Digital Predistortion of 5G Multiuser MIMO Transmitters Using Low-Dimensional Feature-Based Model Generation

Abstract

In this article, we present a novel digital predistortion (DPD) system which can be updated quickly and efficiently in response to the dynamic reconfiguration of multiuser multipleinput multiple-output (MIMO) transmitters. By identifying the shared properties of different power amplifiers (PAs) with two feature extraction stages, nonlinear behaviors of the PAs are encoded into low-dimensional features. Using the extracted features as input, a novel DPD generator is employed to synthesize DPD model coefficients directly. In this way, when the power levels or beam angles change, the DPD model can be updated fast and accurately without capturing the output data or applying linear system identification algorithms. To capture slow-varying dynamics of the PAs, the features can also be calibrated efficiently in the background. Computational complexity and operational latency can thus be reduced significantly. Simulation and experimental results demonstrate that the proposed DPD approach achieves excellent linearization performance with low complexity, making itself a promising linearization solution for 5G massive MIMO transmitters.

A Digitally assisted nonlinearity suppression architecture using over‐the‐air beam cancelation

“…To adapt to the change of beam direction and ensure the normal reception of UE in all directions, the coefficients of the nonlinear model need to be updated. A simplified forward modeling method 15 is adopted to update the coefficients.…”

Training Data Selection and Dimensionality Reduction for Polynomial and Artificial Neural Network MIMO Adaptive Digital Predistortion

“…Regarding multi-antenna DPD, the authors in [30] use sparse estimation techniques to reduce the basis of MIMO Volterra-based polynomial models for moderate input-output crosstalk conditions. In [31], a piecewise closed-loop DPD including a pruning algorithm for faster adaptation is introduced, while in [32] singular vector decomposition (SVD) is applied for dimensionality reduction of multiuser MIMO arrays.…”

Bandwidth, Power and Carrier Configuration Resilient Neural Networks Digital Predistorter

“…Such predistorters require frequent updates, hence this approach is often considered as computationally demanding [1], [2]. On the other hand, the use of scalable digital predistortion systems can be perceived as a better approach for ensuring continuous match between the predistorter's and the amplifier's nonlinear functions [3], [4], [5], [6], [7], [8]. Scalable predistortion systems aim at minimizing the number of predistorter parameters (for example polynomial function coefficients) that need to be updated following a change in the PA operating conditions.…”

“…The second box of the model, namely the memory polynomial, was then updated to fine tune the predistortion function and ensure its scalability while updating only a small portion of the overall model parameters. Feature-based modeling has also been investigated for the design of scalable predistorters [4], [5], [6]. In [4], a complexity reduced adaptation technique was proposed to enable digital predistorters to track changes in PA behavior with reduced complexity updates.…”

“…This technique was based on a pre-training of the digital predistorter over a wide range of operating conditions in order to extract common features of the PA behavior, and identify through principal components analysis technique the model parameters that need to be transformed to track changes in the PA behavior. This approach was then extended to the case of multi-input multi-output transmitters [5]. More recently, the concept of model features extraction was also applied for the linearization of active arrays power amplifiers for various beam conditions [6].…”