Forward behavioral modeling of concurrent dual-band power amplifiers using extended real valued time delay neural networks


Summary: The digital pre-distortion (DPD) technique is a widely accepted linearization approach that provides high accuracy in synthesizing the pre-distortion function and leads to a higher efficiency by allowing PAs to operate near saturation. DPD relies on the introduction of an exact inverse nonlinear pre-distorter before the PA to compensate for nonlinearity.

Many nonlinear models have been proposed for the characterization of PAs, including the memory polynomial model, the Volterra model, Wiener and Hammerstein models, and neural networks (NNs) models. In particular, NN models, with their excellent approximation capability, are becoming an increasingly attractive solution for PA behavioral modeling. To the best of our knowledge, there is no precedent NN model that has been applied for the behavioral modeling in concurrent dual-band PAs.

On account of the distortion caused by cross-modulations, an extended real-valued time-delay NN (RVTDNN) model is proposed to approximate the nonlinear behavior of concurrent dual-band PAs. The proposed model is trained and validated with different segments of the overall test data, and the experimental results demonstrate the advantage of this method.