Application of optimal delays selection on parallel cascade hammerstein models for the prediction of RF-power amplifier behavior

In this work two different black-box modeling strategies are combined in order to achieve improved RF power amplifier (PA) models based on measured input/output signals. The model validation is here done within the low input back-off (IBO) region of the power amplifier, thus showing significant nonlinear operation. The figure of merit (FoM) applied is the normalized mean square error (NMSE). A rectangle constellation modulation, 16-QAM, is used as input signal, and the nonlinearities caused by the low IBO operating conditions of the PA distort the output constellation. The PA output signal used to estimate the model coefficients presented high distortion levels, requiring a model capable to also care for memory effects, as it will be seen in this article. It has been observed that the combination of a selection of the optimal delays values with parallel-cascade Hammerstein pseudo-inverse based models can improve the identification accuracy, leading to precise models. This condition is important regarding the design of efficient pre-distorters, once its performance is very sensitive to the model quality.