Synchronization And State Estimation For Feedback Neural Networks With Limited Transmission Channels

Neural network is a new interdiscipline involving many fields such as biology,mathematics,computer,electronics and so on,which has broad development prospects in signal processing,pattern recognition,automatic control,secure communication,artificial intelligence and other fields.Since the network communication has the advantages of rapid speed,low cost and high reliability,it is common to use the network to realize the data exchange in the synchronization and state estimation for neural networks.The network bandwidth is usually limited,which would lead to a series of networked phenomena,such as transmission delays,packet dropouts,quantizations,communication protocols and so on.Therefore,this paper aims to investigate the synchronization and state estimation for neural networks with limited transmission channels.The main work of this paper can be summarized into the following three parts:1.Based on the master-slave method,the synchronization problem for neural networks with network-induced phenomena is studied.More specifically,on the one hand,the H_?synchronization for a class of continuous neural networks with network-induced delays and packet dropouts is investigated.For the purpose of reducing the conservatism,Wirtinger inequality is employed to deal with the delayed terms of neural networks.Finally,the delay-dependent synchronization criterions are derived.On the other hand,the synchronization method for discrete-time neural networks with uniform quantizations,packet dropouts and norm-bounded disturbances is designed.Several sufficient conditions are given,which ensure that the synchronization errors are exponentially ultimately bounded in mean square.Furthermore,in the example simulation part,the relationship between the upper bound of the synchronization errors and quantizations,packet dropouts is analyzed.2.In order to prevent data collisions and reduce the communication burden,communication protocols are introduced into the state estimation problem for neural networks and the following two questions are investigated.The first one is the state estimation for discrete-time neural networks subject to stochastic protocol,where the neural network model contains energy bounded disturbances,state-and disturbance-dependent noises.Moreover,the phenomenon of estimator parameter imprecision is considered,which is induced by numerical roundoff errors,tuning uncertainties and so on.The second one is the state estimation problem for coupled neural networks with Round-Robin protocol,which simultaneously considers time-varying parameters,multiplicative noises and random coupling strengths.By using of recursive Riccati difference equations,the time-varying estimator is derived to guarantee that the estimation error satisfies the H_?performance constraint over a finite horizon.3.Combining the synchronization technique of neural networks,a chaotic secret communication method is investigated.Firstly,according to the known mathematical model of chaotic neural network,the chaotic signal generator is designed.Then,by means of matrix inequality technique,the parameters of synchronization controller are obtained.Finally,based on the synchronization of neural networks,the chaotic hiding secret communication scheme is designed.By using of Multisim,the effectiveness of the designed circuit and secret communication method is validated.