A new sparse design framework for broadband power amplifier behavioral modeling and digital predistortion

Abstract

In this paper, we propose a new sparse framework for the design of the behavioral model and digital predistorter of a broadband power amplifier (PA). We start by formulating the Volterra kernel to multidimensional memory polynomial by considering the high‐order dynamic truncation of the Volterra model. Then we show how an estimate of the most significant coefficients may be obtained using a matching pursuit (MPT) algorithm by exploiting the sparsity of the model. After the indices of the nonzero coefficients are roughly estimated, the block exact Householder inverse QR‐decomposition‐based recursive least squares (HIQRD‐RLS) algorithm is utilized to estimate the sparse model complex coefficients. For broadband nonlinear PAs, the proposed approach is demonstrated to achieve the best performance among the well‐known traditional approaches in terms of in‐band and out‐of‐band specifications. The proposed approach is also validated by evaluating the digital predistortion (DPD) performance on a Class‐AB PA in terms of adjacent channel power ratio (ACPR). © 2014 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.

Compressive sensing‐based adaptive sparse predistorter design for power amplifier linearization

“…Finally, to compare the usefulness and effectiveness of the SGD-based adaptive SP greedy algorithm, the batch mode CS pruning technique called RSAMP 25 is also used for DPD model performance evaluation. Contrary to the batch mode CS pruning techniques, [15][16][17]25 the SGD-based SP greedy algorithm can achieve similar sparsity level adaptively in the unstructured pruning manner, and suitable for adaptive DPD application. According to the results calculated in Table 2, it is worthy of note that the sparse G2DMP model consists of only 60 coefficients and led to the more better NMSE performance in comparison with the other models.…”

“…13 The basis pursuit and greedy algorithms are the 2 major algorithmic approaches for sparse signal reconstruction. [15][16][17] However, these reported CS greedy algorithms in their ordinary mode of operation have an inherent batch mode of operation, and hence are not suitable for adaptive DPD schemes. 14 Inherently, the PA systems admit sparse representations; that is, the PA models and DPD models can be characterized by a small number of nonzero parameters.…”

“…As one of the efficient greedy algorithms, the novel subspace pursuit (SP) method can provide low computational complexity and high reconstruction accuracy as well as the robustness to cope with inaccurate measurement. [15][16][17] In practice, to properly compensate for the distortion caused by the characteristics change of the PAs in a short time period, the real-time adaptive parameter updating of the DPD must be conducted. Because most of the parameters do not play an important role in PA modeling and DPD, we need to search and select the most significant ones among them.…”

Power amplifier behavioral model adaptive pruning using conjugate gradient‐based greedy algorithm

“…Power amplifier (PA) behavioral model pruning is well recognized to be a challenging problem in the communication field [1][2][3]. To identify such systems, researchers often try function approximations using mathematical models like the Volterra or Wiener series.…”

“…Recently, compressed sensing (CS) theory has gained a significant amount of research interest from the sparse nonlinear system identification perspective. For example, the CS-based greedy algorithm is used for constructing a variety of behavioral models and digital predistorters of PA for pruning their model terms, which can significantly reduce the number of coefficients while achieving comparable performance in Refs [1][2][3]. However, the inherent batch mode of these algorithms limits their application in adaptive environments because of absent tracking capabilities.…”