The core insight behind c-MVEP lies in how the stimulus is modulated. In a conventional c-VEP system, each target flickers according to a binary maximum length sequence, an m-sequence generated by a linear feedback shift register. These sequences possess exceptional autocorrelation properties: the correlation of the sequence with itself is 1 when there is no shift, and only −1/K for any other shift, where K is the sequence length. This near-perfect distinctness is what allows a classifier to identify which of many simultaneously flickering targets a user is looking at, even when all targets are driven by shifted versions of the same underlying code. In the new paradigm, the researchers took the same 31-bit m-sequence, generated with feedback taps at register positions 5 and 2 corresponding to the primitive polynomial p(x) = x⁵ + x² + 1, but instead of mapping the binary states to black and white luminance, they mapped them to the size of a radially zooming stimulus. A ‘0’ in the code corresponded to the stimulus shrinking to 50 percent of its original size, while a ‘1’ corresponded to the stimulus expanding to its full 100 percent size.

Directly applying a binary square-wave modulation to motion parameters, however, would have produced abrupt, discontinuous transitions between the minimum and maximum stimulus sizes. Such discontinuities could lead to perceptually rough motion and might even be experienced as flickering rather than smooth movement, undermining the entire purpose of the paradigm. To address this, the researchers smoothed the sharp edges of the m-sequence using a raised sine function. The original binary sequence was first upsampled by a factor of 18, corresponding to the number of display frames per bit given the monitor’s 360 Hz refresh rate and the 20 Hz presentation rate of the m-sequence. Each transition between consecutive bits was then identified, and within a transition window equal to one bit duration, the sharp edge was replaced by a smooth sinusoidal function that connected 0 and 1. Rising edges used the function s(t), while falling edges used its reverse, 1−s(t). Outside these transition windows, the sequence remained unchanged, ensuring that the event durations and overall timing of the original m-sequence were preserved. The correlation properties of the m-sequence were minimally altered by this smoothing, as confirmed by the researchers’ supplementary analysis, meaning the smoothed sequence retained the near-orthogonal distinctness that makes code modulation so effective for multi-target decoding.

To fully characterize the new paradigm, the team designed a two-by-two factorial comparison across four conditions: the novel c-MVEP, the established c-VEP, the steady-state motion visual evoked potential (SSMVEP), and the steady-state visual evoked potential (SSVEP). This design allowed two orthogonal comparisons. First, comparing motion to flickering, specifically c-MVEP against c-VEP and SSMVEP against SSVEP, assessed how replacing flicker with movement affects comfort, signal characteristics, and performance. Second, comparing code-modulated to steady-state stimulation, specifically c-MVEP against SSMVEP and c-VEP against SSVEP, tested whether the superior decoding performance typically observed for c-VEP also applies to motion-based interfaces. The offline experiment, conducted with nine participants, presented a single stimulus in the center of a screen while EEG was recorded from sixteen gel-based passive scalp electrodes positioned over the parieto-occipital cortex according to the international 10-5 system. The reference and ground electrodes were placed at Cz and AFz respectively, and signals were sampled at 600 Hz with impedance reduced below 5 kΩ using abrasive electrode gel.

The offline results revealed that c-MVEP evoked temporal and broadband spectral responses remarkably similar to those of c-VEP, with a comparable signal-to-noise ratio (SNR). Neither code-modulated condition showed frequency-specific oscillations, which was expected given their non-periodic stimulation structure. Instead, the energy was distributed broadly across frequency bands, with a peak at the 20 Hz presentation rate observed in both conditions. Descriptively, the mean SNR at 20 Hz was 2.11 ± 0.52 dB for c-MVEP and 3.06 ± 1.20 dB for c-VEP, indicating that the motion-based code modulation produced a response of similar strength to the flicker-based original. The c-MVEP response, however, was more focused in the lower frequency range, whereas c-VEP energy spread more beyond 10 Hz. Spatially, both motion-based stimulations peaked at the Oz electrode but spread across multiple neighboring electrodes, whereas both flicker-based stimulations were more localized at Oz. This broader spatial distribution for motion-based paradigms is consistent with prior findings that SSMVEP activation extends toward middle temporal vision areas, while SSVEP remains concentrated in occipital regions.

For the frequency-based conditions, the offline analysis showed clear harmonic oscillations for both SSMVEP and SSVEP. The SSMVEP response was characterized by distinct harmonics at 10 Hz (SNR 2.01 ± 0.94 dB) and 15 Hz (SNR 1.15 ± 0.73 dB), while SSVEP produced stronger and more distinct harmonics at 20 Hz (SNR 10.54 ± 1.81 dB) and 30 Hz (SNR 6.86 ± 1.20 dB). The overall SSVEP SNR of 8.75 ± 2.14 dB substantially exceeded the SSMVEP SNR of 2.46 ± 0.98 dB, confirming that flicker-based frequency stimulation produces a stronger neural response than motion-based frequency stimulation. This asymmetry between the two frequency-based paradigms foreshadowed the performance gap that would emerge in the online experiment, where the same pattern of flicker outperforming motion was observed across all decoding metrics.

