Order reduction of wideband digital predistorters using principal component analysis

Abstract

This paper presents how to apply order reduction in wide-band digital predistortion (DPD) linearizers using the principal component analysis (PCA) technique. This method is tested in a wireless backhauling transmitter where four 28 MHz adjacent subcarrier transmission of M-QAM signals are considered. The DPD has to counteract not only the PA nonlinear behavior, but also its dynamics. This may results critical when considering wideband signals since the number of coefficients required to model memory effects can grow dramatically. By applying the PCA technique, the number of essential parameters can be significantly reduced. In addition, a strategy to minimize the computational cost of finding the optimal coefficients is also presented. A test-bed for evaluating the DPD linearization performance of the RF subsystem when PCA is applied was deployed and experimental results are presented in this paper.

3-D Distributed Memory Polynomial Behavioral Model for Concurrent Dual-Band Envelope Tracking Power Amplifier Linearization

“…This is the case of the multi-dimensional Volterra-based models such as the 3D-BBE Volterra model [10] or 2D P-A Volterra model [11]. Alternatively, or in a complementary way, model order reduction techniques such as the ones based on the SVD [21]- [23] or techniques based on the PCA theory [28] are applied without assuming any a priori physical structure of the model.…”

“…An example on how to apply the PCA theory to reduce the model order of a DPD was published by the authors in [28]. With this technique we can perform a change of basis where the number of required coefficients decreases by a certain reduction factor (RDF), i.e., # coeff.…”

“…This may increase the DPD computational complexity, unless additional punning strategies are considered. As reported in [28], further reduction of the terms participating in the linear combination that produces the new basis can be performed without significantly degrading the overall DPD performance. As a consequence the PCA approach can be useful not only to improve the data matrix conditioning but also to reduce the whole DPD system computational complexity.…”

Comparison of model order reduction techniques for digital predistortion of power amplifiers

“…Several model order reduction techniques have been published in recent past years, mostly based on some kind of greedy or iterative search algorithm [2]- [4] that, given a minimization criterion, allows selecting the most relevant basis functions from the original data matrix. Alternatively, model order reduction can be achieved using techniques based on the transformation of the original data matrix into a new basis of orthogonal components [5], [6]. Because the components of the resulting transformed matrix are independent, the adaptation process is significantly simplified.…”

Peak cancellation and digital predistortion of high-order QAM wideband signals for next generation wireless backhaul equipment

“…LxM data matrix (n=1, 2,··· L); with 0 1 λ ≤ ≤ being the weighting factor and where e is the Lx1 vector of the error defined as is the linear gain of the PA.Moreover, to reduce the computational complexity introduced by the DPD block, the principal component analysis (PCA) technique was used. Further details on this model order reduction technique can be found in[3]. Block diagram of the direct learning approach.…”