This paper proposes a federated learning-based mobility-aware DBA for industrial AGVs. Under 5 Gbps uplink, it achieves near-zero delay and handover interruption, outperforming conventional schemes. Under tighter bandwidth, latency increases but remains lower than benchmarks.

With the continuous expansion of network scale and the explosive growth of 5G applications, new demands and challenges are encountered in network management and operations. Since operational efficiency directly influenced network utilization and service quality, it becomes imperative to enhance the level of network management through intelligent means and reduce traditional inefficient and repetitive tasks. As AI technology became deeply integrated with communication networks, the introduction of large-scale communication network models is recognized as a key pathway to promote network management and operations. In response, an intelligent operational human-computer interaction system based on a large language models (LLMs) was developed. This system integrated capabilities such as knowledge-based question answering, human-computer interaction, data analysis, and solution generation through a collaboration mechanism between large and small models. Deployment and application of the system in live networks demonstrated that it not only significantly improved operational efficiency and reduced maintenance costs, but also minimized network failures through predictive maintenance, thereby enhancing user experience and strengthening corporate competitiveness. The system was designed with high replicability and adaptability, indicating broad application prospects and practical value.

We simulated real network scenarios to test the transmission capability of the current 800G OTN equipment. Experimental results show that current 800G OTN equipment can meet the requirements of data center interconnection.

With the exponential growth of global data, a higher requirement for traditional communication capacity is proposed. With the continuous improvement of fiber manufacturing technology, the capacity of traditional solid core single-mode fiber has approached its limitation. Hollow-Core Fiber (HCF) uses air core to guide light and has excellent properties, such as full-band low loss, ultra-low nonlinearity, ultra-low dispersion and ultra-low latency, which make it become one of the strong candidates for the upgrading of the fiber communication industry. Therefore, this article introduces the development of HCF and corresponding technologies such as parameter testing and transmission technology. It may provide reference for the high-capacity, long-distance, and low latency upgrade of the next generation optical communication systems.

Accurate and generalized channel modeling is a fundamental requirement for the design and optimization of a next generation hollow-core fiber (HCF) communication systems. Neural networks have recently emerged as powerful tools for channel emulation, but existing models may still be limited in fully characterizing the unique stochastic impairments of HCFs. These impairments include phase noise arising from inter-modal interference (IMI) and surface scattering. Conventional real-valued neural networks typically process the in-phase and quadrature (I/Q) components of the optical signal as independent variables. This separation may reduce the ability of the network to capture the intrinsic phase-related characteristics of the channel. In this paper, we propose a complex-valued diffusion (CV-diffusion) model for high-fidelity channel emulation. By treating the I/Q components as a unified complex entity, the emulator preserves phase correlations and jointly captures amplitude-phase distortions. In addition, the progressive diffusion formulation decomposes channel modeling into tractable steps, improving generalization across varying channel conditions. An experiment was conducted in a 20-Gbaud dual-polarization 64-quadrature amplitude modulation (DP-64QAM) 40-channel wavelength-division multiplexed (WDM) hollow-core-fiber (HCF) communication system, and the results showed that the proposed CV-diffusion emulator outperformed a real-valued conditional generative adversarial network (CGAN) emulator and a real-valued diffusion (RV-diffusion) emulator. At an OSNR of 22.5 dB, the proposed CV-diffusion emulator achieved its maximum performance gain, with modeling-accuracy improvements of 38.6% and 31.1% compared with the CGAN and RV-diffusion emulators, respectively. Our experimental results demonstrate that the proposed CV-diffusion emulator is a promising candidate for accurate and generalized channel modeling in HCF optical communication systems.

