Efficient Stepsize Selection Strategy for Givens Parametrized ICA Applied to EEG Denoising

In this letter, an efficient stepsize selection strategy for Givens parametrization-based independent component analysis (ICA) approaches is presented. The method is an extension of the pioneering deflationary ICA algorithm introduced by Delfosse and Loubaton (DelL
^R
). The methodology underlying the computation of such a stepsize is based on sampling the contrast function using the information carried by its frequency components. Epileptic ElectroEncephaloGraphy (EEG)-based simulations show that the proposed approach stands for the best extraction accuracy/numerical complexity tradeoff with 15 times faster execution time. This is compared to a deterministic implementation of the DelL
^R
technique and its recently proposed Jacobi-based variant, namely the JDICA method.