Periodic Reporting for period 2 - ONFIRE (Future Optical Networks for Innovation, Research and Experimentation)
Berichtszeitraum: 2019-07-01 bis 2021-09-30
Efficiency and cost-effectiveness of the future transport networks will play a key role in enabling the digital equality. Prices for ICT equipment and connection time are required to decrease. Otherwise, cost will remain a significant barrier to attain the aforementioned equality, since digital exclusion affects some of the most vulnerable and disadvantaged groups in society. The ONFIRE project addresses these challenge via two research themes to take advantage of flexibility and modularity in the hardware and software domains: i) disaggregation in optical networks; and ii) cognition in optical networks. This combination leads to reduce in capital expenditure (CAPEX) through the use of cost-effective optical systems for optical disaggregated networks with white boxes, and well as lowering operational expenditures (OPEX) through the network optimization brought by Artificial intelligence (AI) mechanisms. The specific technological-oriented objectives of the ONFIRE project are:
Objective I: Leverage on the flexibility and modularity that have been fostered by means of software techniques, exploiting software-defined networking (SDN) concepts, aiming at deploying an AI-enabled control scheme across heterogeneous optical transport networks.
Objective II: Leverage on the flexibility from elastic interfaces and programmable white boxes, transforming the operation of today’s networks infrastructure and reducing over-provisioning and margins, in order to increase overall network equipment utilization.
Objective III: Develop cost-effective subsystems for non-intrusive in-band monitoring of advanced modulation formats to guarantee both quality of transmission (QoT) and quality of service, QoS (on a per flow basis) for end-to-end services.
Objective IV: Use data mining techniques to leverage massive monitoring data from the physical layer, provided by advanced mechanisms in coherent interfaces, white boxes and monitoring subsystems, to continuously adjust and re-optimize network settings supported by SDN control allowing full network programmability with accurate performance prediction.
All objectives have been successfully achieved.
Objective I: Leverage on the flexibility and modularity that have been fostered by means of software techniques, exploiting software-defined networking (SDN) concepts, aiming at deploying an AI-enabled control scheme across heterogeneous optical transport networks.
Objective II: Leverage on the flexibility from elastic interfaces and programmable white boxes, transforming the operation of today’s networks infrastructure and reducing over-provisioning and margins, in order to increase overall network equipment utilization.
Objective III: Develop cost-effective subsystems for non-intrusive in-band monitoring of advanced modulation formats to guarantee both quality of transmission (QoT) and quality of service, QoS (on a per flow basis) for end-to-end services.
Objective IV: Use data mining techniques to leverage massive monitoring data from the physical layer, provided by advanced mechanisms in coherent interfaces, white boxes and monitoring subsystems, to continuously adjust and re-optimize network settings supported by SDN control allowing full network programmability with accurate performance prediction.
All objectives have been successfully achieved.
To achieve these technical objectives, ONFIRE deploys a three-year research programme centred on two European industrial PhDs.
Individual Research Project on data plane technologies for disaggregated optical networks
Fabiano Locatelli (ESR1) joined on May 2018. He has been involved in WP1 activities (disaggregated optical networks). ESR1 objectives included the investigation of novel cost-effective subsystems for non-intrusive in-band optical performance monitoring (OPM), agnostic to the optical transmission waveforms. To this aim, we theoretically studied a front-end monitoring scheme, able to deliver arbitrary spectral resolution, that could work regardless of the modulation format of the considered monitored signals. The possibility to integrate the proposed monitoring scheme with a specific monitoring agent, also allowed us to address another of the main ONFIRE objectives, which was the integration of the developed data plane technologies with the control plane. We also proposed a closed control loop approach to optimize specific figures of merit related to the superchannel subchannels. Another of the objectives addressed by ESR1 within the ONFIRE project was the identification of specific techniques to extract from the monitored physical layer information adequate figures of merit to be leveraged by the control plane.
Individual Research Project on cognitive optical networks
Ankush Mahajan (ESR2), joined ONFIRE project on September 2018. He has been involved on conducting WP2 activities (cognitive optical networks). ESR2 objectives includes physical layer monitoring schemes, ML and data mining techniques to leverage massive physical layer monitoring information to adjust and re-optimize the network settings for aggregated (traditional) and next generation disaggregated optical transport networks. For effective optical network planning, operation and optimization, it is necessary to estimate the Quality of Transmission (QoT) of the connections. Network designers are interested in accurate and fast QoT estimation for services to be established in a future or existing network. Typically, QoT estimation is performed using a Physical Layer Model (PLM) which is included in the QoT estimation tool or Qtool. A design margin is generally included in a Qtool to account for the modeling and parameter inaccuracies, to re-assure an acceptable performance. The PLM with the addition of a design margin is typically referred to as the QoT estimation tool or Qtool. PLM accuracy is highly important as modeling errors translate into a higher design margin which in turn translate into wasted capacity or unwanted regeneration. In this regard, ESR2 first outlined the limitations on applying ML to assist traditional Qtools. Furthermore, with the aim of improving the Qtool estimation accuracy in multi-vendor networks, ESR2 proposed PLM extensions. ESR2 also aimed at investigating and solving the issue of accuracy limitation of Qtool in dynamic optimization tasks.
