(eng): This paper derives analytic closed form expressions that characterize the performance of the algorithm FastICA, which is one of the most successful algorithms for independent component analysis. Second, the Cramér-Rao lower bound for linear ICA is derived as an algorithm independent limit oh the achievable separation quality. The FastICA algorithm is shown to approach this limit in certain scenatios. Extensive computer simulations supporting the theoretical findings are included.

(cze): V clanku jsou odvozeny teoreticke vyrazy ktere charakterizuji presnost separace nezavislych komponent pomoci algoritmu FastICA. Dale je odvozena Rao-Cramerova mez jako univerzalni hranice separovatelnosti. Je ukazano, ze algoritmus FastICA se teto hranici v urcitych situacich blizi. Soucasti clanku jsou rozsahle pocitacove simulace ktere potvrzuji odvozenou teorii.