Discover how high-fidelity network digital twin can enhance the provisioning and performance of network slicing for low-latency applications. Our O-RAN-based service management and orchestration (SMO) solution combines a digital twin service and AI and interfaces with RAN automation to evaluate, create and manage network slices at scale with data-driven KPI predictions. ========================= Frequently asked questions ========================= Q1) What is the objective of the wireless digital twin? The wireless digital twin is an accurate representation of the physical wireless network which is then used to evaluate a wide range of what-if scenarios ahead of time to predict KPIs and enhance network planning / dimensioning as well as improve network efficiency. While an indoor 5G network has been used in this demonstration, digital twins are key to our vision for AI-native 6G. Q2) What are the components of the digital twin? The key components of the digital twin in this work are the: [a] radio propagation model, [b] network model, and [c] neural KPI predictor. The radio model and network model are site-specific and deployment-specific respectively, i.e., “twins” of their physical counterparts. The neural KPI predictor extracts the intelligence out of these high-fidelity models. Q3) How is a digital twin different from a simulation? The digital twin used here is an exact representation of the physical network instance, from a radio propagation and network perspective, as opposed to statistical models that typically characterize simulations. The fidelity of the digital twin is verified using observations from the physical network data. The KPI predictor is also updated and fine-tuned using physical network data. Q4) How do we envision the architecture of digital twin-assisted KPI prediction? In this realization, the digital twin resides at the SMO. The neural KPI predictor is trained on digital twin datasets across various what-if scenarios, updated using physical network data. Digital twin predictions are invoked using an rApp designed for this purpose. In addition, a generative AI (gen AI) module is used to interface with the digital twin predictions for automation / human interface purposes. Q5) How is the digital twin-assisted KPI prediction used for network slicing? The digital twin and KPI predictor help us estimate network slice-specific KPIs before deploying the slice in the physical network. The digital twin simulates across an ensemble of possible physical realizations, i.e., distributions of network slice performance. This way, the network operator can assess resources and KPIs accurately, avoid surprises and make efficient use of system resources. Q6) How did we construct the digital twin’s 3D model of the indoor space? Lidar scans, drone photography and other techniques were used as data sources. Publicly available software libraries were used to build the model. See our demo from MWC 2024 for more details: https://youtu.be/h9cfOFi5TfQ Q7) Can you provide some details on the indoor deployment, for example, on the user locations, traffic mapping, etc.? A superset of user locations is considered, i.e., 24 fixed physical locations. Out of these, multiple random subsets of users are chosen for each “drop”, with different traffic mapping for each drop, also at random, and statistics are combined over multiple drops. Q8) Are the two frequency layers used with carrier aggregation (CA)? No. The two frequencies are handled by independent transmission and reception points (TRPs) or radio units (RUs) that are not operating in CA mode. Q9) Are device locations used to predict KPIs? No, the KPI predictor does not use device location information, i.e., location is not an inference input. Q10) Why does the throughput seem to go up with fewer TRPs or RUs for energy savings? The air interface resource utilization achieved in this demo was low and the visible increase in throughput is a statistically small difference. At higher resource utilization, throughput is expected to decrease with energy savings. Q11) How are network slices operated? The scheduler serves low latency communications (LLC) traffic with hard priority. Network resources are not physically partitioned.

 15188      60