A step-by-step guide to include key electroencephalography (EEG) parameters in the study of human performance applied to air traffic control

The study of human performance of air traffic controllers (ATCOs) is an interesting line of research to improve operational safety. In recent years, there has been an increase in the number of techniques available to develop this research based on massive data analysis. This study presents the use of certain electroencephalography (EEG) parameters to study the human performance of ATCOs. Although software and applications are now available to calculate these parameters, there are often problems in understanding the detailed process used to calculate them. The parameters presented in this study are intended to overcome this limitation and are applicable in real air traffic control (ATC) situations. Six parameters are analysed: excitement, stress, relaxation, boredom, engagement, and attention. As an application case for the parameters, a total of 50 data samples obtained during the development of real-time simulations on a highly realistic ATC platform are analysed. From these data, the above-mentioned EEG parameters and their trends are calculated. In addition, the evolution of these parameters is studied in relation to two other variables that characterise the operational situation of the sector during the simulations: the taskload based on ATC events and the number of simultaneous aircraft in the sector per minute.


Introduction
Air traffic controllers (ATCOs) play a key role in air traffic management (ATM).For this reason, several studies have been conducted over the years to evaluate their performance.Thanks to advances in data analysis, it is now possible to use a variety of techniques based on the analysis of neurophysiological data reflecting the state of ATCOs.Some examples of these techniques are the study of brain activity, eye-tracking, heart rate variability, or galvanic skin response.This study will use the electroencephalography (EEG) technique to study brain activity.By looking at electrical signals, EEG techniques are used to identify normal and abnormal brain activity [1].
This technique enables the activity of groups of neurons in specific regions to be detected.To do this, electrodes are placed on the scalp of the subject.These electrodes are placed in specific positions determined by the so-called International 10-20 EEG system.This system is commonly used in EEG to correlate the locations on the outside of the scalp with the underlying cortical areas [2].To know its exact position, each electrode is identified with an alphanumeric code.Originally, these techniques were very complex and only available in the medical field to study the brain and detect brain pathologies.However, a wide range of wireless devices are now available that can be used in real-life operational situations.One of the major advantages of these devices is the continuous recording of data at regular intervals, without interrupting the free movement of the participants.As a result, they have been the choice for use in a wide range of research disciplines.
Focusing on aviation, interesting studies have been carried out using EEG techniques in various fields.For example, the study of microsleeps in commercial pilots was the focus of [3].In this study, a method for classifying microsleeps was proposed using different EEG-derived features.In [4], EEG data and machine learning techniques were used to discriminate the workload of a pilot during a flight.The results demonstrated the potential of these techniques to discriminate between periods of low, medium, and high workload.In the field of air traffic control (ATC), [5] presents how an EEG-based classifier can be used to trigger different adaptive automation tools on the ATCO's radar screen, depending on the level of workload detected.In [6], an EEG-based situational awareness (SA) detection algorithm for ATCOs is developed.The results showed that EEG is a promising approach for real-time and accurate measurement of SA in ATCOs.
The examples shown above demonstrate that EEG techniques are of great interest for the study of human performance of various aviation professionals.The recorded data are continuous, and the frequency of recording is very high.The availability of these techniques makes it possible to analyse the recorded data without interfering with the main task The use of techniques such as EEG enables a more accurate understanding of the range of human emotions [7].This is a major advantage.
This study focuses on the human performance of air traffic controllers.It uses data recorded during 50 samples of real-time simulation exercises.In these exercises, participants were responsible for the control of one or more sectors during the en-route phase of flight.A total of 16 participants were included in the study.Its main objective is to establish a set of EEG parameters that can describe the state of the ATCO during the development of the simulations and that are related to variables that characterise the operational situation of the sector.
The remainder of this paper is structured as follows.Section 2 presents the motivation for the study and the development of the experiment.Subsequently, Section 3 summarises the steps followed in the methodology.Section 4 presents the six final selected EEG parameters and describes the software developed to calculate them.Section 5 explains the variables considered in the study.Section 6 then presents the process followed to analyse the parameters.This includes both graphical and numerical tools.Finally, Section 7 presents the main conclusions of the study and some ideas for future research.

Description of the study
This section presents the key ideas that led to the study.First, the motivation for the study is explained considering previous work.This is followed by a summary of the EEG equipment and EEG data.

