This work has been submitted to the IEEE for possible publication.

Copyright may be transferred without notice, after which 1

this version may no longer be accessible.

6G Digital Twin Networks: From Theory to Practice

Xingqin Lin†, Lopamudra Kundu†, Chris Dick†, Emeka Obiodu†, and Todd Mostak§
†
NVIDIA, §HEAVY.AI
Emails: {xingqinl, lkundu, cdick, eobiodu}@nvidia.com, todd@heavy.ai

Abstract— Digital twin networks (DTNs) are real-time replicas

of physical networks. They are emerging as a powerful technology
for design, diagnosis, simulation, what-if-analysis, and artificial
intelligence (AI)/machine learning (ML) driven real-time
optimization and control of the sixth-generation (6G) wireless
networks. Despite the great potential of what digital twins can
offer for 6G, realizing the desired capabilities of 6G DTNs requires
tackling many design aspects including data, models, and
interfaces. In this article, we provide an overview of 6G DTNs by
presenting prominent use cases and their service requirements,
describing a reference architecture, and discussing fundamental
design aspects. We also present a real-world example to illustrate
how DTNs can be built upon and operated in a real-time reference
development platform – Omniverse. Figure 1: Digital twins enable virtual replicas of physical reality.
Telecommunications (IMT) systems towards 2030 and beyond
I. INTRODUCTION
(aka. 6G), which lists DT as an important use case and discusses
Digital twins (DTs) are a key part of the industrial metaverse technologies to support DTN [8]. The ITU telecommunication
that simulates physical assets such as factories, cities, and standardization sector (ITU-T) released a recommendation that
transportation systems in the virtual world. With recent describes requirements and architecture for DTNs [9]. The
advances in software platforms, artificial intelligence Institute of Electrical and Electronics Engineers (IEEE)
(AI)/machine learning (ML), and high-performance computing interfacing cyber and physical world working group also
accelerated by graphics processing units (GPUs), DTs have proposed a standard series IEEE 2888, e.g., P2888.1 defines
gained traction in many different areas, such as smart cities, sensor interface between cyber (i.e., DT) and physical world
manufacturing, and retail [1]. Meanwhile, wireless networks [10].
have become more intricate over time [2]. The complexity of Despite that the existing work presents vivid visions of DTNs
the sixth-generation (6G) wireless networks will grow due to for 6G and identifies opportunities, we are at the very beginning
their scale, multi-vendor components, and the need to support of making DTNs a reality for 6G. Realizing the desired
diverse use cases [3]. This trend calls for innovative tools and capabilities of 6G DTNs requires tackling many design aspects,
platforms like DTs to facilitate the design, analysis, and which have not been adequately addressed in the existing work.
operation of wireless networks for 6G and beyond. This article aims to bridge the gap in the existing literature by
A DT network (DTN) is a digital replica of the full life cycle not only providing an overview of DTNs for 6G with a focus
of a physical network. It uses data and models to create a on key design aspects, but also describing how DTNs can
physically accurate network simulation platform, which actually be built and operated in Omniverse [11] – a scalable,
provides up-to-date network status and predicts future network multi-GPU, real-time reference platform for building and
state [4][5]. It also provides interfaces for interactions with the operating metaverse applications.
physical network and network applications/users. Unlike
conventional network simulators, the DTN supports two-way II. USE CASES AND REQUIREMENTS
communication between the physical network and the virtual
twin network to achieve real-time interactive mapping and A. Use Cases
closed-loop decisions [6]. DT applications are destined to become ubiquitous across
DTNs will be a foundational 6G technology to realize the every industry. Increasingly, before anything is made in the
vision of a cyber-physical continuum between the physical physical world, it will be first simulated in a virtual world and
world and its digital representation. Figure 1 shows an example a DT will be created. Furthermore, everything that moves will
of creating virtual replicas of physical reality with DT one day be autonomous, and these autonomous machines will
technology. Inspired by the great potential of what DTs can be taught and trained in virtual worlds. For 6G networks and
offer, recently there has been much work painting visions of network operations, DTNs can support many use cases
DTNs for 6G [4]-[7]. There are also emerging initiatives in including, but not limited to, the examples illustrated in Fig. 2
standards bodies which develop initial guidelines for DTNs. and described in the sequel.
