Digital predistortion of D-band power amplifiers using AI-based nonlinearity compensation

This paper presents the first demonstration of a neural network-based digital predistortion (DPD) architecture for compensating nonlinearities in real D-band power amplifiers (PAs). The proposed method employs two separate models to independently correct amplitude (AM/AM) and phase (AM/PM) distortions. Training data is generated through simulation using a Rapp model under varying nonlinearity levels, and further validated using measurements from real D-band hardware. The AM/AM network predicts the ideal input magnitude required to produce a linearized output, while the AM/PM network compensates for the associated phase shifts. Evaluation results demonstrate improvements in linearity, with the DPD-corrected output closely matching the behavior of an ideal limiter. Model compression techniques are applied to reduce memory and computational requirements with minimal impact on performance, making the solution suitable for real-time and embedded deployment.