Combinational of GAN and LSTM-Based DNN for Automatic Optimization of Active and Passive Devices

Future high-data-rate communication systems include concurrent utilization of active and passive devices as power amplifiers (PA) and antennae. Hence for these high dimensional designs, determining an accurate starting point along with achieving acceptable outcomes are required effectively. This paper is devoted to presenting the implementation of two types of neural networks: generative adversarial network (GAN) and long short-term memory (LSTM) deep neural networks (DNNs) for both PA and antenna devices. The benefit of implementing GAN for the PA side is to estimate the load-pull contours on the Smith chart as optimal gate and drain impedances. In addition, the GAN is employed on the antenna side to predict the radiation pattern outcomes for the determined frequency. From another point of view, the LSTM-based DNN with the utilization of the Thompson sampling efficient multi-objective optimization (TSEMO) is presented for predicting the optimal design parameters leading to achieving the targeted specifications for both active and passive devices. The presented methodology is validated by designing and optimizing a PA with a multipleinput and multiple-output (MIMO) antenna operating at an approximate bandwidth of 2.68 GHz.