The International Telecommunication Union (ITU) Network simulation and planning: DTNs improve network
radiocommunication sector (ITU-R) is finalizing the report on simulation and planning process using large-scale, physically
future technology trends of terrestrial International Mobile accurate digital replicas of city blocks. This allows network
 This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which 2
this version may no longer be accessible.
deployment teams to test and optimize installation and
placement of base stations and their configurations in a virtual
world first, thereby reducing the expenditure of resources and
budget in the physical world. 6G standardization may also
exploit DTNs to carry out simulation rather than relying on
stochastic channel models and hexagonal deployments. In
particular, DTNs with ray tracing are well positioned to
simulate the deployment of reconfigurable intelligence surfaces
(RISs).
Network operation and management: One of the objectives
of the softwarization of 6G is the capability to change virtually
all aspects of the networks through software, from the physical
layer (e.g., site specific optimizations to even bespoke user-
specific waveforms) all the way to the core network. Even the
topology of the fronthaul network will become software
defined. The DTN can be used to create a virtual drive-test to
maintain system robustness and resilience, prior to applying the
Figure 2: Use cases and requirements of 6G digital twin
configuration changes in the real network. The DTN can also networks.
be used to predict future outages based on network behavior,
prompting, if necessary, corrective actions to circumvent the that identifies network misconfigurations, link bottlenecks, and
outage. Likewise, the DTN enables the network operators to security issues to enable prediction of future performance. In
evaluate multiple scenarios to understand actual customer many instances, network failure is a rare occurrence, and so it
experience and minimize churn. This can be used for real-time is difficult to produce a large enough dataset to train a neural
capacity planning for events or to better assess network network. While not feasible in the real network, faults can be
performance before onboarding new customers. deliberately injected into the cyber sibling for the purpose of
Data generation by simulation: In AI-native 6G networks, generating a rich dataset for ML or other data analytics
access to real-world data is crucial. However, it is not always processes. Also, DTNs can enable high-fidelity predictions to
possible to obtain a suitably rich dataset that reflects diverse set reduce randomness in a system, which is a key enabler for ultra-
of network operating conditions, e.g., operating environment, reliable communications.
weather, and network operational parameters. Synthetic data is B. Requirements
annotated information that a simulation running in a DTN
generates as an alternative to real-world data. Put another way, The foundation of 6G DTNs is built upon a set of design
synthetic data is created in a cyber sibling of the physical world principles shaped by various use case requirements, including
rather than being measured or collected from the real world. It reliability & latency, scalability, agility, generalizability,
may be artificial, but synthetic data reflects real-world data in a security, and interpretability, as illustrated in Fig. 2.
Reliability & latency: Trustworthiness is at the core of the
statistical sense. DTNs capture the precise geometry and
DTN’s fundamental characteristics. Rendering high degree of
material properties of objects in the environment to produce
stability throughout the lifecycle of a DTN requires robustness
datasets for training AI/ML models used for optimization and
in its operational infrastructure, including proactive handling of
network control [12].
latency critical adverse situations involving human error,
AI training and inference: AI/ML applied in image and video
equipment failure or malicious attacks with real-time response,
processing employs models that are trained offline with datasets
capability of ensuring robust data collection, storage,
such as ImageNet, Cityscapes, and the like. Inference engines
modeling, and information exchange between the DTN and
are deployed with these a priori generated model parameters.
other network entities, high degree of availability, and
We anticipate that while this paradigm will be applied in some
capabilities for disaster recovery (such as backup provision, and
areas of wireless, it is more likely that online training will play
ability to restore critical historical states/data points).
