In this section, we describe how to set up your EEG lab, with a strong emphasis on obtaining good EEG high-density data quality.

Data quality is crucial for EEG research. Researchers depend on clean and precise data to draw valid conclusions about brain function and behavior. If the collected EEG data contains artifacts or noise, it can introduce errors and bias, leading to invalidation of the hypothesis or misleading results. EEG recordings are susceptible to various artifacts, including muscle activity, eye blinks, environmental interference, and electrode impedance problems. These artifacts can introduce confounding factors, masking or distorting the true brain signals of interest. High-quality data minimizes the presence of artifacts, enhancing the accuracy of the observed brain activity and reducing confounds.

Cleaning up bad data or removing artifacts is a time-consuming and complex task. Poor data quality increases the workload on researchers, as they have to spend significant time and effort identifying and removing unwanted signals. Importantly, very often, it is not even possible to remove artifacts from EEG data, and doing so can decrease statistical power. By ensuring high-quality data from the beginning, the analysis process becomes more efficient, saving time and resources.

Table of contents

• Which EEG system to use
  • Active vs. passive electrodes
  • What is CMS/DRL correction?
  • Dry vs. wet electrodes
• Maximizing EEG data quality - tips and grand-mother recipes
  • Environmental noise
  • Applying gel
  • Checking impedances
  • Scanning electrode position
  • EEG synchronisation

Which EEG system to use

There are several reputable EEG (electroencephalography) research systems available on the market. Most are being made by companies that have been there for decades.

• Brain Products: The company offers various amplifier options and software solutions. Offers a convenient way to check the impedance on the cap with LEDs.

• BioSemi: This company offers a single active electrode system that provides excellent signal quality, low noise, and is known for its ease of use. Because this company only offers one system and provides no analysis software, it is often cheaper than other solutions. This company is known for its CMS/DRL noise reduction (see below), although other companies use it too. It is also known for not directly measuring electrode impedance (instead using a custom offset), which annoys some researchers (for good reasons).

• Neuroscan by Compumedics: This is the oldest company, comparable to Brainamps, with more focus on clinical research.

• g.tec: g.tek provides wireless and mobile EEG systems designed for neurofeedback and brain-computer interface (BCI) applications.

• ANT Neuro: A company created more than two decades ago.

• EGI/Magstim (formerly Philips Neuro): Known for its water-based electrode system (as compared to gel-based), this company has changed hands several times. The hardware is solid, and their gel caps are considered some of the best in the industry. Innovation at this company has slowed down in the past decade since it changed hands twice. However, BIOSEMI is still using the same amplifiers. EEG does not evolve that fast.

These are the main companies for EEG research, and we have personally worked with all of these systems. Amplifying EEG data is not rocket science, and the quality of the data is more linked to the cap than the EEG amplifier, which uses decade-old technology.

What about consumer-grade EEG systems, for example:

• Emotiv: For example, the Epoc+ 16-channel consumer-grade EEG system offers a more affordable and accessible option. It is suitable for basic brain activity monitoring and is often used in gaming and entertainment applications. This is a water-based system. Although the EEG amplifier is probably OK (although there is one amplifier that scans each channel in sequence, in comparison to the research grade system above, which has one amplifier per electrode), the quality of the signal coming from the spongy electrodes is low (still EEG, just bad EEG). We do not recommend this system for research.

• Muse: Muse offers different headsets, but all of them have a low channel count (less than 5). Even though this is considered a dry system, the company advises wetting electrodes with water to increase signal quality. Like BIOSEMI, this system uses the CMS/DRL correction.

• OpenBCI: Sells an open-sourced design for low budgets. The same remark applies to the Emotiv solution.

The main issue with consumer-grade EEG is the low signal quality when recording on the scalp (when recording directly on the skin (forehead), the signal quality is likely comparable to research-grade systems). The main reason is that the signal quality on the scalp relies on using gel-based electrode systems (which these consumer systems do not want to use).

What about the clinical EEG systems (Nihon_Kohden and the like)? These systems offer similar signal quality as the research system, although the technology is often older (they do not use active electrodes, for example). The issue with such systems is that they are limited to low-density recordings (less than 32 channels) and make it difficult to record experimental events synchronized with the EEG.

We are not mentioning here research-level dry electrode systems (some of the companies listed above have a dry electrode system solution), and there are other companies that specialize in that (Cognionics, Quasar, Wearable sensing, etc.). We do not believe the data quality is on par with gel systems (see below) and, therefore, would not recommend such systems.

This is by no means an exhaustive list, and there are many other reputable EEG systems available.

Active vs. passive electrodes

Active and passive EEG electrodes are two types of electrodes used in electroencephalography (EEG) to measure brain activity. Here are the key differences between active and passive EEG electrodes:

• Design. Active electrodes: These electrodes have built-in amplification circuitry, which amplifies the weak electrical signals received from the brain before transmitting them to the EEG recording system. Passive electrodes: These electrodes do not have built-in amplification circuitry and directly transmit the low-voltage electrical signals from the brain to the EEG recording system.

