1: \begin{abstract}

2: Transform-domain least mean squares (LMS) adaptive filters encompass the class of algorithms in which the input data are subjected to a unitary transform followed by a power normalization stage and an LMS adaptive filter. Because of the data-independent nature of conventional transformations, such a transformation improves the convergence of the LMS filter only for certain classes of input data. However, for input data from unknown classes or a specific set of classes, it is difficult to decide which transformation to use. This decision necessitates a learning framework that obtains such a transformation using input data, which improves the condition number after transformation with minor additional computation. It is hypothesized that the underlying data topology affects the selection of the transformation. With the data modeled as a weighted graph and the input autocorrelation matrix known or computed beforehand, we propose a method, PrecoG, that obtains the desired transform by recursively estimating the graph Laplacian matrix. Additionally, we show the efficacy of the transformation as a generalized split preconditioner on a linear system of equations with an ill-conditioned real positive definite matrix. PrecoG shows significantly improved post-transformation condition number as compared to the existing state-of-the-art techniques that involve unitary and non-unitary transforms. 

3: \end{abstract}

4: