Sampling rate · Grey Matter

The sampling rate is how many times per second a recording measures its signal, and it sets the fastest event a method can capture: a brain read slower than its spikes can only see what those spikes leave behind.

Sampling rate. The sampling rate is how many times per second a recording measures its signal, and it sets the fastest event a method can capture: a brain read slower than its spikes can only see what those spikes leave behind.

To capture a waveform, a signal has to be sampled at more than twice its highest frequency (the Nyquist rate), and in practice several times more to keep its shape. An action potential lasts about a millisecond and its sharp parts carry energy up to several kilohertz, so spike recordings are sampled at 20 to 30 kHz: Neuropixels samples its spike band at 30 kHz and its slower local field band at 2.5 kHz. EEG and MEG, which record summed synaptic currents that rarely carry useful power above a few hundred hertz, are sampled at hundreds of hertz to a few kilohertz.

The method can be slower than its sampling. fMRI takes a whole-brain image every second or two, but the blood-oxygen signal it reads rises and falls over several seconds, so the true limit is the slowness of the blood, whatever the scanner does.

Timing is information. Which neuron fired first, by a few milliseconds, carries meaning in sensory and motor circuits, and only methods sampled fast enough to resolve spikes can read it.

Rate multiplies into data. Every electrode sampled at 30 kHz with 10 bits produces 300,000 bits per second, which is where the recording bandwidth problem begins.

Questions: Do neural data laws protect a raw recording, or only what is decoded from it? The raw recording, at whatever rate it was sampled. Colorado and California define neural data as information generated by measuring the activity of the nervous system, so the protection attaches to the measurement itself, before any decoding. That choice matters because what can be extracted from a recording depends on how it was sampled and on decoders that will keep improving, so a stored recording can reveal more in a few years than it did when it was collected. Why must spikes be sampled tens of thousands of times a second while EEG makes do with hundreds? A spike lasts about a millisecond and its sharp edges carry frequencies of several kilohertz, and a signal has to be sampled at more than twice its highest frequency to be captured, in practice several times more to keep its shape. Neuropixels therefore samples its spike band at 30 kHz, and spike sorting depends on those shapes to tell neighbouring neurons apart. EEG records summed synaptic currents that change slowly and are filtered further by the skull, with little useful power above a few hundred hertz, so hundreds of samples a second to a few thousand are enough.