Nonlinear acoustic system identification using a combination of Volterra and power filters

The paper proposes a nonlinear system identification method that uses a combination of adaptive linear, Volterra and power filters. Adaptation of the kernels is made using a Normalized Least Mean Square algorithm. The method is applied in echo cancellation, where several sources of nonlinearities exist: the overdriven amplifier, the small loudspeaker at high volume, the room with different absorbent walls. Functions with nonlinear characteristics are chosen to model these distortions. The evaluation is made in terms of Echo Return Loss Enhancement. Results show that the overall convex combination approach performs better or at least as well as the best single adaptive filter.