Research On Multi-input And Multi-output Photonic Chaos Reservoir Computing System And Its Applications

The recursive neural network with closed-loop feedback structure can realize the functions of optimization calculation,pattern recognition and associative memory,but it is difficult to train and convergence.Therefore,the classical reservoir computing system,which simplifies the training process,arises at the historic moment and however,the classical reservoir has complex structure,difficult hardware implementation and high cost.As a hardware-based machine learning system,the chaotic reservoir computing system composed of nonlinear nodes and delay feedback loops has excellent performance in pattern recognition,real-time series prediction and other tasks.In addition,since the bandwidth of the photonic system is much larger than that of the electronic system,the optoelectronic chaos reservoir computing systems or all-optical chaos reservoir computing systems designed and implemented by photonic devices,will have a high bandwidth,fast signal processing speed,easy to parallel processing,simple structure,easily implementation by hardware,low energy consumption,simple training process and high cost performance,etc.,and hence,it has broad potential development and application prospects.In the last several years,a couple of studies about photonic chaos reservoir computing system structures and their applications in time series prediction,optical packet header recognition are reported,and however,all are single-channel or at most two-channel information processing system,cannot be applied to the parallel processing of multiple input and multiple output,and can only cope with shorter optical packet headers,and thus,can merely recognize less patterns.Therefore,in this dissertation around the extension of port numbers of photonic chaos reservoir computing systems,and upgrade their capacities of the signal processing and their applications,multiple-input and multiple-output photonic chaos reservoir computing system structures,modeling,performance evaluation,and their applications in signal waveform recognitions,optical packet header recognitions of future optical packet switching networks,digital speech recognitions,and etc.,are explored systematically and deeply.The innovative works of this dissertation are as follow.(1)In view of the waveform recognition algorithms being complicated in current,ultilizing the electronic chaos reservoir computing system based on the Mackey-Glass,nine different waveforms totally of sine wave,triangle wave and square wave with three different frequencies are recognized effectively.In addition,the dynamic characteristics of the time-delayed nonlinear system realized by Mackey-Glass nonlinear equation and the influences of the number of virtual nodes,feedback strength factor and input signal gain on waveform recognition are analyzed.The system has the advantages of simple structure,easy hardware integration,and simple training process.(2)Aiming at the problems of the application of the multi-input and multi-output signal processing and the expansion of processing capacity,a structure and implementation method of multi-input and multi-output optoelectronic chaos reservoir computing system based on the MZM are proposed.In order to verify its effectiveness and application,taking four-input and four-output optoelectronic chaos reservoir computing system,and eight-input and eight-output optoelectronic chaos reservoir computing system as examples,the simultaneous recognitions of multi-channel 3-bit/8-bit/16-bit/32-bit and 128-bit optical packet headers compatible with IPv6 network protocol address length are realized.The influences of the feedback strength factors,the SNR(signal-to-noise ratio)of input signals and other parameters on the signal recognition results are analyzed.After that,the multi-input and multi-output reservoir computing system is used to realize the simultaneous recognitions of four-channel digital speeches.In addition,the optoelectronic chaos reservoir computing system based on the MZM is built experimently.The optimal reservoir states are analyzed by adjusting the parameters,and the 3-bit optical packet header recognition is realized.(3)For the problem of the signal processing speed limited by the electronic devices in the optoelectronic hybrid reservoir computing system,a structure and implementation method of multi-input and multi-output all-optical reservoir computing system composed of a semiconductor laser and multiple optical delayed-feedback loops are proposed.In order to verify its effectiveness and application,taking a three-input and three-output system for example,three channels of 3-bit/8-bit/16-bit/32-bit and 128-bit optical packet headers compatible with IPv6 network protocol address length are recognized at the same time.Furthermore,the influences of the feedback strength factors on the signal recognition results are analyzed,and the best operational state of the reservoir is optimized.Besides,under the optimal values of the system parameters,the simultaneous recognitions of three-channel digital speeches are realized,and the recognition results are compared with those of other reservoir computing systems in references.(4)Using the bidirectional light-emitting characteristics of semiconductor ring lasers,two structures and implementation methods of parallel all-optical reservoir computing systems based on self-feedback and cross-feedback are proposed respectively and their system parameters are optimized by modeling.Then,in order to verify their effectiveness and application,4 bits,8 bits and 32 bits optical packet headers for two channels are recognized simultaneously by these two systems.Next,in view of the problems that the above two parallel all-optical reservoir computing systems can only process two-channel signals and can not process more channels signals in parallel,a structure and implementation method of multi-input and multi-output photonic chaos reservoir computing system based on semiconductor ring laser array are proposed.In order to verify its effectiveness,taking four channels as an example,the simultaneous parallel recognition of 8-bit optical packet headers is realized.The system can process all-optical parallel information and has the advantages of fast signal processing speed and easy integration and etc.