Research On Adaptive Photonic Reservoir Computing And Its Application

Recently,with the rapid progression of the information society,the requirement for improving information processing rates and reducing power consumption is increasing.There is an urgent need for an efficient and energy-saving information processing architecture.Reservoir computing(RC),which utilizes the high-dimensional dynamic characteristics of optical nonlinear systems,is a novel brain-heuristic computing method that is easy to train and implement.Due to the inherent advantages of light in information processing,such as ultra-fast processing speed,ultra-high bandwidth and parallel processing,reservoir computing has been a frontier hotpot of information science and neuroscience,and extensively applied in time series prediction,wireless channel equalizing,automatic control and other areas.However,the shortcomings of RC are gradually exposed: reservoir computing leverages the virtual node states to implement a high-dimensional mapping of the input information,and its performance greatly relies on the training method adopted to obtain the readout weights.Up to present,the least squares and the ridge regression are commonly used in RC,and well performance of reservoir computing can be obtained.Since the readout weights remain invariable once the training is completed,these RC structures are only applicable to parameter-invariant tasks.While for the parameter-variant tasks,even though a satisfactory initial performance can be achieved,the long-term performance would deteriorate if RC cannot update the readout weights according to the variation of task states.Aiming at the shortcomings of traditional RC in terms of computing performance and adaptability,this thesis conducts research on adaptive photonic reservoir computing and its application.The main research contents are as follows:1.An adaptive photonic RC structure based on Kalman filtering(KF)algorithm is proposed.In this scheme,KF algorithm is used as the training method of RC,and the readout weights can be adaptively updated according to the task states.With respect to the conventional RC structure,the adaptive photonic RC structure based on KF algorithm shows better prediction and equalization performance in the time-series prediction task and nonlinear channel equalization task.Meanwhile,since the readout weights can be updated adaptively according to the variation of channel states,the proposed RC structure can process the parameter-variant tasks effectively.2.By optimizing chaos through bandwidth enhancement and delay suppression techniques,a bandwidth,randomness and complexity enhanced random chaotic mask(Echaotic mask)is generated experimentally.Leveraging it as the input mask signal,the response laser can have faster transient response and richer dynamic characteristics,as such the high-dimensional mapping ability of RC is improved.3.In addition,the adaptive photonic RC structure based on KF algorithm and Echaotic mask is used as a nonlinear equalizer,and the application of reservoir computing to the equalization of the optical single sideband(SSB)transmission system is studied and analyzed.An optical SSB PAM4 transmission system based on DDMZM is established,and the equalization performance of different equalizers is compared.It is verified the proposed photonic RC structure can effectively compensate the linear and nonlinear interference in an optical transmission system,as such the bit error rate of the transmission system is reduced.