With the continuous growth of large language model (LLM) sizes, individual datacenters are increasingly insufficient to support LLM training. Cross-datacenter training has become a feasible solution, where optical networks play a crucial role in data transmission. However, current optical networks suffer from severe over-provisioning issues, primarily due to the conflict between the bursty nature of LLM traffic and the fixed bandwidth of optical network connections. Therefore, this paper presents CrossOptic, a novel, to our knowledge, framework for optimizing cross-datacenter LLM training over optical networks. The proposed solution addresses the critical challenges of bandwidth under-utilization in optical networks and excessive communication overhead in distributed LLM training. CrossOptic integrates two synergistic components: adaptive GPU resource orchestration (AGRO) for communication-minimizing task placement and traffic-interleaved connectivity provisioning (TICP) for bandwidth-efficient flow aggregation. Through comprehensive simulations using realistic GPT model workloads, CrossOptic outperforms conventional dedicated connection approaches by 38.1% in terms of optical resource utilization. For a GPT model with 529.6B parameters, compared with the training methods using data parallelism and pipeline parallelism between datacenters, CrossOptic reduces the iteration time by 55% and 14%, respectively. The framework achieves significant cost efficiency in the optical infrastructure while supporting the computational demands of LLM training.
{"title":"Task placement and traffic interleaving for cross-datacenter LLM training over optical networks","authors":"Qiaojun Hu;Wei Wang;Chongzhu Huang;Xiaoyu Wang;Yajie Li;Yongli Zhao;Yanlei Zheng;Yanxia Tan;Jie Zhang","doi":"10.1364/JOCN.579324","DOIUrl":"https://doi.org/10.1364/JOCN.579324","url":null,"abstract":"With the continuous growth of large language model (LLM) sizes, individual datacenters are increasingly insufficient to support LLM training. Cross-datacenter training has become a feasible solution, where optical networks play a crucial role in data transmission. However, current optical networks suffer from severe over-provisioning issues, primarily due to the conflict between the bursty nature of LLM traffic and the fixed bandwidth of optical network connections. Therefore, this paper presents CrossOptic, a novel, to our knowledge, framework for optimizing cross-datacenter LLM training over optical networks. The proposed solution addresses the critical challenges of bandwidth under-utilization in optical networks and excessive communication overhead in distributed LLM training. CrossOptic integrates two synergistic components: adaptive GPU resource orchestration (AGRO) for communication-minimizing task placement and traffic-interleaved connectivity provisioning (TICP) for bandwidth-efficient flow aggregation. Through comprehensive simulations using realistic GPT model workloads, CrossOptic outperforms conventional dedicated connection approaches by 38.1% in terms of optical resource utilization. For a GPT model with 529.6B parameters, compared with the training methods using data parallelism and pipeline parallelism between datacenters, CrossOptic reduces the iteration time by 55% and 14%, respectively. The framework achieves significant cost efficiency in the optical infrastructure while supporting the computational demands of LLM training.","PeriodicalId":50103,"journal":{"name":"Journal of Optical Communications and Networking","volume":"18 2","pages":"137-149"},"PeriodicalIF":4.3,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082191","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
With the development of services such as video streaming and cloud computing, the demand for network capacity and quality of transmission would be gradually increased. Space-division multiplexing elastic optical networks (SDM-EONs) provide a promising approach to support the greater capacities of networks due to their sufficient resources with multi-core transmission existence. Unfortunately, inter-core crosstalk results in significant degradation of spectrum resource utilization and quality of transmission in SDM-EONs. To avoid this dilemma, this paper focuses on the crosstalk mitigation benefit of counter-propagation between the adjacent cores and presents a crosstalk impact fragmentation-aware metric (CIFA) under bidirectional spectrum where the state of spectrum slots is defined as unoccupied (“0”) and occupied (“1” for positive and “−1” for opposite transmissions). In this metric, the crosstalk impact on adjacent cores reflects the utilization of counter-propagation, and the fragmentation is considered to estimate spectrum utilization. Furthermore, benefiting from the presented metric, this paper proposes a bidirectional-propagation-based crosstalk-avoided spectrum-efficient resource allocation (BCSRA) scheme in SDM-EONs, which aims to avoid the crosstalk and fully utilize the spectrum resources in networks. In this scheme, the algorithm assigns the lightpath with the minimum CIFA to the incoming request, which minimizes the impact on future requests. Simulation results show that the proposed BCSRA scheme can effectively reduce the blocking probability and enhance resource utilization while avoiding inter-core crosstalk in SDM-EONs.
