TwinLight: A Multi-Timescale Physical-Layer-Aware T-API-Compliant Optical Domain Controller Digital Twin

TwinLight: A Multi-Timescale Physical-Layer-Aware T-API-Compliant Optical Domain Controller Digital Twin Carlos Natalino(1) , Piotr Lechowicz(1) , Renato Ambrosone(2) , Enrico Miotto(2) , Vittorio Curri(2) , Paolo Monti(1) (1)

arXiv:2611.09359v1 [physics.optics] 6 Nov 2026

Department of Electrical Engineering, Chalmers University of Technology, Gothenburg, Sweden. carlos.natalino@chalmers.se (2) Politecnico di Torino, Torino, Italy. Abstract We demonstrate TwinLight, an open-source optical-network digital twin acting as an optical domain controller. Its T-API REST and telemetry streaming interfaces expose a physical-layer engine that combines a GNPy steady-state baseline with four analytical transients. ©2026 The Author(s) Review Disaggregated and software-defined optical networks increasingly rely on digital twins (DTs) to de-risk control-plane decisions12 . A faithful software replica of the optical domain helps validate software-defined networking (SDN) control logic, explore what-if scenarios, prototype routing, modulation, and spectrum assignment (RMSA) algorithms, and generate training data for machine learning (ML) models [1, 2]. This requirement becomes more relevant as autonomous and agentic strategies begin to operate in closed loop, observe physics-consistent consequences, and be validated against the same northbound interfaces that a production controller would expose [3, 4]. The tooling landscape that has emerged to meet these needs resolves into three disjoint families of solutions [5]. First, physical-layer tools such as GNPy compute steady-state quality of transmission (QoT) with high fidelity [6, 7], but are not designed to expose the persistent state and standardized northbound interfaces that a controllerfacing optical domain controller (ODC) must provide. Second, network emulators such as MininetOptical capture topology, embed optical performance monitoring (OPM), and expose NETCONF and gNMI interfaces aligned with OpenConfig and OpenROADM device models, yet operate exclusively in steady state [8]. Within the same family, data-driven digital twins such as OCATA refine steady-state QoT estimation through telemetrydriven model tuning, but expose ad-hoc northbound interfaces and likewise lack an analytical model of temporal impairments [9]. A third family combines physical-layer-aware QoT with RMSA logic and recalculates mutual impact when a new lightpath perturbs the signal-to-noise ratio (SNR) of existing services [10, 11], but lacks both a standards-compliant northbound interface and an1 ECOC 2026 paper ID Tu2-Ex1/1440. 2 The code of the tool described in this paper is available at

https://github.com/carlosnatalino/TwinLight.

alytical models of temporal physical impairments. Considered in isolation, none of these families can serve as an ODC for a production orchestrator. Two further developments sharpen the requirements that a controller-facing DT must satisfy. The first concerns physics, including physical effects that span multiple timescales. The community has converged on the Gaussian noise (GN) family of models as the working description of nonlinear interference in coherent transmission, and on GNPy as its de facto implementation, with recent field validation against partially-disaggregated multivendor deployments [6, 12, 13]. Real coherent links, however, undergo impairments that no steady-state snapshot can capture. Channel adddrop events trigger microsecond-scale power excursions in cascaded Erbium-doped fiber amplifiers (EDFAs) [14]. The state of polarisation drifts on second-to-day timescales [15], and polarizationdependent loss (PDL) adds a stochastic optical signal-to-noise ratio (OSNR) penalty over the same range [16–18]. Accumulated dispersion couples with the local-oscillator linewidth to produce equalization-enhanced phase noise (EEPN) on the minute scale [19], while diurnal thermal cycles drift the chromatic dispersion (CD) and modulate span loss [20, 21]. Together, these mechanisms span nine orders of magnitude in time. The second concerns the application programming interfaces (APIs) that the DT exposes. The Open Networking Foundation (ONF) TransportAPI (T-API) defines a representational state transfer (REST) interface that an ODC must expose to a hierarchical orchestrator, which may integrate optical services with other technology domains such as packet-switching and wireless networks. T-API exposes topology, connectivity, photonic-media, and path-computation contexts that allow comprehensive control of the infrastructure associated with the ODC [22, 23]. T-API streaming, aligned with the OpenConfig gRPC network management interface (gNMI) subscription model, has emerged

