Digital subscriber lines (DSL) are fundamentally limited by near-end crosstalk (NEXT). A computationally efficient method to cancel the NEXT in high sampling rate DSL systems is to first detect the major NEXT and then deploy frequency-domain least-mean-square (FDLMS) adaptive filters to cancel them. However, when the environment is non-stationary, the mean residual error (MRE) rises increasing the noise to the system. A new technique for improving the tracking performance of the NEXT cancelling system is described. In the technique, the step-size of each adaptive filter is made proportional to the magnitude of the NEXT that is to be cancelled. Having an improved tracking performance lowers the MRE in a non-stationary environment, thereby translating into a higher data throughput.