Tradeoff between User Experience and BCI Classification Accuracy with Frequency Modulated Steady-State Visual Evoked Potentials

Abstract

Steady-state visual evoked potentials (SSVEPs) have been widely employed for the control of brain-computer interfaces (BCIs) because they are very robust, lead to high performance, and allow for a high number of commands. However, such flickering stimuli often also cause user discomfort and fatigue, especially when several light sources are used simultaneously. Different variations of SSVEP driving signals have been proposed to increase user comfort. Here, we investigate the suitability of frequency modulation of a high frequency carrier for SSVEP-BCIs. We compared BCI performance and user experience between frequency modulated (FM) and traditional sinusoidal (SIN) SSVEPs in an offline classification paradigm with four independently flickering light-emitting diodes which were overtly attended (fixated). While classification performance was slightly reduced with the FM stimuli, the user comfort was significantly increased. Comparing the SSVEPs for covert attention to the stimuli (without fixation) was not possible, as no reliable SSVEPs were evoked. Our results reveal that several, simultaneously flickering, light emitting diodes can be used to generate FM-SSVEPs with different frequencies and the resulting occipital electroencephalography (EEG) signals can be classified with high accuracy. While the performance we report could be further improved with adjusted stimuli and algorithms, we argue that the increased comfort is an important result and suggest the use of FM stimuli for future SSVEP-BCI applications.

Use of Sine Shaped High-Frequency Rhythmic Visual Stimuli Patterns for SSVEP Response Analysis and Fatigue Rate Evaluation in Normal Subjects

“…We designed visual stimulus patterns at high-frequency range and obtained similar results (>90%) (Tables 1 , 2 ). Our results for single trials analysis are in the range of 73–74% (Tables 4 , 5 ) and are comparable with (Dreyer et al, 2017 ) that reported 33–87%. Our single trial accuracy rate was also comparable with a recent low frequency study (Zhang et al, 2012b ).…”

“…In these studies, a high frequency carrier was modulated by another frequency and the SSVEP was evoked at the difference of these two frequencies that covered low frequency range for improving the accuracy rate. This method reduced subjective perceptibility as well as fatigue rate; however, the accuracy rate in these studies was lower than the simple stimulus used in recent studies (Dreyer and Herrmann, 2015 ; Dreyer et al, 2017 ).…”

“…In this study, we observed higher but insignificant trend in VAS values over time indicating higher subjective fatigue from beginning to the end of experiment (Supplementary Figures S6 , S7 ). As subjective fatigue was expected to increase over time in our experiment, compared to few other recent SSVEP studies (Diez et al, 2011 , 2013 ; Dreyer and Herrmann, 2015 ; Dreyer et al, 2017 ) we kept acceptable rest time (2 min) to minimize it. However, the aim of our study was not to evaluate fatigue at various time points for each pattern, rather we were more interested in knowing the fatigue rate between simple and rhythmic groups as well as its variation within these two groups.…”

“…Using high frequency range stimuli can reduce the subjective fatigue rate compared to the low frequency (Diez et al, 2011 ). Studies that have examined the fatigue rate included simple high frequency (Diez et al, 2011 , 2013 ) while others have used the advantages of high frequency range as carrier signal (Chang et al, 2014 ; Dreyer and Herrmann, 2015 ; Dreyer et al, 2017 ). This was done to decrease visual perceptibility and consequently decrease the fatigue rate.…”

“…For example, use of high frequency for the reduction of fatigue (Sakurada et al, 2015 ); however, it decreases the BCI performance (Volosyak et al, 2011 ). As it is important to reduce fatigue rate and also have high accuracy rate for a real BCI application, several studies have used high frequency for fatigue reduction and low frequencies for improving the accuracy rate by introducing amplitude and frequency modulated visual stimulation methods (Chang et al, 2014 ; Dreyer and Herrmann, 2015 ; Dreyer et al, 2017 ). In these studies, a high frequency carrier was modulated by another frequency and the SSVEP was evoked at the difference of these two frequencies that covered low frequency range for improving the accuracy rate.…”