External Clients T-API streaming / gNMI Client (telemetry subscriber)

SDN Orchestrator (T-API client)

Web UI

HTTP polling gNMI

T-API REST

TwinLight T-API Server tapi-common

gRPC Server GnmiServicer Capabilities | Subscribe (ONCE / STREAM / POLL)

tapi-topology

/internal/* OPM, services, spectrum

connectivity

photonic-media

/admin/* snapshot, restore

PathDecodeMiddleware

TapiContext Topology Graph (NetworkX DiGraph)

TAPI Models (Topology, Node, Link, NEP, SIP, ConnSvc)

Spectrum State (per-link slot arrays)

Baseline Cache (OpmBaseline per svc)

EDFA State Tracker

Route Cache (k-shortest paths)

SIP→GNPy UID mapping

GNPy Topology JSON

OOPT GNPy API (steady-state QoT)

Analytical Transients

Physics Engine BER/Q Conversion (M-QAM Gaussian)

Modulation Params

Fig. 1: TwinLight architecture.

as the telemetry counterpart [24]. A DT that aims to operate under a production orchestrator must therefore implement both protocols natively. Novelty The overall architecture of TwinLight is shown in Fig. 1. Northbound, the DT exposes two interfaces: (i) a T-API REST interface that can be used to interface with orchestrators and (ii) a T-API streaming interface to telemetry subscribers. The DT also provides a built-in web user interface that uses the same T-API REST surface, as well as internal endpoints for administrative tasks. A single stateful context manages the topology graph, the connectivity services, the per-link spectrum state, and the per-service QoT baselines, so that physics, state, and standard interfaces remain consistent without inter-process synchronization. The physics engine combines a GNPy steady-state baseline with the four analytical transient models and a modulation-format-aware bit-error-rate conversion, feeding back into the same context that the northbound interfaces serve. TwinLight’s novelty lies in three compounding contributions. First, TwinLight is a DT that serves as a complete ODC, simultaneously exposing T-API and telemetry streaming interfaces over a live physics engine, with both steady-state QoT and transients. The same physics engine is exposed through T-API and the built-in web user interface, such that control logic, telemetry consumers, and human operators share a single source of truth. Second, TwinLight provides a PDL treatment appropriate for coherent receivers. The model cascades a per-hinge polarization transfer across reconfigurable optical add-drop multiplexer (ROADM) and EDFA hinges of each lightpath following the hinge formulation of [16]. Each hinge is assigned a Maxwell-distributed magnitude [25] seeded deterministically from its identifier, and its alignment drifts with a slightly incommensu-

rate period, so that the joint trajectory becomes quasi-periodic and ergodically fills the joint PDL ensemble without an explicit Monte-Carlo pass. The accumulated PDL is then mapped to a coherentreceiver OSNR penalty using the experimentally characterized treatment of [17, 18]. The mapping is sensitive to a configurable amplifier spontaneous emission (ASE)-distribution scenario that reflects the transmitter-, distributed-, and receiverside cases reported in the same literature. Third, TwinLight composes four transient impairment models onto the GNPy steady-state baseline through a QoT combiner that maintains generalized signal-to-noise ratio (GSNR) and OSNR as independent metrics and routes each impairment to its physically correct counterpart. EDFA reservoir dynamics, polarisation effects, and the thermal loss component of the environmental drift act on both GSNR and OSNR. EEPN, a digital signal processing (DSP)-domain impairment, acts only on GSNR. Thermal CD drift updates the reported CD metric directly and modulates GSNR indirectly through the EEPN term, the only path by which a DSP-compensated dispersion change reaches a coherent receiver. The combiner produces a six-metric OPM tuple at every time instance, comprising OSNR, GSNR, pre-forward error correction (FEC) BER, Q-factor, CD, and polarizationmode dispersion (PMD), on top of a baseline that respects the physical separation between linear, nonlinear, polarisation, phase-noise, and environmental contributions. From the same combined state, the combiner also synthesizes a constellation diagram and an eye diagram. The diagrams are parameterized by the GSNR, the combined transmitter and local-oscillator linewidth, and the PMD, following the Gaussian-mixture representation introduced by OCATA [9, 26] and the standard PMD-driven jitter mapping [15]. The artefact thereby exposes physics-consistent observability at three abstraction levels: aggregated metrics, scatter statistics, and time-domain waveforms, all derived from the same physics state. ECOC Relevance Network operators and vendors of SDN orchestrators repeatedly face the problem of integrating new control logic against a realistic, standardscompliant ODC. Such integration requires access to laboratory hardware that is expensive, scarce, and routinely committed to production-readiness validation rather than open experimentation. A free, scriptable DT that exposes the same northbound interfaces as a real ODC removes that bottleneck and lets integration work start before any optics are powered on. Academic researchers working on RMSA, control-plane algorithms, agentic artificial intelligence (AI), ML-based QoT estimation, and failure detection and prediction gain

