What an EEG band can—and cannot—tell us

An EEG band is a named frequency range. Looking at a band asks how much of a measured signal varies at those rates, or what remains after other frequencies are attenuated. It does not reveal a separate stream containing a particular thought.

EEG records voltage differences at electrodes. A recording mixes activity from multiple sources, along with noise and artifacts. Frequency analysis is useful precisely because it offers another way to describe that mixture—not because it separates every source cleanly.

A label depends on a convention

Delta, theta, alpha, beta and gamma are familiar names, but their exact boundaries can vary with the source and task. A study should state its chosen intervals. Our EEG band explorer gives the numerical boundaries and uses synthetic tones so the filtering behavior can be checked.

A component just below a named boundary does not abruptly become a different biological process when its frequency moves just above it. Band names are useful summaries, not hard divisions in nature.

Filtering changes what we can inspect

A filter has a frequency response: it retains some components more strongly than others. Practical choices include transition width, attenuation, filter length and phase behavior. These can change the apparent timing or shape of an event. A sharp-looking event can also produce ringing after filtering.

The explorable selects Fourier bins in a short periodic synthetic record. Its clean reconstruction is a property of that controlled example. Real EEG requires choices about padding, edges, artifacts and the purpose of the analysis; the demonstration is not a validated preprocessing pipeline.

Power is not a diagnosis

Band power summarizes signal energy in a chosen interval. Absolute and relative power answer different questions. Relative power divides by a specified total, so a proportion can rise because the numerator grows or because power elsewhere falls. Always state the denominator's frequency range.

Brain rhythms have contextual associations. For example, posterior alpha is often visible during relaxed wakefulness with the eyes closed. That observation does not justify assigning a mental state from a band-power number alone. Electrode location, reference, task, age, recording quality and other signals can affect interpretation. Muscle activity can contaminate faster frequencies.

The question to carry into a project

Before asking whether a band helps a classifier, specify the prediction target, the recording conditions and how performance will be evaluated on unseen people. For my sleep-staging project, those choices are documented in the completed internal study; its detailed results are held for later release. No sleep-stage accuracy is claimed here.

The American Epilepsy Society's normal EEG chapter supplies clinical context for rhythms. MNE's filtering tutorial explains practical filter design and its pitfalls. The lesson to take from both is to keep acquisition and analysis choices attached to any interpretation.