The Research Of Digital Pre-distortion Technology On Power Amplifier With Cascade Neural Networks

This paper belongs to the category of digital predistortion, neural network is applied to the communication system, using cascade BP neural network to fit the model of power amplifier. The traditional BP neural network structure were optimized. The traditional BP neural network the steepest descent method is improved, effective solution to the traditional BP neural network in the pre distortion design in learning efficiency is not high, the improved neural network learning speed and accuracy of model fitting.Aiming at the problems encountered in the process of cascade BP neural network to fit the power amplifier in this paper, the following solutions and test methods:1. BP neural network learning efficiency is low, unable to meet part of the pre distortion is designed for the demand of practical application, this paper introduces the concept of quasi Newton method, in the modeling of the power amplifier model, iterative cycle from 538, reduce to less than 50 times.2. In view of the neural network structure is more complex, in its implementation will consumes more computation and storage resources, this paper structure of neural network in unnecessary were reasonable deletion, and the structure of neural network is optimized, will cascade BP neural network resource consumption is reduced by more than one-third.3. In the validation of the improved BP neural network, the cascade BP ne ural network is used to model the memory polynomial power amplifier model, and the actual power amplifier model which is used to identify the memory is used to simulate the power amplifier. In the bandwidth of WCDMA 5MHz signal processing, ACPR can effectively be reduced from-30 dB to-57 dB, and the NMSE below-50.3155 dB.4. In the class E power amplifier for a test platform based on experiments, the output power of the power amplifier is set at about 32 d Bm. The center frequency of the input signal set 1.7GHz and 2.0GHz, 2.2GHz, broadband signal bandwidth setting 5MHz, and the experimental results show that, the improved cascade BP neural network can effectively ACPR from-32.2dB reduced to below-47.4dB.