Research On Narrowband Interference Suppression In Spread Spectrum System Based On Neural Network

The narrowband interference prediction and rejection technology is used to solve the problem that the strong narrowband interference brings on the system unable to work in the spread spectrum communication. This paper predicts the narrowband interference from the received signal by used of a strongly nonlinear tool-neural network, and counteracts it from the received signal to eliminate the influence of the strong narrowband interference.Firstly, the paper simply summarizes the spread spectrum communication and neural network, mostly discusses the basic principle, working method and main character of the spread spectrum communication, and introduces the principle, character and study arithmetic of the neural network and its application in nonlinear prediction. Then, the paper summarizes the character and generation method of several narrowband interferences, presents the rules to scale the prediction and rejection arithmetic of narrowband interference, and researches various narrowband interference rejection technologies on time-domain. Lastly, the paper mainly introduces the narrowband interference rejection technology based on neural network and its learning arithmetic improvement, and processes simulation experimentation based on Matlab language.Aiming at the fact that the conventional time-domain adaptive linear and nonlinear interference rejection technologies have the disadvantage of large error and slow convergence speed, the paper presents the narrowband rejection technology based on recurrent neural network, then particularly introduces the configuration and real time recurrent learning (RTRL) arithmetic of the recurrent neural network, and lastly gives out the simulation and performance comparison between the RTRL arithmetic and the conventional time-domain interference rejection technology. We discover that the recurrent neural network interference rejection technology based on RTRL arithmetic improves a lot on interference-noise rate (JNR) improvement and signal-noise rate (SNR) lose, but its convergence speed is slow.Aiming at the disadvantage of slow convergence speed of RTRL arithmetic in the recurrent neural network, we suggest using the extended Kalman filter (EKF) learning arithmetic to improve the interference rejection technology based on the recurrent neural network, and process simulation and performance comparison. We discover that the recurrent neural network interference rejection technology based on EKF learning arithmetic improves a lot on JNR improvement and SNR lose, and more important it improves the convergence speed of interference prediction largely. But because of the error from the practical noise statistic character supposition and linearization, the EKF learning arithmetic behaves badly, and even emanative.Aiming at the disadvantage of error from linearization and demand of knowing noise statistic character of EKF learning arithmetic in the recurrent neural network, we suggest using the robust Kalman filter (RKF) learning arithmetic to compensate the error from linearization and unknown noise statistic character, and process simulation and performance comparison. We discover that compared with other interference rejection arithmetic, the recurrent neural network interference rejection technology based on RKF learning arithmetic improves largest on JNR improvement and SNR lose, its convergence speed is soon, and it successfully solve the disadvantage of large error and slow convergence speed of the common time-domain linear and nonlinear interference prediction and rejection technology.