Study On Photonic Reservoir Computing Based On Nonlinear Dynamics Of Semiconductor Lasers

Reservoir computing(RC)has attracted much attention as a framework of neural networks.Different from the complex training of traditional neural networks,RC has simple structure and easy training,so it is widely used in speech recognition,time series prediction,nonlinear channel equalization and other tasks.At present,the research of RC mainly was focused on the time-delay photonic reservoir computing,among which the research of all-optical reservoir computing has been very extensive,and the research of optoelectronic reservoir computing was focused on the structure composed of electro-optic modulator(EOM)as nonlinear elements.On the use of semiconductor lasers(SL)as a nonlinear element,the report on the construction of optoelectronic feedback RC system based on SL was very limited.In this thesis,the optoelectronic feedback RC composed of SL were studied theoretically.Two types of new RC structures were proposed and studied,including parallel RC based on mutually-coupled SLs with optoelectronic feedback,RC based on dual optoelectronic feedback SL.In addition,we also proposed and explored the characteristics of RC with double optical injection and optical feedback.The detailed work and research results are summarized as follows:(1)A new structure of parallel RC for optoelectronic feedback mutually-coupled SLs was proposed.The structure can realize two tasks at the same time(Santa Fe chaotic time series prediction task and nonlinear channel equalization task)parallelly.Numerical simulation shows that,the performance of positive coupling is better than negative coupling,and the optimization of coupling strength depends on the scale factor.Since the states of virtual nodes are richer,it has more advantages than single optoelectronic feedback RC in multi-step prediction task,and the more predicted steps,the more obvious advantage.The influence of parameter mismatch was discussed.Under optimized condition,the lowest estimated rate can reach ?1.668 Gb/s.(2)A new structure of RC based on dual optoelectronic feedback SL was proposed.The influence of delay difference on prediction performance was analyzed.The nonlinear dynamic characteristics of SL system under different normalized current and feedback intensity were analyzed.Finally,the memory capacity of the system was also studied.The results show that,when the feedback intensity is within a certain range,the system can obtain a lower normalized mean square error(NMSE).The dual optoelectronic feedback system has advantages over single optoelectronic feedback system when performing multi-step prediction task.The memory capacity(MC)is very sensitive to normalized current and increases with it,with little dependence on scale factors.When the two feedback intensities are positive at the same time,the MC is better,and the maximum value of 25.2 can be reached.Compared with single optoelectronic feedback system,dual optoelectronic feedback system has certain advantages in term of MC.(3)A new structure of RC based on dual optical injection and optical feedback SL was proposed.The effects of normalized current,feedback intensity,injection intensity,scale factor and frequency detuning on the prediction performance were studied.The results show that,the injection intensity,phase difference,scale factor and frequency detuning have certain symmetry when performing the prediction task and classification task;after optimizing the parameters,one SL can process two different tasks in parallel at the same time,which is of significance to the research of RC system based on optical feedback SL.