Implementation of EEG Signal Decomposition and Feature Extraction Through Efficient Wavelet Transforms

In the diagnosis of brain and mental disorders and anomalies, Electroencephalogram (EEG) analysis plays a critical role. There are five bands of EEG signal namely Alpha, Gamma, Beta, Delta and Theta. This research proposes an enhanced method of EEG signal analysis based on FPGA employing Wavelet Packet Transform (WPT) and Discrete Wavelet Transform (DWT) for decomposition followed by feature extraction. Using the decomposed EEG signal, the mean, median, variance, and peak-to-peak values can be computed. The EEG signal consists of five distinct stages: alpha, beta, gamma, theta, and delta waves. Four levels of DWT and three levels of WPT are used to split the signal. The 8 tap transpose FIR filter with daubechies 4 wavelet is employed. In DWT, approximate coefficients from the filters are down sampled to four levels whereas in WPT both the approximate coefficients and detail coefficients are down sampled to three levels. The coefficients of the high pass filter are detailed, but those of the low pass filter are approximations. The proposed decompositions have been implemented using the Basys 3 Field Programmable Gate Array (FPGA). The DWT uses a total of eight filters: four high pass filters and four low pass filters. The WPT algorithm performs the decomposition using a single high pass filter and seven low pass filters, each with a unique initial approach. The various hardware resources utilized for DWT and WPT are estimated. The latency of WPT is less than DWT making it faster at the expense of resources utilized.