Recently, the technique of artificial neural networks (NNs) has drawn the attention of researchers in the field of power amplifier (PA) modeling due to its successful implementation and favorable results in pattern recognition, signal processing, system identification, and control [1–4]. As a result of its adaptive nature and its universal approximation capability [5–9], the NN approach has been investigated as one of the modeling and predistortion techniques for PAs and transmitters [5–14]. Different NN topologies and training algorithms that also take into account memory effects have been proposed. NN models can be either static (i.e., memoryless) or dynamic. A static NN model can be augmented to take into account memory effects and derive a dynamic model suitable for broadband nonlinear transmitters; such a model is often designated as a time-delay neural network (TDNN) [1, 3, 14].
This chapter presents major static and dynamic NN-based behavioral models and offers a comparison of different NN topologies and training algorithms that can be used for the identification of both forward and reverse models of nonlinear PAs and transmitters.
7.2 Basics of Neural Networks
NNs have been widely used as powerful tools for modeling nonlinear dynamic systems [15, 16]. The application of NNs to system modeling and identification is motivated by their universal approximation property, where a feedforward network with a finite number ...