A novel metamaterial power divider optimized by artificial neural network

Abstract

An approach for enhancing the operating bandwidth of the classic dual-band power divider (PD) is proposed by using the metamaterial (MTM) units. To overcome the limitation of the EM simulation and improve the design efficiency, we propose an artificial neural network (ANN) approach that enables inverse prediction of the MTM-PD geometry from its desired physical response. In the ANN approach, the convolutional neural network and the long short-term memory neural network are combined to learn the relationship between the geometric of the proposed MTM-PD and its corresponding physical responses and then accurately predict the geometric parameters of MTM-PD. The predicted MTM-PD includes the low frequency (LF) band of 1.90–2.43 GHz and the high frequency (HF) band of 4.61–5.25 GHz. Compared to the classic dual-band PD, the bandwidth of the LF band has been enhanced by five times. The measured results confirm that the predicted MTM-PD has both the LF band and HF band with a bandwidth of 0.5 GHz, verifying the reliability of our ANN approach. It demonstrates the potential of using ANN for designing microwave devices and solving electromagnetic problems.