A Custom Hardware CCA Engine for Real-time SSVEP-based BCI Applications

Canonical Correlation Analysis is widely exploited in different signal processing applications such as detection of the brain internal activity in an SSVEP based Brain-Computer Interface system. The main disadvantage of the CCA algorithm is its computationally intensive nature which demands numerous complex matrix transformations including inversion, eigenvalue and eigenvector extraction, multiplication, and decomposition. These complexities prevent its efficient hardware implementation while its software implementation also is not so fast to be exploited in real-time applications. In this paper, some simplifications are applied to an 8-channel CCA algorithm to make its hardware implementation feasible. The effects of these simplifications on the output accuracy of the algorithm are also demonstrated in the paper. A custom fully hardware fixed-point CCA engine is then developed based on the simplified algorithm to achieve higher performance and less power consumption. The FPGA synthesis results of the design are also included in the paper.