Inter-domain Space Digital Twin

• Inter-domain space digital twins are virtual replicas that integrate heterogeneous domains with unified spatio-temporal frameworks and semantic standards.
• They employ cross-domain data fusion and bidirectional control via federated coordination to maintain real-time consistency across systems.
• Applications span 6G network emulation, lunar communications, smart mobility, and industrial logistics, demonstrating broad multi-domain utility.

An inter-domain space digital twin is a virtual replica that simultaneously represents and integrates multiple spatial domains—such as terrestrial GIS, aerospace telemetry, indoor-building BIM, marine or ocean sensing, and underground infrastructure—into a single coherent system, and, in networked settings, can also be treated as a federation of domain-specific digital twins coordinated across RAN, transport, core, cloud or edge, application, and non-terrestrial segments under unified orchestration (Ali et al., 2023, Yu et al., 2023, Tran et al., 2 Jun 2025). Relative to a traditional digital twin, which mirrors a specific physical asset or process without explicit geospatial context, and a single-domain spatial digital twin, which embeds geospatial context for one domain, the inter-domain form emphasizes cross-domain data fusion, unified spatio-temporal reference frames, semantic interoperability, and bidirectional control across heterogeneous cyber-physical systems.

1. Definition and formalization

The conceptual core of the field is shared across several formalisms. In the IoFDT framework, a single digital twin is defined as

$DT_i = (\mathcal P_i,\mathcal C_i,f_i,g_i),$

where $\mathcal P_i$ is the physical state space, $\mathcal C_i$ is the cyber representation space, $f_i:\mathcal P_i \to \mathcal C_i$ maps observations to digital states, and $g_i:\mathcal C_i \to \mathcal P_i$ maps digital states back to control or actuation. An inter-domain federated digital twin is then

$F = (\mathcal D,\mathcal H,\mathcal V),$

where $\mathcal D=\{DT_1,\dots,DT_N\}$ is a collection of domain-specific twins, $\mathcal H$ is the set of horizontal edges connecting twins of similar functional or hierarchical level, and $\mathcal V$ is the set of vertical edges connecting twins across abstraction levels. Federation enforces cross-twin consistency within tolerance:

$$\forall (DT_i,DT_j)\in \mathcal H \cup \mathcal V,\quad \|f_i(p_i)-f_j(p_j)\| \le \epsilon_{ij}.$$

This formulation makes explicit that inter-domain operation is not merely co-location of models; it is consistency-constrained coordination among multiple twin instances (Yu et al., 2023).

A spatially explicit formulation sharpens the geospatial dimension. An inter-domain space digital twin is defined as a virtual replica that simultaneously represents and integrates multiple spatial domains under unified spatio-temporal reference frames and semantics. Its key differentiators are cross-domain data fusion, a unified spatio-temporal model harmonizing coordinate reference systems and time bases, and semantic interoperability through ontologies and metadata schemas (Ali et al., 2023).

TwinArch provides a domain-independent abstraction with a related tuple,

$\mathcal P_i$0

where $\mathcal P_i$1 denotes the physical twin instances, $\mathcal P_i$2 the digital representations, $\mathcal P_i$3 the physical-to-digital mapping, and $\mathcal P_i$4 the digital-to-physical mapping. The stated consistency property,

$\mathcal P_i$5

frames the twin as a bidirectional system rather than a passive model. This domain-neutral meta-model is intended to be specialized for aerospace, network, urban, or other domains without altering the architectural core (Somma et al., 10 Apr 2025).

2. Architectural organization across domains

A network-centered instantiation of the concept is the multi-domain NDT architecture for 6G and beyond. In that architecture, a single layered ecosystem spans RAN, ORAN, transport, 5G Core, cloud or edge, application verticals, NTN, and quantum networks. Each domain is represented by domain-specific twin modules. The DT-RAN module virtualizes geo-referenced cells, antenna patterns, and channel ray-tracing; the NTN-DT module mirrors satellite links, orbital dynamics, and ground-station paths; the 5G Core twin models UPF, AMF, SMF, PCF, UDM, and related functions; cloud and edge twins reflect compute and storage resource pools; and application twins map blockchain ledgers, healthcare workflows, manufacturing lines, and vehicular control (Tran et al., 2 Jun 2025).

