S‐Parameters Modeling Investigation for GaAs pHEMT Based on LSTM and ELM

Based on long short term memory (LSTM) neural network and extreme learning machine (ELM) neural network, two small‐signal models of gallium arsenide (GaAs) pseudo‐crystalline high electron mobility transistor (pHEMT) are established and analyzed in this paper. To verify these models, 10‐fold cross‐validation and feature ranking are performed on 600 sets of the scattering parameters (S‐parameters) for GaAs pHEMT. Experimental results show that the LSTM is better than other S‐parameters when predicting ReS 12 , with the mean square error (MSE) of 6.2483 × 10 −05 , the mean absolute error (MAE) of 6.38 × 10 −03 , and the coefficient of determination ( R 2 ) of 9.1013 × 10 −01 . On the other hand, the ELM model achieves better MSE, MAE, and R 2 values of 1.3946 × 10 −06 , 7.6491 × 10 −04 , and 9.9697 × 10 −01 , respectively. Meanwhile, MSE and MAE of ELM are smaller than LSTM when predicting all the S‐parameters. In addition, it shows a smaller fluctuation range of the error curve for ELM, indicating that it is more stable than LSTM. Therefore, ELM is an effective method to model the small‐signal of GaAs pHEMT.