# SSVEP-003: Filter bank canonical correlation analysis for implementing a high-speed SSVEP-based brain-computer interface > Generated from a local paper card. Do not edit this page directly; edit the source card and rerun `vp run docs:generate`. > Internal wiki boundary: paper-card evidence and user-provided PDF downloads are available for private reading. Extracted full-paper text is not published. > Source: `library/paper_cards/SSVEP-003.md` ## Paper Access * Internal PDF: Download Paper * DOI / official page: [10.1088/1741-2560/12/4/046008](https://doi.org/10.1088/1741-2560/12/4/046008) * Deployment boundary: these PDF links are intended only for a private/protected internal wiki. ## SSVEP-003: Filter bank canonical correlation analysis for implementing a high-speed SSVEP-based brain-computer interface ## Metadata * ID: SSVEP-003 * Title: Filter bank canonical correlation analysis for implementing a high-speed SSVEP-based brain-computer interface * Year: 2015 * DOI / URL: 10.1088/1741-2560/12/4/046008 * Local PDF: see Paper Access section above * Text artifact: local-only path withheld from docs site * Review status: `extracted` ## Study Type * Track: SSVEP * Task: high-speed SSVEP speller decoding with filter-bank CCA * Participants or dataset: 19 healthy participants; 12 in offline experiments, 10 in online experiments, with 3 overlapping participants * Hardware: Neuroscan Synamps2, 1000 Hz sampling, 23.6-inch LCD at 1920 x 1080 and 60 Hz * Channels or sensors: 9 occipital/parieto-occipital electrodes: Pz, PO5, PO3, POz, PO4, PO6, O1, Oz, O2; vertex reference; impedance under 10 kOhm ## Methods * Paradigm: 40-target 5 x 8 SSVEP speller using 8-15.8 Hz frequency coding with 0.2 Hz spacing * Signal processing or model: FBCCA with Chebyshev type I zero-phase IIR filter bank; standard CCA in each sub-band; weighted combination of squared correlations * Training/calibration: designed as training-free or no individual calibration for the main FBCCA approach * Online/offline: both offline and online experiments; online trials used 1.25 s stimulation plus 0.55 s gaze shifting ## Results * Metrics: offline M3 peak accuracy 89.47% and ITR 145.52 bits/min; standard CCA 76.80% and 113.85 bits/min; online mean accuracy 91.95 +/- 7.22%, ITR 151.18 +/- 20.34 bits/min, 33.3 characters/min * Main findings: filter-bank treatment of fundamental and harmonic SSVEP components improved high-speed SSVEP target detection relative to standard CCA * Reported limitations: fixed speller matrix rather than free scene selection; needs larger and patient samples; low-frequency stimulation may cause visual fatigue; real-world comfort, mobility, and usability remain open ## Relevance To This Project * Supports: FBCCA as a strong baseline for multi-target SSVEP decoding in a scene-candidate selection module * Conflicts with: no direct conflict, but the protocol is a fixed gaze-based speller, not object-level robotic grasping * Design implication: the first SSVEP experiment can compare scene-object selection against a known 40-target FBCCA baseline, but must separately test object motion, target layout, and fatigue ## Extracted Evidence | Claim | Status | Evidence Note | Page/Section | | --- | --- | --- | --- | | FBCCA improves multi-target SSVEP detection by using harmonic information across filter-bank sub-bands. | verified | The paper proposes FBCCA and compares sub-band weighting methods for SSVEP target recognition. | Abstract; Methods | | High-speed 40-target SSVEP spelling is feasible with 9 occipital/parieto-occipital channels. | verified | The online setup used 9 electrodes and reported mean accuracy above 90% with ITR around 151 bits/min. | Methods; Results | | FBCCA is a suitable candidate baseline for scene-aware SSVEP target selection. | inferred | The method supports many frequency-coded targets, but the paper tests fixed speller cells rather than detected scene objects. | Methods; Online experiment | | Dynamic YOLO-box SSVEP performance remains unverified. | needs confirmation | The paper does not evaluate moving boxes, object detection confidence, robot tasks, or gaze-free object selection. | Future work / limitations | ## Open Questions * How many object candidates can be assigned clean frequencies without visual crowding or harmonics? * Does a gaze-shift interval remain acceptable when targets are scene objects rather than fixed character cells? * Should the first prototype use FBCCA as the default baseline before testing TRCA or deep SSVEP models?