a bigger role in 6G and beyond. One example is deep
Scalability: Scale of a DTN can vary over a wide range,
reinforcement learning (DRL) based closed-loop control. A
depending on the dimension and complexity of its physical
neural network embedded in the DRL agent will be continually
counterpart. A physical network may grow or shrink in scale
updated in response to the environment. Latency will be
and the DTN should automatically adjust itself. Such flexible
important and so training and inference speeds will be critical.
scalability is an essential attribute of DTNs across various
Therefore, for DRL based real-time control in 6G, it is essential
network domains.
to utilize a DTN that natively resides on a programmable
Agility: Serving various network applications on-demand
hardware architecture equipped with hardware accelerators that
requires a DTN to have sufficient flexibility in its
support training and inference pipelines. In short, faster-than-
functionalities as needed. Ability for cross domain interaction,
real-time simulation capabilities of DTNs will be critical for AI-
information exchange and service cooperation between
based network optimization and control in real time.
multiple DT entities are some of the important attributes of a
What-if-analysis: 6G networks will be dynamic, with
DTN towards meeting needs of various operational stages of a
network operators being able to modify key network parameters
physical network.
and architecture in the time scale of minutes and hours, rather
than months or years. In this case, the DTN is a virtual sandbox
 This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which 3
this version may no longer be accessible.
Generalizability: Applicability of DTNs across network
equipment of various network applications and topologies
requires a certain degree of compatibility. Specifically, data
collection, storage, modeling, and interfaces need to be
generalizable to support multi-vendor, multi-standard
interoperability and ensure backward compatibility between
updated and older versions of DTNs.
Security: Guarantee of adequate protection from potential
attacks as well as defense mechanism for threat prevention
should be imprinted in the DNA of a robust DTN. A
comprehensive security and privacy consideration for DTN
elements including data, models, interactive interfaces, and
Figure 3: A reference 6G digital twin network architecture based
overall network infrastructure is warranted to ensure integrity
on ITU-T recommendation Y.3090.
of a DTN.
Interpretability: Augmenting DTNs with easy-to-use asset The 6G network application layer includes various network
management services aids user-friendly management of overall applications that exploit the capabilities exposed by the 6G twin
lifecycle of the twin entities like network equipment, links, layer. The network applications provide intent to the 6G twin
traffic flows, etc. Ability to observe discernable changes in the layer which then emulates the required services and, if needed,
functionalities of a DTN through display tools improves sends control messages to the 6G physical network layer.
interpretability of the DTN’s behavior with associated events Examples of the network applications include network
like instance creation, aggregation, update, and termination. operations, administration, and maintenance (OAM), network
optimization, and network visualization.
III. REFERENCE ARCHITECTURE There will be diverse 6G DTNs with different capabilities to
meet the demands of a wide range of use cases. It is vital to
In this section, we describe the architectural aspects of 6G
develop common components, generally applicable
DTNs that enable interactive virtual-real mapping and control.
frameworks, standardized interfaces, and universal platforms
Based on the ITU-T recommendation Y.3090 [9], a reference
and tools to enable interoperability and extensibility of the 6G
6G DTN architecture may consist of three layers: 6G physical
DTNs. These developments will gradually evolve into unified
network layer, 6G twin layer, and 6G network application layer.
tools to build the diverse 6G DTNs on universal platforms
Figure 3 provides an illustration of the reference 6G DTN
hosting data, models, and management functions.
architecture.
