Code-Modulated Motion VEP (c-MVEP)

• c-MVEP is a BCI paradigm employing pseudo-random, m-sequence driven motion stimuli to generate broadband visual evoked potentials.
• It uses smoothed binary transitions for continuous motion, serving as a flicker-free alternative to c-VEP and SSVEP paradigms.
• Performance metrics show intermediate accuracy and response times, highlighting potential for further decoder optimization.

Code-modulated motion visual evoked potential (c-MVEP) is a visual brain-computer interfacing (BCI) paradigm in which pseudo-random sequences modulate object motion rather than luminance flicker. In the formulation introduced in "Beyond Flickering: Introducing Code-Modulated Motion Visual Evoked Potentials for Brain-Computer Interfacing," c-MVEP uses an m-sequence-driven radial zooming stimulus and is positioned as a motion-based alternative to code-modulated visual evoked potential (c-VEP), while also being compared with steady-state motion visual evoked potential (SSMVEP) and steady-state visual evoked potential (SSVEP). In offline and online experiments, c-MVEP exhibited time-domain characteristics similar to c-VEP, evoked a broadband response with comparable signal-to-noise ratio (SNR) but more concentrated in the lower frequency range, and achieved intermediate online BCI performance: lower than c-VEP and SSVEP, but higher than SSMVEP (Scheppink et al., 15 May 2026).

1. Stimulus formalization and code design

The c-MVEP stimulus is driven by a pseudo-random sequence generated by a 5-bit linear feedback shift register (LFSR) with primitive polynomial

$p(x) = x^5 + x^2 + 1$

and initial register state $\mathbf r(0)=(1,1,1,1,1)$. The sequence length is $K=2^5-1=31$ bits. The offline prototype sequence is

$c_0 = 0001101110101000010010110011111.$

For the online 4-class BCI, four circular shifts of $c_0$ by 1, 3, 5, and 8 bits were used as the four class codes. The presentation rate is $20\,\mathrm{Hz}$, so one bit lasts 18 frames on a $360\,\mathrm{Hz}$ display, with $F_b=18$ (Scheppink et al., 15 May 2026).

Binary transitions are smoothed to produce a continuous motion driver. The transition window length is $L=\alpha F_b$ with $\alpha=1$. A rising edge $\mathbf r(0)=(1,1,1,1,1)$0 is replaced by

$\mathbf r(0)=(1,1,1,1,1)$1

and falling edges are replaced by $\mathbf r(0)=(1,1,1,1,1)$2. The resulting continuous sequence in $\mathbf r(0)=(1,1,1,1,1)$3 drives radial zooming between 50% and 100% of the original stimulus size.

The temporal structure is explicitly cycle-based. One m-sequence cycle lasts $\mathbf r(0)=(1,1,1,1,1)$4. Offline stimulation uses 3 cycles, giving $\mathbf r(0)=(1,1,1,1,1)$5 per trial, with an inter-trial interval of $\mathbf r(0)=(1,1,1,1,1)$6. Online calibration uses the same $\mathbf r(0)=(1,1,1,1,1)$7 duration per trial, whereas testing uses dynamic stopping. This design places c-MVEP within the code-modulated VEP family while replacing flicker with motion as the stimulation channel.

2. Electrophysiological characteristics

Offline EEG preprocessing consists of a notch filter at $\mathbf r(0)=(1,1,1,1,1)$8, a bandpass of $\mathbf r(0)=(1,1,1,1,1)$9–$K=2^5-1=31$0 using a zero-phase 4th-order Butterworth filter, baseline correction, and epoching. For time-domain analysis, segments are cut into non-overlapping cycles $K=2^5-1=31$1, yielding $K=2^5-1=31$2. The cycle duration is $K=2^5-1=31$3 for c-MVEP and c-VEP, and $K=2^5-1=31$4 for SSMVEP and SSVEP (Scheppink et al., 15 May 2026).

At Oz, the grand-average ERP for c-MVEP shows a pronounced negative peak at approximately $K=2^5-1=31$5. The c-VEP waveform differs subtly but shares broad features. By contrast, SSMVEP and SSVEP show clear sinusoidal ERPs. In frequency-domain analysis, the power spectral density (PSD) is computed using a Hamming-windowed FFT zero-padded to $K=2^5-1=31$6, giving $K=2^5-1=31$7. For c-MVEP and c-VEP, $K=2^5-1=31$8 is defined as the median of $K=2^5-1=31$9 bins excluding the $c_0 = 0001101110101000010010110011111.$0 adjacent bins; for SSMVEP and SSVEP, it is the median of $c_0 = 0001101110101000010010110011111.$1 bins excluding the $c_0 = 0001101110101000010010110011111.$2 adjacent bins.

