ABSTRACT Behavioral models for Power Amplifier (PAs), particularly those based on Neural Networks (NNs), often suffer from high computational complexity when handling wideband signals. In this paper, we present a novel behavior model for wideband GaN PAs based on a Two‐Sided Temporal Convolutional Neural Network (TST‐CNN). The model uses a specialized convolutional layer to process the input data in order to extract the essential basis functions needed to accurately capture the dynamic memory effects and static nonlinearities of the PA. These enhanced features are then fed into a Fully Connected (FC) layer to establish the predictive model. Due to the efficient feature extraction enabled by the convolutional structure, the proposed architecture successfully handles strong memory effects without a significant increase in model complexity. The experimental results, obtained at a carrier frequency of 2.14 GHz using a 100 MHz wideband signal emulating 5th Generation (5G) wireless technologies, demonstrate that the TST‐CNN outperforms current state‐of‐the‐art models by achieving an Adjacent Channel Power Ratio (ACPR) of −52.1 dB. Furthermore, the TST‐CNN requires only 237 coefficients, resulting in a high reduction in computational complexity compared with existing approaches, which makes it a very effective solution for future wideband communication systems.
Rezgui et al. (Fri,) studied this question.