This thesis investigates a steady-state visually evoked potential (SSVEP)–based brain-computer interface (BCI) for controlling a computer cursor. The system employs two frequency-detection methods: filter-bank canonical correlation analysis (FBCCA) and ensemble task-related component analysis (eTRCA). Performance was evaluated using EEG data from a public dataset (MAMEM) and a self-recorded dataset. Classification accuracy, information transfer rate (ITR), and signal-to-noise ratio (SNR) were analyzed.

Results indicate that the MAMEM dataset, featuring a 256-channel high-density electrode setup, yielded superior classification accuracy—averaging 70.3% with FBCCA and reaching up to 93.3% peak accuracy in some subjects. In contrast, the self-recorded dataset, acquired with an 8-channel OpenBCI headset, showed lower performance, with FBCCA average accuracies ranging from 33.3% to 71.1%, and a maximum observed peak accuracy of 93.3% in isolated trials. FBCCA proved particularly effective in real-time conditions due to its calibration-free design, achieving real-time classification accuracies of 67.86% and 62.5% in two separate test sessions. In addition, system latency was assessed, and it was found that the complete signal processing pipeline executes within 0.232 seconds. eTRCA achieved higher offline accuracy on the MAMEM data when sufficient training trials were available but underperformed on the self-recorded dataset due to limited calibration data. Real-time cursor control was successfully demonstrated using two target frequencies (8.57 Hz and 12.00 Hz), confirming the practical viability of SSVEP-based BCIs. Future work should aim to improve signal quality, enhance spatial resolution through better electrode placement, and reduce the need for user-specific calibration to enable more reliable and accessible real-world applications.