2.7. Electrodermal Activity Preprocessing

As discussed above, EDA has been measured by a non-invasive device. Concretely, the E4 Empatica bracelet measures the skin conductance (SC) in the form of EDA signals. These measurements are composed of two signals: a first signal that varies slowly, called the tonic driver or skin conductance level (SCL), and the second that varies rapidly, called the phase driver or skin conductance response (SCR). The SCL signal establishes the base level of the signal, while the SCR is directly associated with the activity of the sweat motor system which, in turn, is directly associated with the parasympathetic nervous system.

Within the process of processing the EDA signals, different phases are crossed during which the signals are transformed. These phases are usually preprocessing, filtering, artefact removal and discrete deconvolution. The preprocessing process is in charge of establishing the segments acquired in each of the phases of the experiment. Then, it is necessary to filter the SC signals to eliminate the artefacts and interference recorded during the acquisition phase. In our case, two different filters have been used: first, a low-pass filter with a 4 Hz cutoff frequency, and second, a Gaussian filter to smooth the signal and attenuate artefacts and noise.

The next step is the deconvolution process to separate the SCR from the SCL signals. This method makes it possible to minimize the effects that race, sex and age contribute to the SC signal. Figure 2 shows an outline of how this process has been performed. As can be seen, it is the SCR driver that can be used to detect the arousal level of the participant. For this sake, the MATLAB library called Ledalab 3.4.9 has been successfully used [47]. Mathematically, the sudomotor nerve activity can be considered a Driver containing a train of impulses that develop over time. This response is integrated in SC and, consequently also in SCR and SCL. The result is represented by a convolution (*) of the driver with the impulse-response function (IRF), which describes the flow of the impulse response over time, as shown in Equation (1).

Flowchart of the deconvolution process

The SC signal is composed of signals SCL and SCR, as shown in Equation (2).

Thus, by deconvolution of Equation (3), the tonic signal driver is obtained as:

At this point the resulting signals can be used in the following process, which is feature extraction and analysis.