Behavioral Modeling of Dynamic Nonlinear Distortions in 5G Wireless Transmitters Using Cascaded Augmented Real-Valued Neural Networks

Neural networks are increasingly adopted for performance enhancement in wireless communication infrastructure for 5G and 6G applications. This paper proposes a modular two-box neural network-based system for the behavioral modeling of dynamic nonlinear distortions observed in wireless transmitters. The proposed model, labeled cascaded augmented real-valued artificial neural networks (CAR-VANN), uses a first neural network with an augmented but memoryless input vector feature to model memoryless nonlinear behavior. This model is designed for low-complexity and coarse estimation of the nonlinear distortions. The second neural network, which aims to fine-tune the model output and boost its accuracy, is a conventional augmented real-valued time-delay neural network (ARVTDNN). Experimental validation shows that the CAR-VANN model can achieve the same performance as the ARVTDNN with a significant reduction in the number of parameters (between 35% and 52%). Accordingly, this model can be considered a viable alternative for the computationally efficient modeling of dynamic nonlinear distortions in 5G systems, reducing the computational complexity associated with neural networks-based models without compromising their performance.