Interpreting Signals

During amplification, electrodes work together with one another to record and compare the very faint signals that the brain produces. One way they compare signals is by using a common reference. When an electrode measures activity, it compares the difference of activity with a common reference electrode to calculate and amplify the intensity of the signal. The activity of the common reference electrode may affect how the electrodes calculate the difference and therefore the intensity. The BCI can re-reference with another reference electrode to mitigate this problem. There is also the bipolar method, where electrodes are amplified together. This allows the BCI to locate high activity areas better than just one electrode. The most popular and consistent approach is by using a common average reference, where activity from all of the electrodes are measured and averaged, and activity from the target electrode is compared and amplified with that.

During preprocessing the BCI prepares the recordings even more to be interpreted.The recordings are filtered several times. The BCI may only look for frequencies in a certain range, cutting off frequencies higher or lower than that range. It also needs to filter out signals known as ‘artifacts,’ which are brain signals produced during events like motor movement and blinking. It also needs to filter out any other external and environmental sources that may have affected the BCI recording.

In this step, the BCI attempts to extract information out of the recording by breaking it down. Two examples are the Fourier Transform and Wavelet Transform.

​

The Fourier Transform is able to change recordings from the time domain to the frequency domain. This allows the BCI to see what frequencies are within a certain wave pattern. The Fourier Transform has its own variations. Whereas regular Fourier Transform calculates for a theoretically continuous and infinite interval, Discrete Fourier Transform calculates for a limited interval. Short-Time Fourier Transform aims to make up for the loss of the time domain during Fourier Transform by converting brain waves in chunks of certain time intervals, and performing a Fourier Transform for each and recording that data alongside the time interval. The Fast Fourier Transform aims to lessen the amount of calculations needed by exploiting the symmetric properties of waves. It decreases the time from n^2 to n(log(n)).

The Wavelet Transform runs through a graph using a base wavelet, which is a wave of a certain shape. The wavelet runs through the graph over and over again, changing its frequency each time. In the Discrete Wavelet Transform, the graph gets decomposed into its high resolution and low resolution ranges. The lower range chart once again gets split into its high resolution and low resolution ranges. This repeats for a certain amount of levels, creating a decomposition tree. In a wavelet packet tree, both the high and low resolution charts get split. Higher ranges tend to have more detail and lower ranges tend to have smoother detail.

In the classification step, the BCI finds patterns in the brain waves that are associated with various functions. These are called event-related potentials. An example is the p300 wave. The p300 wave is associated with an identification of a specific stimulus within a general pool. It is called that because the peak of the wave comes approximately 300 milliseconds after the stimulus. Each person has unique brain waves, so a person would have to undergo multiple trials. There are various methods that the BCI can use to classify and identify waves. An example is a convolutional neural network (CNN). This is a form of AI deep learning that utilizes many convoluted and pooling layers to calculate the likelihood of an image being a certain thing.