A Comparative Analysis of Behavioral Models for RF Power Amplifiers.pdf

348 IEEE TRANSACTIONS ON MICROWAVE THEORY AND TECHNIQUES, VOL. 54, NO. 1, JANUARY 2006 A Comparative Analysis of Behavioral Models for RF Power Amplifiers Magnus Isaksson, Student Member, IEEE, David Wisell, Member, IEEE, and Daniel Rönnow, Member, IEEE Abstract—A comparative study of nonlinear behavioral models plex for circuit-level or transistor-level simulations [17]. Phys- with memory for radio-frequency power amplifier (PAs) is ical models are based on a physical description of the compo- presented. The models are static polynomial, parallel Hammer- nents of the PA and how they interact and are typically formu- stein (PH), Volterra, and radial basis-function neural network (RBFNN). Two PAs were investigated: one was designed for the lated as equivalent circuit models. When used for analyzing a third-generation (3G) mobile telecommunication systems and PA, physical models can give insight into the cause of the non- one was designed for the second-generation (2G). The RBFNN linear properties and memory effects [18]. Both behavioral and reduced the total model error slightly more than the PH, but the physical models are thus useful in the design process. error out of band was significantly lower for the PH. The Volterra Behavioral models are used for digital predistortion (DPD) was found to give a lower model error than did a PH of the same nonlinear order and memory depth. The PH could give a lower of PAs. In DPD, an inverse model of the PA’s response function model error than the best Volterra, since the former could be is used (either model-based or as a look-up table). The signal identified with a higher nonlinear order and memory depth. The is predistorted b