The Notch Filter in EEG

What Is a Notch Filter?

A notch filter is a signal-processing filter that strongly attenuates a narrow range of frequencies while allowing frequencies below and above that range to pass with comparatively little change. The rejected region is called the stopband, and its center is often called the notch frequency. Because the stopband is narrow, the filter is intended for a known, concentrated interference rather than for broad-spectrum noise.

In EEG, the interfering component is often electrical mains contamination introduced through the recording environment, cables, amplifiers, or electrode impedance differences. A notch filter can reduce a prominent 50 Hz or 60 Hz component, depending on the local electrical standard and recording conditions. It does not remove every source of artifact, and it does not distinguish electrical noise from brain activity by meaning; it acts on frequency content.

How a Notch Filter Operates (and What It Doesn't Do)

A notch filter works by creating a sharp dip in what engineers call the filter's magnitude response, which is just a plot of how much a filter allows each frequency to pass through.

At most frequencies, the response sits near full strength, meaning the signal passes through unchanged. At the target frequency, usually 50 or 60 hertz, the response drops sharply, attenuating that frequency band while leaving neighboring frequencies mostly intact. The result, in theory, is a cleaned signal where the power line hum is gone but the brain's own oscillations remain.

What a notch filter does not do is equally important. It is not a general-purpose artifact remover.

Broadband contamination, such as the muscle activity that spreads across a wide range of frequencies during jaw clenching, or the large deflections caused by eye blinks, passes right through a notch filter because that contamination is not confined to a narrow band.

The notch filter also performs no spectral decomposition. It does not break the signal into its component rhythms the way time-frequency analysis does when researchers want to see how power in the theta, alpha, or beta bands changes over time. A notch filter simply removes one narrow slice of frequency content and leaves the rest of the spectral structure exactly where it was.

Band Stop Notch Filter Design Basics

Band-stop design begins with a clear definition of the unwanted frequency and the amount of neighboring spectrum that can be sacrificed. A notch centered at 60 Hz with a very narrow bandwidth behaves differently from one that attenuates 55–65 Hz. The appropriate choice depends on interference stability, signal bandwidth, sampling frequency, and the scientific or clinical question.

Quality factor, commonly written as Q, expresses the relationship between center frequency and bandwidth. A high-Q filter has a narrow stopband relative to its center frequency; a lower-Q design affects a wider region. Greater attenuation and sharper transitions may require higher filter order, but higher order can also increase ringing, computational cost, and sensitivity to implementation details.

Design specifications are best evaluated through both frequency and time-domain behavior. Frequency plots reveal attenuation, bandwidth, and passband ripple, while impulse or step responses show ringing and transient behavior. For EEG, an apparent improvement in the spectrum should be considered alongside waveform morphology, event timing, and any downstream feature extraction.

Notch Filter vs Band Stop Filter: What's the Difference?

A notch filter is generally understood as a narrow form of band-stop filter. Both attenuate frequencies inside a stopband and pass frequencies outside it, but the notch usually targets a specific, isolated frequency or a very small interval.

A broad band-stop filter can be appropriate when interference occupies a wider, well-defined region. A notch is more conservative when the objective is to remove a single line component while retaining nearby activity. In EEG, that apparent conservatism does not eliminate the need for review, because a narrow filter can still affect harmonics, phase, or spectral estimates around the target.

The following comparison summarizes the typical distinction. Actual behavior depends on the filter specification and implementation, so the labels should not substitute for examining the response curve.

The practical difference is therefore one of selectivity. A notch filter is not automatically safer, and a broad band-stop filter is not automatically excessive; each must be judged against the signal features that matter for the recording.

Using a Notch Filter in EEG Signal Processing

EEG signals contain low-amplitude voltage fluctuations generated by distributed neural activity, along with physiological and environmental artifacts. A notch filter may be used during preprocessing when a narrow spectral peak is consistent with mains interference.

It is one operation within a larger workflow that can include:

• Rereferencing

• Bad-channel assessment

• Artifact correction

• Segmentation

• Baseline handling

The timing of filtering matters. Applying a filter before epoching can produce edge effects at the boundaries of the continuous recording, while applying it separately to short epochs can create different transients and frequency responses.

Researchers also distinguish offline processing from real-time processing: noncausal or zero-phase methods may be practical for stored data but cannot be used in the same way when decisions must be made immediately.

A notch should be evaluated against the analysis objective. It may be reasonable for reducing a stable line-frequency peak, but it cannot correct electrode motion, muscle activity, eye movements, poor contact, or broadband amplifier noise. The EEG recording context, montage, reference, sampling rate, and artifact profile all influence whether a narrow frequency filter is informative or misleading.

Common Notch Filter Frequencies (50Hz and 60Hz)

The most familiar notch frequencies in EEG are 50 Hz and 60 Hz, corresponding broadly to regional electrical power systems. The correct target is determined by the recording environment, not by a universal preference. A spectrum can show which line component is present, although harmonics and equipment-specific interference may complicate the picture.