The 6G physical network layer refers to the target real-world
IV. FUNDAMENTAL BUILDING BLOCKS
6G network, consisting of physical network elements and their
operating environment. Depending on the need, a 6G DTN may With a reference 6G DTN architecture in place, the next
focus on different parts of the target 6G physical network (e.g., question is how we can realize it. The inception of a DTN
a radio cell, a radio access network, a transport network, and a begins with the blending of three fundamental building blocks:
core network) or the entire end-to-end 6G network. The 6G data, models, and interfaces. The degree of reliability with
physical network layer exchanges data and control messages which a DTN can mirror its physical counterpart depends on the
with the 6G twin layer. quality of data upon which its models are built, and the accuracy
The 6G twin layer is the core of a 6G DTN. It consists of of the models. Interfaces around a DTN, on the other hand,
three domains: data domain, model domain, and management enable its seamless interaction with the underlying physical
domain. Data domain is a data repository subsystem responsible network and associated network applications.
for collecting data from the 6G physical network in order to A. Data
build accurate and up-to-date models in the twin. The functions
of the data repository subsystem may include collection, Data is the basis for building a 6G DTN. Detailed, up-to-date
storage, service, and management of data. Model domain is a data enables the construction of high-accuracy models in the
service-mapping subsystem comprising of models to represent DTN. With AI/ML at its core, the efficacy of a 6G DTN relies
the real-world objects in the 6G physical network based on the fundamentally on both the quality as well as the quantity of data
collected data. There are two types of models: basic and collected from different parts of the physical network
functional. Basic models represent network elements and infrastructure.
topology to describe the target physical network. Functional What-to-collect (data type): In general, the collected dataset
models refer to analytical models for extracting insights from needs to be comprehensive enough to constitute a holistic
the DTN. Examples include network planning, traffic analysis, representation of the physical network. At the same time, proper
fault detection, network emulation, and prediction. analysis of specific network topology to determine required
Management domain is a DT entity management subsystem data type is crucial in order to avoid redundant data collection,
responsible for the management function of the 6G twin layer. which may result in unnecessary cost of compute and storage
Examples include model management (e.g., model creation, resources. Some of the data type examples include data
configuration, update, and monitoring), and security representing operational and provisional status of a physical
management (e.g., authentication, authorization, encryption, network, network telemetry, performance measurement, data
and integrity protection). from network services/applications and lifecycle management
(LCM) of network entities, and user data. Striking the right
 This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which 4
this version may no longer be accessible.
balance between data quality and quantity is essential in of the underlying data management framework, including the
building a sustainable and trustworthy DTN. LCM of data collection, storage, maintenance, and retrieval.
When-to-collect (time/frequency): In order to swiftly adjust Data management can either be an integral part of the overall
to the dynamic variation of a live network, the DTN requires DTN management framework or can be an independent entity
constant monitoring, performance reassessment, and timely in the twin layer. Automation is arguably the most desired
adjustment. Collection of data to feed a DTN, therefore, cannot feature of data management, along with other capabilities like
be a one-time event. The frequency and pattern of the time secure handling, storage, deposition and destruction of sensitive
series of data collection depend primarily on the data type. For network data (e.g., data related to user privacy), traceability and
example, data related to faults/events are usually collected in an accurate records of historical data transactions, and guarantee
event-triggered manner, whereas network performance specific of data integrity (i.e., accuracy, completeness, and consistency).
data are usually collected at a regular interval, with potential
B. Models
aperiodic trigger enablement depending on pre-defined
conditions/intents. Some types of data that statistically vary DTs of a physical network can be modularized at different
over shorter periods of time (e.g., port/interface statistics) levels, depending on the use case requirement. For example,
require higher collection frequency. On the other hand, some DTNs can be of individual physical network domains (like
other types of data that are critical to network’s operational radio access/core/transport network), or of a combination of
health (e.g., flow status and link failures) need to be gathered in collocated network entities/functionalities (like DT of cell
a time sensitive manner. site/near edge/far edge/cloud), or at the extreme case, a single
How-to-collect (mechanism/tool): Building a comprehensive DTN for the entire end-to-end physical network. The degree of
DTN requires network data with diverse characteristics, which, modularity, in turn, determines specific aspects of a DTN. For
in turn, necessitates usage of various data collection example, a DTN responsible for a cell site can be operating at a
mechanisms and tools. The state of physical devices may be high-level of detail using high-fidelity ray tracing employing
collected by the associated sensors that send periodic or event- multi-bounce specular reflection, 3D diffraction, and diffuse
triggered updates to data collectors. The state of physical sites scattering to enable site-specific optimizations in the base
may be captured by using lasers and drones besides connected station’s physical and medium-access control layers. A DTN in
sensors. Integrated sensing and communication utilizing the cloud would be responsible for end-to-end service analysis
terahertz (THz) or optical frequency bands can perform high- and network operation including resource management and, for
resolution sensing, localization and tracking, and 3D imaging example, AI/ML approaches for network slicing.