a reproducible physical-layer baseline that can generate realistic training data on demand. The broader optical-networking community, which has spent the last decade converging on the GN model and GNPy as the open steady-state physics baseline [6, 7] and on T-API as the open northbound interface [22, 24], gains a concrete embodiment of that convergence in a single artifact. TwinLight is best suited for a live demonstration because its value proposition is fundamentally aligned with the Demo Zone format. Visitors can inspect how T-API works in real time, establish and decommission lightpaths, change physical-layer parameters, and observe the cascading effects. Demo Implementation The demo is built to convince a visitor of three things, in order. A T-API provisioning call against the DT triggers QoT-aware RMSA that behaves the way an ODC RMSA module would. A telemetry subscription on the resulting service streams real-time OPM the way an operational telemetry pipeline would. Operator-level events produce physics-consistent time-varying OPM that can be visually correlated with the underlying transient model on screen. Two pre-loaded topologies are selectable from the user interface. A small one supports fast iteration and easy visualization, while a continentalscale topology with tens of ROADMs and roughly two hundred fiber spans [6] supports realistic experimentation at scale. Both run in the C-band on an 88-channel grid with 50 GHz spacing, with three coherent modulation formats selectable per service, following the standard GNPy implementation. Snapshot and restore endpoints instantly reset state between visitors, so a full demonstration cycle remains manageable even with continuous traffic. The demo setup is intentionally minimal. A poster contains an overview of the demonstration. A single (presenter’s) laptop hosts the entire DT, including its web interface. An external monitor displays the web interface in a large format, while the laptop screen shows a presentation with a few slides covering the architecture and theory behind TwinLight. There is no optical hardware in the demo. Fig. 2 shows the preliminary web interface, which shows all relevant aspects of the DT. One of the tabs shows the loaded topology with pan and zoom controls, and selecting a link displays the associated devices and their respective physicallayer parameter values. Another tab shows all currently established services. Selecting a specific service highlights its hop-by-hop path, along with the ROADM and EDFA hinges that contribute to its PDL penalty, thereby making the hinge model visible. From a service, you can navigate to its

Fig. 2: TwinLight preliminary web interface.

monitoring page. The monitoring page shows a current-value card with a sparkline and a historical chart for each of the six OPM metrics, alongside the constellation and eye diagrams synthesized from the same physics state. The three views together expose the perturbation of QoT at progressively finer levels of abstraction, from aggregated metrics through scatter statistics to time-domain waveforms. The web interface also exposes a tab that allows the audience to request a lightpath between two nodes. The user can select a modulation format or leave it as the best available for the path. After adding a new lightpath, the audience can then inspect its real-time OPM metrics. Finally, the web interface also contains tabs that allow the audience to inspect the network inventory. This includes a complete list of equipment, a page that allows the user to compute the k-shortest paths between two nodes, and one that shows the spectrum allocation for all links in the network. Conclusions TwinLight unifies a GNPy steady-state baseline with four analytical transient models via standardscompliant T-API REST and telemetry-streaming interfaces. By behaving as a complete ODC to any conformant orchestrator, the tool is positioned to accelerate the validation of SDN control logic, the training and testing of agentic and ML-driven optical-network algorithms, and the empirical study of physical-layer dynamics under realistic provisioning loads. We release TwinLight as open source so that the community can adopt, extend, and integrate it into pipelines that today rely on laboratory hardware, typically subject to production-readiness validation. Open extensions include multi-band physics, telemetry streaming coverage for connectivity-service objects, and a fault-prediction pipeline backed by the same synthesized observability.

