Linearization of Multi-band RF Transmitters Based on Direct Learning Architecture for LTE/5G Applications

This paper presents a multi-band extension of the Digital PreDistortion (DPD) based on the Direct Learning Architecture and pruning structure. While DPD is widely used to linearize RF power amplifiers by acting as an inverse dynamic function, conventional implementations are generally restricted to single-band transmissions. This limitation constrains the spectral efficiency and capacity of modern telecommunication networks. In this work, we propose an adaptation of the Direct Learning Architecture to a multi-input/multi-output (MIMO) DPD framework to compensate both inter-band and intercarrier intermodulation in multi-band RF system. Particular attention is given to the Generalized Memory Polynomial (GMP) model, selected for its high accuracy and modeling precision. We also propose a reformulation of the doubly orthogonal matching pursuit approach to select the active terms of the GMP model in each band, while integrating the DLA approach to enhance linearization performance. The effectiveness of the proposed linearization approach is validated through simulations using a dual-band nonlinear RF transmitter designed for LTE/5G sub-6G signals.