{"title":"BCSRA: bidirectional-propagation-based crosstalk-avoided spectrum-efficient resource allocation scheme in SDM-EONs","authors":"Bowen Bao;Ming Wei;Xiaoliang Li;Yan Ding;Jiajia Zhao;Qiuyan Yao;Hui Yang;Bijoy Chand Chatterjee;Eiji Oki","doi":"10.1364/JOCN.573325","DOIUrl":"https://doi.org/10.1364/JOCN.573325","url":null,"abstract":"With the development of services such as video streaming and cloud computing, the demand for network capacity and quality of transmission would be gradually increased. Space-division multiplexing elastic optical networks (SDM-EONs) provide a promising approach to support the greater capacities of networks due to their sufficient resources with multi-core transmission existence. Unfortunately, inter-core crosstalk results in significant degradation of spectrum resource utilization and quality of transmission in SDM-EONs. To avoid this dilemma, this paper focuses on the crosstalk mitigation benefit of counter-propagation between the adjacent cores and presents a crosstalk impact fragmentation-aware metric (CIFA) under bidirectional spectrum where the state of spectrum slots is defined as unoccupied (“0”) and occupied (“1” for positive and “−1” for opposite transmissions). In this metric, the crosstalk impact on adjacent cores reflects the utilization of counter-propagation, and the fragmentation is considered to estimate spectrum utilization. Furthermore, benefiting from the presented metric, this paper proposes a bidirectional-propagation-based crosstalk-avoided spectrum-efficient resource allocation (BCSRA) scheme in SDM-EONs, which aims to avoid the crosstalk and fully utilize the spectrum resources in networks. In this scheme, the algorithm assigns the lightpath with the minimum CIFA to the incoming request, which minimizes the impact on future requests. Simulation results show that the proposed BCSRA scheme can effectively reduce the blocking probability and enhance resource utilization while avoiding inter-core crosstalk in SDM-EONs.","PeriodicalId":50103,"journal":{"name":"Journal of Optical Communications and Networking","volume":"18 2","pages":"114-125"},"PeriodicalIF":4.3,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082267","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This paper presents a novel, to the best of our knowledge, ultra-low latency fronthaul networking approach based on nanosecond (ns)-precision optical switching with advanced synchronization. We propose and demonstrate a comprehensive control architecture that integrates precision time protocol (PTP)-based time alignment with semiconductor optical amplifier (SOA)-based optical switches, enabling coordinated, jitter-minimized switching with accuracy as fine as 100 ns. Two primary synchronization methods are explored: a distributed hardware approach employing dedicated timing devices for simultaneous switch configuration, and “SyncNet,” a software-defined networking (SDN) controller utilizing time information inspired by reverse PTP. Experimental validation on a commercial Open RAN 5G system shows that our fronthaul switching framework achieves exceptional reliability, evidenced by only 0.00024% packet loss during link recovery while maintaining steady throughput (uplink: 1.03 Mbps and downlink: 1.01 Mbps) over repeated switching events. The proposed architecture supports dynamic reconfiguration for rapid link recovery and energy-efficient, traffic-adaptive resource allocation. Our results establish a robust foundation for next-generation fronthaul networks, delivering deterministic performance and flexibility aligned with 6G requirements. This work directly addresses the escalating needs for ultra-low latency, high reliability, and energy efficiency in emerging mobile and AI-driven applications.
