Simplified Augmented Real-Valued Time-Delay Neural Network for Digital Predistortion

The conventional augmented real-valued time-delay neural network (ARVTDNN) approach offers state-of-the-art performance in power amplifier linearization but suffers from high complexity. This paper introduces a simplified ARVTDNN (SARTDNN) model to compensate for power amplifier distortions with lower complexity. It involves two steps: a neural network followed by a static nonlinear system. By learning the static nonlinear effects separately, the neural network requires fewer neurons than in ARVTDNN, reducing complexity while maintaining similar performance. The simulation results indicate that the proposed model offers a better performance/complexity tradeoff than ARVTDNN while satisfying the 3GPP LTE requirements.