• Noise Reduction. Active electrodes: Due to their built-in amplification circuitry, active electrodes can amplify the weak brain signals closer to the scalp, minimizing the effect of noise and interferences. Passive electrodes: Since they lack amplification circuitry, passive electrodes are vulnerable to picking up more noise and interference, which can affect the quality of the recorded EEG signals.

• Input Impedance. Active electrodes: Typically have a very high input impedance, meaning they draw very little current from the scalp, reducing the contact impedance between the electrode and the scalp. Passive electrodes: These electrodes often have higher contact impedance, as they do not have amplification circuitry. This can result in a higher contact impedance and potentially lead to signal distortions.

• Signal Quality. Active electrodes: With their built-in amplification circuitry and low noise characteristics, active electrodes tend to provide higher signal quality and better signal-to-noise ratio compared to passive electrodes. Passive electrodes: Due to their susceptibility to noise and interferences, passive electrodes may have lower signal quality and a lower signal-to-noise ratio.

• Complexity and Cost. Active electrodes: Active electrodes are typically more complex and sophisticated, as they require additional circuitry for signal amplification. This complexity can lead to higher manufacturing costs. Passive electrodes: Passive electrodes are simpler in design and do not require additional amplification circuitry, making them less complex and cheaper to manufacture.

• Scalp Preparation. Active electrodes: Since active electrodes have a higher input impedance, they typically require less scalp preparation, such as cleaning or application of conductive gel, to ensure good electrical contact. Passive electrodes: Due to a potentially higher contact impedance, passive electrodes may require more thorough scalp preparation, including cleaning and gel application, to achieve optimal electrical contact.

In summary, active electrodes have built-in amplification circuitry, provide better noise reduction, higher signal quality, and have higher input impedance compared to passive electrodes. However, they are more complex and expensive. Passive electrodes, on the other hand, lack built-in amplification circuitry and may pick up more noise and interferences, resulting in lower signal quality. They are simpler and cheaper but require more thorough scalp preparation for optimal contact.

In practice, in the last decade, most EEG manufacturers have sold active electrode systems, which should be preferred for most applications. However, the difference in signal quality with a passive system is small (so you should continue using your passive system if you have one).

What is CMS/DRL correction?

CMS/DRL correction stands for common mode sense/difference mode sense-driven reference level correction. It is a technique used to reduce the influence of common mode noise on the recorded brain signals.

EEG signals are susceptible to various types of noise, including common mode noise, which refers to noise that is present on both the active electrode (recording electrode on the scalp) and the reference electrode. This common noise can interfere with the accurate measurement of brain activity.

CMS/DRL correction involves using additional electrodes placed on the subject’s body (or on the scalp) to measure the common mode and difference mode signals. The common mode signal represents the noise present on both the active and reference electrodes, while the difference mode signal represents the actual brain signal of interest.

By utilizing these extra electrodes, the CMS/DRL correction technique effectively subtracts the common mode noise from the recorded EEG signal, leaving primarily the brain activity. This correction helps improve the signal quality and increases the accuracy of EEG analysis by reducing the impact of external interference and noise sources.

Overall, CMS/DRL correction is a valuable method for enhancing the signal-to-noise ratio and obtaining more reliable EEG measurements. The Biosemi company is known for this system although it might also be used by other companies (the Muse system also uses it). This type of correction is not needed when the EEG signal is recorded in a faraday cage.

Dry vs. wet electrodes

Dry electrode versus “wet” cap vs. “water-based” cap. Dry electrodes and “wet” caps are both types of electrodes used in electroencephalography (EEG) to measure brain activity. However, they differ in terms of their contact with the scalp and the use of conductive materials.

• Dry electrodes: Dry electrodes do not require the use of conductive gels or creams to establish contact with the scalp. They are typically made of carbon or metal and have a system built within the electrode to improve conductivity. Advantages: Easy to use and less messy compared to wet electrodes. Faster setup time as there is no need to apply conductive gels. Disadvantages: Generally, lower signal quality due to reduced contact with the scalp. Sometimes rigid, so heavier. More susceptible to environmental noise and movement artifacts. Reduced signal stability over long periods of time. Increased pressure to keep contact that can lead to subject headaches

• “Wet” caps (also known as gel electrodes): “Wet” caps use conductive gels or creams to establish a good electrical connection between the electrode and the scalp. These caps consist of several electrodes embedded in a flexible cap or a headgear. Advantages: Better contact with the scalp, resulting in higher signal quality and reduced noise. More stable signals. Disadvantages: Takes time to set up (about 30 minutes to 1 hour). Subjects have gel in their hair and often need to wash their hair at the end of the experiment. Requires a time-consuming cleaning process.

