Does EmotivPRO automatically remove artifacts from the EEG data collected?

Artifacts

When using EEG headsets, some signals can interfere with brainwave measurements. These unwanted signals, called “artifacts,” come in two main types:

Intrinsic Artifacts: These are caused by normal biosignals originating from your body, such as:

• Facial, neck and jaw muscle activity: Smiling, clenching your teeth, or frowning, blinking, winking, chewing, speaking, turning your head (neck muscles). Each muscle group is located closer to some EEG sensors and much more distant from others, so the signal detected at each location is different, making the artifacts more difficult to remove. In fact Emotiv uses signal processing and machine learning methods to untangle the distribution of muscle signals to deduce which groups are activating, and therefore to identify your facial expressions!
• Ocular activity: Each of your eyeballs has a high concentration of nerves across the rear surface (retina, optic nerves) and almost no nerves across the front surface. In effect, your eyeball acts as a large dipole with an imbalance of electrical charge from front to back. When your eyes rotate in their sockets, the dipole field changes direction to point towards where you are looking, and this is detected as a change in the background biopotentiall which is angled differently relative to each EEG sensor - which means it is not a common signal across sensors. Additional signal artifacts are generated by the muscles controlling your eye rotation.
• Cardiac signals: Your heart is a significant source of raw muscle signals which can sometimes be detected directly by some or all EEG channels, in the same way that an electrocardiogram is recorded. The characteristic P-Q-R-S-T complexes may be directly observed occasionally in some EEG channels. A second type of cardiac artifact arises from large blood vessels which expand and contract as the heart pumps blood through your arteries. Arterial walls are muscluar, and generate secondary signals as they expand and contract in sync with our heartbeat. Finally, if you happen to place a sensor directly adjacent to a significant artery, the sensor may be mechanically displaced by the changing shape and size of the vessel, leading to rhythmic movements of the sensor across the skin surface which can change contact impedance and induce spurious voltages over a cyclic pattern.

These actions create muscle, eye and other biosignals signals that can mix with brainwave data. Usually these biosignals are significantly larger than brain signals, making detection of brain activity difficult unless some form of filtering and source separation are undertaken.

Intrinsic artefacts fall into specific, predicatable categories and there are many preprocessing tools which can be applied to selectively remove them. The most common method is Independent Components Analysis (ICA, available in many libraries such as EEGLab, NME and others), and Artefact Subspace Reconstruction methods (ASR, rASR, more computationally efficient than ICA). These models rely on breaking a time-series signal into different components, then reassembling the signal from a subset of these components which are not associated with different types of artifacts.

Emotiv EEG data is delivered to the host PC in as clean a form as possible, but without removing the intrinsic biosignal artifacts which may be of interest to different users, and which also enhances the ability of ICA and rASR methods to remove known classes of intrinsic artifacts because their signals are not distorted by on-device filtering.

Extrinsic Artifacts: These come from outside sources, such as:

• Sensors slipping, the headset moving on your head or being bumped
• Radiated electric fields from appliances, computers and other equipment, transformers and electrical wiring, particularly at the electrical power line frequency (50/60 Hz) and harmonic multiples of these frequencies. Power line noise is often the strongest source of artifacts in EEG signals. 
• All modern EEG systems use analog-to-digital signal converters which operate at a fixed sampling frequency. A well-known phenomenon with digital sampling is aliasing, which occurs when the sampling system encounters a signal which has frequency components higher than 50% of the sampling frequency (the Nyquist frequency). For example, when sampling at 128Hz, the Nyquist frequency is 64Hz, just higher than 60Hz power line frequency. However the harmonics of 60Hz: [120Hz,  180Hz, 240Hz, …] “wrap around” the Nyquist frequency and appear as fake or “aliased” signals at 8Hz, 24Hz, 16Hz and so on, because the digital system samples a part of every second, third, fourth … cycle of these high-frequency signals. High harmonics of power line radiation are present because the currents and radiated fields in power systems are rarely perfect sine waves. Typically there is substantial radiated power detectable up to about the 10th harmonic. These aliased high-frequency signals are indistinguishable from real oscillations at lower frequencies within the typical range of brain signals, so they must be removed from the incoming signal before it is presented to the sampling system.
• Static electric fields from charged objects and people nearby: Accumulating electrostatic charge can result in potential differences of many thousands of volts between you and other people and surrounding objects. For example a positively charged object will draw negative charges in your body and head towards that object, and negative charges to be repelled, resulting in an uneven distribution of body potential beneath different EEG sensors. Emotiv devices use AC-coupled sensing (analog high-pass filtering), with a single reference point, to decouple uneven static charge distribution to a significant extent. However if you or any of these charged sources moves around, charge moves around your body causing a changing potential, which may be fast enough to be transmitted through the filters.
• Your electrostatic potential can change slowly or instantaneously if you take charge on or rapidly discharge yourself, such as by walking on carpet or touching metal objects, perhaps generating a spark. Your body potential can change by tens of thousands of volts in an instant, a few seconds, or longer periods. These changes can temporarily overwhelm the body potential cancellation circuits in wearable EEG systems, resulting in massive spikes and slower recovery in the EEG signals.
  Laboratory-based EEG systems can be protected against many of these artifacts, for example by restricting the subject’s movement, electrically screening the laboratory, attaching a grounding lead to the subject to prevent electrostatic build-up, very high sampling frequency and so on.
  
  Wearable, battery operated wireless EEG systems cannot rely on these measures and therefore must use a range of mitigation strategies. Data transmission rate must be balanced against battery life, because wireless transmitters are quite power-hungry.

Reducing Interference

EEG headsets are designed to minimize unwanted noise. Most extraneous noise sources such as static electricity and electromagnetic interference (e.g., 50/60 Hz noise and harmonics from power lines) appear as Common Mode Noise, where the underlying body potential is oscillating in approximately the same way across all sensors. 

Emotiv devices use a single-point reference sensor (CMS) to measure the body potential, combined with an active cancellation system in the analog domain (CMS signal is inverted and fed back to the DRL sensor to cancel the Common Mode oscillations and derive a low-noise EEG reference level for the differential input amplifiers. High-pass (AC coupling) and low-pass analog filters (anti-alias analog filter), significant oversampling at 2048Hz, followed by successive sub-Nyquist digital filtering, 50/60Hz dual notch filtering and down-sampling to the data transmission frequency (128 or 256Hz) in the digital domain in the DSP processor  in the headset prior to transmission. These measures attenuate most extrinsic noise sources to undetectable levels when the headset is correctly filtered and contact impedances are low.

Motion artefacts are minimised by our mechanical design which independently supports each sensor and adjusts to the size and shape of each user.

How EmotivPRO Handles Data

The EEG data in EmotivPRO is recorded exactly as received from the headset. The software doesn’t automatically remove artifacts from muscle or eye movements because data cleaning techniques (like ICA) work better on raw, unfiltered data. However, as outlined above, Emotiv headsets apply carefully crafted signal processing which helps produce clean signals when the headset has good contact, making brainwave data easier to analyze.