Recently, the Future Ready Optical Network technology (FRONT) team led by Professor Weisheng Hu of Shanghai Jiao Tong University worked closely with the Microwave Photonics team of Peking University to deal with the challenges of limited capacity and signal-to-noise ratio (SNR) of the cloud radio access network (C-RAN) in post-5G era. The joint research team from the State Key Laboratory of Advanced Optical Communication Systems and Networks creatively proposed a hybrid digital-analog fronthaul architecture based on cloned optical frequency combs, which can simultaneously support the transmission of high-capacity and high-order format wireless signals. On August 17th, 2023, the relevant research results entitled "Clone-comb-enabled high-capacity digital-analogue fronthaul with high-order modulation" formats, was published online in Nature Photonics. Fig. 1. Article screenshot BackgroundAt present, billions of mobile end users around the world rely on high-speed information transmission through wireless communication and optical fiber communication. The convergence architecture of optical fiber and wireless communication has become a challenging direction. As the ‘last mile’ of optical fiber and wireless access network, the fronthaul directly connects the mobile end user and has a crucial impact on the capacity and fidelity of the wireless signal. With the globalization and intergenerational evolution of wireless services, the bandwidth and signal-to-noise ratio of fronthaul need to be further improved. As an emerging technology, hybrid digital-analogue radio-over-fiber architecture can achieve exponential improvement of the signal-to-noise ratio at the expense of linearly increased bandwidth, which combines the advantages of both digital and analog fronthaul solutions. The disadvantages of low spectral efficiency in traditional digital interfaces and insufficient signal-to-noise ratio of analog interfaces are successfully overcome. In parallel, high-speed coherent detection has become the mainstream technology of optical fiber communication. However, due to the bottleneck of carrier phase estimation, the application of coherent detection to digital-analogue radio-over-fiber interface suffers from enormous challenges in both technical and cost aspects. Fig. 2. Digital-analogue radio-over-fiber fronthaul architecture based on cloned optical frequency combs ResultsIn order to solve the bottleneck of coherent detection application, and achieve an order of magnitude improvement in bandwidth and signal-to-noise ratio simultaneously, the joint research teams fully leveraged their research foundation accumulated for many years in radio-over-fiber and optical frequency comb technologies. They proposed the idea of applying the cloned optical comb to the digital-analogue radio-over-fiber fronthaul. Specifically, by cloning an optical comb that is identical to the original optical comb at the remote end, the carrier phase estimation can be directly omitted. Self-homodyne coherent detection of massive parallel channels can be supported simultaneously. Based on the scheme, a digital-analogue radio-over-fiber link with 12 wavelength-division-multiplexing channels is successfully demonstrated. 1024-QAM signal transmission is achieved with 240 GHz aggregated wireless bandwidth. The experimental results show that the common public radio interface (CPRI)-equivalent rate of 1.17 Tb/s in a single wavelength channel exceeds the capacity requirement of 6G fronthaul. At the same time, the joint research team also experimentally demonstrated the transmission of ultra-higher-order modulation formats up to 65536-QAM, which improves the signal-to-noise ratio by more than 600 times compared to typical analog radio-over-fiber interfaces. Through theoretical analysis and experimental validation of the scalability of the cloned optical frequency comb, the joint research team achieved a record 32.8 Tb/s CPRI-equivalent capacity, which can support 5,600 5G new radio (NR) users at the same time. Compared with other technologies that have been publicly reported in the field of optical-wireless access networks, the joint research team has achieved an order-of-magnitude improvement in bandwidth and signal-to-noise ratio, ranking in the leading position among the global research groups. This breakthrough is of great significance and can potentially promote the development of the next generation of radio access networks. Fig. 3. Experimental results of 65536-QAM format transmission; The comparison in terms of bandwidth and SNR. (The obtained results are marked in red and purple symbols) Paper informationThe co-first authors of the paper are Chenbo Zhang (PhD candidate at the School of Electronics, Peking University) and Yixiao Zhu (associate professor at the Department of Electronic Engineering, Shanghai Jiao Tong University). Co-corresponding authors are Xiaopeng Xie (assistant professor at the School of Electronics, Peking University) and Yixiao Zhu (associate professor at Shanghai Jiao Tong University). Professor Weisheng Hu, Professor Lilin Yi, Associate Professor Qunbi Zhuge, and PhD candidate Yicheng Xu (Future Ready Optical Network Technology team of Shanghai Jiao Tong University), Professor Zhangyuan Chen, Professor Weiwei Hu, PhD candidates Bibo He, Jingjing Lin, and Rongwei Liu (Microwave Photonics team of Peking University) all participated in the research work and the writing of the paper. This work is an important result of the cooperation between Shanghai Experimental Area (supported by Shanghai Jiao Tong University) and Beijing Experimental Area (supported by Peking University) of the State Key Laboratory of Advanced Optical Communication Systems and Networks. Link to the paperhttps://www.nature.com/articles/s41566-023-01273-2 Journal InformationNature Photonics is a peer-reviewed top journal dedicated to publishing high-quality research across all areas of photonics, including the generation, manipulation and detection of light, with a 2022 impact factor of 35.0.

Main headline:Important progress made by Prof. Lilin Yi's team at Shanghai Jiao TongUniversity in the rapid and accurate modeling of mode-locked fiber lasersSubheading:This technology can significantly improve the efficiency of laser design andoptimization, and is expected to become a universal method for modelingmode-locked lasers.Prof. Lilin Yi's team at theDepartment of Electronic Engineering at the School of Electronic Informationand Electrical Engineering of Shanghai Jiao Tong University, in collaborationwith Prof. Minglie Hu's team at Tianjin University, used a recurrent neuralnetwork to achieve rapid and accurate modeling of mode-locked fiber lasers. Therunning speed is 146 times faster than the traditional split-step Fouriermethod. The prior information feeding method is proposed to achievegeneralization for parameters excluded by the signal waveform, such as thecavity length and small signal gain (corresponding to pump power), enabling thefast and accurate modeling of different states (soliton, soliton molecule,etc.) evolution. The relevant research results were published in theinternational optics journal Laser & Photonics Reviews in March 2023 underthe title "Fast Predicting the Complex Nonlinear Dynamics of Mode-LockedFiber Laser by a Recurrent Neural Network with Prior Information Feeding."ResearchbackgroundModeling mode-locked fiber lasersis of great significance for laser design and optimization. The traditionalmodeling method based on the split-step Fourier method to solve the nonlinearSchrödinger equation uses small-step iteration calculation, with highcomputational complexity and long operation time. The dynamics of themode-locked laser are very complex due to the coupling of multiple factorsinside the laser (dispersion, nonlinearity, loss, gain). The generation ofsolitons often requires hundreds of round trips inside the cavity, and thepropagation distance of light during the establishment of solitons can reach a kilometerlevel. To achieve long-distance accurate prediction, this study proposes to uselong short-term memory recurrent neural networks (LSTM) to model themode-locked laser. The split-step Fourier method needs to iteratively calculatemultiple times at the set step length to obtain a waveform of one round trip,while LSTM only needs to perform forward propagation once to obtain a waveformof one round trip, thus significantly improving the operation speed.Figure 1. a) The AI model withprior information feeding for intra-cavity evolutionary process inference.SSFM, split-step Fourier method; LSTM, long-short term memory. b) The workflowfor using the AI model.Thecavity length and small signal gain (corresponding to pump power) are the twobasic parameters for evaluating the working state of the mode-locked laser. Theformer determines the laser's repetition rate, while the latter determines thelaser's output state (such as whether it is mode-locked). However, these twoparameters are laser characteristics and are not included in the waveform dataof each round trip. Therefore, how to inform LSTM of the cavity length andsmall signal gain and let LSTM predict different dynamic processes under thesame initial waveform (fixed pulse is used for initial state to accelerate thedynamic process) has become a major challenge. To address this, this studyproposes the Prior Information Feeding method (see Figure 1(a)), which uses afully connected layer to raise the dimension of the cavity length and smallsignal gain parameters, and then adds them to the waveform data to form theinput of the subsequent LSTM layers, thus achieving generalization for thecavity length and small signal gain. When using AI to model mode-locked lasers,first use the split-step Fourier method to produce the dataset, then divide thedataset, use the training set to train the AI, and finally evaluate the AIperformance on the test set.InnovationachievementsFigure 2 shows that AI canaccurately predict the dynamic process of soliton establishment, with theintensity and phase in both the temporal and spectral domains accuratelypredicted, and the spectral beating phenomenon during the process of solitonestablishment accurately restored by AI. By further increasing the small signalgain, the generation of soliton molecules can be observed, as shown in Figure3. AI can also accurately predict the dynamic process of the establishment of solitonmolecules, and even performs well in 500-800 round trips beyond the trainingrange (the training dataset only contains waveform data of 500 round trips),and the prediction deviation of the spacing and relative phase of solitonmolecules is also small (see Figure 3(d)).Figure 2. The soliton formationcomparison between the SSFM and the AI model.Figure 3. The soliton moleculeformation comparison between the SSFM and the AI model.Table1 compares the running time of AI and the split-step Fourier method (500round-trip calculations). The average running time of the split-step Fouriermethod on an AMD Ryzen 7 5800H CPU is 13.145 s. When using a 2-layer LSTM, themodeling error (NRMSE) of AI is only 0.102 and the average running time on anAMD Ryzen 7 5800H CPU is 2.106 s, which is nearly 6 times faster than the split-stepFourier method. Furthermore, when accelerating the 2-layer LSTM AI model withRTX2080Ti GPU and CUDA, the average running time is only 0.09 s, which isnearly 146 times faster than the split-step Fourier method.Table 1. Comparison between AImodels with different numbers of LSTM layers and the SSFMThisstudy demonstrates that AI can be used to achieve fast and accurate modeling ofcomplex mode-locked lasers, while achieving generalization of parameters suchas cavity length and small signal gain. The researchers of this work statedthat this technology can greatly improve the efficiency of laser design andoptimization, and is expected to become a universal method for mode-lockedlaser modeling.PaperInformationAuthors: Guoqing Pu (Postdoctoralresearcher at Shanghai Jiao Tong University), Runmin Liu (PhD student atTianjin University), Hang Yang (PhD student at Shanghai Jiao Tong University),Yongxin Xu (PhD student at Shanghai Jiao Tong University), Weisheng Hu (Prof.at Shanghai Jiao Tong University), Minglie Hu (Prof. at Tianjin University),and Lilin Yi (Prof. at Shanghai Jiao Tong University).Fromleft to right: Guoqing Pu, Lilin Yi, Weisheng HuShanghai Jiao Tong University isthe first affiliation for this research. Guoqing Pu is the first author, and Prof.Lilin Yi is the corresponding author.Funding:This research work was supported by the National Natural Science Foundation ofChina Major Research Instrumentation Project (62227821) and the OutstandingYouth Fund Project (62025503).Paper Link (or click "Read theoriginal article"):https://onlinelibrary.wiley.com/doi/10.1002/lpor.202200363Journal Information:Laser & Photonics Reviews is an internationally peer-reviewed opticaljournal dedicated to publishing high-quality review papers and research work inthe field of optics, with an impact factor of 10.947.Source | Department of ElectronicEngineeringAuthor | Guoqing Pu

The group of Professor Weiwen Zou from the Department of Electronic Engineering of Shanghai Jiaotong University has developed a novel photonic tensor processing chip that realizes high-speed tensor convolution operations. The related results were published in Nature Communications in December 2022 under the title "High-order tensor flow processing using integrated photonic circuits". The research is solely completed by Shanghai Jiao Tong University, with Shaofu Xu, assistant professor, and Jing Wang, Ph.D. student, as co-first authors, and Prof. Weiwen Zou, at the Intelligent Microwave Lightwave Integration Innovation Center (imLic), Department of Electronic Engineering, as the corresponding author.From left to right：Shaofu Xu, Jing Wang, Sicheng Yi, and Weiwen ZouBackgroundsThe tensor form stacked from multidimensional data is an efficient form of data processing, which is conducive to discovering the intrinsic structural features in data and is widely used in radar, communication, artificial intelligence, life science and other fields. With the development of information technology and the explosive growth of data generation rate, the multidimensional stacking of massive data and its efficient and fast processing become an important scientific challenge nowadays. To meet this challenge, traditional electrical processors usually adopt the Generalized