A Research On Linearization Based On Neural Network In Power Amplifier

RF power amplifier is the key to modern communications equipment. it is widely used in wireless communications, radar, satellite communications, navigation and electronic warfare equipment and other systems. With modern communication technology, especially the third generation of the rapid development of mobile communication systems, people put forward higher requirements about amplifier linearity. As the RF power amplifier often work in higher power efficiency close to saturation region, while these region relatively easy to produce saturation nonlinear distortion, so in order to reasonably resolve this contradiction we must have better technology to improve the power amplifier nonlinear problem.This paper studied the network structure of artificial neural networks, working methods and several commonly used neural network technology, detailing the basic principles of wavelet neural network and wavelet neural network structure. Analysis of the power amplifier's nonlinear characteristics, study and compare The traditional linear power amplifier technology such as power regression method, negative feedback, predistortion and feedforward and so on. Discuss the limitations of these methods. Due to the wavelet network has good localization characteristics and strong approximation and tolerance, and wavelet network convergence speed and simple algorithm, so in this paper application of continuous wavelet network constructing the RF power amplifier wavelet neural network model. The wavelet network model have input, wavelet and output layer. In this paper mainly studied wavelet neural network model of the structure. Wavelet network parameters of initialization method not only reduced training steps but also reduced the possibility of misconvergence. In practice, adopted a gradient algorithm for network parameters adjustment to allow until when error within the range of training, or stop when training steps over the value set up before. The simulation experiment verifies the feasibility of linearization technique.