Individual Research Project on data plane technologies for disaggregated optical networks
Fabiano Locatelli (ESR1) joined on May 2018. He has been involved in WP1 activities (disaggregated optical networks). ESR1 objectives included the investigation of novel cost-effective subsystems for non-intrusive in-band optical performance monitoring (OPM), agnostic to the optical transmission waveforms. To this aim, we theoretically studied a front-end monitoring scheme, able to deliver arbitrary spectral resolution, that could work regardless of the modulation format of the considered monitored signals. The possibility to integrate the proposed monitoring scheme with a specific monitoring agent, also allowed us to address another of the main ONFIRE objectives, which was the integration of the developed data plane technologies with the control plane. We also proposed a closed control loop approach to optimize specific figures of merit related to the superchannel subchannels. Another of the objectives addressed by ESR1 within the ONFIRE project was the identification of specific techniques to extract from the monitored physical layer information adequate figures of merit to be leveraged by the control plane.
Individual Research Project on cognitive optical networks
Ankush Mahajan (ESR2), joined ONFIRE project on September 2018. He has been involved on conducting WP2 activities (cognitive optical networks). ESR2 objectives includes physical layer monitoring schemes, ML and data mining techniques to leverage massive physical layer monitoring information to adjust and re-optimize the network settings for aggregated (traditional) and next generation disaggregated optical transport networks. For effective optical network planning, operation and optimization, it is necessary to estimate the Quality of Transmission (QoT) of the connections. Network designers are interested in accurate and fast QoT estimation for services to be established in a future or existing network. Typically, QoT estimation is performed using a Physical Layer Model (PLM) which is included in the QoT estimation tool or Qtool. A design margin is generally included in a Qtool to account for the modeling and parameter inaccuracies, to re-assure an acceptable performance. The PLM with the addition of a design margin is typically referred to as the QoT estimation tool or Qtool. PLM accuracy is highly important as modeling errors translate into a higher design margin which in turn translate into wasted capacity or unwanted regeneration. In this regard, ESR2 first outlined the limitations on applying ML to assist traditional Qtools. Furthermore, with the aim of improving the Qtool estimation accuracy in multi-vendor networks, ESR2 proposed PLM extensions. ESR2 also aimed at investigating and solving the issue of accuracy limitation of Qtool in dynamic optimization tasks.
• We proposed ML based closed control loop QoT estimator or Qtool retraining scheme to optimize the launch power of transponders (TPs) in an optical network. The proposed schemes reduce the optimization time by ~94% by reducing the number of monitoring calls requirements.
• We proposed a ML-based approach to estimate erbium-doped fiber amplifier (EDFA) gain ripple penalties with only optical monitoring information. By using monitored Dynamic Gain Equalizer (DGE) attenuation profile and signal-to-noise (SNR) at the coherent receiver’s information to train ML model, we estimated the QoT accurately for new connections with >70% reduction in the related margin.
• We proposed a multi-vendor Qtool and integrated with convex algorithms to optimize TPs launch power. We observed ~0.6dB higher SNR estimation accuracy and ~35% min. margin savings compared to a traditional Qtool.
• We proposed an iterative Qtool retraining scheme to achieve better and more realistic optimization objectives. We applied our proposed method to solve the dynamic version of the TPs launch power optimization problem. With the proposed Qtool retraining, we showed an improvement of ~8.5 dB and ~0.64 dB in the sum of margins and the lowest margin, respectively, over optimization with a one-time trained Qtool scheme.
• We developed two spectral processing techniques, to retrieve signal- and filter-related parameters from the optical spectra. We found out that the monitoring strategy in which the optical monitors are placed at the ingress ports of the nodes, if properly combined with our proposed processing technique, reduced the estimation of the SNR penalty introduced by an optical filter, up to the 92%.
• We proposed a closed control loop process to optimize the spectral distances of the subchannels in a superchannel. Our objective was to identify the subchannel frequencies set that optimized specific functions of their QoT. For a superchannel with four subchannels and uniform signal characteristics, with respect to the case where the subchannels were equally spaced, we achieved an improvement for the total superchannel SNR and for the minimum subchannel SNR of 1.45 dB and 1.19 dB, respectively.
• Two summer schools and three symposia were organized as network-wide activities for the training program.
• We proposed a ML-based approach to estimate erbium-doped fiber amplifier (EDFA) gain ripple penalties with only optical monitoring information. By using monitored Dynamic Gain Equalizer (DGE) attenuation profile and signal-to-noise (SNR) at the coherent receiver’s information to train ML model, we estimated the QoT accurately for new connections with >70% reduction in the related margin.
• We proposed a multi-vendor Qtool and integrated with convex algorithms to optimize TPs launch power. We observed ~0.6dB higher SNR estimation accuracy and ~35% min. margin savings compared to a traditional Qtool.
• We proposed an iterative Qtool retraining scheme to achieve better and more realistic optimization objectives. We applied our proposed method to solve the dynamic version of the TPs launch power optimization problem. With the proposed Qtool retraining, we showed an improvement of ~8.5 dB and ~0.64 dB in the sum of margins and the lowest margin, respectively, over optimization with a one-time trained Qtool scheme.
• We developed two spectral processing techniques, to retrieve signal- and filter-related parameters from the optical spectra. We found out that the monitoring strategy in which the optical monitors are placed at the ingress ports of the nodes, if properly combined with our proposed processing technique, reduced the estimation of the SNR penalty introduced by an optical filter, up to the 92%.
• We proposed a closed control loop process to optimize the spectral distances of the subchannels in a superchannel. Our objective was to identify the subchannel frequencies set that optimized specific functions of their QoT. For a superchannel with four subchannels and uniform signal characteristics, with respect to the case where the subchannels were equally spaced, we achieved an improvement for the total superchannel SNR and for the minimum subchannel SNR of 1.45 dB and 1.19 dB, respectively.
• Two summer schools and three symposia were organized as network-wide activities for the training program.