Significance Against the background of global digital transformation, as the solid foundation of new digital infrastructure, optical networks have great impact on the secure and reliable operation of upper-layer service networks in terms of robustness and security. AI applications, particularly those powered by large language model (LLM), have driven a significant surge in demand for intelligent computing power. This necessitates the establishment of an intelligent computing network that fosters synergy between computing and networking through highly-efficient collaboration between computing resources and network infrastructure to enhance computing capabilities. However, the increasing global network incidents, the escalating international cybersecurity events, and the restructuring of global industrial supply chains are collectively bringing new and great challenges to optical network resilience. We need to build a resilient optical network in all dimensions and multiple levels, characterized by high reliability, high security, robust protection, and independent controllability, to ensure that the optical network maintains its functionality and guaranteed service quality when facing various potential failures. Progress Based on the evolving requirements of optical networks, trends in technological development, and the diverse challenges they face, we propose the concept of resilient optical networks from multiple perspectives such as high reliability and security, and design a comprehensive, multi-layered architectural framework for resilient optical networks. Under the architectural framework, we provide a detailed analysis of several key technical advancements and innovation practices, including: a highly reliable optical transport technology resilient to multi-point concurrent failures in cross-layer, multi-domain, and large-scale networking scenarios, which can achieve 99.9999% availability; OTN framing and encryption integrated transmission technology, enabling encrypted service transmission at rates of 400G/800G; an optical link threat proactive identification technology achieved by the convergence of optical communication and sensing, enabling a proactive prevention mechanism for optical networks; physical layer encryption based on channel noise, enabling endogenous security protection for optical networks; multi-parameter status awareness technology in real time enables on-demand service assurance for optical networks aimed at differentiated requirements; lossless transmission technology enables intelligent computing power long-distance transmission and efficient collaborative training; the independent controllability of key technologies from chips and components to equipment can significantly enhance the safety of optical network supply chain. Conclusions and Prospects Optical networks underpin global interconnectivity, enabling everyone to access and utilize the internet safely, affordably and efficiently. In the face of increasingly severe threats, such as extreme weather events, earthquakes, even man-made damages, many network systems remain fragile and vulnerable. Therefore, it is essential to comprehensively enhance resilience of optical network across multiple aspects, including optical fiber cables, chip and equipment, optical transmission, networking, network management and control, to build highly reliable, secure, robust and resilient optical network with independently controllable supply chain. This will continuously strengthen the robustness of information and communication networks, ensuring high-level operational security of optical networks to support the high-quality development of the digital economy.

In this work, a field-trial digital-twin assisted optical transmission performance optimization in a C+L-band 400G 16QAM optical system is carried out. After optimization using the GN model and gradient descent algorithm, both OSNR and BER exhibit improved and flatter performance.

