Complexity-aware-normalised mean squared error 'CAN' metric for dimension estimation of memory polynomial-based power amplifiers behavioural models

The memory polynomial model is widely used for the behavioural modelling of radio-frequency non-linear power amplifiers having memory effects. One challenging task related to this model is the selection of its dimension which is defined by the non-linearity order and the memory depth. This study presents an approach suitable for the selection of the model dimension in memory polynomial-based power amplifiers' behavioural models. The proposed approach uses a hybrid criterion that takes into account the model accuracy and its complexity. The proposed technique is tested on two memory polynomial-based behavioural models. Experimental validation carried out using experimental data of two Doherty power amplifiers, built using different transistor technologies and tested with two different signals, illustrates consistent advantages of the proposed technique as it significantly reduces the model dimension by more than 60% without compromising its accuracy.