Multi-objective neural network modeling and applications to microwave power amplifiers

Abstract

In this paper, two methods for accurate behavioral modeling of circuits applicable to the power amplifiers (PAs) are proposed. Firstly, an X-band PA based on WIN's 0.15 μm GaAs pHEMT process is presented. Taking this PA as an example, a modeling method based on the Particle Swarm Optimization algorithm optimized Back-Propagation Artificial Neural Network (PSO-BP ANN) is proposed and verified to realize a multi-output PA behavioral model. To solve the problem of increasing error caused by the inconsistency of basic property (i.e. dimension) between the output nodes of the BP ANN, the Non-dominated Sorting Genetic Algorithm-II optimized BP ANN (NSGA–II–BP ANN) modeling method is proposed. Finally, the two modeling methods proposed above are compared in terms of modeling accuracy, and the accuracy of NSGA–II–BP ANN is further verified based on the circuit measured results.