本文提出了一种新颖的，据我们所知，基于纳秒（ns）精度的先进同步光交换的超低延迟前传网络方法。我们提出并演示了一种综合控制架构，该架构集成了基于精确时间协议（PTP）的时间校准和基于半导体光放大器（SOA）的光开关，实现了精确到100 ns的协调、抖动最小化的开关。本文探讨了两种主要的同步方法：一种采用专用定时设备进行同步交换机配置的分布式硬件方法，以及“SyncNet”，一种利用受反向PTP启发的时间信息的软件定义网络（SDN）控制器。在商用Open RAN 5G系统上的实验验证表明，我们的前传交换框架实现了卓越的可靠性，链路恢复期间的丢包率仅为0.00024%，同时在重复交换事件中保持稳定的吞吐量（上行：1.03 Mbps，下行：1.01 Mbps）。该架构支持动态重构，以实现快速链路恢复和节能、流量自适应的资源分配。我们的研究结果为下一代前传网络奠定了坚实的基础，提供了符合6G要求的确定性性能和灵活性。这项工作直接解决了新兴移动和人工智能驱动应用中对超低延迟、高可靠性和能效不断升级的需求。
{"title":"Nanosecond-level optical switching for flexible and resilient fronthaul networks","authors":"Vaigai Nayaki Yokar;Ali Mehrpooya;Sen Shen;Yiran Teng;Wanxin Zhao;Shuangyi Yan;Dimitra Simeonidou","doi":"10.1364/JOCN.578768","DOIUrl":"https://doi.org/10.1364/JOCN.578768","url":null,"abstract":"This paper presents a novel, to the best of our knowledge, ultra-low latency fronthaul networking approach based on nanosecond (ns)-precision optical switching with advanced synchronization. We propose and demonstrate a comprehensive control architecture that integrates precision time protocol (PTP)-based time alignment with semiconductor optical amplifier (SOA)-based optical switches, enabling coordinated, jitter-minimized switching with accuracy as fine as 100 ns. Two primary synchronization methods are explored: a distributed hardware approach employing dedicated timing devices for simultaneous switch configuration, and “SyncNet,” a software-defined networking (SDN) controller utilizing time information inspired by reverse PTP. Experimental validation on a commercial Open RAN 5G system shows that our fronthaul switching framework achieves exceptional reliability, evidenced by only 0.00024% packet loss during link recovery while maintaining steady throughput (uplink: 1.03 Mbps and downlink: 1.01 Mbps) over repeated switching events. The proposed architecture supports dynamic reconfiguration for rapid link recovery and energy-efficient, traffic-adaptive resource allocation. Our results establish a robust foundation for next-generation fronthaul networks, delivering deterministic performance and flexibility aligned with 6G requirements. This work directly addresses the escalating needs for ultra-low latency, high reliability, and energy efficiency in emerging mobile and AI-driven applications.","PeriodicalId":50103,"journal":{"name":"Journal of Optical Communications and Networking","volume":"18 2","pages":"126-136"},"PeriodicalIF":4.3,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082268","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We perform a field trial of end-to-end transmission of heterogeneous signals: digital coherent optical (DCO) channels up to 600 Gbps and analog radio-over-fiber (ARoF) carriers at 60 GHz (mmWave) and 210 GHz (sub-THz), across the HEAnet live production metro network, leveraging optical spectrum-as-a-service and a laboratory-based passive optical network. We then perform wavelength-dependent performance characterization of ARoF carriers across the 50 GHz ITU-T channel grid within the C-band, measuring the pre-FEC bit-error rate (BER). The channel whose BER lies within the acceptable FEC threshold is selected as the optimum spectral slot for co-launching ARoF with background WDM traffic. This analysis enables optimization of the spectrum allocation, enhancing performance, throughput, and spectral efficiency, and shows the seamless coexistence of analog and digital traffic services.
{"title":"Field trial and spectral allocation analysis of transparent heterogeneous digital coherent and analog radio over fiber mmWave/sub-THz signals over a live production network","authors":"Devika Dass;Colm Browning;Dan Kilper;Amol Delmade;Agastya Raj;Eoin Kenny;Liam Barry;Marco Ruffini","doi":"10.1364/JOCN.575391","DOIUrl":"https://doi.org/10.1364/JOCN.575391","url":null,"abstract":"We perform a field trial of end-to-end transmission of heterogeneous signals: digital coherent optical (DCO) channels up to 600 Gbps and analog radio-over-fiber (ARoF) carriers at 60 GHz (mmWave) and 210 GHz (sub-THz), across the HEAnet live production metro network, leveraging optical spectrum-as-a-service and a laboratory-based passive optical network. We then perform wavelength-dependent performance characterization of ARoF carriers across the 50 GHz ITU-T channel grid within the C-band, measuring the pre-FEC bit-error rate (BER). The channel whose BER lies within the acceptable FEC threshold is selected as the optimum spectral slot for co-launching ARoF with background WDM traffic. This analysis enables optimization of the spectrum allocation, enhancing performance, throughput, and spectral efficiency, and shows the seamless coexistence of analog and digital traffic services.","PeriodicalId":50103,"journal":{"name":"Journal of Optical Communications and Networking","volume":"18 2","pages":"A150-A158"},"PeriodicalIF":4.3,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082266","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Key management strategies are a critical yet often overlooked aspect of integrating quantum key distribution (QKD) networks as a service into critical infrastructure. It has a considerable impact on the efficiency of QKD network services, thereby shaping its suitability for diverse applications. In this paper, we examine the effectiveness of key management strategies developed through practical testbeds, identifying their strengths and weaknesses. A novel, to the best of our knowledge, organization of key storage to enhance key construction efficiency and overall service performance is introduced. Using simulation tools, the proposed strategy is evaluated against existing approaches, demonstrating superior performance and effectiveness.
