The signal is mostly not-brain
Here is a blunt truth that surprises most beginners: a raw EEG trace is dominated by artifacts — non-brain electrical junk that is often far larger than the brain signal you actually want. Imagine trying to hear one person whispering across a crowded, noisy room. The whisper is the brain. Almost everything else is the room.
Why so noisy? The electrodes sit on your scalp, centimeters of skin and bone away from the neurons. Anything electrical that is physically closer — your eyes, your face muscles, your heart, even the wall power — gets a louder seat at the table. The brain's vote is real, but it is quiet.
The usual suspects
Most of the junk comes from a short list of repeat offenders. Learn to recognize these five and you have named most of what fights your brain signal:
- Eye blinks and movements (EOG) — the eyeball is like a tiny battery, so every blink or glance throws a big, slow swing into your frontal channels.
- Muscle tension (EMG) — clenching your jaw, frowning, or even a tense neck produces fast, spiky bursts that smother nearby electrodes.
- Heartbeat (ECG) — your heart's electrical pulse is strong enough to leak into the recording as a steady, rhythmic blip in time with your pulse.
- Head and cable motion — a nod, a wobble, or a swinging wire physically jostles the electrodes and smears the trace with large, messy drifts.
- Power-line hum — every nearby wall outlet radiates a steady electrical buzz at 50 Hz across much of the world (60 Hz in the Americas), which the leads pick up like an antenna.
How we fight back
The good news: each suspect leaves a fingerprint, and we have a matching tool for each. The first move is filtering by frequency. Power-line hum lives at one exact pitch (50 or 60 Hz), so a narrow notch filter can mute that single tone. Slow eye drifts and fast muscle buzz live at the far edges of the spectrum, so trimming the very lowest and very highest frequencies clears a lot of mess in one pass.
The second move is rejection: when a chunk of recording is hopelessly contaminated — say, a giant blink or a cough — we simply throw that slice away rather than trust it. A short gap of honest silence beats a stretch of confident garbage.
Why it matters for a BCI
This is not housekeeping you can skip. A blink can masquerade as a command. If your brain-computer interface learns that a big frontal swing means "move the cursor," it has not learned to read your intent — it has learned to detect your blinks. The device will look brilliant in the lab and fall apart the moment a real user blinks for any ordinary reason.
Honest artifact handling is the difference between a real BCI and a glorified blink-detector. Once the signal is genuinely clean, you have something trustworthy to work with — and the next rung is where the magic happens: turning that clean brain signal into a decision the computer can act on, which is the job of decoding.