Iterative Optimal Combination of Information for Power Amplifiers Linearization: A Bayesian Approach

This paper presents a Bayesian framework for the linear estimation problem in digital predistortion (DPD) systems applied to radio-frequency power amplifiers (PAs). Unlike conventional least-squares (LS) approaches, which treat model coefficients as fixed parameters and require the full dataset to be processed simultaneously, the proposed methodology leverages both the mean and precision of the coefficient distribution to enable an optimal combination of prior information with new observations. This probabilistic formulation naturally leads to a maximum a posteriori (MAP) estimator that supports iterative updates, allowing data to be incorporated sequentially while reducing the computational and memory demands typically associated with Volterra-based behavioral models. The resulting algorithm enables real-time, batch-wise learning that is well-suited for FPGA-based implementations and adaptive predistortion schemes. Experimental validation on a 5G-NR waveform with a bandwidth of 100 MHz using a Doherty PA demonstrates that the iterative Bayesian estimator achieves modeling and linearization performance comparable to classical methods, even when operating with significantly fewer samples per iteration. The results confirm that the proposed approach effectively reduces complexity while preserving linearization quality, enabling computation-efficient DPD adaptation in practical transmitter architectures.