{"title":"Design of a key management system for efficient key supply in quantum key distribution networks","authors":"Emir Dervisevic;Amina Tankovic;Enio Kaljic;Miroslav Voznak;Miralem Mehic","doi":"10.1364/JOCN.577670","DOIUrl":"https://doi.org/10.1364/JOCN.577670","url":null,"abstract":"Key management strategies are a critical yet often overlooked aspect of integrating quantum key distribution (QKD) networks as a service into critical infrastructure. It has a considerable impact on the efficiency of QKD network services, thereby shaping its suitability for diverse applications. In this paper, we examine the effectiveness of key management strategies developed through practical testbeds, identifying their strengths and weaknesses. A novel, to the best of our knowledge, organization of key storage to enhance key construction efficiency and overall service performance is introduced. Using simulation tools, the proposed strategy is evaluated against existing approaches, demonstrating superior performance and effectiveness.","PeriodicalId":50103,"journal":{"name":"Journal of Optical Communications and Networking","volume":"18 2","pages":"98-113"},"PeriodicalIF":4.3,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082264","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
We propose a DCN architecture that directly connects server racks distributed among DCs connected to metro-access nodes consisting of ROADMs. ELSs are introduced to coherent TRXs to prevent laser diodes being operated under high-temperature conditions on ToR electrical packet switches. A single-wavelength ELS for intra-DCIs is distributed to several hundred TRXs by splitting its optical power. A full C-band WDM ELS for inter-DCIs supplies each wavelength to a TRX. Both intra- and inter-DCIs are executed using an identical colorless coherent TRX, and the wavelength assignment and optical-path switching are carried out using OCSs. We experimentally evaluated the feasibility of the proposed architecture through BER measurements in a four-node ring network with four DCs per node and about 300 racks per DC, where ToR pairs communicated using 112 Gb/s DP-QPSK signals. The intra-DCI results indicate a sufficient received power margin of 11.6 dB despite a 2.4 dB penalty induced by in-band XT occurring in the OCSs. We conducted 42 channel dense WDM transmission experiments involving flat bandwidth loading with ASE noise for inter-DCIs, where in-band XT in each node and DC was ignored since our analytical estimations showed that its impact was negligible. The inter-DCI results indicate a sufficient OSNR margin of 10.9 dB despite the OSNR penalties induced by the linear/nonlinear impairments of 100 km transmission and spectrum narrowing by eight wavelength-selective switches.