Significance The exponential growth in global data traffic,driven by the rapid development of the digital economy and diversified emerging services,presents unprecedented challenges and opportunities for optical transport networks(OTNs)as fundamental communication infrastructure.Traditional fixed-rate,fixed-grid optical transport technologies demonstrate limitations in spectrum utilization efficiency and service adaptation flexibility,rendering them inadequate for future differentiated service provisioning requirements.Flexible-rate grid-less optical transport technology has emerged as a solution,enabling next-generation high-efficiency,intelligent,and elastic optical transport networks through advancements in key technologies,including multi-format signal generation,flexible-rate signal reception,and multi-granularity elastic wavelength switching. Progress This paper systematically examines the core innovations of flexible-rate grid-less optical transport technology through an in-depth exploration of three dimensions:the transmitter,receiver,and switching node. At the transmitter side,traditional optical networks utilize fixed modulation formats and coding rates,limiting their adaptability to flexible-rate requirements of diverse services.Flexible coded modulation technology addresses this limitation by enabling on-demand transmission of multi-format optical signals through dynamic adjustment of multi-dimensional parameters,including modulation order(e.g.,QPSK,16QAM,64QAM),symbol rate,and forward error correction(FEC)coding rate.The core technologies comprise flexible coding and flexible modulation.Flexible coding dynamically adjusts coding parameters and strategies according to transmission conditions and service requirements,ensuring efficient and reliable optical signal transmission.Flexible modulation incorporates probabilistic constellation shaping(PCS)and time-domain hybrid modulation(TDHM).PCS optimizes power efficiency,enabling flexible rate adaptation without bandwidth increase,while TDHM achieves high-efficiency transmission and dynamic adaptation by combining different modulation signals in the time domain.These technologies enable optimal selection of modulation formats and coding schemes under varying transmission distances and channel conditions,establishing the foundation for dynamic and flexible optical signal generation. At the receiver side,traditional optical transport equipment faces limitations due to single-rate reception,restricting effective processing of diverse modulation formats.Modulation format identification(MFI)and adaptive equalization technologies serve as essential components for enhancing receiver processing capability and adaptability.MFI technology precisely analyzes received optical signal characteristics to automatically identify modulation formats(e.g.,QPSK,16QAM,or 64QAM),providing vital information for subsequent signal processing.Adaptive equalization employs advanced digital signal processing(DSP)algorithms to dynamically compensate for signal impairments caused by chromatic dispersion,polarization mode dispersion,or nonlinear effects during transmission.Real-time adjustment of equalization parameters enables the receiver to optimize demodulation schemes based on signal characteristics and channel conditions,ensuring efficient signal recovery across diverse transmission scenarios.Additionally,deep learning-based MFI methods enhance identification accuracy and real-time performance,providing intelligent solutions for flexible-rate optical signal reception. In terms of switching nodes,traditional fixed-grid wavelength-division multiplexing(WDM)systems exhibit limitations in bandwidth allocation and increasing spectrum fragmentation.Grid-less flexible optical switching technology overcomes these constraints through dynamic scheduling and reconfiguration of wavelength-level or sub-wavelength-level optical paths,eliminating conventional 50 GHz fixed channel spacing restrictions and enabling dynamic adjustment of channel bandwidth and center frequency.This technology utilizes advanced devices such as optical cross-connects(OXCs)and wavelength-selective switches(WSSs),combined with intelligent routing algorithms and resource virtualization techniques,substantially improving spectrum utilization and service adaptability.A software-defined networking(SDN)-based centralized control mechanism enhances switching node intelligence and automation,enabling real-time network traffic variation perception and dynamic spectrum resource allocation.Channel bandwidth can be expanded for high-throughput scenarios requiring high-order modulation signals,while narrower channels improve spectral efficiency for low-rate services.This on-demand flexibility minimizes resource wastage and strengthens network support for diversified services. Conclusions and Prospects The optical transmission network increasingly aggregates and carries massive differentiated data generated from various application scenarios,driven by rapid emerging business development.Flexible rate grid-less optical transmission technology represents a crucial enabler for all-optical infrastructure.The convergence with cutting-edge technologies such as quantum communications and photonic neural networks may transform optical networks from"elastic adaptation"to"cognitive autonomy".However,challenges persist in addressing energy efficiency degradation from dynamic modulation and coding strategy switching,real-time optimization of multi-rate adaptive equalization algorithms,and coordinated operation of high-dimensional optical switching nodes.Academia and industry must strengthen collaboration in device fabrication,algorithm architecture,and operational frameworks to advance optical communication networks toward intelligent all-optical networking.

In this Letter, we propose a novel, to the best of our knowledge, end-to-end (E2E) learning scheme leveraging a time-frequency decoupling network (TFDnet) for joint probabilistic shaping (PS) and pre-equalization in hollow-core fiber (HCF)-based wavelength division multiplexing (WDM) systems. The TFDnet emulator effectively models HCF transmission channels by decoupling signal impairments into high-frequency, linear, and nonlinear distortions. Furthermore, a TFDnet emulator-based E2E strategy for joint PS and pre-equalization is presented with the aim of compensating the signal impairment for the HCF-based WDM systems. An experiment is conducted on a 30-channel HCF-based WDM system over a 10 km HCF. The experimental results demonstrate that the proposed TFDnet-based joint PS and pre-equalization scheme achieves the same bit-error rate (BER) performance with optical signal-to-noise ratio (OSNR) improvements of 1.0 dB and 1.6 dB compared to conditional generative adversarial network (CGAN)-based and traditional joint PS and pre-equalization scheme, respectively, under a 20% hard-decision forward error correction (HD-FEC) threshold. These results highlight the potential of the proposed scheme for ultrahigh-capacity HCF communication systems.

With the increase sizes of training datasets and models, the bottleneck in distributed machine learning (DML) training has shifted from computation to communication. To address this bottleneck, we propose an all-optical switching network architecture for accelerating the communication phase of DML training. Experimental results validate packets with error-free and 385 ns server-to-server low-latency communication at traffic load of 0.9. Small-scale DML training experiment deployed in the proposed architecture shows that Resnet50, Resnet101, and Vgg19 can be accelerated by 1.16x to 1.48x compared to electrical switching network. The proposed architecture demonstrates a 58.9% enhancement in cost efficiency and a 60.9% improvement in power efficiency compared to the 3-tier fat-tree architecture.

In this letter, we demonstrate a real-time S+C+L band optical transmission system with a total transmission capacity of 84 Tbit/s. 105 wavelength division multiplexing (WDM) channels with 150 GHz grid are used to transmit 135 GBaud probabilistic constellation shaping 16 quadrature amplitude modulation (PCS-16QAM) signals. The system only uses doped fiber amplifiers for signal amplification. We adopt self-developed S-band integrated tunable laser assembly (ITLA) with high output power to generate S-band signal. The real-time net data rate of S+C+L band is 800 Gbit/s and the total transmission distance is 300 km. To the best of our knowledge, it is currently the highest real-time net data rate reported in the S band and the farthest transmission distance of real-time S+C+L band transmission systems.

Real-time 115.2Tb/s and 230.4Tb/s transmission over 125 μm cladding G.654.E-compatible MCF is successfully demonstrated by C+L band FIFO-less MC-EDFA, in which the impact of inter-core crosstalk on transmission performance is investigated.

Targeting industrial PON scenario, we propose a bandwidth allocation algorithm based on multi-branch LSTM-Attention model. Compared to IPACT algorithm, the proposed algorithm is load-independent and maintains low latency (around 1ms), especially under high loads.

We report a real-time 80 lambda x800 Gbps/carrier optical transmission system based on C6T + L6T band spectra and PCS-16QAM modulation format. The span loss is compensated only by the erbium-doped fiber amplifiers (EDFAs). The transmission performance of the 800 Gbit/s system based on G.654.E fiber and G.652.D fiber was tested and compared. The experimental results show that the 80 lambda x800 Gbps/carrier signals in G.654.E fiber and G.652.D fiber are able to transmit 1050 and 600 km, respectively, under standard span loss. The transmission distance of 1050 km, to the best of our knowledge, is the farthest transmission distance reported for the 800 Gbps/carrier system based on C + L band spectra under standard span loss, which provides guidance for the future optical network upgrade from 400 to 800 Gbit/s. (c) 2025 Society of Photo-Optical Instrumentation Engineers (SPIE) [DOI:10.1117/1.OE.64.5.058101]

32 channels of single-carrier 800 Gbit/s signals to transmit over 3000 km in G.654.E fiber has been proven. The transmission loss is only compensated through EDFA, and the capacity-distance product can reach 76.8 Pbit•km/s.

Health assessment is essential for ensuring reliable operation and proactive maintenance in optical networks. In this work, we demonstrate digital twin-assisted health assessment for a 4-span S+C+L-band transmission system, where optical power and OSNR assessment are carried out.

We report a field trial of 3000-kilometer long-distance RDMA traffic transmission based on OTN lossless flow control and end-network collaborative congestion control, and the port bandwidth utilization rate has increased from 20% to over 90%.

This paper introduces the key technologies of 800G optical transmission system and demonstrates a field trial of hybrid 400G and 800G transmission. It verifies the possibility and reliability of multi-rate and multi-channel spacing transmission system in the current network.

A routing section of the communication cable in the live network is used, combined with distributed optical fiber sensing equipment, for long-term monitoring, and through data recording, to achieve a variety of dynamic event response analysis.

The high dynamics of the topology structure,the high variability of the links between space and ground,and the highly lim-ited network resource in the space-ground heterogeneous network pose significant challenges.To address these challenges,a space-ground heterogeneous joint QoS assurance scheme was proposed based on deep analysis of space-ground heterogeneous network development situation.The cross-domain network resource orchestration,construction of a space-ground QoS system,dynamic policy control of heterogeneous joint QoS,and on-demand adaptation and deployment of space-ground network resources technologies were discussed.