On Accelerating the Convergence of the LMS Algorithm