A comprehensive signal-analytic examination is made of the detection of cognitive load from the measurement of pupil diameter oscillation. Earlier such computational methods failed to provide a real-time variant of frequency-based approaches and failed to justify various design choices. In this work, the problem is approached from first principles, considering what pupillary oscillations should be expected, both under rest and under (cognitive) load. Then three detection techniques are derived based on the Fast Fourier Transform (FFT), the Discrete Wavelet Transform (DWT), and the Butterworth filter to compute the Low/High Frequency (LF/HF) ratio. The work also covers the minimum signal duration required for each method, the theoretical range of the LF/HF ratio, drawing from physiological and signal processing perspectives, and the minimum time window needed to produce output. It is shown that the Butterworth filter is better suited for real-time implementation than the other two.
Andrew Duchowski (Fri,) studied this question.