Modeling, analysis and classification of a PA based on identified Volterra kernels

This article presents the modeling of a microwave power amplifier (PA) in the almost linear and compression operation modes. An in-band quasi-white noise real-valued signal is used as input for the identification process to excite every possible source of nonlinearity. A segment of the input-output measurement data is processed to generate an initial parallel cascade Wiener model (PCWM). The model is cross-validated with the entire measurement signal. The first order Volterra kernel is extracted in order to obtain an estimation of the amplifier´s memory. A new model is generated and its Volterra kernels up to the second order are estimated to apply the structural classification methods (SCM). The result of this process is a suitable block-structure for the final amplifier model. The optimized model is intended to be numerically robust having a high identification percentage based on a variance figure of merit. This resulting model can be used for simulation of linearization systems or even in further identification processes.