From microvolts to numbers
Imagine trying to hear a single voice whispering across a noisy stadium. That is roughly what a brain–computer interface faces. The electrical signals your neurons produce are tiny — measured in microvolts, millionths of a volt — while the world around them hums with electrical noise. Before a computer can do anything useful with these signals, they have to make a short journey called the acquisition chain.
The chain has three big jobs. First amplify: boost the faint signal until it stands above the noise floor. Then sample: take measurements many times every second, like film frames freezing a moving scene. Finally digitize: convert each measurement into a discrete number. By the end, a smooth ripple of voltage has become a stream of numbers — the raw material for everything else a BCI does.
Amplification
A scalp EEG signal is often just tens of microvolts — far too small to measure directly. So the very first device in the chain is an amplifier: an electronic component whose whole purpose is to make a signal bigger. The amount it multiplies by is called gain. A gain of 10,000, for instance, turns a 20-microvolt whisper into a 0.2-volt signal the rest of the system can comfortably handle.
But an amplifier is honest to a fault: it boosts noise just as eagerly as it boosts your brain signal. This is where impedance — how strongly the electrode-to-skin connection resists the flow of current — quietly matters. A high-impedance, poorly attached electrode lets stray interference leak in and get amplified alongside the real signal, producing a signal artifact. Keeping impedance low (clean skin, good gel or paste, snug contact) is one of the simplest ways to get cleaner recordings.
Sampling and the Nyquist rule
Once amplified, the signal is still a smooth, continuously changing voltage. A computer cannot store something infinitely smooth, so it takes samples — snapshots of the voltage at regular instants. The number of snapshots per second is the sampling rate, measured in hertz (Hz). Think of it like filming a spinning wheel: take enough frames per second and the motion looks right; take too few and the wheel can appear to crawl, freeze, or even spin backward.
That spinning-backward illusion has a serious cousin in signal recording called aliasing: if you sample too slowly, a fast wiggle in the brain signal disguises itself as a slow one, and you can never tell them apart afterward. The remedy is a simple rule discovered by engineers Harry Nyquist and Claude Shannon: you must sample at least twice as fast as the fastest frequency you care about.
f_s >= 2 x f_max # f_s = sampling rate (samples per second, Hz) # f_max = highest frequency in the signal you want to keep # Sample at least twice the fastest frequency, or it aliases.
Quantization and resolution
Sampling decides *when* you measure; quantization decides *how finely* you record each measurement. A computer stores numbers in discrete steps, so the smooth voltage at each snapshot gets rounded to the nearest available step — like reading a temperature only to the nearest whole degree. The number of steps is set by how many bits the converter uses: each extra bit doubles the steps. An 8-bit converter offers 256 levels; a 16-bit converter offers 65,536.
More bits means finer resolution — the rounding error gets smaller, so subtle details in the brain signal survive instead of being flattened away. Because EEG features can be vanishingly small, BCI systems typically use 16 to 24 bits, capturing voltage differences far smaller than a single microvolt. With amplification, sampling, and quantization all done, the brain's faint electrical whisper has finally become a clean stream of numbers, ready for the decoding steps ahead.