Chaotic synchronization based on neural filter

Because of the high sensitivity of chaotic systems to their initial conditions, synchronization of chaotic systems with uncertain parameters has been a challenging problem especially in noisy environment. Since synchronization of the transmitter and receiver systems involves recursive estimation, recursive nonlinear filters are called for and the extended Kalman (EKF) filter and unscented Kalman (UKF) filter have been applied. However, such suboptimal filters incur high synchronization errors and provide no capacity for uncertain environment, which motivated the use of the neural filter for chaotic synchronization in this paper. The neural filter, which is a recurrent neural network, can approximate the minimum-variance to any degree. Furthermore, the neural filter can adapt to a uncertain environment without online filter weight adjustment, which is computationally efficient. Numerical experiments show that the chaotic synchronization scheme based on the neural filter outperforms those based on EKF and UKF by a large margin.