Threshold-crossing and feature-based compression

The most widely deployed on-implant data reduction: instead of streaming the full broadband voltage, the device detects when the signal crosses an amplitude threshold and transmits only the times of those events (threshold crossings), or coarse features such as counts per bin and spike-band power. This exploits the fact that much decodable information in intracortical BCIs lives in spike timing and rate.

The reduction is dramatic — broadband at tens of kilohertz and many bits per sample collapses to sparse events — and multiple studies show cursor and reach control from threshold crossings that rivals full spike-sorted decoding. The honest cost is loss of waveform: without spike shapes one cannot separate two neurons on the same site, retrospectively re-sort, or robustly follow a unit as it drifts, so the strategy trades long-term flexibility for immediate bandwidth.