• “Water-based” caps (e.g., EGI or Emotiv company): “water-based” caps use saline water instead of conductive gels or creams to establish an electrical connection between the electrode and the scalp. These caps consist of several electrodes embedded in a flexible cap or a headgear. Advantages: Quick set up time. Subjects may have slightly wet hair at the end of the experiment but might not need to wash their hair right away. Disadvantage: Water bridges can form (direct connection between two channels), which can create later problems during data analysis. This is especially a problem for high-density systems (128 channels or more). Electrodes dry up and may need to be re-wet every 20 minutes.

In general, because data quality is essential, research should be conducted with a wet gel-based system. Then, in case this is not possible, a water-based system is the second best choice (for example, in clinical applications or applications with kids where it is not possible to spend 30 minutes adjusting a cap). Dry electrodes are the last choice, although it makes sense to use them when embedded in an audio headset, a VR system, or an earbud.

Maximizing EEG data quality - tips and grand-mother recipes

Environmental noise

Recording EEG data in a Faraday cage will lead to better signal quality and no line noise when done properly. In practice, most EEG experiments (>80%) are conducted outside a Faraday cage. However, when recording outside of a Faraday cage, minimizing environmental noise in the recording chamber is critical (note that this is also critical inside a Faraday cage).

• Some researchers are obsessed with the quality of the ground signal in a building. We know of at least one EEG researcher expert in signal analysis who had an electrician create a second ground just for his EEG experiment. This is especially relevant for EEG systems that do not run on battery and are plugged into the wall.

• The subject and the experimenter should be isolated in a separate room. If they communicate with an intercom, make sure the intercom is far away from the subject. Do not use an intercom that relies on the electrical circuit to transmit the signal.

• Remove metal objects touching the subject (or very close). For example, the subject should not be sitting on a metal chair. If the subject needs to respond to stimuli, make sure the button press or mouse pap is nonmetallic.

• Remove any non-shielded battery or backup power. Non-shielded batteries can create very large noise artifacts, especially when plugged into the wall. Laptops, if held far away from the subject, are probably OK. Laptops should not be plugged into the wall and run on battery if present with the subject in the recording room.

• It is probably a good idea for the subject to remove anything that could be used as an antenna on his body (e.g., a metal watch).

• Compute screens and light are the major sources of 50 Hz noise in the EEG. These should be placed as far as possible from the subject. Reduction of the 50 Hz noise is very important, as in practice, the lower the 50 noise is, the lower the noise is at other frequencies. One technique we have used is to have one electrode on a pole and go around the room to try to detect the source of noise (outlet, screen, etc.). EMF detectors can probably also be used.

Applying gel

Using the syringe filled with gel, push down on the electrode well (so the gel doesn’t spread to other parts of the cap), and part the participant’s hair with the needle until you reach the skin. Then squeeze a small amount of gel into each well. Press firmly but not so firmly that the subject experiences pain. DO NOT PUT TOO MUCH GEL IN, otherwise, the gel will spread between electrode wells across the scalp, merging multiple distinct EEG signals into one.

Slightly irritating the scalp of the subject by asking them to brush their hair for 5 minutes may decrease electrode impedance and increase signal quality.

Checking impedances

Decreasing electrode impedance and maximizing data quality (even for high-impedance systems using active electrodes) is critical. The extra 10 minutes you will spend making sure each electrode has a good connection is critical. In addition, ensuring that the reference electrode has the best possible is important. If there is noise in this electrode, it will reflect on all others.

Scanning electrode position

Scanning electrode position is easy (a smartphone and an app can construct detailed 3-D models) and can improve source location (even in the absence of the subject MRI). It should be done systematically even if you are not sure you are going to use that information (see this page for more information).

EEG synchronisation

Synchronizing EEG with experimental events is critical and needs to be performed with millisecond precision (in psychophysics, a 10-millisecond difference in reaction time is considered large). Here are a few tips.

• If you are using visual stimuli, check the screen you are using. You should likely be using a low-latency, high-frequency (200 Hz or so) gaming screen. Standard LCD screens at 60 Hz can have all types of strange behavior: for example, the screen may be divided into 4 in hardware with the 4 sections of the screen shown in sequence – which is invisible to the eye but not to the retina or visual system). Most psychophysics software allows to check the latency of stimuli presentation. However, researchers should still check with a photodiode that the visual stimulus is synchronized with the experimental event signal (coming out of the experimental computer and entered into the EEG event box).

• Wireless EEG systems are subject to buffering, and this needs to be taken into account. This is partially handled by systems like the lab-streaming layer (LSL, developed in our lab).

• Only rely on experimental event signals directly sent to the EEG system. For example, the LSL event stream will only offer a precision of a few milliseconds, which is considered substandard for EEG (this is OK when the event does not need to be precisely timed). Even if you are using LSL, make sure the subject responses are logged as an additional channel in the EEG (or directly logged along with the EEG depending on the system). For event-related studies, EEG should also always be recorded with at least a 1Khz sampling rate for that reason (otherwise, events will not be recorded with 1kHz precision).

Note: This page only reflects Arnaud Delorme’s personal opinion

---