Motivation of the study
As mentioned above, EEG technology has developed significantly in recent years.As a result, there are many companies, software, and applications designed to record and analyse this data.The use of these tools developed by companies is of great interest in the initial stages of the study of EEG parameters and also for those professionals who work with this type of data for the first time.
However, to develop a more exhaustive study, explain the parameters obtained and follow the process of calculating the parameters, they have a number of limitations.This study aims to overcome these limitations.
One of the main limitations of commercial software is that it often provides proprietary EEG parameters calculated from raw data using different algorithms.The problem is that the algorithms used by the manufacturers are not easily traceable.On the other hand, it is difficult to obtain a detailed definition of the parameters provided by the software.Although it is true that many manuals provide examples of parameter interpretation, it can be complex to extrapolate such a definition to the ATC domain.
In many cases, not only is the algorithm used to calculate the parameters unknown but also the process by which these parameters are calculated.Therefore, if additional variables are to be introduced into the calculation process, there is not enough information to do so.In summary, it is difficult to interpret the values obtained from the parameters predefined by the manufacturers, and therefore, to increase their use in real ATC operational conditions.
To overcome all the limitations mentioned above, the aim of this study is to select a set of EEG parameters and demonstrate its applicability in real ATC situations.The whole process will be explained, including the selection of the parameters, the development of the software to calculate them, the detailed analysis of the parameters and the results obtained.To this end, it is essential to explain in a clear and simple manner each of the steps taken to succeed in the application of these parameters in real ATC situations.

Development of the experiment
To develop this study, real-time simulations were conducted on an ATC control platform that replicates the radar screen of en-route ATCOs.The exercises were designed to include a series of defined ATC events.
Each participant completed three simulation sessions.The first was an initial training session and a series of familiarisation exercises with the platform.The next two were data collection sessions.In each session, the participant simulated two exercises with a one-hour break between.The simulation sessions were separated by one week according to the designed calendar.The simulation exercises were divided into four different categories according to their level of difficulty.The participants always simulated the exercises in increasing order of difficulty.
All participants were ATCO students, and the sample of participants was selected based on their performance in previous exercises and tests throughout their training.The sample of participants was chosen to be as homogeneous as possible.The participants had an average age of 21 years, had all received the same ATCO training, and had already some knowledge of the operation of an ATC position and general concepts of airspace organisation and techniques for conflict detection and resolution.Although they had completed practical exercises during their training, this experiment was their first contact with the SkySim simulation platform used in the study.Therefore, they received documentation and training on the platform and performed trial exercises prior to the simulation campaign.
During the exercises, raw EEG data was recorded using the EmotivPro software.Although this software enables the calculation of its own parameters, called performance metrics, they will not be used here.The aim of this study is to use the raw data to define other EEG parameters that are easier to define and whose calculation process is known.The Emotiv Insight headset was used to record the data.This is a wireless headset with a total of five electrodes.AF3 and AF4 (in the frontal area), T7 and T8 (in the temporal lobe), and Pz (in the parietal lobe).All electrodes were placed according to the 10-20 system.Potential differences were measured with respect to the two reference electrodes CMS and DRL on the left mastoid.The sampling rate of the headset is 128 Hz, that is, 128 samples per second.

Methodology
The aim of this section is to present the methodology followed in the study and the main findings from each stage.Figure 1 shows a schematic diagram of these steps.The starting point was to define the objective of the study to be conducted.As mentioned in the previous section, in this case the aim is to validate the use of a set of EEG parameters that can be applied in real ATC operational situations.
The purpose of this study is to introduce the use of EEG parameters to ATC professionals who do not necessarily have previous experience working with this type of data.Therefore, these parameters should be derived from models with validation from previous studies.To this end, a systematic review of previous studies and publications was conducted, both within the ATC domain and in other disciplines.As a result of this research, a total of six EEG parameters were selected: excitement, stress, relaxation, boredom, engagement, and attention.
The next step was to develop software to automate the process of calculating these parameters using a series of formulae and transformations.Since it is necessary to work with Fourier transforms to move from the frequency domain of the signals to the time domain, it was decided to work with the MATLAB software.
To validate the feasibility of the selected EEG parameters and the developed software, the above was applied to data from a total of 50 simulation exercise samples using a real-time simulation platform.From the raw data recorded during the development of the exercises, their EEG parameters could be calculated.Subsequently, in the analysis phase, the behaviour of these parameters was analysed.Furthermore, their evolution was compared with that of two other variables that characterise the operational situation of the ATC sector during the exercise.These two variables are the taskload based on ATC events and the number of simultaneous aircraft in the sector.
Based on the conclusions obtained from the above analysis, a complete proposal is made on the possibility of using these EEG parameters in real ATC situations.Furthermore, based on the results obtained in this study, a number of future research lines are identified to further progress in the use of EEG parameters for the study of human performance of ATCOs.