Acknowledgements This work has been supported by the Horizon Europe ECO-eNET project, funded by the SNS JU under grant agreement No. 101139133.

[14]

A. Bononi and L. Rusch, “Doped-fiber amplifier dynamics: A system perspective”, Journal of Lightwave Technology, vol. 16, no. 5, pp. 945–956, May 1998. DOI: 10.1109/50.669054

[15]

R. Vilalta et al., “Applying Digital Twins to Optical Networks with Cloud-Native SDN Controllers”, IEEE Communications Magazine, vol. 61, no. 12, pp. 128–134, Dec. 2023. DOI: 10.1109/MCOM.003.2300105

J. P. Gordon and H. Kogelnik, “PMD fundamentals: Polarization mode dispersion in optical fibers”, Proceedings of the National Academy of Sciences, vol. 97, no. 9, pp. 4541–4550, Apr. 25, 2000. DOI: 10.1073/pnas.97. 9.4541

[16]

D. Wang et al., “Digital Twin of Optical Networks: A Review of Recent Advances and Future Trends”, Journal of Lightwave Technology, vol. 42, no. 12, pp. 4233–4259, Jun. 2024. DOI: 10.1109/JLT.2024.3401419

S. Zarkosvky and M. Shtaif, “Statistical distribution of polarization-dependent loss in systems characterized by the hinge model”, Optics Letters, vol. 45, no. 5, pp. 1224– 1227, Mar. 1, 2020. DOI: 10.1364/OL.385525

[17]

E. Etezadi et al., “Demonstration of DRL-based intelligent spectrum management over a T-API-enabled optical network digital twin”, in European Conference on Optical Communications (ECOC), Oct. 2023, pp. 1730– 1733. DOI: 10.1049/icp.2024.1801

A. D’Amico et al., “Experimental Probing and Modeling of the PDL Impact on the Optical Signal-to-Noise Ratio”, in Optical Fiber Communications Conference and Exhibition (OFC), Mar. 2023, W1E.6. DOI: 10.1364/OFC. 2023.W1E.6

[18]

E. Miotto, A. D’Amico, R. Ambrosone, F. Aquilino, S. Straullu, and V. Curri, “Statistical Assessment of System Margin in Metro Networks Impaired by PDL”, in Optical Fiber Communications Conference and Exhibition (OFC), Mar. 2025, M3F.5. DOI: 10.1364/OFC.2025.M3F. 5

[19]

W. Shieh and K.-P. Ho, “Equalization-enhanced phase noise for coherent-detection systems using electronic digital signal processing”, Optics Express, vol. 16, no. 20, pp. 15 718–15 727, Sep. 29, 2008. DOI: 10.1364/OE.16. 015718

[20]

T. Kato, Y. Koyano, and M. Nishimura, “Temperature dependence of chromatic dispersion in various types of optical fiber”, Optics Letters, vol. 25, no. 16, pp. 1156– 1158, Aug. 15, 2000. DOI: 10.1364/OL.25.001156

[21]

C. Berkdemir and S. Özsoy, “The temperature dependent performance analysis of EDFAs pumped at 1480 nm: A more accurate propagation equation”, Optics Express, vol. 13, no. 13, pp. 5179–5185, Jun. 27, 2005. DOI: 10.1364/OPEX.13.005179

[22]

C. Manso et al., “TAPI-enabled SDN control for partially disaggregated multi-domain (OLS) and multi-layer (WDM over SDM) optical networks [Invited]”, Journal of Optical Communications and Networking, vol. 13, no. 1, A21–A33, Jan. 2021. DOI: 10.1364/JOCN.402187

[23]

“TR-547 TAPI v2.4.0 Reference Implementation Agreement”, Open Networking Foundation, TR-547, Dec. 2022.