{"title":"Proposal and feasibility study of a converged all-optical intra-/inter-data-center network with external laser sources to distribute optical carriers to coherent transceivers","authors":"Ritsuki Hamagami;Masamichi Fujiwara;Shin Kaneko;Jun-ichi Kani;Tomoaki Yoshida;Tatsuya Shimada","doi":"10.1364/JOCN.571907","DOIUrl":"https://doi.org/10.1364/JOCN.571907","url":null,"abstract":"We propose a DCN architecture that directly connects server racks distributed among DCs connected to metro-access nodes consisting of ROADMs. ELSs are introduced to coherent TRXs to prevent laser diodes being operated under high-temperature conditions on ToR electrical packet switches. A single-wavelength ELS for intra-DCIs is distributed to several hundred TRXs by splitting its optical power. A full C-band WDM ELS for inter-DCIs supplies each wavelength to a TRX. Both intra- and inter-DCIs are executed using an identical colorless coherent TRX, and the wavelength assignment and optical-path switching are carried out using OCSs. We experimentally evaluated the feasibility of the proposed architecture through BER measurements in a four-node ring network with four DCs per node and about 300 racks per DC, where ToR pairs communicated using 112 Gb/s DP-QPSK signals. The intra-DCI results indicate a sufficient received power margin of 11.6 dB despite a 2.4 dB penalty induced by in-band XT occurring in the OCSs. We conducted 42 channel dense WDM transmission experiments involving flat bandwidth loading with ASE noise for inter-DCIs, where in-band XT in each node and DC was ignored since our analytical estimations showed that its impact was negligible. The inter-DCI results indicate a sufficient OSNR margin of 10.9 dB despite the OSNR penalties induced by the linear/nonlinear impairments of 100 km transmission and spectrum narrowing by eight wavelength-selective switches.","PeriodicalId":50103,"journal":{"name":"Journal of Optical Communications and Networking","volume":"18 2","pages":"83-97"},"PeriodicalIF":4.3,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026455","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In passive optical networks (PONs), the optical power budget serves as a key determinant of system scalability. A higher power budget allows for larger split ratios over a given transmission medium, thereby enabling more end users to share the same infrastructure and reducing the cost per subscriber. In this paper, we introduce a carrier-assisted complex-valued double-sideband direct detection PON architecture that eliminates the need for a local oscillator laser at the optical network unit of coherent PON while maintaining a high optical power budget. The system leverages a deep-learning-enabled optimal direct-detection receiver to achieve high launch power and receiver sensitivity without the complexity of coherent detection. Using the proposed architecture, we experimentally demonstrate an optical power budget of 42 dB for 100 Gb/s downstream transmission over 20 km standard single-mode fiber (SSMF), meeting the 15% soft-decision forward error correction (SD-FEC) threshold of ${2} times {{10}^{- 2}}$. To the best of our knowledge, this represents the highest optical power budget for a 100 Gb/s direct-detection PON using SSMF. Furthermore, by employing our self-developed 20 km anti-resonant hollow-core fiber, we demonstrate an optical power budget of 49 dB under the 15% SD-FEC threshold, marking a record-high optical power budget achieved to date for 100 Gb/s PON.
{"title":"Demonstration of a 100 Gb/s complex-valued double-sideband direct-detection PON with a 42/49 dB optical power budget over a 20 km SSMF/AR-HCF","authors":"Xingfeng Li;Xu Zhang;Hui Chen;Yao Lu;Honglin Ji;Zhaopeng Xu;Peng Sun;Siyue Jin;Tonghui Ji;Shuchao Mi;Bo Wu;Chao Li;Qibing Wang;Zichen Liu;Jie Luo;Lei Zhang;Lei Wang;Zhixue He;Shaohua Yu","doi":"10.1364/JOCN.575553","DOIUrl":"https://doi.org/10.1364/JOCN.575553","url":null,"abstract":"In passive optical networks (PONs), the optical power budget serves as a key determinant of system scalability. A higher power budget allows for larger split ratios over a given transmission medium, thereby enabling more end users to share the same infrastructure and reducing the cost per subscriber. In this paper, we introduce a carrier-assisted complex-valued double-sideband direct detection PON architecture that eliminates the need for a local oscillator laser at the optical network unit of coherent PON while maintaining a high optical power budget. The system leverages a deep-learning-enabled optimal direct-detection receiver to achieve high launch power and receiver sensitivity without the complexity of coherent detection. Using the proposed architecture, we experimentally demonstrate an optical power budget of 42 dB for 100 Gb/s downstream transmission over 20 km standard single-mode fiber (SSMF), meeting the 15% soft-decision forward error correction (SD-FEC) threshold of <tex>${2} times {{10}^{- 2}}$</tex>. To the best of our knowledge, this represents the highest optical power budget for a 100 Gb/s direct-detection PON using SSMF. Furthermore, by employing our self-developed 20 km anti-resonant hollow-core fiber, we demonstrate an optical power budget of 49 dB under the 15% SD-FEC threshold, marking a record-high optical power budget achieved to date for 100 Gb/s PON.","PeriodicalId":50103,"journal":{"name":"Journal of Optical Communications and Networking","volume":"18 1","pages":"35-41"},"PeriodicalIF":4.3,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026435","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yousuf Moiz Ali;Jaroslaw E. Prilepsky;Nicola Sambo;Joao Pedro;Mohammad M. Hosseini;Antonio Napoli;Sergei K. Turitsyn;Pedro Freire
Machine learning-based failure management in optical networks has gained significant attention in recent years, but severe class imbalance, where normal instances far outnumber failure cases, remains a considerable challenge. While pre- and in-processing techniques have been widely studied, post-processing methods are largely unexplored. We present a direct comparison of pre-, in-, and post-processing approaches for class imbalance mitigation in failure detection and identification using an experimental dataset. For failure detection, post-processing, particularly threshold adjustment, yields the highest F1 score improvement of up to 15.3%, while random under-sampling offers the fastest inference. In failure identification, generative AI methods deliver the most significant performance gains up to 24.2%, whereas post-processing has a limited impact in multi-class settings. When class overlap exists and latency is critical, over-sampling methods like synthetic minority over-sampling technique (SMOTE) are most effective; without latency constraints, meta-learning excels. In low-overlap scenarios, generative AI approaches provide the best performance with minimal inference time.