to collect data for building and updating the models of wireless Arguably, the most challenging design aspect of these
propagation environments. Data collection capabilities can be multimodal DTNs is their models that represent accurate digital
embedded in the 6G physical network to collect status about versions of physical entities constituting the real network peers.
network elements, topology, traffic, etc. Technologies enabling The generic requirements of a typical DTN model are manifold.
simultaneous and collaborative data collection from The crux of a DTN model meeting the diverse set of
heterogeneous network nodes in a performant and time- requirements is built upon two pillars: physically accurate
synchronized manner are critical in establishing a high-quality modeling and its functional manifestation using AI-powered
data collection process to serve as the solid foundation of an simulation, visualization and control.
ultra-reliable DTN. Physically accurate modeling: High-fidelity models are
Where-to-store (repository): Collection of massive amounts fundamental building blocks for 6G DTNs. The models need to
of network data comes with the burden of complex storage accurately capture the up-to-date characteristics of their
requirements. To that end, construction of a unified data physical counterparts, including geometries, materials,
repository for myriad network data becomes one of the key properties, lighting, behaviors, and rules, among others. One
prerequisites for creating a successful DTN. The salient features key requirement of the models for the 6G DTN is the need of
of a hyperconverged data repository include support for physically accurate simulation of radio wave propagations in
heterogeneous database, provision for efficient data services true-to-reality environments. This is particularly pertinent
(such as data augmentation and federation, historical data considering that 6G will utilize THz technology with ultra-wide
reporting snapshot and rollback, and fast database query), and signal bandwidth, integrate sensing with radar or lidar
assurance of high data availability, among others. technology, use extremely large antennas, and incorporate RIS.
How-to-access (retrieval): Acceleration is the key for Therefore, it is vital to accurately model the effects of radio
efficient operations (e.g., data ingestion, search, query, and wave reflection, diffraction, scattering, and multipath
visualization) on the DTN data repository which has large- propagation in the environment. Traditional software ray
volume data and fast data streams ingestion from the physical tracing techniques may have difficulty in performing efficient,
network. To this end, it is crucial to exploit the massive parallel physically accurate simulation of the radio wave propagations
capabilities provided by GPU and the corresponding tools such in the complex 6G systems. Ray tracing at-scale is extremely
as Dask to accelerate the three-phase data processing - extract, computationally intensive. At least two components are
transform, and load. This is particularly important when the required to bring radio frequency (RF) ray tracing at-scale to
database operations need to be real-time to synchronize the reality in a DTN: 1) Programmable hardware accelerators
DTN with its physical counterpart, derive lightning-fast data supporting the mathematical operations at the heart of ray
analytics, and achieve immediate closed-loop decisions. tracing and 2) optimized libraries for the ray tracing pipeline.
How-to-maintain (management): Long term stability, An example of a ray tracing hardware acceleration (RTX) are
adaptability, and performance of a DTN depend on the efficacy the ray tracing cores found in the many hundreds of streaming
 This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which 5
this version may no longer be accessible.
multiprocessors that comprise the NVIDIA Ampere class GPU interconnects is crucial in enabling real-time interactive
and that are capable of performing many trillions of floating- mapping between the DTN and its physical embodiment.
