RAN Digital Twins Drive the Future of Wireless Network Performance and Utility

Twinning the Radio Access Network

Mobile devices around us contain an enormous amount of rich information, including sensing, telemetry, applications, patterns of life, performance measurement and more. The heart of the DT-RAN is to leverage better data and models for our mobile networks to maintain an accurate and real-time model for the world around us that reflects the real world much more accurately than we could with simplified aggregate models. In the past, it was sufficient to design for a generic “urban” or “rural” or another such generic environment—however, as we seek smarter, denser, more performant networks—it is essential to leverage this information for localized models which can be used for optimization, coordination, improved resilience and efficiency and can enable new applications.

Core enabling network data for RAN Digital Twins:

• Channel measurement: Includes key metrics such as channel impulse responses, receive signal levels and interference levels
• Performance metrics: Throughput, latency and performance metrics are essential for optimizing network utility.
• Positioning and telemetry data: Provides insights into device locations, movement and environmental context.
• Mobile sensor data: Incorporates lidar, imager, and video to enhance the accuracy of environmental models.
• Spectrum awareness: Radar returns, integrated sensing data and communication insights add valuable context.
• Environmental factors: Weather conditions, including humidity, smog, rain, snow and sea, contribute to accurate modeling.

By leveraging this wide range of real-time data, the DT-RAN maintains an accurate and dynamic model of the network environment. This model can optimize performance, improve coordination and enhance network resilience and efficiency.

Driving Network Performance Using DT-RAN

Wireless performance is a crucial aspect of the mobile user experience, coverage and data rates and for network operators to realize the full value and return on their infrastructure investment. Network parameters have been tuned for years using techniques like self-organizing networks (SON) but have been known to be combinatorically large and complicated to optimize thousands of parameters of thousands of sectors with ever-changing loads and conditions. As techniques such as neural receivers, neural beamforming and parameter-heavy AI models throughout the whole RAN stack are introduced, this problem becomes even more complex, necessitating tuning and optimization on an accurate and simulated version of the network. DT-RAN is a core enabler of this, allowing for realistic modeling and optimization on top of a digital twin network. We have focused heavily on enabling performance within the physical layer (L1) in its capabilities, such as neural uplink (PUSCH) receiver models in OmniPHY 5G, which can either generalize or specialize to different network conditions.

Recently, we have worked heavily on the tasks of gathering and modeling macro-cell channel impulse response data, which models propagation effects accurately under an NTIA-funded public wireless supply chain innovation fund (PWSCIF) grant. This work is highlighted in the report’s modeling of wireless propagation environments section, which presents numerous methods, including generative AI models (as shown in Figure 1) as well as real-time ray-tracing and differentiable ray-tracing models, which comprise powerful new tools for accurately modeling communications systems.

Figure 1: Generative AI-based Impulse Response DT-RAN Model

These allow for accurate test and evaluation (T&E) and performance characterization and provide essential models for parameter and machine learning model tuning for real-world localized propagation conditions. This is important for transparent AI/ML models within today’s 5G OpenRAN networks but also is increasingly critical for emerging 5G Advanced networks with new 3GPP Rel 18 and 19 features and forthcoming 6G features in Rel 20+, which use additional machine learning features throughout the stack. These features must be optimized and validated in realistic digital twin simulation models.

DeepSig’s OmniPHY End-to-end learned physical layer modem solution provides a further prototype for the future of the 6G air interface. It can directly leverage the DT-RAN for optimization at additional stack levels, including beamforming, modulation, pilot, and frame design, in an end-to-end manner, which provides additional performance and efficiency benefits.

Spectrum Awareness With RAN Digital Twin

Figure 2: Spectrum Awareness for DT-RAN with OmniSIG Software

Spectrum sensing has entered the modern-day RAN through the deployment of citizens’ broadband radio service (CBRS), which allows for the sharing of spectrum in the U.S. between federal agencies and both private and public networks. One of the core enablers is spectrum sensing, which can help avoid interference and allow for real-time adaptation to allow for denser and fuller spectrum usage while guaranteeing priority access for spectrum owners and applications. This paradigm continues to expand within CBRS 2.0, and additional spectrum bands for sharing are considered, as well as the expansion of localized clutter models, which allow for more dense spectrum sharing and re-use and form a sort of DT-RAN light model.

Figure 3: RAN-Based Emitter Sensing for Spectrum Awareness

Aside from formal spectrum sharing mechanisms such as CBRS, spectrum sensing provides a critical tool in which adjacent private networks, sources of electromagnetic interface (EMI), and even wireless cyber threats can be detected, characterized, and tracked within the DT-RAN to provide a more intelligent internal model for how networks can adapt, respond, mitigate, and optimize around such constraints. Over the next few years, as the use of DT-RAN grows, we expect spectrum sensing to grow hand in hand with it within the RAN to provide ground truth feedback within and between networks and wireless services to help extract maximum value from our scarce shared spectrum resources.

Simulation and Generative RAN Models

Mobile telemetry throughout the network provides a valuable resource, collating signal quality performance metrics (KPIs) such as received signal strength (RSSI) and signal quality (RSRQ) alongside metrics such as throughput, call drop rate or outage, with latency, and power levels with location, channel response and other statistics. This information can enable high-quality localization, location-based services, intelligent connectivity planning, and continuous improvement of network configuration or allocation. By leveraging DT-RAN, performance expectations can be guaranteed in simulation to provide certain levels of performance and coverage across new and unseen locations, conditions or paths, offering unprecedented levels of assurance and reliability.

Traffic Modeling and RAN Optimization

Finally, DT-RAN can be used to enable network operators to understand user-level applications, content, and traffic demands as a fine-grained model within their network.  This could lead to improved network and content delivery and even optimization of semantic communications on their network, jointly optimizing both channel impairments and content transport, which offers significant advances in efficiency and capacity.  This level of intelligence across the true user demands of each layer of the access network allows for levels of optimization and efficiency, which 6G will be able to exploit as has never been possible before.

Key Takeaways

OpenRAN is a key enabler for DT-RAN, as it allows for concepts and standardization of interfaces. This will unleash significant new amounts of mobile network data, including pervasive network sensing, channel response and telemetry data. These, in turn, can lead to powerful new RAN digital twin models. These can help optimize numerous layers of the RAN stack and deliver unprecedented performance optimization, content delivery, network efficiency, and the foundation for an intelligent 6G network.

The report, Digital Twin RAN: Key Enablers, was published on October 16, 2024, and includes contributions from industry leaders such as NVIDIA, Bell CA, CMCC, Dell, Ericsson, Nokia, Qualcomm, Rakuten, Verizon and VIAVI. Download it here.