Effects of finite register length on fast ICA, bacterial foraging optimization based ICA and constrained genetic algorithm based ICA algorithm

Independent component analysis (ICA) technique separates mixed signals blindly without any information of the mixing system. Fast ICA is the most popular gradient based ICA algorithm. Bacterial foraging optimization based ICA (BFOICA) and constrained genetic algorithm based ICA (CGAICA) are two recently developed derivative free evolutionary computational ICA techniques. In BFOICA the foraging behavior of E. coli bacteria present in our intestine is mimicked for evaluation of independent components (IC) where as in CGAICA genetic algorithm is used for IC estimation in a constrained manner. The present work evaluates the error performance of fast ICA, BFOICA and CGAICA algorithms when they are implemented with finite length register. Simulation study is carried on both fixed and floating point ICA algorithms. It is observed that the word length greatly influences the separation performance. A comparison of fixed-point error performance of the three algorithms is also carried out in this work.