Above these domain twins sits a unified orchestration and control layer implemented via the O-RAN SMO, NFV MANO, and Kubernetes, connected to non-RT RIC and near-RT RIC for closed-loop control. This layer hosts global network-wide policies, end-to-end slice management, intent-based AI training, and run-time reconfiguration. Its telemetry plane ingests real-time control-plane and user-plane metrics, including signal strength, KPI counters, temperature, and vibration, via O1/O2, E2, A1, and X-/N-/NG- interfaces. Time-series databases and data lakes align, cleanse, and store live and historical data, while models pull updates at sub-second cadence to maintain digital-real synchrony. For consistency maintenance, cross-domain digital-real loops enforce atomic commit across multiple twins via a two-phase commit in the SMO. Update frequencies are explicitly differentiated: RAN and NTN twins at 10 ms, transport twins at 100 ms, core twins at 1 s, and cloud or edge twins at 5 s.

Plotinus shows how such an architecture can be realized as a microservice system for space-air-ground networks. Physical-layer emulation, topology generation, path computation, traffic injection, real-world integration, and visualization each run as autonomous services communicating through REST or gRPC APIs and a shared state bus, while NS-3 and TapBridge connect the emulated network to live traffic. TwinArch complements this implementation style with separable architectural views: a Module Twin View defining domain entities and their static relationships, a Component Twin View refining them into deployable software components and connectors, a Traceability Twin View linking module and component abstractions, and a Dynamic Twin View capturing runtime message sequences (Gao et al., 2024, Somma et al., 10 Apr 2025).

3. Federation, semantics, and data infrastructure

Inter-domain operation requires explicit federation logic. The IoFDT architecture is organized into a Physical Layer of sensors, actuators, and processes, a Twin Layer of individual $\mathcal P_i$6 nodes, and a Federation Layer whose federation manager oversees horizontal and vertical interactions, coordination, and service orchestration. Horizontal interaction among peers uses consensus or averaging rules such as

$\mathcal P_i$7

or distributed averaging,

$\mathcal P_i$8

to reduce disagreement cost

$\mathcal P_i$9

Vertical interaction aggregates child-twin states into higher-level summaries,

$\mathcal C_i$0

and can then push high-level commands back down:

$\mathcal C_i$1

The same framework also introduces a federation-level network-slice optimization in which resources $\mathcal C_i$2 are allocated to maximize aggregate utility under a total resource budget (Yu et al., 2023).

The spatial data substrate needed by such federation is commonly described through a four-layer model. At the acquisition layer, terrestrial sensors, aerospace telemetry, indoor or BIM feeds, marine or underwater sensors, and time-synchronization clients provide raw streams. All sources are stamped in UTC and, if needed, corrected for clock drift via Precision Time Protocol; incoming spatial data are reprojected on the fly to ECEF or WGS84. A representative metadata record is

$\mathcal C_i$3

where the record carries a domain tag, coordinate reference system, ISO 8601 UTC timestamp, 3D spatial extent, ontology term, and pointer to a domain-specific schema. At the storage and analytics layer, PostgreSQL/PostGIS, MongoDB with GeoJSON, Cassandra with geo-spatial extension, SpatialHadoop, and GeoSpark support storage, indexing, batch analytics, and stream analytics across points, lines, polygons, 3D solids, and time series (Ali et al., 2023).

Middleware and semantics complete the integration path. OGC-compliant services such as WMS, WFS, and WMTS, OGC SensorThings API, MQTT topics, CesiumJS, ArcGIS Server, Azure Digital Twins, and AWS IoT TwinMaker abstract heterogeneous sources behind standard interfaces. In the NDT setting, NTN twins and terrestrial twins share a common YANG-based schema, identified with 3GPP TR 38.821, exposed through REST or gRPC APIs in the SMO, and all domain twins conform to a unified information model of network slices and service-level objectives. A URLLC slice spanning satellite and 5G RAN can therefore be defined once in the SMO and instantiated concurrently in the RAN twin, transport twin, and NTN twin. GOST extends this interoperability stack with knowledge-based semantics, using OWL ontologies and RDF or knowledge graphs as a common language for heterogeneous entities, attributes, and relationships, and with semantic compression and semantic communication to transmit task-relevant features rather than full raw states (Tran et al., 2 Jun 2025, Qiu et al., 18 Dec 2025).

