Experimental paradigm and general approach for EEG data analysis. (A) Motion detection task. (B) EEG time series were subject to (1) whole trial analysis and (2) moving window analysis.

SSVEP analysis at the whole trial level. (A) Grand average SSVEP at Oz. (B) Fourier spectrum of the data in Figure 2(A). (C) Target amplitude across all electrodes is significantly larger than distractor amplitude at p = 2.6 x 10-4. (D) Topographical distributions of target and distractor amplitude. (E) Correlation between target SSVEP amplitude and task performance (left) and between distractor SSVEP amplitude and task performance (right). Both correlation values are not significant.

MVPA decoding analysis of distractor processing at the whole trial level. (A) Pair-wise decoding accuracies between pleasant vs neutral, unpleasant vs neutral, and pleasant vs unpleasant are 57.86% ±9.86%, 55.14% ±8.17%, and 59.45% ±9.73%, respectively, which are all significantly above chance level of 50% (red dashed line) at p=3.2 x 10-4, p=3.0 x 10-3, and p=3.0 x 10-5). (B) Three-way decoding accuracy is 41.09% ±6.25% which is significantly higher than the chance level of 33% (red dashed line) at p=3.9 x 10-7. (C) Decoding accuracy vs task performance. The correlation of r =-0.0313 (p = 0.8769) is not significant. (D) Distractor decoding accuracy vs distractor SSVEP amplitude. The correlation of r = 0.1531 (p = 0.4458) is not significant.

Temporal dynamics of target and distractor processing. (A) (i): Target amplitude time series from the moving window approach for a representative subject (left) and its Fourier spectrum (right). (A) (ii): The average target amplitude spectrum across 27 subjects. (B) (i): Distractor decoding accuracy time series from the moving window approach for a representative subject (left) and its Fourier spectrum (right). (B) (ii): The average distractor decoding accuracy spectrum across 27 subjects.

Target-distractor competition analysis. (A) Phase polar histogram for the relative phase between target process time series and distractor processing time series (1 Hz). The average relative phase is 0.51π. (B) Kolmogorov-Smirnov test showed that the relative phase distribution is not different from uniform distribution. (C) Temporal relation between target processing and distractor processing for (i) a high performer (accuracy=83.84%; relative phase=0.877π) and (ii) a low performer (accuracy=33.33%; relative phase=0.053π). (D) Task performance vs 1 Hz relative phase. The significant positive correlation (r=0.6041, p=0.0008) means that the more separated the target and distractor sampling within the 1 Hz oscillation cycle the better the behavioral performance. CDF: Cumulative distribution function.

Simulation results. (A) The signal containing a 4.29 Hz component and a 6 Hz component where the 6Hz signal’s magnitude is about half that of the 4.29Hz signal. The amplitude is modulated at 1 Hz. No noise is added. (B) Low level of noise is added to the signal in Figure S1(A) where the SNR = 12.72 dB. Sidebands are still seen. (C) Middle level of noise is added to the signal in Figure S1(A) where the SNR = 5.38 dB. Sidebands become difficult to see. (D) High level of noise is added to the signal in Figure S1(A) where the SNR = 2.24 dB, sidebands become more indistinguishable from the noise floor. Red dots indicate the location of the main frequency components and the locations where the sidebands should appear.

Experimental data. (A) The time course of the SSVEP and its Fourier spectrum from a subject with high SNR. The sidebands can be observed. (B) The time course and its Fourier spectrum from a subject with low SNR. The sidebands are indistinguishable from the noise floor. (C) The averaged Fourier spectrum from 5 highest SNR subjects and 5 lowest SNR subjects. Again, for subjects with high SNR, the sidebands are identifiable, whereas for subjects with low SNR, the sidebands are not identifiable.

Beating frequency analysis. (A) The SSVEP time course from one subject (left), the Fourier spectrum of the amplitude envelope from the subject (middle), and average Fourier spectrum of the amplitude envelops from all subjects (right). In both spectra we observed the spectral peal at the beating frequency 1.71 Hz. Importantly, we also observed a much stronger spectral peak at around 1 Hz which cannot be explained by the beating phenomenon. (B) Simulated signal by combining a 4.29 Hz component and a 6 Hz component (left). The spectral peak corresponding to the beating frequency 1.71 Hz is clearly seen but there is no spectral peak at 1 Hz. This simulation further demonstrated that the 1 Hz spectral peak in (A) does not arise from a pure signal processing perspective.

SSVEP amplitude analysis at the whole trial level. (A) Target amplitude vs distractor amplitude, where the correlation is r = 0.7992 (p = 0.000006), suggesting the 6 Hz signal amplitude is strongly influenced by the 4.29 Hz signal amplitude. (B) Target amplitude vs distractor decoding accuracy, where the correlation is r = 0.0536 (p = 0.7908), suggesting that the decoding accuracy as an index of distractor processing is not influenced by the 4.29 Hz target amplitude.

Moving window analysis. (A) The relative phase between the target amplitude time series and the distractor amplitude time series. (B) Kolmogorov-Smirnov test showed that the relative phase distribution is significantly different from the uniform distribution. (C) Relative phase vs task performance. r=0.1940 (p=0.3322) means that there is no significant correlation between amplitude relative phase and task performance.