Real-time Feature Extraction for Multi-channel EEG Signals Time-Frequency Analysis

This paper presents a model-based Field Programmable Gates Array (FPGA) design for real-time feature extraction of Electroencephalogram (EEG) signals, which can be used for brainwaves bands classification to track and detect mental status in Brain Computer Interface (BCI) applications and consciousness studies. An model-based design approach is used to implement Short-time Fourier Transform (STFT) and extract 20 frequency feature components for classification. These 20 features are divided into 5 groups corresponding to 5 different brainwaves bands. Each feature is defined as the average power spectrum of a number of adjacent frequency components. A hardware model is designed using Xilinx System Generator and implemented on FPGA. Fixed-point is used instead of floating-point to increase operating speed for meeting timing requirement of the real-time system. The design is implemented on a Xilinx Zedboard at 50MHz clock rate, and can be used for up to 128-channel EEG signals feature extraction at 250 Hz sample rate.