Algorithm for Model Dimensions Estimation of Memory Polynomial-Based RF Transmitters / Power Amplifiers Behavioral Models

In this paper, a novel accurate model dimensions estimation algorithm suitable for memory polynomial based behavioral models is proposed. For a predefined model performance criterion, the proposed algorithm selects the nonlinearity order and the memory depth for a trade-off between performance and computational complexity. Indeed, the nonlinearity order is first chosen to minimize the conditioning number of the Vandermonde matrix. Then, the memory depth is optimized to decrease the overall the number of coefficients in the model. The proposed algorithm is experimentally validated on a high power 3G Doherty amplifier prototype driven by various WCDMA multi-carrier signals. This validation showed that the model performance was maintained while its complexity was reduced by 40% and its robustness improved by 20%.