An Improved FastICA Algorithm Using Correction Factors And Lanczos Algorithm For EEG Signals

Independent Component Analysis (ICA) is a powerful technique for source separation, particularly in scenarios where prior knowledge is unavailable. However, the conventional FastICA algorithm often suffers from slow convergence and a high iteration count, limiting its applicability in real-time scenarios. This paper proposes an improved FastICA algorithm that addresses these limitations by reducing computational time and iteration count. Specifically, incorporating an inexact correction factor reduces the average number of iterations by 5.13%. By setting the correction term to 0.993 and introducing a random initial matrix with parameter values of 96, the algorithm achieves a 29.20% improvement in computational efficiency. Furthermore, applying the Lanczos algorithm for whitening further improves performance. When the parameter values of the random initial matrix is adjusted to 72, the number of iterations is reduced by 28.65%. Combining a correction term of 0.999 with parameter values of the random initial matrix are 72 yields an overall efficiency improvement of approximately 35.11%, demonstrating a significant advancement over the original FastICA algorithm.