point math operations per second. There are several ray tracing Another important aspect of network-bound interfaces is
frameworks available including DX12, Vulkan, and OptiX. ‘decoupling,’ i.e., making the interfaces agnostic of the type of
Take OptiX for example. Based on compute unified device information being transported and specificities of physical
architecture (CUDA), it is an application framework for entities at its endpoints. Decoupling enables the DTN to use a
achieving optimal ray tracing performance on a GPU that takes harmonized interface towards the physical network, without
advantage of RTX hardware accelerators. OptiX enables the requiring data specific or network entity/topology specific
developers to concentrate on wave propagation rather than on customization.
low-level ray tracing optimization. Creating the geometry for Application-bound interfaces: Delivering requirements
the DT at interactive rates has become feasible based on capture (related to network management, optimization, and protocol
using instant neural graphics primitives [13]. validation) or intents of network applications to a DTN is
AI-powered simulation, visualization, and control: fulfilled through application-bound interfaces. Application-
Interactive virtual-real mapping and control is a key bound interfaces support the capability to convey the DTN’s
distinguishing characteristic of 6G DTN, compared to the features, including common data models and their performance
traditional link, system, or network simulators. However, real- to authenticated, third-party network applications. Speed,
time interaction and control is challenging due to the large latency, and bandwidth requirements for application-bound
amount of data, stringent latency requirements, and interfaces are less stringent and less demanding than network-
computation-intensive simulations. AI-powered functions are bound interfaces, which makes these interfaces realizable in
key to achieve an autonomous feedback loop between the 6G relatively lightweight ways with reduced complexity.
physical network and its DTN. As one example, AI techniques Application-bound interfaces provide a layer of abstraction
can leverage historical and real-time data to create and evolve towards the DTN, implying that the differences between
models in the 6G twin layer to constantly run simulations for network applications are transparent to the DTN, and the same
anomaly detection and network optimization, leading to set of interfaces serves seamless interaction between the
intelligent decision making for the target 6G physical network. endpoints, without requiring application-specific tailoring.
AI models can also be used to constantly run predictive “what- Intra/inter-DTN interfaces: Depending on the use case,
if” simulations in the 6G twin layer to determine in advance the scalability and flexibility requirements of a DTN drive the
consequences of hypothetical actions, which allows for impact degree of modularity, resulting in either a single DTN model or
analysis of changes, prevention of outage events, and finding multiple DTN models. For the latter, inter-DTN interfaces
optimal settings for future operations. In addition, AI enable efficient communication between various DTN models
techniques can create photorealistic 3D graphics to visualize the that are either co-located or distributed. For a DTN built upon
6G DTN and illustrate findings, facilitating the involvement of federated learning-based ML models, both communication
humans in the decision-making processes. between distributed model objects (during training) and
aggregation of trained objects can be facilitated through intra-
C. Interfaces
DTN interfaces. While standardization of inter-DTN interfaces
Open and standard interfaces between twin layer and between multiple DTN models is crucial for multi-vendor
physical network layer/network application layer are essential implementation, intra-DTN links between multiple model
towards proliferating a multi-vendor, interoperable DTN objects leading to a DTN model ensemble can be proprietary.
ecosystem. In general, the guiding principles of interface
protocols around DTNs include extensibility, backward V. OMNIVERSE FOR DIGITAL TWIN NETWORKS
compatibility, easy usage and accessibility, capability of
handling high concurrency of dataflow, and mechanisms to The ambitious goals and complexity of 6G demand DTN
ensure secure and reliable communication pathway between implementation frameworks that provide a high-level
DTNs and other network entities. programming abstraction, large scale modeling, AI/ML training
Network-bound interfaces: As the bridge between a DTN and and inference capabilities, tools to interact with the twin
its underlying physical network, network-bound interfaces through application programming interfaces, and visualization
provide efficient information exchange between the endpoints, capabilities. Such implementation frameworks enable
including both control and user plane data. Network-bound developers to rapidly implement DTs and let them focus on
interfaces carry various types of network data (e.g., provisional only the essential complexity of the problem. One such example
and operational data) collected via different methods such as is the Omniverse platform [11].
