The building blocks of a brain-computer interface
The six-step BCI pipeline from user to output: data collection, pre-processing, feature extraction, prediction, and why the feedback loop that closes the cycle is harder than it looks.
A BCI pipeline is a cycle, not a linear process: the system influences the user, the user changes what they do, and that changes what the system reads. The article walks through the six stages: the user performing a mental task, EEG data collection, pre-processing to remove noise, feature extraction to isolate relevant frequency and spatial patterns, machine learning prediction, and the output that closes the feedback loop.
Each step has its own traps. Device selection is the most consequential early decision. Pre-processing choices cascade through to classification accuracy. And the feedback loop at the end is harder to design for than it looks: wrong predictions change the user's mental state, which creates a different signal, which affects the next prediction. Published on Medium, July 2022.