EEG parameters considered in the study
This section presents the six parameters selected from the literature review.These six parameters were obtained from three different models: arousal -valence, engagement index, and R parameter.
The arousal-valence model is the most commonly used model to classify emotions using EEG.The model is based on the arousal and valence parameters.The valence parameter ranges from unpleasant to pleasant, and the arousal parameter ranges from relaxation to excitement [8].If each of these parameters is positioned as the axes of a coordinate system, it is possible to represent the position of different emotional states on a circle.This diagram, known as Russell's two-dimensional affective model, and the formulae used to calculate the two parameters above, can be seen in Figure 2. Depending on the sign of the two parameters, four different quadrants can be identified.In this study, a different parameter was used as the reference for each of these quadrants.Starting from the quadrant of positive arousal and positive valence and going counterclockwise, the four parameters are: excitement, stress, boredom, and relaxation.
To obtain these parameters from the raw data, it is first necessary to calculate the Energy Spectral Density (ESD) values that appear in the formulae.This parameter describes how the energy is distributed along the different frequency bands that compose the signal.
The next model used is the engagement index, called engagement in this study.It measures the engagement or involvement in the task of the participant at a given point in time.This parameter has been used in previous studies to differentiate high-intensity events from normal task performance [9].Its formula is shown in Figure 3.
The last parameter to be considered is attention.It is obtained by dividing the ESD associated with the alpha waves by the ESD associated with the beta waves.Beta waves occur when the brain is active.Therefore, the lower the R parameter, the more dominant the beta waves and the more active the brain is [10].

Software designed to automate the process of calculating the parameters
As can be seen in Figure 2 and Figure 3, the calculation of the parameters requires the calculation of several ESD values.As a preliminary step in calculating these values, it was necessary to calculate the power spectral density (PSD).The ESD is calculated by integrating the PSD.A MATLAB code was developed to automate the calculation of these parameters.In this case, this software was chosen because it has a spectrogram function that allows the PSD value of the signals to be calculated.
The code has been automated in such a way that the user only has to import the data and specify the minutes between which the recording is to be analysed.The software performs all the calculations and outputs a table with the EEG parameters.The number of data calculated depends on certain parameters specified in the code.In this case, to have the same number of EEG parameters as the independent variables in each exercise, the EEG parameters were calculated for each minute.

Variables to be considered in the study
The variables considered in the study are divided into two groups: dependent and independent variables.The independent variables are those used to characterise the operational situation of the ATC sector during the simulation exercises.On the other hand, the dependent variables are the six EEG parameters mentioned in the previous section.
The first of the independent variables is the taskload, which represents the task demand of the ATCO.This parameter is defined per minute as a function of the ATC events that occur during the simulation.Each of the events has a taskload score and an average duration.The events considered are identification, takeover, handover, speed change, flight level change, re-routing, cruise-cruise conflict, and overtaking conflict.
The taskload parameter is of great interest in the study, as it considers the different behaviour of each participant and their cognitive processes in relation to air traffic management in their sector of responsibility.Once the simulation campaign had been completed, a detailed process was carried out to define the taskload distribution profiles per participant and exercise.This data has been used as an independent variable in the study of each exercise simulated by each participant.The second independent variable is the number of simultaneous aircraft in the sector per minute.This value could be calculated considering that each time the identification event occurs, a new aircraft appears.In contrast, when the handover event occurs, an aircraft leaves the sector.As each exercise had a duration of 45 minutes, there are 45 data per record for the independent variables.The EEG parameters were calculated to have a database of the same dimension.

Analysis and results
The analysis of the independent and dependent variables was carried out in two stages: one using graphical representations and the other using numerical analysis.This section aims to summarise the main results obtained using both approaches.
Before developing the analysis, a pre-processing and filtering of the data was carried out.The quality of electrode contact during the recordings was used as a variable.This enabled the filtering of those recordings with better contact quality.A total of 63 recordings were collected during the simulation campaign.After the filtering process, the 50 recordings used in the analysis were obtained.

