Abstract: This paper presents two new approaches to model nonlinear distortions which commonly arise in small loudspeakers used in mobile terminals. One is based upon a new, quantized frequency domain approach and another upon a parallelized polynomial filter approach. Both models are derived from practical studies of the inputoutput characteristics of real mobile terminal loudspeaker and artificial sinusoidal signals. The models are then used to predict the nonlinear distortions in real speech signals. Comparisons to ground-truth distortions are performed to validate the models and confirm that both produce reliable predictions of non-linear distortions. In contrast to existing approaches, the new models are computationally efficient and are suitable for the real-time compensation of non-linear distortions.

Download: