Noise constrained LMS algorithm in transform domain

In a recent paper the authors proposed a new Variable Step LMS algorithm for transform domain. In that algorithm the step-size structure was analyzed from a new point of view. The step-size has two components: a global component, that is the same for each filter coefficient and a local time-variable component, given by the power normalization. Making the global component also time-variable, the speed of convergence was greatly increased. In this paper we introduce a new method for adaptation of the global component by using some informations about noise variance. The novelty of this algorithm consists in low complexity and also the trade-off between a small misadjustment and a fast convergence speed is completely eliminated.