on-demand, subscription-based, event-triggered, direct Taking advantage of the Omniverse platform, the wireless
measurement, and indirect observation. The interfaces support industry is making progress towards building DTNs. For
versatile data requirements from DTN models, resulting in example, Ericsson has built a city-scale 5G DTN in Omniverse
varying data transfer speed requirements ranging from real-time [13]. The 5G DTN helps accurately simulate the interplay
(~ms) to near-real time (~second) and to non-real time between the physical network and the environment for
(~minute). It is imperative to support management and control maximum performance and coverage. As another example,
signaling through these interfaces for the DTN to configure data HEAVY.AI, a startup that offers an eponymous GPU-
collection protocol, time/frequency of collection, and other accelerated analytics platform that enables users to interactively
configurations of network elements in the physical network. query, visualize, and power data science workflows over multi-
Realization of network-bound interfaces using low-latency
 This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which 6
this version may no longer be accessible.
in many facets of wireless research, development, and
deployment in the future.

VI. CONCLUSION
The fast growth of network scale towards 6G and the
stringent performance requirements of diverse use cases call for
innovative tools and platforms. The unique capabilities of
DTNs make them a powerful technology for the design,
analysis, diagnosis, simulation, and control of 6G wireless
networks. In this article, we have provided an overview of
DTNs for 6G and presented a real-world example to illustrate
how DTNs can be built and operated in Omniverse. The
innovations and applications of DT technology in wireless
Figure 4: An example of building digital twins with Omniverse. networks are as vast as the imagination. It will be exciting to
see how DT technology will improve lives and transform
billion record datasets, has recently started building HeavyRF, industries on the path to 6G and beyond.
a network DT application in Omniverse [15].
As illustrated in Fig. 4, HeavyRF integrates Omniverse with
REFERENCES
the HeavyDB structured query language (SQL) backend,
leveraging the latter for real-time processing of lidar and other [1] F. Tao, H. Zhang, A. Liu and A. Y. C. Nee, “Digital Twin in
Industry: State-of-the-Art,” IEEE Transactions on Industrial
terrain data, as well as executing ray-traced RF propagation
Informatics, vol. 15, no. 4, pp. 2405-2415, April 2019.
simulation itself. Accurate 5G and 6G network planning [2] X. Lin et al., “5G New Radio: Unveiling the Essentials of the
requires simulation to be performed on unprecedentedly high- Next Generation Wireless Access Technology,” IEEE
resolution terrain data, due to the higher transmitter densities Communications Standards Magazine, vol. 3, no. 3, pp. 30-37,
required as well as the high degree of clutter attenuation Sep. 2019.
associated with the use of higher transmission frequencies. [3] W. Saad, M. Bennis and M. Chen, “A Vision of 6G Wireless
HeavyRF can run real-time RF simulations on lidar data at Systems: Applications, Trends, Technologies, and Open
resolutions as granular as 10 cm, which are two orders-of- Research Problems,” IEEE Network, vol. 34, no. 3, pp. 134-142,
magnitude higher than the common 10 m plus resolutions used May/June 2020.
[4] H. X. Nguyen, R. Trestian, D. To and M. Tatipamula, “Digital
by conventional RF planning solutions for 4G networks,
Twin for 5G and Beyond,” IEEE Communications Magazine, vol.
creating true-to-life representations of the physical landscape 59, no. 2, pp. 10-15, February 2021.
that a telco operates in. In addition, it can incorporate building- [5] L. U. Khan, W. Saad, D. Niyato, Z. Han and C. S. Hong, “Digital-
level customer and demographic information to tie the results Twin-Enabled 6G: Vision, Architectural Trends, and Future
of network simulation with an operator’s key business drivers. Directions,” IEEE Communications Magazine, vol. 60, no. 1, pp.
