| As an vital device in the wireless communication system,power amplifier(PA)has inherent nonlinear characteristics,and the distortion caused by it will cause interference to adjacent channels of the communication system and increase the signal error rate.Digital Predistortion(DPD)technology,as one of the mainstream linearization technologies,has been widely used because of its low cost,flexible configuration,and good linearization performance.In order to meet the increasing data transmission needs of users,wireless communication systems are gradually developing towards high-frequency bands and large bandwidths.The peak-average ratio of the PA under the excitation of complex modulation signals shows strong nonlinear characteristics.The traditional DPD model is difficult.To efficiently fit the strong nonlinearity of PA,a two-stage cascaded DPD method is proposed.Compared with one stage DPD method,the multistage DPD method that adopts a cascade of several models can further compensate the strong nonlinear distortions.However,the multistage method requires more complex identification process.To accelerate the identification process and improve the generalization performance,we propose a novel two stage DPD based on generalized advanced neural network(GANN)model.In this framework,GANN model is inserted into the DPD architecture as the first stage.The residual nonlinearity of the power amplifier(PA)is compensated by a simplified GMP model in the second stage.The generalization of neural networks can be effectively improved by injecting the embedding vector into GANN.The proposed method is performed on a Doherty PA driven by the orthogonal frequency division multiplexing(OFDM)signals for experimental validation.The proposed two stage DPD based on GANN is validated to compensate strong nonlinear distortions within low output power back-off and achieve the similar good linearization as the other multistage method while having low complexity of the identification process. |
PDF Full Text Request