Radio Frequency Nonlinear System Modeling And Structure Optimization Based On Deep Learning

Nonlinearity of RF power amplifier is one of the core problems in modern communication system.Because the signal of the new generation communication system has the characteristics of complex modulation,high peak-to-average power ratio and large bandwidth,the non-linearity of power amplifier is aggravated,which further affects the efficiency and quality of the system output signal.At the same time,due to the rapid development of deep learning technology in recent years,which can effectively cope with massive data and extract the high-dimensional characteristics of data,deep learning technology has received wide attention in the physical layer of wireless communication.Therefore,for the RF power amplifier with strong non-linearity in the communication system,a deep learning technique is proposed to model and linearize the RF system.The main purpose of this paper is to complete an adaptive digital pre-distortion method based on lookup table using in-depth learning technology.The basis of the full text is to obtain a high-precision power amplifier behavior model.Therefore,this paper first studies the non-linear characteristics of power amplifier,determines the influence of non-linearity of power amplifier on output signal,and models and extracts parameters for strong memory effect.Secondly,in order to alleviate the phase ambiguity in the signal transmission process,a phase correction module based on one-dimensional convolution neural network is proposed to preprocess the digital pre-distortion input data.Then,according to the reciprocity of the external characteristics between the power amplifier and the digital pre-distortion behavior model,the same model structure is shared with the power amplifier in the process of extracting the digital pre-distortion parameters.Finally,an adaptive digital pre-distortion method based on the lookup table method is designed.In order to increase the number of digital pre-distortion behavior models in the lookup table,this paper uses ADS simulation software to obtain signals from other amplifiers and establishes corresponding digital pre-distortion behavior models.Finally,based on the one-dimensional convolution neural network,the matching method between the input signal of the linearized system and the model in the lookup table is completed,and the target of adaptive digital pre-distortion for the input signal of various types of power amplifiers is achieved.At the same time,the method proposed in this paper also lays a foundation for the further application of machine learning technology in the physical layer of wireless communication.