This work presents a new artificial neural network (ANN) model formulation for RF high-power transistors which includes the S-parameters of the active device. This improves the small-signal extrapolation capability, and the OFFstate impedance approximation, making it suitable for Doherty power amplifier (DPA) design. This extrapolation capability plays a key role in the correct Doherty load modulation prediction, since, at low power levels, the peaking PA is subjected to active loads that cannot be synthetized with a passive load-pull system, forcing the model to extrapolate. Thus, the proposed model formulation is able to solve the issues that are normally observed when ANN-based models are used in complex PA architectures as the Doherty PA. To validate the proposed behavioral model, a 700-W asymmetrical LDMOS DPA, centered at 1.84 GHz, was simulated and measured. Index Terms-Artificial neural network (ANN), behavioral model, Doherty, load modulation, passive load-pull, power amplifier. I. INTRODUCTION T HE Doherty power amplifier (DPA) is, nowadays, the wireless base-station workhorse in what RF signal amplification is concerned [1]-[4]. Accurate nonlinear models for the state-of-the-art high-power transistors are very difficult to obtain mostly because of the thermal issues and the distributed nature of these devices [5]. Therefore, the conventional Doherty design process is normally based on load-pull and S-parameter measurements. From these measurements, the optimal power load (Z pwr) is determined and the optimal efficiency termination for a particular VSWR, defined on this chosen power load, is selected. This VSWR imposes the back-off Manuscript