A 50 Hz or 60 Hz setting may also influence harmonics at multiples of the fundamental frequency, depending on the filter or acquisition system. A filter centered at the fundamental does not necessarily eliminate every harmonic unless those frequencies are separately addressed or the system documents a harmonic-rejection behavior. Harmonic removal can further alter signal content and should be treated as an explicit processing choice.

The frequencies below are common reference points rather than a complete prescription. They describe the usual reason for selecting each target and the principal caution associated with it.

These values should be documented with the acquisition location, filter bandwidth, attenuation, and whether harmonics were included. Reporting only “notch filtered” makes later interpretation and replication more difficult.

Common Mistakes When Applying a Notch Filter

The most frequent error is treating a notch filter as a general-purpose cleanup step. A narrow filter has a specific target, so it should be connected to an observed or well-justified interference source. If the artifact is unstable, broad, or dominated by transients, a fixed notch may provide limited benefit while still changing the signal.

Another mistake is ignoring the relationship between filtering and the rest of the preprocessing pipeline. Reference changes, resampling, epoch boundaries, and artifact rejection can all affect the apparent spectrum.

A filter applied without checking the raw recording and its acquisition parameters may conceal a problem that would have been better addressed through electrode maintenance, shielding, grounding, or channel-quality review.

A compact review sequence can help keep the decision auditable:

• Inspect the unfiltered spectrum and waveform for a stable, narrow interference component.

• Confirm the local mains frequency and the recording system's sampling and filter settings.

• Compare candidate bandwidths using both spectral attenuation and time-domain waveforms.

• Record the filter type, order, center frequency, bandwidth, phase mode, and processing stage.

The Risk of Signal Distortion

Researchers suggest that the notch filter, despite being widely used, carries “the risk of potentially severe signal distortions.”

To test this, a 2019 study built synthetic test signals and also worked with real EEG and MEG recordings, including EEG data collected in an unshielded setting where power line noise was especially strong. Using a Butterworth notch filter, a common IIR filter design, they found that line noise was removed effectively. The interference at 50 or 60 hertz disappeared from the frequency spectrum as expected. But the same filtering process introduced distortions in the time domain, meaning the shape of the waveform over time was altered in ways that went beyond simply removing the unwanted frequency.

When the same tests were run using spectrum interpolation instead, the method matched the notch filter's ability to remove line noise, but the study reports it showed “less distortions in the time domain in many common situations.”

The advantage was not universal or absolute. It appeared across many common situations, which leaves room for cases where the two methods might perform more similarly, or where other factors could shift the outcome.

Why Removing Power Line Noise from EEG Deserves More Than a Default Fix

A notch filter provides a precise way to attenuate a narrow frequency component, making it relevant to EEG recordings affected by stable mains interference. Its usefulness depends on accurate frequency identification, an appropriate bandwidth, awareness of phase and edge effects, and comparison with the unfiltered signal.

Used as one documented step within broader signal-quality and preprocessing practices in neuroscience set ups, it can reduce interference without unnecessarily discarding neighboring information.

References

1. Leske, S., & Dalal, S. S. (2019). Reducing power line noise in EEG and MEG data via spectrum interpolation. NeuroImage, 189, 763–776. https://doi.org/10.1016/j.neuroimage.2019.01.026

Frequently Asked Questions

What is the main job of a notch filter in EEG recordings?

A notch filter is a narrow band-stop filter that suppresses one specific frequency—typically the 50 or 60 Hz power line hum—while preserving the rest of the EEG spectrum. It targets only this fixed-frequency electrical noise, not broadband artifacts like muscle activity or eye blinks.

Why are 50 Hz and 60 Hz notch filters common?

These frequencies correspond broadly to the alternating-current power standards used in different regions. Electrical interference from the recording environment can appear as a prominent line component at one of these frequencies and sometimes at its harmonics.

Why not just use independent component analysis (ICA) to remove line noise?

ICA separates different source signals across electrodes, not specific frequency bands, so it is not designed to target a single, fixed-frequency sinusoid like power line hum. The notch filter remains the standard tool because it directly attenuates that one frequency, whereas ICA may not isolate line noise as a separate source.

What kind of signal distortion has been directly observed with notch filters?

A study using both synthetic signals and real EEG data found that a Butterworth notch filter could introduce time-domain distortions, changing the shape of the waveform beyond simply removing the hum. These distortions were documented as potentially severe, though the precise impact on specific neural measures like ERPs was not examined in that same study.

What is the safest way to use a notch filter if you must use one?

If a notch filter is applied, gentler parameters—such as a slightly wider bandwidth and less steep attenuation—may reduce the risk of excessive time-domain ringing, though the exact harm to downstream analyses has not been quantified in the reviewed data. It is also wise to visually compare raw and filtered traces and to document every preprocessing step to catch unexpected distortions.

Can a notch filter remove real brain activity?

Yes. If meaningful neural activity overlaps the target frequency or its affected bandwidth, filtering can reduce that activity. This possibility is especially relevant when analyzing higher-frequency rhythms or harmonics near the notch.