The main Oz peaks distinguish the paradigms. For c-MVEP, the peak at $c_0 = 0001101110101000010010110011111.$3 is $c_0 = 0001101110101000010010110011111.$4; for c-VEP, the corresponding peak is $c_0 = 0001101110101000010010110011111.$5. For SSMVEP, the fundamental at $c_0 = 0001101110101000010010110011111.$6 is $c_0 = 0001101110101000010010110011111.$7, with harmonics at $c_0 = 0001101110101000010010110011111.$8 of $c_0 = 0001101110101000010010110011111.$9 and $c_0$0 of $c_0$1. For SSVEP, the fundamental at $c_0$2 is $c_0$3, with harmonics at $c_0$4 of $c_0$5 and $c_0$6 of $c_0$7. c-MVEP and c-VEP therefore evoke broadband spectra, whereas SSMVEP and SSVEP evoke narrowband peaks at the stimulation frequency and harmonics.

Spatially, c-MVEP has peak SNR at Oz but spreads to PO$c_0$8, Pz, O2, and PO$c_0$9. c-VEP is highly focal at Oz. SSMVEP is broader over temporal-occipital areas, whereas SSVEP is confined to occipital sites. The study states that the c-MVEP pattern suggests engagement of motion-sensitive areas. A plausible implication is that c-MVEP recruits a wider cortical network than flicker-based code modulation, although the data as presented are spatial-topographic rather than source-resolved.

3. Detection and classification framework

The online decoder uses template-matched canonical correlation analysis (CCA). For each class, the template is defined as

$20\,\mathrm{Hz}$0

Trials are stacked as $20\,\mathrm{Hz}$1 and the corresponding templates as $20\,\mathrm{Hz}$2. The optimization is

$20\,\mathrm{Hz}$3

The spatial filters $20\,\mathrm{Hz}$4 and $20\,\mathrm{Hz}$5 are then applied to a new trial $20\,\mathrm{Hz}$6 and each class template $20\,\mathrm{Hz}$7, giving $20\,\mathrm{Hz}$8 and $20\,\mathrm{Hz}$9. Classification is performed by

$360\,\mathrm{Hz}$0

The paper notes that this is equivalent to seeking $360\,\mathrm{Hz}$1, that is, time-domain CCA (Scheppink et al., 15 May 2026).

Dynamic stopping is integrated into online testing. The stopping criterion uses a margin $360\,\mathrm{Hz}$2, with a minimum window of $360\,\mathrm{Hz}$3. The window expands in $360\,\mathrm{Hz}$4 steps and then slides up to a maximum of $360\,\mathrm{Hz}$5 for c-MVEP and c-VEP or $360\,\mathrm{Hz}$6 for SSMVEP and SSVEP, with an overall limit of $360\,\mathrm{Hz}$7. This procedure operationalizes c-MVEP as a practical asynchronous decision process rather than a fixed-window offline decoder.

4. Quantitative BCI performance

Performance is reported in terms of accuracy $360\,\mathrm{Hz}$8, average selection time $360\,\mathrm{Hz}$9, and information transfer rate (ITR) in bits per minute:

$F_b=18$0

where $F_b=18$1 classes and $F_b=18$2 includes the inter-trial interval of approximately $F_b=18$3–$F_b=18$4 (Scheppink et al., 15 May 2026).

In the online 4-class BCI, the average $F_b=18$5 standard error values are:

• c-MVEP: $F_b=18$6, $F_b=18$7, $F_b=18$8
• c-VEP: $F_b=18$9, $L=\alpha F_b$0, $L=\alpha F_b$1
• SSMVEP: $L=\alpha F_b$2, $L=\alpha F_b$3, $L=\alpha F_b$4
• SSVEP: $L=\alpha F_b$5, $L=\alpha F_b$6, $L=\alpha F_b$7

Statistical analysis uses Friedman and Wilcoxon tests with Bonferroni-adjusted $L=\alpha F_b$8. The accuracy ordering is reported as c-VEP $L=\alpha F_b$9 SSVEP $\alpha=1$0 c-MVEP $\alpha=1$1 SSMVEP, with c-MVEP $\alpha=1$2 SSMVEP ($\alpha=1$3) and SSVEP $\alpha=1$4 c-MVEP ($\alpha=1$5). For selection time, the ordering is c-VEP $\alpha=1$6 SSVEP $\alpha=1$7 c-MVEP $\alpha=1$8 SSMVEP. For ITR, the ordering is c-VEP $\alpha=1$9 SSVEP $\mathbf r(0)=(1,1,1,1,1)$00 c-MVEP $\mathbf r(0)=(1,1,1,1,1)$01 SSMVEP.