Matrix Multiplication (GeMM) strategy to transform high-order tensor operations into matrix multiplication operations. Multi-level looped operations are transformed into parallel operations, so that increasing the number of processing cores is effective to increase the throughput of tensor processing. However, the GeMM strategy relies on a large number of data replications, which requires additional memory usage and repeated communication overhead between memory and processor. It is one of the core bottlenecks to improve the speed of tensor operations on multidimensional data.Innovation resultsThis study proposes an innovative interdisciplinary research idea, using integrated photonic circuits to build a tensor processor. It can take advantage of the high-speed characteristics of photonics to increase the clock frequency of processing to tens of GHz and also can use the multiple degrees of freedom of optics to directly represent multiple dimensions of tensor data. This idea eliminates the need for tensor-to-matrix transformation and enables tensor processing from input to output in a flow fashion (as shown in Figure 1). Based on this idea, the group designed and developed a photonic tensor processing chip (shown in Figure 2), which integrates the three degrees of freedom of optical wavelength, space and time delay. It successfully verified the high-speed tensor convolution operation with a clock frequency of 20 GHz on multi-channel images (shown in Figure 3), with an on-chip computing density of 588 GOPS/mm2. Increasing the scale of photonic integration will upgrade computing density over 1TOPS/mm2. The research group used the chip to build a convolutional neural network for video action recognition. The convolutional layers in the network were completed on the photonic tensor processing chip. A recognition accuracy of 97.9% was finally achieved on the KTH video data set, close to the ideal recognition accuracy of 98.9%. In summary, this research shows that the photonic integrated chip can realize tensor flow processing at ultra-high clock frequency. It solves the problem of extra memory occupation and memory accesses. It provides a new technical way to build advanced information systems such as high-performance computing and broadband signal processing.Figure 1 Architecture of the photonic tensor processing chipFigure 2 The photonic tensor processing chip. (a) Packaged chip module. (b) Microscopic photo of the chip. (c) Wavelength response of the wavelength multiplexer. (d) Wavelength response of the microring array.Figure 3 Result of multi-channel image convolution.Related worksProf. Weiwen Zou's group has conducted in-depth research in the direction of photonics-based computing and has developed the principle of optoelectronic processing and optoelectronic hybrid integration approaches to provide an important foundation for the successful development of this photonic tensor processing chip. Recently, they have conducted a series of researches on the core principle of photonic computing, innovative architecture, and advanced applications. Related papers have been published such as Opt. Express 2022, 30(23): 42057, Opt. Lett. 2022, 47(24): 6409. This research has also contributed to the successful development of the microcomb-based integrated photonic processing unit (led by Prof. Xingjun Wang's group at Peking University with the participation of Prof. Weiwen Zou's group), also published in Nature Communications recently，Nature Communications volume 14, Article number: 66 (2023).Article informationShaofu Xu, Jing Wang, Sicheng Yi, and Weiwen Zou, High-order tensor flow processing using integrated photonic circuits, Nature Communications 13, 7970 (2022), funded by National Key Research and Development Program (2019YFB2203700) and National Natural Science Foundation of China (T2225023, 62205203).Link to the paper: https://www.nature.com/articles/s41467-022-35723-2.Nature Communications is an international peer-reviewed multidisciplinary open access journal dedicated to publishing high-quality research in a variety of fields including biology, medicine, health, physics, chemistry, and earth sciences.Editor on Duty：Zhaoying WuResponsible Editor：Qianqian Jiang