A novel low-power DLMS adaptive filter with sign-magnitude learning and approximated FIR section

In this paper we propose a novel approximate implementation for the Delayed Least Mean Square (DLMS) filter, able to improve the power consumption while preserving the learning capabilities. In order to minimize the switching activity, we exploit the magnitude of the error signal to update the filter coefficients. Moreover, the FIR section of the adaptive filter is approximated by using a novel approximate fused multipliers-adder tree, exploiting a partial products cancellation and correction technique. Simulation results show that the convergence properties of the proposed filters are practically unchanged with respect to the original DLMS algorithm. Syntheses in 28 nm technology show a power saving of 53.7% that surpass the state of the art.