4. Physical, network, and simulation models

The analytical core of an inter-domain space digital twin is the coupling of physical propagation, topology dynamics, queuing, and reliability. For NTNs, the twin must simulate LEO, MEO, and GEO orbital motion through the standard two-body equations

$\mathcal C_i$4

with circular-orbit period

$\mathcal C_i$5

Propagation and service models include one-way delay

$\mathcal C_i$6

free-space link budget

$\mathcal C_i$7

and total end-to-end latency

$\mathcal C_i$8

Each satellite beam is abstracted as a virtual RU, with twin modules tracking steerable antenna bore-sight, per-beam power, and per-beam user loads, while the NTN-DT continuously measures simulated SNR and BER and computes a space-to-ground reliability curve for each orbit pass. For transport and joint RAN or NTN modeling, the surveyed framework also uses M/M/1 relations

$\mathcal C_i$9

ray-tracing path loss

$f_i:\mathcal P_i \to \mathcal C_i$0

and orbital handover planning framed as maximizing $f_i:\mathcal P_i \to \mathcal C_i$1 subject to visibility windows computed from ephemeris (Tran et al., 2 Jun 2025).

The lunar communications literature pushes this modeling further into cislunar reliability engineering. In the inter-domain lunar twin, the received power on a generic hop is modeled as

$f_i:\mathcal P_i \to \mathcal C_i$2

with free-space path loss

$f_i:\mathcal P_i \to \mathcal C_i$3

The channel adopts a Rician fading model,

$f_i:\mathcal P_i \to \mathcal C_i$4

and scenario outage probability is defined by

$f_i:\mathcal P_i \to \mathcal C_i$5

Closed forms are given for one-hop, two-hop, and three-hop lunar scenarios, enabling a decision engine that, at each epoch, selects the lowest-power scenario among those satisfying the highest achievable reliability target and otherwise falls back to a reserve scenario (Cetin et al., 21 Jul 2025).

Plotinus supplies a packet-level, cross-layer complement to these analytical models. Its topology generator emits a dynamic graph

$f_i:\mathcal P_i \to \mathcal C_i$6

the physical-layer emulator computes beam gain, interference, and per-link capacity, and the channel model includes

$f_i:\mathcal P_i \to \mathcal C_i$7

$f_i:\mathcal P_i \to \mathcal C_i$8

$f_i:\mathcal P_i \to \mathcal C_i$9

and

$g_i:\mathcal C_i \to \mathcal P_i$0

Routing can be optimized through shortest-path or multi-commodity-flow objectives, and cross-domain handovers are treated as link-state changes plus rule updates, with incremental re-optimization and sub-second handover without IP-level disruption (Gao et al., 2024).

5. Implementations, platforms, and representative applications

A broad implementation ecosystem has emerged around these models. The surveyed NDT stack includes GPU-accelerated ray tracing for RAN twins, MATLAB or Simulink orbital propagators for NTN twins, surrogate GNN models for large-scale RAN or NTN joint interference prediction, Unity or 3DMax for site-level visualization, Colosseum for integrated O-RAN or NTN experiments, GNS3 plus free5gc for core-twin prototyping, and WireNet for security digital twins. The same survey identifies an open-source O-RAN NDT emulator, Colosseum, in which RAN, transport, and NTN twins coexist for security experiments; optical NDT systems with IQ-constellation twins for ROADMs and EDFAs integrating soft-failure emulation with optical link twins; and a DT-based UAV swarm framework for DRL-driven coordination in which each micro-twin is hosted at an edge server (Tran et al., 2 Jun 2025).

Plotinus is a more explicit blueprint for a space-air-ground digital twin system. Its services can be hot-swapped across physical-layer plugins such as GEO, LEO, UAV, and beam-hopping, path-computation plugins such as shortest-path and SDN-controlled methods, topology plugins driven by static configuration, TLE, or real-time telemetry, and traffic or real-device plugins spanning VMs, containers, and hardware nodes via TapBridge. Time is divided into discrete slots, such as 100 ms or 1 s, and a lightweight clock service synchronizes state publication and NS-3 reconfiguration at each slot boundary. This gives the system real-time emulation with live network traffic rather than an offline-only simulator (Gao et al., 2024).

