Data-Driven BiLSTM Prediction Model for Dual-Band RF Power Amplifier

This article proposes a Bidirectional Long Short-Term Memory (BiLSTM) model to capture nonlinear distortions in dual-band RF power amplification (PA) systems. It is important for advanced wireless communication systems, especially those with high peak-to-average power ratio (PAPR) waveforms, to characterize and linearize RF power amplifiers (PAs) despite deep learning’s success in various fields. This study employs BiLSTM neural networks to address memory effects in nonlinear PAs. This paper designed and tested a dual-band PA operating at 2.4 GHz, using the BiLSTM model to predict output power and linearity. Our BiLSTM-based modeling architecture effectively addresses non-causality concerns, achieving an R-Square ($R^{2}$) of 0.89, mean squared error (MSE) of 0.013, and mean absolute error (MAE) of $\mathbf{0. 0 7}$. Experimental results validate our approach’s effectiveness, demonstrating accurate PA characterization by the trained networks. The BiLSTM model also shows promising performance in linearizing tested PAs, addressing their inherent unpredictability. This work advances deep learning integration to enhance PA performance in future wireless communication systems.