[24]

V. Karunakaran et al., “TAPI-based Telemetry Streaming in Multi-domain Optical Transport Network”, in Optical Fiber Communications Conference and Exhibition (OFC), Mar. 2024, M3Z.9. DOI: 10.1364/OFC.2024.M3Z. 9

[25]

A. Mecozzi and M. Shtaif, “The statistics of polarizationdependent loss in optical communication systems”, IEEE Photonics Technology Letters, vol. 14, no. 3, pp. 313–315, Mar. 2002. DOI: 10.1109/68.986797

[26]

D. Sequeira, M. Ruiz, N. Costa, A. Napoli, J. Pedro, and L. Velasco, “OCATA: A deep-learning-based digital twin for the optical time domain”, Journal of Optical Communications and Networking, vol. 15, no. 2, pp. 87–97, Feb. 2023. DOI: 10.1364/JOCN.477341

References [1]

[2]

[3]

[4]

Y. Song et al., “Synergistic Interplay of Large Language Model and Digital Twin for Autonomous Optical Networks: Field Demonstrations”, IEEE Communications Magazine, vol. 63, no. 6, pp. 90–96, Jun. 2025. DOI: 10.1109/MCOM.001.2400342

[5]

V. V. Deshpande and S. Mohani, “A Comparative Study of Network Simulators with Focus on OTN (Optical Transport Networks)”, in IEEE Engineering Informatics, Nov. 2023. DOI: 10.1109/IEEECONF58110.2023.10520628

[6]

A. Ferrari et al., “GNPy: An open source application for physical layer aware open optical networks”, Journal of Optical Communications and Networking, vol. 12, no. 6, pp. C31–C40, Jun. 2020. DOI: 10.1364/JOCN.382906

[7]

V. Curri, “GNPy model of the physical layer for open and disaggregated optical networking [Invited]”, Journal of Optical Communications and Networking, vol. 14, no. 6, pp. C92–C104, Jun. 2022. DOI: 10.1364/JOCN.452868

[8]

B. Lantz, A. A. Díaz-Montiel, J. Yu, C. Rios, M. Ruffini, and D. Kilper, “Demonstration of Software-Defined Packet-Optical Network Emulation with Mininet-Optical and ONOS”, in Optical Fiber Communications Conference and Exhibition (OFC), Mar. 2020, M3Z.9. DOI: 10. 1364/OFC.2020.M3Z.9

[9]

[10]

[11]

P. Gonzalez, S. Ghasrizadeh, S. Barzegar, H. Shakespear-Miles, M. Ruiz, and L. Velasco, “Demonstrating Lightpath Operation with the OCATA Digital Twin in Multiband Optical Networks”, in International Conference on Transparent Optical Networks (ICTON), Jul. 2025. DOI: 10.1109/ICTON67126.2025.11125283 C. Natalino et al., “Optical Networking Gym: An opensource toolkit for resource assignment problems in optical networks”, Journal of Optical Communications and Networking, vol. 16, no. 12, G40–G51, Dec. 2024. DOI: 10.1364/JOCN.532850 V. V. Garbhapu, C. Ware, and M. Lourdiane, “Physicallayer-aware network simulator for future optical functionalities”, in International Conference on Network of the Future (NoF), Oct. 2023, pp. 28–36. DOI: 10.1109/ NoF58724.2023.10302753

[12]

A. Carena, G. Bosco, V. Curri, Y. Jiang, P. Poggiolini, and F. Forghieri, “EGN model of non-linear fiber propagation”, Optics Express, vol. 22, no. 13, pp. 16 335– 16 362, Jun. 30, 2014. DOI: 10.1364/OE.22.016335

[13]

G. Borraccini et al., “Experimental Demonstration of Partially Disaggregated Optical Network Control Using the Physical Layer Digital Twin”, IEEE Transactions on Network and Service Management, vol. 20, no. 3, pp. 2343– 2355, Sep. 2023. DOI: 10.1109/TNSM.2023.3288823

Related documents

• Optical digital twins for disease prevention, diagnosis, therapy, and intervention.#433829
• Trans-Domain Digital Twin: Conceptual Foundations, Architecture, and Research Outlook#1536189
• Digital Twins Need Feedback#1014749
• NotLine: A Non-Intrusive Automated Platform to Build a Digital Twin#319014
• Missing Dimensions: Integrating Human and Social Systems into Digital Twin Engineering#1000098