Research On Digital Predistortion Technology Based On Wavelet Network In WCDMA System

In the mobile communication system, in order to ensure the signal coverage, a RF power amplifier is usually used to amplify the signal at the base stations and repeaters. To ensure the efficiency, a power amplifier usually works at the nonlinear area near the saturation point, therefore the transmitted signal would be distorted severely. As a result, the bit error rate and adjacent-channel interference would increase. In the GSM system, the constant envelope modulation mode like GMSK is used, thus the signal is not sensitive to nonlinear distortion. But in the WCDMA system, in order to improve the spectrum efficiency, some modulation modes, such as QAM and QPSK, and code division multiple access transmission technology are widely used, which leads to wide-band and high peak-to-average power ratio (PAPR) signal. Therefore the WCDMA system is very sensitive to nonelinearities caused by power amplifiers, and highly linear RF power amplifiers is needed in the transmitter. Consequently, the study on the linearization techniques of power amplifiers is of great significance.In the linearization techniques of power amplifiers, digital predistortion technique has advantages like good linearity, suitable for wide bandwidth, high stability and flexibility, so it is considered to be the most promising linearization technique. The basic principle of the technology is to build a predistorter reciprocal to the characteristic of the power amplifier at the front-end of the power amplifier, which is used to compensate the distortion produced by the nonlinearity of the power amplifiers, thus realizing the linearization of the whole system. So the predistorter can be regarded as a nonlinear system which is used to realize the nonlinear mapping from input to output. The wavelet neural network(WNN) combines the good time-frequency localization feature of the wavelet transform and the excellent self-learning ability of the neural network, which has strong capability in nonlinear approximation. So this paper considers to apply the wavelet neural network to the modeling of the predistortion. As the power amplifiers in the WCDMA system has memory effect, the current output not only depends on the current input, but also depends on the states of the history signal. So this paper designed a time-delayed wavelet neural network predistortion model. In addition, as the tranditional back propagation (BP) algorithm can easily get into local minimum. Besides, its convergence speed is slow and it often results in oscillating, so an improved BP algorithm was used to train the network.In order to evaluate the performance of the proposed algorithm in the paper, the WNN-based digital predistortion algorithm was simulated and compared with BP neural network-based digital predistortion algorithm. The simulation results showed that compared with the system without predistortion, the adjacent-channel power of the WNN-based digital predistortion system is decreased by17dB. Compared with BP neural network-based digital predistortion system, the adjacent-channel power is decredased by7dB. In addition, and the predistortion algorithm based on the wavelet network has better precision.