Improved performance of FLL using a more robust error term.