From brain to file: the EEG signal path, step by step

Part 3 of 28 · Beginner · EEG from zero: the series

Between the neurons and your data file, the EEG signal passes through seven steps. Each one changes it. This lesson follows the signal along that path, so you know what your numbers have been through.

The one-sentence idea

The signal gets weaker and noisier on its way out of the head. The machine cleans, boosts and samples it into numbers.

Explain it simply

Think of a data pipeline with a source system, a network and an ingest job.

• The source is the cortex, where many neurons work in step (lesson 2).
• The network is the skull and scalp. It is lossy: the message arrives weaker and blurred.
• The ingest job is the EEG machine. It boosts the message, removes noise, samples it and writes it to disk.

Every step leaves a trace in the data. If you know the steps, you can tell real brain activity from things the pipeline added.

Go one level deeper

Step	What happens	What it does to the data
1. Cortex	Many aligned neurons in step make a field	The source signal
2. Skull and scalp	Current spreads through tissue to the scalp	Much weaker; smeared over a wide patch of scalp
3. Electrode	A metal contact picks up the voltage on the skin	Also picks up mains hum, slow drift and other noise
4. Amplifier	Takes the difference from a reference electrode, then boosts it	Noise that both electrodes share mostly cancels
5. Filter	Removes frequencies outside the band you want	Drift and very fast noise are reduced
6. Digitiser	Measures the voltage at a fixed rate, rounds each reading to a step	A smooth wave becomes a list of numbers
7. File	Stores the numbers with channel names and settings	A table: one row per sample, one column per channel

The skull is the weak link. It conducts current far less well than the brain or the scalp.

So the signal spreads out sideways before it reaches the skin. This spreading through tissue is called volume conduction.

At the scalp the signal is tiny: usually tens of microvolts (millionths of a volt). A AA battery gives 1.5 volts, about 30,000 times more than a 50-microvolt brain wave.

Every channel is a difference. The amplifier never measures one electrode alone. It measures the voltage between an electrode and a reference.

Mains hum reaches both, so subtracting one from the other removes most of it. Lesson 8 covers this in detail.

Filters come in two places. Before digitising, an anti-aliasing filter removes anything too fast to sample. After recording, you usually filter again in software.

The digitiser sets the grid. Two settings matter:

• Sampling rate: readings per second, often 256 to 1,000 for research and clinical systems. To keep frequencies up to 40 Hz, sample at more than twice that, so above 80 per second. This is the Nyquist rule (lesson 9).
• Resolution: how fine the steps are. Many systems use 16 or 24 bits.

Visual: from brain to file

Watch the trace at each step. The noise joins at the electrode, most of it leaves at the amplifier, and the filter removes the rest of the drift. Sizes are relative, and the signals are simulated.

Why it matters for data analysis

• Your file is a table. 19 channels sampled 256 times a second for 30 minutes is 19 × 256 × 1,800 ≈ 8.8 million numbers. Stored as 16-bit values, that is about 17.5 MB.
• Settings are part of the data. The reference, the filters and the sampling rate change what the numbers mean. Keep them with the data, like a schema.
• Filters can't be undone. Once a band is removed, it is gone. Check the recording's filter settings before you study very slow or very fast activity.
• Noise has a source. Hum, drift and blinks enter at known steps. Lesson 7 shows how to spot each one.
• Space is blurred. Because of volume conduction, one electrode hears a wide area. Lesson 18 shows what that means for locating a source.

Key terms

Term	Meaning
Volume conduction	Current spreading through the tissues of the head to the scalp
Microvolt (µV)	A millionth of a volt; scalp EEG is usually tens of µV
Reference	The electrode every channel's voltage is measured against
Differential amplifier	An amplifier that boosts the difference between two inputs
Anti-aliasing filter	A filter that removes frequencies too fast for the sampling rate
Digitiser	The part that turns voltage into numbers (an analogue-to-digital converter)
Sampling rate	How many readings are taken per second, in hertz (Hz)

Check yourself

1. Why is the signal at the scalp weaker and more spread out than in the cortex? The skull conducts current poorly. The current spreads sideways through tissue (volume conduction) before it reaches the skin.
2. Why does mains hum mostly cancel in the amplifier? The amplifier measures the difference between an electrode and the reference. Hum reaches both about equally, so it drops out when you subtract.
3. You want frequencies up to 40 Hz. Is 64 readings a second enough? No. You need more than twice the highest frequency: more than 80 per second. 256 would be safe.

Go further

• Nunez P. L., Srinivasan R. (2006). Electric Fields of the Brain: The Neurophysics of EEG, 2nd ed. Oxford University Press. Volume conduction and the skull.
• Sinha S. R. et al. (2016). American Clinical Neurophysiology Society Guideline 1: Minimum technical requirements for performing clinical electroencephalography. J Clin Neurophysiol 33(4): 303–307.
• MNE-Python. Overview of EEG and MEG data processing. How the file looks in Python.

---

For learning only, not medical advice.

Previous: Neurons and electricity · Next: Brain waves: delta, theta, alpha, beta, gamma