Description
Brain-Computer Interface (BCI) has been studied deeply and extensively for the purpose of restoring sensorimotor functions. By recording the brain signals using electrodes, BCI helps individuals to make use of their own brain signals and its electrical activity. Through techniques such as decoding, translating and actuating, these recorded brain signals can be used to control external devices. Here we show that the decoding of brain signals can be achieved in an efficient manner using on-chip implementation unlike most of the state of the art techniques which depend on offline analysis. This on-chip implementation also paves way for developing a real-time BCI system that can predict volitional movement intentions. In this work we mainly concentrate on the main building block of our decoding architecture which is a hardware friendly version of Principal Component Analysis (PCA). PCA is mainly used for dimensionality reduction and it also aides the neural network classifiers which is the block next to the feature extraction in increasing the classification accuracy. Experimental results on FPGA using publicly available electrocorticography (ECoG) data show that our proposed algorithm converges to less than 0.1 in terms of mean square error (MSE) in less than 10 iterations when compared with MATLAB SVD based algorithm.
