Individual-Activation-Factor Memory Proportionate Affine Projection Algorithm With Evolving Regularization

The individual-activation-factor memory proportionate affine projection algorithm (IAF-MPAPA) provides a good solution for echo cancelation. However, the IAF-MPAPA with fixed regularization factor requires a tradeoff between fast convergence rate and low steady-state misalignment. In this paper, the mathematical relationship between the regularization factor and the steady-state mean square error (MSE) of the IAF-MPAPA was deduced. The mathematical formula of the steady-state MSE indicates that it is inversely proportional to the value of regularization factor. Then, inspirited by the evolutionary method, the IAF-MPAPA with evolving regularization (ERIAF-MPAPA) was proposed. The ERIAF-MPAPA increases or decreases the regularization factor by comparing the power of output error with a threshold which contains the information of the steady-state MSE. For highly sparse impulse responses, simulation results demonstrate that the proposed ERIAF-MPAPA offers better convergence performance than other proportionate-type APAs in terms of convergence rate and steady-state misalignment.