WINNIIO

WHITE PAPER

From Dropped Calls to Holographic Cities

Why Physics-Based Network Digital Twins Are the Foundation of High-Fidelity Spatial Societies

Nicolas Waern

CEO, WINNIIO AB / Life Atlas

Co-Chair, Digital Twin Consortium Telecom Working Group

May 2026

Classification: Public

Abstract

The telecommunications industry is approaching an inflection point. Current 5G networks struggle with basic mobility management: 2–5% of handovers fail in dense urban environments, costing operators billions annually in churn, NOC triage, and degraded experience. These same networks are expected to support holographic communication, persistent AR overlays, and city-scale spatial computing within the next decade.

This paper argues that physics-based network digital twins are not merely an optimization tool for today’s RAN—they are the foundational infrastructure for high-fidelity holographic societies. The same ray-tracing engine that predicts where a phone call drops is the engine that will compute where a hologram can persist. The same 3D city model that reveals RF shadows will anchor spatial content to physical reality. The same reinforcement learning agent that tunes handover parameters will orchestrate the holographic rendering pipeline.

We present a working implementation using NVIDIA Sionna (open-source GPU ray-tracing), Japan’s PLATEAU CityGML (government open 3D city data), and CesiumJS (spatial visualization), demonstrating the complete pipeline from building geometry to RF simulation to handover analysis to business impact quantification.

1. The Handover Problem Is a Physics Problem

When a mobile device moves between cell towers, the network must execute a handover—transferring the connection from one cell to another without dropping it. In theory, this is a solved problem. In practice, it fails billions of times per day worldwide.

The root cause is not software. It is physics. The radio signal between a tower and a device is shaped by every building, every window, every tree, every vehicle between them. A 120-meter glass tower in Nishi-Shinjuku reflects, refracts, and diffracts RF energy in patterns that no empirical model can predict.

Yet the industry’s dominant planning tools—Atoll (Forsk), ASSET (Aircom), Planet (Infovista)—still rely on empirical propagation models derived from Okumura-Hata measurements taken in 1968 Tokyo. These models treat the city as a statistical abstraction: average building height, average street width, average clutter. They draw smooth coverage circles that bear little resemblance to actual RF conditions at street level.

The result: handover boundaries are tuned by trial and error. Network engineers adjust Cell Individual Offset (CIO) parameters based on post-event failure logs, iterating through 15-minute batch cycles. By the time a problem is identified, thousands of subscribers have already experienced degraded service.

2. Ray-Tracing: From Gaming to Infrastructure

The solution exists. It was developed for a different industry.

GPU ray-tracing—the technology that makes video game reflections look photorealistic—is mathematically identical to RF propagation simulation. A photon bouncing off a glass facade follows the same laws of reflection and refraction as a 3.7 GHz radio wave. The only differences are wavelength, material properties, and the addition of diffraction.

NVIDIA recognized this convergence and released Sionna: an open-source (Apache 2.0) GPU-accelerated channel simulator built on PyTorch. Sionna implements ITU-R P.2040 radio material models, UTD wedge diffraction, and can trace millions of rays through arbitrary 3D geometry in seconds on a consumer GPU.

The implication is profound: the same computational infrastructure that will render holograms in real-time can simultaneously compute the RF environment that carries the holographic data. Ray-tracing is not two separate technologies. It is one engine with two applications.

3. The 3D City as Shared Infrastructure

Ray-tracing requires geometry. The accuracy of the RF simulation is bounded by the accuracy of the 3D scene.

Japan’s Ministry of Land, Infrastructure, Transport and Tourism (MLIT) has, through its PLATEAU Project, created the most comprehensive open 3D city dataset in the world: CityGML LOD2 models for 250+ cities, covering 23 million buildings with measured heights and roof geometry.

This 3D city model serves multiple purposes simultaneously:

RF propagation simulation — building geometry determines signal reflection, diffraction, and shadowing

Spatial content anchoring — AR/holographic content needs physical surfaces to attach to

Urban planning simulation — new construction impact on RF and spatial services

Autonomous navigation — vehicle and drone path planning with RF coverage awareness

Emergency response — communication-aware evacuation routing

The 3D city model is not a telecom asset. It is a civilizational asset that happens to make telecom optimization dramatically better as a first application.

4. What Holographic Societies Demand

A holographic society—one where persistent, high-fidelity spatial content is woven into everyday life—imposes network requirements that are qualitatively different from today’s mobile broadband:

The gap between today’s performance and holographic requirements is not bridgeable by hardware alone. You need predictive, physics-aware network intelligence that knows, before the user moves, what the RF environment will look like at their destination—and has already prepared the handover.