Graphical analysis
The graphical analysis included the study of the change in the magnitude of the EEG parameters, as well as their evolution compared to the evolution of the taskload parameter.At this stage, an exploratory data analysis of the EEG parameters was also performed, and they were analysed using descriptive statistics.Figure 4 shows an example of the graphs analysed.In this case, two graphs are presented for the data of Participant 6 and Exercise 3. p On the left is a boxplot of the six EEG parameters and, on the right, a representation of the evolution of the excitement parameter together with the taskload (both variables expressed as the variation of their value between one minute and the previous one).
Analysis of outliers is always an important step in the study.In this case, a detailed analysis was carried out to assess whether the outliers identified in the boxplots were a consequence of data collection, or whether they were part of the study data in response to a situation in the ATC sector.In all cases, it was found that the outliers were a consequence of situations in response to various ATC events.Therefore, it was concluded that these data should be included in the numerical study as additional values in the sample to be analysed.
Several key findings emerged from this graphical analysis.The scatter plot of the data was found to follow the typical pattern of an experiment.In terms of outliers, the boxplots showed that the EEG parameters had either no or very few outliers.It was also possible to establish graphical relationships between the evolution of the parameters and the taskload.As a result, it was confirmed that the behaviour of each of the EEG parameters was different, justifying the inclusion of each of them in the study.

Numerical analysis
To complement the graphical analysis, a numerical analysis was also carried out.In this case, the tools used were regressions and correlations and the use of the ANOVA test.These tools were used with both the taskload parameter and the number of simultaneous aircraft in the sector.
The results of the regressions allowed the correlation between the different EEG parameters with the taskload or the number of aircraft per participant and exercise to be obtained.Figure 5 shows an example of these results.Each of the tables shows the correlations obtained with the simultaneous number of aircraft for each participant and exercise included in the study.The dependent variable in the left table is attention, whereas the variable to be analysed in the right table is engagement.The conclusions drawn from these tables can be generalised to the rest of the analysis.The best regression results are obtained for the engagement and attention parameters.An ANOVA test was performed to examine the significance of the changes in the magnitude of the parameters.As in the previous case, it was possible to confirm that the results were sufficiently significant in the case of attention and engagement parameters.A point should be made regarding the use of the ANOVA test.In general, one of the conditions for the application of the ANOVA test is the condition of normality.However, recent studies have shown that the ANOVA test is robust to nonnormality of the data.Some of them impose as an additional condition that the sample sizes are equal or very similar (a feature that is met in this study).Meanwhile, other studies, for example [11], have performed various tests showing that the test is robust even if this condition is not met.Based on these findings, the ANOVA test was used in this study because it has the advantage of being a very widely used test and the results are easy to explain.
From the results obtained, it can be concluded that all the EEG parameters presented are of interest.However, based on the results of the correlations, it is proposed to use attention and engagement to establish direct relationships with taskload and the number of simultaneous aircraft, and to use the other four as indicators of the ATCO state using the arousal-valence model.

Conclusions and future work
The study has demonstrated the relevance of the six presented EEG parameters and established a complete methodology for their calculation and application, analysing their behaviour both individually and in relation to the parameters of taskload and the simultaneous number of aircraft in the sector.The literature review process has allowed the selection of six EEG parameters to be used in the study of human performance in real ATC operational situations.
In addition, a MATLAB code was developed to automate the process of calculating these EEG parameters from the raw data recorded during the development of the simulations.
As a conclusion of the analysis, it is proposed to use the engagement and attention parameters for the regression analysis (and possible future predictions) and to use the parameters derived from the arousal-valence model as an indicator of the current EEG state of the ATCO.
Finally, the results obtained have demonstrated the great potential of using these parameters to study the human performance of ATCOs in real operational environments.
As future steps, it would be interesting to extend the study to other simulation exercises and to new participants.As this study was the first stage in the use of these parameters, ATCO students were used to carry out the study with as homogeneous a sample of participants as possible.The results obtained may vary if the experiment is repeated with ATCOs in service or with different levels of experience.For this reason, future work is already underway to repeat the simulation campaign to compare the study results obtained with different participant samples.
It would also be interesting to repeat the regression analysis and the application of the ANOVA test directly on the arousal and valence parameters, to analyse whether different conclusions are reached.

Figure 1 .
Figure 1.Methodology followed in the study.

Figure 2 .
Figure 2. EEG parameters from the arousal-valence model considered in the study.

Figure 3 .
Figure 3. Models and calculations to obtain the EEG parameters of engagement and attention.

Figure 4 .
Figure 4. Examples of graphs obtained as part of the graphical analysis for Exercise 3 of Participant 6: boxplot of the EEG parameters (left) and evolution of parameter and taskload (right).

Figure 5 .
Figure 5. Examples of regression results: between attention (left) and engagement (right) and the number of simultaneous aircraft in the sector per minute.