signals had already passed through a clinical amplifier, and author marks served as the reference. Results cover 90 seconds per patient and at most four pairs per patient; they do not establish performance in a live implant.

SEEG · hippocampal ripples · causal detection · analog-to-digital conversion · quantization · sampling rate · closed-loop stimulation · NREM sleep