{"title":"From data to decision: a multi-stage framework for class imbalance mitigation in optical network failure analysis","authors":"Yousuf Moiz Ali;Jaroslaw E. Prilepsky;Nicola Sambo;Joao Pedro;Mohammad M. Hosseini;Antonio Napoli;Sergei K. Turitsyn;Pedro Freire","doi":"10.1364/JOCN.576774","DOIUrl":"https://doi.org/10.1364/JOCN.576774","url":null,"abstract":"Machine learning-based failure management in optical networks has gained significant attention in recent years, but severe class imbalance, where normal instances far outnumber failure cases, remains a considerable challenge. While pre- and in-processing techniques have been widely studied, post-processing methods are largely unexplored. We present a direct comparison of pre-, in-, and post-processing approaches for class imbalance mitigation in failure detection and identification using an experimental dataset. For failure detection, post-processing, particularly threshold adjustment, yields the highest F1 score improvement of up to 15.3%, while random under-sampling offers the fastest inference. In failure identification, generative AI methods deliver the most significant performance gains up to 24.2%, whereas post-processing has a limited impact in multi-class settings. When class overlap exists and latency is critical, over-sampling methods like synthetic minority over-sampling technique (SMOTE) are most effective; without latency constraints, meta-learning excels. In low-overlap scenarios, generative AI approaches provide the best performance with minimal inference time.","PeriodicalId":50103,"journal":{"name":"Journal of Optical Communications and Networking","volume":"18 1","pages":"42-58"},"PeriodicalIF":4.3,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11363018","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026434","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Huitao Zhou;Lizhu Liu;Chen Su;Bojun Zhang;Zhiqun Gu;Jiawei Zhang;Yuefeng Ji
The rise of time-sensitive applications has intensified the need for deterministic transmission across networks. As computing and service functions are shifted toward the edge, metro networks have become the key interconnection layer between access and core domains. However, conventional optoelectronic-switched network architectures, which rely on statistical multiplexing and slow reconfiguration, fail to meet stringent latency and jitter constraints. To overcome these challenges, this paper focuses on a scalable all-optical spine-leaf network architecture by introducing an optical time-slot switching (OTS) framework. OTS offers fine-grained switching and low-latency transmission; however, achieving deterministic transmission over shared optical resources remains challenging due to slot contention and insufficient coordination in existing scheduling schemes. Therefore, we propose a time-aware wavelength and slot allocation (TA-SA) scheme that allocates transmission windows (TWs) according to individual flow parameters, including arrival time, period, and jitter tolerance. We formulate an integer linear programming model and a heuristic algorithm to jointly optimize TW placement and wavelength-slot mapping under strict QoS constraints. To analyze worst-case network performance, deterministic network calculus is adopted to model delay and backlog bounds for different scheduling schemes. Simulation results show that TA-SA not only guarantees end-to-end latency and jitter constraints but also improves average resource utilization by over 34.9% compared to the existing scheduling schemes. Moreover, the proposed scheme significantly reduces buffer occupancy under increasing traffic loads, demonstrating its scalability and effectiveness for deterministic transmission in metro all-optical networks.