These results place c-MVEP between the code-modulated flicker paradigm and the steady-state motion paradigm. The specific comparison reported in the abstract is that the c-MVEP BCI reached a mean accuracy of $\mathbf r(0)=(1,1,1,1,1)$02 with an average selection time of $\mathbf r(0)=(1,1,1,1,1)$03, which was significantly lower than c-VEP ($\mathbf r(0)=(1,1,1,1,1)$04; $\mathbf r(0)=(1,1,1,1,1)$05) and SSVEP ($\mathbf r(0)=(1,1,1,1,1)$06; $\mathbf r(0)=(1,1,1,1,1)$07), but significantly higher than SSMVEP ($\mathbf r(0)=(1,1,1,1,1)$08; $\mathbf r(0)=(1,1,1,1,1)$09).

5. Subjective assessment and user preference

The study includes both offline and online subjective evaluations. In the offline questionnaire, 9 subjects rated the paradigms on a 6-point Likert scale. No clear preference was found for motion-based stimulation, represented by c-MVEP and SSMVEP, over flicker-based stimulation, represented by c-VEP and SSVEP, across comfort, concentration, disturbance, focus loss, likability, and overall rating. The highest overall rating was reported for c-VEP at $\mathbf r(0)=(1,1,1,1,1)$10, followed by SSVEP at $\mathbf r(0)=(1,1,1,1,1)$11, c-MVEP at $\mathbf r(0)=(1,1,1,1,1)$12, and SSMVEP at $\mathbf r(0)=(1,1,1,1,1)$13 (Scheppink et al., 15 May 2026).

The online questionnaire involved 19 subjects and 7 questions plus a forced-choice component. No significant differences among conditions were found for fatigue, comfort, concentration, disturbance, focus loss, likability, or overall rating, with all $\mathbf r(0)=(1,1,1,1,1)$14. In the end-of-study forced choices, 11 of 19 preferred flicker for ease of focus, which was not significant; 14 of 19 stated that flicker caused more eye strain, with $\mathbf r(0)=(1,1,1,1,1)$15; and overall 10 of 19 preferred flicker versus 9 of 19 preferring motion, which was also not significant.

A common assumption is that motion-based stimulation necessarily confers a strong comfort advantage over flicker. The reported data do not support that claim at the tested $\mathbf r(0)=(1,1,1,1,1)$16 bit-rate. The study instead reports no clear preference for motion and no significant differences on the main questionnaire dimensions, even though the forced-choice responses suggest a tendency for more eye strain under flicker.

6. Position within VEP-based BCI research

c-MVEP is explicitly characterized as combining the broad-band, high-SNR characteristics of c-VEP with flicker-free motion stimulation. Its time-domain and frequency-domain ERPs closely resemble c-VEP in being broad-band and m-sequence-driven, but the study states that c-MVEP has slightly lower amplitude and greater inter-subject variability (Scheppink et al., 15 May 2026).

The paradigm also occupies a distinct position relative to steady-state motion stimulation. SSMVEP and SSVEP are described by oscillatory responses at the stimulation frequency and harmonics, whereas c-MVEP and c-VEP are code-modulated and broadband. In practical online BCI use, c-MVEP significantly outperforms SSMVEP but remains below c-VEP and SSVEP. This suggests that substituting motion for flicker in a code-modulated framework preserves much of the characteristic response structure of c-VEP while changing both spatial distribution and attainable performance.

The spatial maps are particularly important for classification of the paradigm. c-MVEP peaks at Oz but spreads into PO$\mathbf r(0)=(1,1,1,1,1)$17, Pz, O2, and PO$\mathbf r(0)=(1,1,1,1,1)$18, while c-VEP remains highly focal at Oz; SSMVEP is broader over MT areas, and SSVEP is occipital-only. The study interprets these differences as indicating engagement of a wider cortical network for motion-based stimulation beyond primary occipital regions. A plausible implication is that decoder designs optimized for focal occipital responses may not fully exploit c-MVEP, which aligns with the paper’s suggestion that further gains may be possible through decoder optimization, including reconvolution CCA, and through motion-parameter tuning.

7. Significance, limitations, and prospective directions

Within the reported evidence, c-MVEP is a viable flicker-free alternative for BCI users intolerant to high-contrast flicker. That conclusion rests on several specific findings: c-MVEP successfully elicits broadband code-modulated responses; its online 4-class performance is clearly above SSMVEP; and it provides a motion-based alternative to c-VEP without requiring steady-state oscillatory entrainment (Scheppink et al., 15 May 2026).

At the same time, the limitations are explicit. c-MVEP does not match c-VEP in accuracy, selection time, or ITR, and it also remains below SSVEP in the online comparisons. Subjective data do not show a strong comfort or preference advantage for motion over flicker at the tested parameters. The paradigm’s broader scalp distribution and greater inter-subject variability further indicate that its current implementation may be less optimized than the established flicker-based alternatives.

The study’s concluding implications are therefore measured rather than categorical. c-MVEP is presented as having great potential and as providing a valuable alternative to c-VEP, but not as a superior replacement under the reported conditions. The paper specifically identifies decoder optimization, such as reconvolution CCA, and motion-parameter tuning as directions through which further gains may be obtained.