一种基于维度加权盲K近邻算法的数字预失真技术

A Digital Predistortion Technique Based on the Dimension Weighted Blind K-Nearest Neighbor Algorithm

传统的数字预失真(DPD)模型通常在所有的输入信号功率上采用单一多项式模型和单一记忆深度对功率放大器(PA)进行线性化矫正。然而,功率放大器在不同的功率水平下会呈现出不同的非线性特性,并产生不同的记忆效应。针对这一问题,该文提出一种基于维度加权盲K近邻(KNN)算法的数字预失真模型,所提模型根据功放当前输入信号以及记忆输入信号的幅度进行维度加权的KNN分类,组成维度加权盲KNN记忆多项式(WKMP)模型,并为每一类输入信号序列建立子模型。所提方法用Doherty功率放大器进行实验验证,使用带宽为30 MHz、频点为2.2 GHz的3载波长期演进(LTE)信号作为输入,反馈端使用122.88 MHz的采样率进行采样。实验结果表明,所提维度加权盲KNN分类方法与记忆多项式(MP)模型结合时,功放正向建模效果和数字预失真效果均超过了广义记忆多项式(GMP)模型,并远超记忆多项式模型的效果,实验结果验证了所提模型的优良性能。

In traditional Digital PreDistortion(DPD) models, the same set of polynomial models and the same memory model are usually used to linearize the Power Amplifier(PA) at all input signal powers. However, the PAs exhibit different nonlinear characteristics and different memory effects at different power levels. In order to solve this problem, a DPD model based on the blind K-Nearest Neighbor(KNN) algorithm with dimension weighting is proposed. The input signal sequence is classified by the proposed model according to the magnitudes of amplifier’s current input signal and the memory input signal with the dimension-weighted KNN classification.And sub-models are established for each type of input signal sequence. The proposed method is verified experimentally by a Doherty PA, a three carrier Long Term Evolution(LTE) signal with a bandwidth of 30 MHz and a frequency point of 2.2 GHz is used as the input, the feedback channel is sampled using a sampling rate of 122.88 MHz. When the dimensional-weighted blind KNN classification method is combined with the Memory Polynomial(MP) model, the forward modeling performance and digital pre-distortion performance for the PA which exceed the performance of Generalized Memory Polynomial(GMP) model and MP model are manifested in the experiment. The excellent performance of the proposed model is verified in the experiment.