{"title":"Deterministic optical time-slot scheduling and performance evaluation via time-aware resource allocation in metro all-optical spine-leaf networks","authors":"Huitao Zhou;Lizhu Liu;Chen Su;Bojun Zhang;Zhiqun Gu;Jiawei Zhang;Yuefeng Ji","doi":"10.1364/JOCN.577375","DOIUrl":"https://doi.org/10.1364/JOCN.577375","url":null,"abstract":"The rise of time-sensitive applications has intensified the need for deterministic transmission across networks. As computing and service functions are shifted toward the edge, metro networks have become the key interconnection layer between access and core domains. However, conventional optoelectronic-switched network architectures, which rely on statistical multiplexing and slow reconfiguration, fail to meet stringent latency and jitter constraints. To overcome these challenges, this paper focuses on a scalable all-optical spine-leaf network architecture by introducing an optical time-slot switching (OTS) framework. OTS offers fine-grained switching and low-latency transmission; however, achieving deterministic transmission over shared optical resources remains challenging due to slot contention and insufficient coordination in existing scheduling schemes. Therefore, we propose a time-aware wavelength and slot allocation (TA-SA) scheme that allocates transmission windows (TWs) according to individual flow parameters, including arrival time, period, and jitter tolerance. We formulate an integer linear programming model and a heuristic algorithm to jointly optimize TW placement and wavelength-slot mapping under strict QoS constraints. To analyze worst-case network performance, deterministic network calculus is adopted to model delay and backlog bounds for different scheduling schemes. Simulation results show that TA-SA not only guarantees end-to-end latency and jitter constraints but also improves average resource utilization by over 34.9% compared to the existing scheduling schemes. Moreover, the proposed scheme significantly reduces buffer occupancy under increasing traffic loads, demonstrating its scalability and effectiveness for deterministic transmission in metro all-optical networks.","PeriodicalId":50103,"journal":{"name":"Journal of Optical Communications and Networking","volume":"18 1","pages":"59-73"},"PeriodicalIF":4.3,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026439","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This tutorial paper presents a survey on several current ongoing research activities about the expected evolution of passive optical networks, focusing on the physical layer (i.e., on the physical media dependent or PMD layer). After the recent public release of the ITU-T G.9804.3 50G-PON standard, there is today ongoing extensive scientific brainstorming on what can be the best optical transmission solutions for the PON “next step,” which will likely be targeting 100G or 200G per wavelength per direction. Elaborating on our OFC2025 tutorial on these topics, we start with a brief overview of the possible IM-DD solutions and their actual limitations for 200G-PON, while the core of the paper discusses the pros and cons of the introduction of advanced modulation formats and coherent detection also in PON, including some hybrid solutions that are currently under investigation in the ITU-T Very High Speed PON initiative. We show that coherent PON can actually reach the 200G per wavelength target, including some spare link budget margin, which can potentially be used for new access architectures. As an example, we present a summary of our recent results on hybrid solutions and on the all-optical convergence between the metro and access segments.
{"title":"High-speed PON solutions: recent evolutions and expected trends [Invited Tutorial]","authors":"Giuseppe Rizzelli;Roberto Gaudino","doi":"10.1364/JOCN.575601","DOIUrl":"https://doi.org/10.1364/JOCN.575601","url":null,"abstract":"This tutorial paper presents a survey on several current ongoing research activities about the expected evolution of passive optical networks, focusing on the physical layer (i.e., on the physical media dependent or PMD layer). After the recent public release of the ITU-T G.9804.3 50G-PON standard, there is today ongoing extensive scientific brainstorming on what can be the best optical transmission solutions for the PON “next step,” which will likely be targeting 100G or 200G per wavelength per direction. Elaborating on our OFC2025 tutorial on these topics, we start with a brief overview of the possible IM-DD solutions and their actual limitations for 200G-PON, while the core of the paper discusses the pros and cons of the introduction of advanced modulation formats and coherent detection also in PON, including some hybrid solutions that are currently under investigation in the ITU-T Very High Speed PON initiative. We show that coherent PON can actually reach the 200G per wavelength target, including some spare link budget margin, which can potentially be used for new access architectures. As an example, we present a summary of our recent results on hybrid solutions and on the all-optical convergence between the metro and access segments.","PeriodicalId":50103,"journal":{"name":"Journal of Optical Communications and Networking","volume":"18 2","pages":"A134-A149"},"PeriodicalIF":4.3,"publicationDate":"2026-01-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146026451","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：