Code-modulated visual evoked potentials (c-VEPs) are established as fast and reliable control signals for reactive brain-computer interfaces (rBCIs). The authors identify three main practical limitations to deployment beyond controlled laboratories: visually intrusive stimulation, dependence on wet electroencephalography (EEG) systems, and unintended activations during non-control or idle periods. To address these issues, they propose a usability-oriented c-VEP rBCI that integrates four design choices: low-salience textured visual stimulation, slow flashing codes, an asynchronous evidence-accumulation decision policy to reduce spurious commands, and semi-dry EEG acquisition to improve wearability and practicality.
Thirty participants evaluated the proposed system. The primary evaluation used a standard-screen 11-class keypad typing task performed in two identical online sessions. Between these two online typing sessions, participants watched a two-minute video that served as an explicit idle condition to probe false-activation rates under non-control behavior. In addition to the screen-based task, feasibility was tested in a virtual reality (VR) proof-of-concept: a shell-game task designed to introduce head movements and immersive rendering constraints representative of realistic VR use.
The system combines textured, low-salience visual stimuli with code modulation to evoke discriminable VEP responses while aiming to reduce peripheral distraction and improve visual comfort. The authors used a semi-dry EEG setup rather than traditional wet electrodes to support more practical and wearable acquisition. The decision-making architecture used an asynchronous evidence accumulation approach intended to be conservative during idle periods to avoid false activations.
Calibration requirements were kept minimal: the system required 33 seconds of calibration data before online operation. Online decoding relied on the calibrated responses combined with the asynchronous evidence-accumulation mechanism to determine selections in the 11-class keypad and in the VR task.
On the standard screen keypad typing task the system achieved a high level of performance. With 33 seconds of calibration data, the reported online accuracy was 99.8%. Mean decoding time for selections in this condition was 4.3 seconds. These metrics reflect the combined effects of the visual stimulation, semi-dry EEG acquisition, and the online decoding strategy used by the authors.
To probe conservative idle behavior, the authors inserted an explicit 2-minute video-viewing period between two identical online sessions. During this tested 2-minute idle condition no spurious commands were recorded. The absence of unintended activations in that interval is attributed to the asynchronous evidence-accumulation decision policy and the conservative thresholds implemented for command issuance.
Feasibility in immersive and movement-prone contexts was tested with a VR shell-game task that included head movements and the constraints of VR rendering. In this VR proof-of-concept the system reached an online accuracy of 100% with a mean decoding time of 5.6 seconds. Participants nonetheless reported higher fatigue in VR compared with the screen condition, despite preserved visual comfort and low peripheral distraction.
Participants provided subjective assessments of the stimulation. Across both screen and VR conditions they reported high visual comfort and minimal peripheral distraction, consistent with the low-salience textured stimulation design. However, the reported fatigue was greater in the VR condition than during the screen-based task, indicating an experiential cost associated with immersive use even when accuracy remains high.
The authors conclude that their c-VEP rBCI demonstrates a combination of desirable properties for practical, wearable systems: short calibration (33 seconds), high accuracy (99.8% on-screen, 100% in VR proof-of-concept), comfortable stimulation, semi-dry EEG acquisition, and conservative asynchronous decision-making that avoided spurious activations in the tested idle interval. They interpret the VR results and idle-state performance as supportive of feasibility for self-paced interaction outside laboratory settings.
Reported limitations and recommendations in the source include the need to evaluate performance in longer idle periods and in real-world, ecologically valid tasks. It is also noted that this work is presented as a preprint and has not been peer reviewed. Details beyond those reported in the source (for example, extended testing durations, wider participant demographics, or comparisons to alternative electrode technologies) were not reported in the article.