Power Amplifier Modeling along with Hyperparameter Optimization of LSTM-based DNN through Multi-Verse Optimizer

Abstract

This work is devoted to presenting an intelligent-based methodology leading to model characteristics of power amplifiers (PAs) as well as sizing through the deep neural network (DNN). The long short-term memory (LSTM) architecture is employed to predict the extended frequency responses and achieve optimal hyperparameters of DNN for accurately sizing the PA. The PA is modeled in terms of scattering parameters, output power, power gain, and efficiency, in which nature-inspired optimization algorithms are executed to determine the optimal numbers of hidden layers with neurons of LSTM-based DNN. The proposed method leads to the construction of a reliable network for predicting the future specifications of any PA with design automation and acceptable convergence. To verify the effectiveness of the proposed method, a PA with 1.4 GHz is optimized, and the simulation outcomes prove that the multi-verse optimizer is strong enough to determine optimal hyperparameters of DNN.