5. The Digital Twin as the Bridge

5.1 Today: Reactive MRO

Current MRO operates in 15-minute batch cycles. Handover failures are logged, categorized, and CIO parameters are adjusted retroactively. The network is always responding to yesterday’s problems. A physics-based digital twin identifies failure zones before any subscriber experiences them — by simulating UE movement through ray-traced coverage maps.

5.2 Near-Term: Predictive MRO via Reinforcement Learning

The digital twin becomes a training environment for ML-based optimization. A reinforcement learning agent explores millions of CIO configurations in simulation—each evaluated against ray-traced coverage—and converges on optimal parameters in hours rather than weeks. The trained model deploys as an xApp on the near-RT RIC via E2 interface, first in shadow mode, then in closed-loop control.

The O-RAN architecture was designed for exactly this pattern. What has been missing is the physics-based simulation environment to train the models. Empirical propagation models produce training data that is statistically plausible but physically wrong—and ML models trained on wrong physics will make wrong decisions in production.

5.3 Future: The Holographic Network Controller

In a holographic society, the network controller must know:

Where the user is (sub-10 cm positioning from the same ray-tracing engine)

Where the user is going (trajectory prediction from mobility traces)

What the RF environment looks like along that path (pre-computed from the digital twin)

Which holographic content is active at which locations (spatial content registry anchored to 3D city model)

How to pre-stage rendering and network resources (proactive resource allocation, not reactive)

Every one of these capabilities is a natural extension of the MRO digital twin pipeline. The holographic network controller is the MRO digital twin, evolved.

6. The Convergence of Three Ray-Tracing Pipelines

All three pipelines consume the same 3D city model. All three use ray-tracing on GPU hardware. An operator who builds a physics-based network digital twin today is simultaneously building one-third of the holographic rendering infrastructure for free.

7. The Open Stack Imperative

This convergence only works on open infrastructure. Our implementation:

NVIDIA Sionna — Apache 2.0, GPU ray-tracing, PyTorch-native

PLATEAU CityGML — Government open data, OGC standard, 250+ cities

CesiumJS — Apache 2.0, 3D Tiles, spatial visualization

Stable Baselines3 — MIT license, RL training

O-RAN interfaces — Open specifications, multi-vendor RIC

Intel COTS hardware — No proprietary ASICs required

Every component is replaceable. The value is in the methodology and orchestration, not in any single component.

8. Implementation: Tokyo MRO Digital Twin

We built a working MRO digital twin for central Tokyo’s 5G network:

22 cell sites across 7 districts, 3 frequency bands (n77, n78, n257)

4,000 buildings with ITU-R P.2040 material classification (concrete, glass)

SRTM 30m terrain with DEM-adjusted tower heights

3GPP TR 38.901 antennas (65° HPBW, 18 dBi, 30 dB F/B)

10⁶ rays per transmitter with UTD wedge diffraction

5 UE mobility traces with RSRP sampling and HO event classification

Calibration pipeline ready for MDT data (RMSE < 8 dB target)

RL environment with MRO reward function

The entire pipeline runs end-to-end on a free Google Colab T4 GPU in approximately 20 minutes.

9. The Economics of Getting There Early

For a network the scale of Rakuten Mobile Japan (50,000 sites, 10M subscribers), reducing handover failure rate from 2.5% to 0.5% through ML-optimized CIO yields:

1.6 million fewer failures per day

Billions of JPY in annual savings (NOC, churn, QoE)

0.1% churn reduction = tens of billions JPY in subscriber LTV

But the strategic value exceeds the operational savings. The operator who builds the digital twin for MRO today has:

A calibrated 3D city model ready for holographic content anchoring

A ray-tracing pipeline that extends to rendering and acoustic simulation

An ML infrastructure for autonomous network optimization

An open, composable architecture that absorbs new capabilities without rearchitecting

A methodology (SMILE / SPIN Twinning) for extending the twin collaboratively

Operators who wait for holographic use cases to emerge before building digital twin infrastructure will find themselves 3–5 years behind those who started with MRO.

10. Conclusion: Reality Is the Persistent Layer

A holographic society is not built by projecting content into empty space. It is built by understanding physical reality so precisely that digital content can coexist with it seamlessly. The 3D city model is the foundation. Ray-tracing is the engine. The network digital twin is the first application.

The tools for solving today’s handover problems and the tools for enabling tomorrow’s holographic experiences are the same tools. The question is not whether to build them, but whether to start now—with immediate ROI from MRO optimization—or later, when the infrastructure gap has become a competitive moat for those who moved first.

The path from dropped calls to holographic cities is not a leap. It is a pipeline. And the first stage of that pipeline is running on a free GPU right now.

Contact

Nicolas Waern — CEO, WINNIIO AB

ceo@winniio.io

Co-Chair, Digital Twin Consortium — Telecom Working Group

© 2026 WINNIIO AB. This document may be distributed freely with attribution.