Research On The Programmable Electro-optic Frequency Comb Inversely Designed By Deep Learning

Optical Frequency Comb(OFC)is a spectrum composed of a series of discrete and equally spaced frequency components,shaped like a comb.OFC is regarded as one of the major research hotspots in the field of optoelectronic information in recent years,which has produced strong driving force and far-reaching influence in the fields of optical communication,optical precision measurement,microwave photonics,and optical sensing.However,OFC usually needs to be designed for different application scenarios.The current OFC generation technology has problems such as low design efficiency,difficulty in obtaining optimal design effect,and lack of programmability of spectrum(comb-line number,comb space,comb-line power,flatness),etc.,which make it difficult to apply to different application scenarios.This paper focuses on the efficient inverse design method and programmable generation technology of OFC,and the specific work is as follows.Firstly,an inverse design method of OFC based on deep learning is proposed,which realizes the efficient inverse design of OFC and achieves better design effect.The working principle of the inverse design scheme is introduced in detail,a perfect physical model is established,and a cascade neural network is constructed.By using the constructed inverse network,the corresponding parameters can be inversely designed in real time according to the desired target OFC.The effect of inverse design and the programmability of OFC are studied by OptiSystem,an optical communication simulation software.The results show that the method of inversely designing OFC based on deep learning can be used to inversely design OFC in 0.1 s and improve the design efficiency of OFC.The generated OFC is highly consistent with the target OFC,and the comb-line number can up to 7 with the flatness of 0.33 d B,which improves the design effect of OFC.The spectral characteristics of OFC,such as comb-line number,comb-line power,side mode suppression ratio and comb spacing,are programmable,which improve the performance of generated OFC.The method can also be used in more complex OFC generators,which provides a new idea for efficient design of OFC.Secondly,a method of generating OFC by an electro-optic modulator driven by arbitrary waveform is proposed,which is inversely designed based on deep learning.The programmable OFC with greater freedom is realized,and the comb-line number is greatly increased.The working principle of the scheme is introduced in detail.A physical model is established,and the influence of the abundant amplitude and phase information in arbitrary waveform on OFC is analyzed.Theoretically,by setting the amplitude and phase of each harmonic in arbitrary waveform reasonably,the OFC with strong programmability and good flatness can be obtained.A dynamic input neural network integrated with physical model is constructed,which reduces the size of the neural network and improves the universality of the model under different target inputs.The trained inverse network can be used to inversely design arbitrary waveform in real time according to the desired target OFC.Matlab is used to study the programmable characteristics of OFC generated by inverse design method,and the error analysis is carried out.The results show that the OFC generation technology of electrooptical modulator driven by arbitrary waveform based on deep learning can generate 501 comb lines with the flatness of 3.14 d B,which further improves the efficiency and effect of inverse design of OFC.The generated OFC is more programmable in terms of comb-line number,comb-line power,and comb spacing.The OFC generator and inverse design method realize the digital controllability of OFC.The highly programmable feature of OFC makes it flexible to be applied in various scenarios and has a broad application prospect.