Real-Time EEG Brainwave Analysis
← Project Index Real-Time EEG Brainwave Analysis

Real-Time EEG Brainwave Analysis

A NeuroSky MindWave headset, a stream of brainwaves over Bluetooth, and a pipeline that turns them into live Delta-through-Gamma band power plus attention and meditation -; wired so an AI can read your brain state as it happens.

Archived Started: Summer 2025 Updated: Summer 2025

Overview

This is the foundation layer of a small constellation of EEG experiments: get clean, real-time brainwave data off a consumer NeuroSky MindWave Mobile and into a form software can reason about. It connects over Bluetooth, captures raw EEG at 512 Hz, runs FFT to split the signal into the standard frequency bands, and surfaces NeuroSky's proprietary attention and meditation meters alongside.

The piece that made it more than a logger is an MCP server that bridges the headset to an AI, so a model can query live brain state and (eventually) adapt to it -; the seed of an adaptive-learning system.

Background

Consumer EEG is a strange, fun middle ground: nowhere near medical-grade (one dry sensor at FP1, lots of artifacts), but cheap, wireless, and good enough to detect broad cognitive states at 70-;85% accuracy. That's enough to play with. The whole MindWave line of work here started from a simple want -; to see my own attention and meditation move in real time and then do something useful with the signal.

How It Works

The data path is straightforward and deliberately documented end to end:

  • Capture -; auto-detect the MindWave on a COM port, decode the ThinkGear binary protocol, pull raw samples at 512 Hz.
  • Process -; 1-second windows (25% overlap), Hanning window to cut spectral leakage, FFT, then sum power within each band: Delta, Theta, Alpha1/2, Beta1/2, Gamma1/2.
  • Augment -; read the hardware's eSense attention and meditation values (0-;100) and total power.
  • Export / stream -; write timestamped CSVs for offline analysis, or stream live through the MCP server for an AI to consume.

Where It Landed

Archived, but it genuinely worked -; there's a pile of real recorded sessions (timestamped CSVs from July 2025) and a frequency-analysis figure to prove the pipeline ran on live data, not just simulated input.

  • Reliable Bluetooth capture and ThinkGear parsing from a real headset.
  • Full eight-band FFT decomposition plus attention/meditation extraction.
  • MCP server bridge so an AI can read the stream -; the hook into the adaptive-interface and pattern-detection experiments.