The simulation itself is run directly in the HeavyDB database 74-80, January 2022.
using the construct of an SQL table function that takes as inputs [6] Y. Wu, K. Zhang and Y. Zhang, “Digital Twin Networks: A
subqueries representing the terrain, transmitter, and building Survey,” IEEE Internet of Things Journal, vol. 8, no. 18, pp.
data, avoiding the overhead that would otherwise be associated 13789-13804, 15 Sept.15, 2021.
[7] H. Ahmadi, A. Nag, Z. Khar, K. Sayrafian and S. Rahardja,
with moving these large datasets between servers/software
“Networked Twins and Twins of Networks: An Overview on the
platforms. The use of SQL to orchestrate the simulation also Relationship Between Digital Twins and 6G,” IEEE
makes it easy to perform inline extract, load, and transform on Communications Standards Magazine, vol. 5, no. 4, pp. 154-160,
the input data and various enrichment operations on the December 2021.
simulation output, such as joining the terrain locations and [8] ITU-R, “Future Technology Trends of Terrestrial IMT Systems
simulated signal strengths to building polygons to measure the Towards 2030 and Beyond,” preliminary draft new report, June
minimum, maximum, and average signal power received for 2022.
each building, as well as capture the principal towers serving [9] ITU-T, “Digital Twin Network – Requirements and
each building. HeavyRF optimizes communication between Architecture,” Recommendation ITU-T Y.3090, Feb. 2022.
[10] IEEE, “Specification of Sensor Interface for Cyber and Physical
HeavyDB and Omniverse to provide a real-time capability to
World,” IEEE P2888.1. [Online]. Available:
manipulate network parameters such as tower locations, https://standards.ieee.org/ieee/2888.1/7676/. Accessed on Nov.
antenna gain and transmissions frequency and then visualize the 21, 2022.
results in full 3D. Figure 5 provides a visualization of HeavyRF [11] NVIDIA, “Digital Twins with Omniverse Platform,” [Online].
for network planning and operation. Users can place new Available: https://www.nvidia.com/en-
towers or change the parameters of existing towers, and see the us/omniverse/solutions/digital-twins/. Accessed on Nov. 21,
resultant changes in simulated RF signal across the scene’s 2022.
terrain, as well as rolled up to the building level and tied to key [12] X. Lin, L. Kundu, C. Dick, and S. Velayutham, “Embracing AI in
customer metrics. 5G-Advanced Towards 6G: A Joint 3GPP and O-RAN
Perspective,” arXiv preprint arXiv:2209.04987, Sep. 2022.
In summary, Omniverse is a powerful platform for delivering
Accessed on Nov. 21, 2022.
DT solutions. As more and more innovative DT applications are
built in Omniverse, we anticipate that it will be a game changer
 This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which 7
this version may no longer be accessible.

Figure 5: A visualization of HeavyRF built with Omniverse for network planning and operation.
[13] T. Müller, A. Evans, C. Schied and A. Keller, “Instant Neural
Graphics Primitives with a Multiresolution Hash
Encoding,” ACM Transactions on Graphics, vol. 41, no. 4, pp.
1-15, Jul. 2022.
[14] Ericsson, “Next-generation Simulation Technology to
Accelerate the 5G Journey,” [Online]. Available:
https://www.ericsson.com/en/blog/2021/4/5g-simulation-
omniverse-platform. Accessed on Nov. 21, 2022.
[15] NVIDIA, “HEAVY.AI Delivers Digital Twin for Telco
Network Planning and Operations Based on NVIDIA
Omniverse,” [Online]. Available:
https://blogs.nvidia.com/blog/2022/09/20/heavy-ai-digital-twin-
omniverse/. Accessed on Nov. 21, 2022.

BIOGRAPHIES
Xingqin Lin is a Senior 3GPP Standards Engineer at NVIDIA.
Lopamudra Kundu is a Senior Standards Engineer at
NVIDIA.
Chris Dick is a Wireless Architect at NVIDIA.
Emeka Obiodu is a Global Marketing Manager at NVIDIA.
Todd Mostak is the CTO and co-founder of HEAVY.AI.