A BCI records electrical activity from populations of neurons,
amplifies and filters it, then a decoder classifies the pattern
into an intended command that drives a cursor, speech synthesiser
or robotic limb.
SNR(dB) = 20·log10(gain / noise)
decode success if SNR·confidence > threshold
- Invasive implant — electrodes penetrate cortex, sitting right next to individual neurons: high SNR, high resolution, but requires surgery (e.g. Neuralink-style arrays).
- Non-invasive EEG — scalp electrodes read summed activity through bone and skin: safe and reversible, but noisier and lower resolution.
- Gain — amplifier strength applied to the recorded signal.
- Noise — electrical and muscular interference competing with the true neural signal.
- Decode threshold — the confidence a command needs to clear before it's executed; higher is safer but slower.