The lunar communication use case demonstrates the concept under explicitly inter-domain conditions. The simulated setup uses ANSYS STK, a lunar surface terminal at Mons Malapert at 86°S, 1 min resolution over 24 h, 10 LLO satellites in 2 planes, 6 GEO relays equally spaced, and three DSN-NSN ground stations at Goldstone, Madrid, and Canberra. Links are Ka-band at 26–27 GHz with $g_i:\mathcal C_i \to \mathcal P_i$1 MHz, and lunar brightness temperatures are evaluated for $g_i:\mathcal C_i \to \mathcal P_i$2 K. The reported results show that no single static architecture achieves continuous high reliability; scenario 3, Moon to GEO to Earth, is the most stable but still occasionally exceeds targets, while scenario 4, Moon to LLO to GEO to Earth, is best on average but more complex. Under $g_i:\mathcal C_i \to \mathcal P_i$3 K and 400 K, dynamic switching between scenarios 3 and 4 allows the twin to maintain $g_i:\mathcal C_i \to \mathcal P_i$4 for most of the day, falling only briefly into moderate or low reliability and invoking scenario 3 fallback when needed (Cetin et al., 21 Jul 2025).

Outside purely space-network settings, inter-domain federated twins have been instantiated in smart mobility and industrial logistics. In smart mobility, Level 1 per-vehicle twins capture LiDAR, camera, and CAN-bus inputs; Level 2 intersection traffic twins fuse vehicle and RSU states; and a Level 3 city-wide mobility twin aggregates Level 2 outputs for route recommendations, with data flow vehicle to RSU to edge DT to cloud DT. In factory-to-farm automated food processing, horizontal federation among multiple farm twins shares weather and yield predictions, while vertical federation to a factory twin schedules transport and processing lines under freshness and logistics constraints. These examples show that the same federation logic applies whether the inter-domain boundary is space-to-ground, vehicle-to-intersection-to-city, or farm-to-factory (Yu et al., 2023).

6. Open problems and competing design paradigms

Real-time synchronization remains a primary unresolved problem. In the NDT literature, maintaining sub-100 ms freshness across highly distributed satellites, small cells, and edge nodes is described as unsolved, and proposed directions include predictive filtering through Kalman or Gaussian-process methods and adaptive sampling with AI. In spatial and co-simulation settings, synchronization is similarly constrained by conflicting clocks, heterogeneous time steps, and distributed simulators; proposed remedies include PTP or NTP for real-time feeds, HLA time management services using the LBTS algorithm, HLA 1516 federation with a common Federation Object Model, and CRDTs for eventual consistency. Scalability is likewise unresolved: end-to-end twins spanning billions of devices or petabyte-scale archives motivate hierarchical and modular architectures, AI-driven model order reduction, edge-cloud bursting, tiered storage, distributed indexing, and auto-scaling microservices (Tran et al., 2 Jun 2025, Ali et al., 2023).

Semantic interoperability, trust, and explainability define a second cluster of open issues. The literature calls for standardized semantics for cross-domain data exchange and ontology alignment, joint optimization of network, compute, and AI objectives under tight SLAs, trust, privacy, and access control in multi-party federations, and causal inference and explanation inside federated workflows. Proposed directions include domain-aware XAI modules, zero-trust principles, privacy-preserving federated learning, blockchain-anchored telemetry provenance, quantum-safe key management, confidential computing with Intel SGX and multi-party computation, and interoperability profiles coordinated with OGC and ISO/IEC (Yu et al., 2023, Tran et al., 2 Jun 2025, Ali et al., 2023).

A recurrent research tension concerns the status of fidelity itself. Conventional digital twins pursue full-scale, high-fidelity mirroring, but the GOST formulation for SAGSIN argues that this assumption breaks down under large spatial scale, sparse sensing, satellites moving at more than 7 km/s, millisecond-scale topology changes, deep heterogeneity, and severe SWaP limits on non-terrestrial nodes. GOST therefore replaces full-mirror fidelity with goal-oriented utility, organized into knowledge-based semantics, data-driven semantics, and goal-oriented principles, and evaluated through infrastructure, data-to-model transformation, model quality, synchronization capability, and application utility. Its core metrics include Age-of-Information,

$g_i:\mathcal C_i \to \mathcal P_i$5

goal-oriented DRL optimization,

$g_i:\mathcal C_i \to \mathcal P_i$6

and semantic compression with task-relevant mutual-information constraints. In a collaborative satellite-UAV tracking case study, GOST achieves 15–30% lower average AoI than both conventional DT and local models across SINR regimes, maintains near-optimal tracking rewards even at low SINR, and uses a cloud-edge lifelong learning mechanism that yields rapid re-convergence under target-motion changes (Qiu et al., 18 Dec 2025). This suggests that future inter-domain space digital twins may increasingly be instantiated as semantically aligned, task-specific, and resource-aware twins rather than as universally exhaustive mirrors of every subsystem.