The online experiment, conducted with nineteen participants after one was excluded for task disengagement, evaluated a four-target BCI using all four stimulation paradigms in a copy-selection task. Participants were instructed to select numbers 1 to 4 according to randomized 12-digit sequences displayed at the top of the screen, with the next digit to be selected highlighted in bold and blue. All conditions used the same template-matching canonical correlation analysis (CCA) classifier, built from user-specific calibration data collected in a training block of 24 trials per condition. The researchers deliberately chose this unified decoding framework across all conditions, rather than employing paradigm-specific state-of-the-art methods, to ensure that observed performance differences reflected the stimulation paradigm rather than the decoder. A dynamic stopping approach was implemented, accepting a classification only when the margin between the maximum correlation and the runner-up exceeded a confidence threshold of 0.3, tested every 0.05 seconds into the trial. The minimum classification window was 0.75 seconds, with an expanding and sliding window approach that extended the window in 0.05-second increments until the threshold was met or the maximum trial duration of 9.3 seconds was reached.

The online results demonstrated that c-MVEP significantly outperformed SSMVEP across all metrics. The c-MVEP BCI reached a mean accuracy of 85.67 ± 14.41 percent with an average selection time of 2.61 ± 1.09 seconds and an information transfer rate (ITR) of 20.11 ± 10.71 bits per minute. In contrast, SSMVEP achieved only 64.91 ± 23.21 percent accuracy with a selection time of 4.18 ± 1.62 seconds and an ITR of 10.13 ± 10.53 bits per minute. Post-hoc Wilcoxon signed-rank tests with Bonferroni correction confirmed that these differences were statistically significant for accuracy, selection time, and ITR (all p < 0.001). However, both flicker-based paradigms outperformed the motion-based ones. c-VEP achieved the highest accuracy at 97.81 ± 3.65 percent with a selection time of just 1.15 ± 0.37 seconds and an ITR of 35.56 ± 5.93 bits per minute, while SSVEP reached 93.42 ± 9.18 percent accuracy with 1.94 ± 1.20 seconds and 28.12 ± 9.78 bits per minute. Notably, when the researchers computed subject-wise motion effect scores, comparing the accuracy reduction from adding motion to each paradigm, they found that the reduction was significantly larger for SSVEP (28.51 ± 24.04 percent) than for c-VEP (12.13 ± 15.08 percent, p = 0.031), indicating that the code-modulated paradigm was more robust against the addition of motion than the frequency-based paradigm.

Perhaps the most striking finding of the study concerned subjective comfort. Contrary to the researchers’ hypothesis that motion-based stimulation would be more comfortable than flickering, the questionnaires revealed no clear preference between the motion- and flicker-based paradigms. In the offline experiment, no significant differences were found between conditions for eye discomfort, ease of concentration, disturbance, loss of focus, system likability, or overall impression. In the online experiment, a significant difference emerged only for disturbance (χ² = 8.475, p = 0.037), but post-hoc pairwise comparisons with Bonferroni correction revealed no significant differences between specific pairs of conditions. In the final forced-choice questionnaire, participants were evenly split between zooming and flickering for ease of focus (8 zooming, 11 flickering), eye strain (5 zooming, 14 flickering), and overall preference (9 zooming, 10 flickering), with none reaching statistical significance. This finding was particularly notable given that the study used a presentation rate of 20 Hz for c-MVEP and c-VEP, rather than the commonly used 60 Hz, and prior work has shown that c-VEP flickering at higher presentation rates is perceived as less fatiguing. The lower rate used here would have been expected to make the motion stimulation preferable, yet no such preference emerged.

The researchers acknowledge several limitations and avenues for future work. The current implementation used a 50 percent zooming amplitude, chosen because too large an amplitude could be perceived as uncomfortable and might produce less smooth stimulation given the limited number of frames available, while too small an amplitude might not evoke a strong enough response. The stimulus size also matters for the evoked response, creating a disadvantage for c-MVEP and SSMVEP compared to c-VEP and SSVEP, since the latter two were always presented at 100 percent size. The researchers suggest that centering the zoom around the size of the flickering stimuli, for instance zooming between 75 and 125 percent rather than between 50 and 100 percent, would make the size component more equitable. They also propose that moving textures on the stimulus rather than the full stimulus itself could provide comfort benefits that the current implementation did not reveal. The decoding approach, a basic template-matching CCA classifier not optimized for any specific paradigm, could be improved by adapting reconvolution-based CCA methods from the c-VEP domain or task-related component analysis from the SSMVEP domain. The offline sample size of nine was small and exploratory, and the four-target layout with horizontally neighboring stimuli may not generalize to practical applications with higher target counts and denser layouts. Nonetheless, the study demonstrates that c-MVEP provides a viable, effective, non-flickering alternative to traditional c-VEP and SSVEP, particularly for users who prefer motion over flickering or for whom flickering stimuli are not suitable. By showing that code modulation can be carried by smooth motion without sacrificing the reliability that makes c-VEP so robust, the researchers have taken the first step toward a flicker-free c-VEP brain-computer interface, one that could be tailored to individual users and provide competitive performance when conventional flickering paradigms do not work or feel less comfortable for specific individuals.

Subject of Research: Neuroscience

Article Title: Beyond flickering: introducing code-modulated motion visual evoked potentials for brain-computer interfacing

Article References: Scheppink, H. A., Herpers, R., Thielen, J., & Volosyak, I. (2026). Beyond flickering: introducing code-modulated motion visual evoked potentials for brain-computer interfacing. Frontiers in Neuroergonomics, 7, Article 1884144. https://doi.org/10.3389/fnrgo.2026.1884144

Image Credits: AI Generated

DOI: 10.3389/fnrgo.2026.1884144

Keywords: brain-computer interface, c-MVEP, c-VEP, SSVEP, SSMVEP, EEG, visual evoked potential, neuroergonomics, Beyond, flickering, introducing, code-modulated