High-Dimension Diffusion for Real-Time Network Assurance using Generative AI
Generative AI for Network Assurance
This award-winning solution showcases our commitment to pushing the boundaries of what’s possible in wireless technology.
It utilizes generative diffusion models to optimize and enhance the performance of 5G networks, ensuring faster, more reliable, and more efficient connectivity for users.
Notable features include data-driven joint characterization of hundreds of KPIs for a global view of the network, flexible forecasting, the ability to zoom in and out of the network with ease, straightforward recalibration, generative high-dimensional diffusion, and deployment on 3GPP Release 16 Network Data and Analytics Function (NWDAF) for truly real-time analytics.
It was awarded the prestigious “Most Integration Potential” award at Deutsche Telekom’s most recent T Challenge.
Solution Highlights:
1. Data-Driven Joint Characterization of hundreds of KPIs in Core, RAN, and OAM Systems for global view of the network.
2. Flexible forecasting for any view in any granularity of time, at any point of time.
3. Zoom in-and-out of network with ease, view the network from observations in Control and Data Plane, Network Functions in Control and Data Plane, to Logical planes connecting any Network Functions, to view of the Core as a whole, to finally view of both Core & RAN for fundamentally global view.
4. Recalibrate with ease, our solution can adapt to commercial demands in all areas, you can remove nodes of network that aren’t interesting to you to observe, while saving compute without any shift in your MLOPS.
5. With Generative High-Dimensional Diffusion, create meaningful insights from large quantities of data such as 5G KPIs and logs with ease.
6. Deployment on 3GPP rel16 Network Data and Analytics Function (NWDAF) for immediate integration for truly real-time analytics.
Deep Dive:
Our solution is intended to provide both telescopic and microscopic views of the network in real-time. At the smallest possible level in Fig 1, our model is a function of the smallest area of interest which could be a network function for example the Access and Mobility Management Function (AMF). Predictions at this level already beat the state-of-the-art, however, with lack of global context, anomalies and events of interest that are not independent cannot be fully realized.
Fig 1: Smallest IP for Local Predictions
Incomplete context means failures and lost revenue, and poor quality of customer service. We overcome this problem by creating context across network functions or areas of interest in the network as seen in Fig 2, this context can be used to perform early detection of anomalies and reduce false-alarms.
False alarms are reduced by understanding an anomaly threshold that our model learns during its offline-training but can also be controlled by operators if they wish.
Fig 2: Early Detection and Forming Network Representation
The Inter-Network Component High-Dimensional Diffusion IP then based on the marginal-distributions, characterizes the joint-distribution of the network, creating a global view to fully understand the anomalies in the network for superior predictions and reliable anomaly-detection as seen in Fig 3.
Fig 3: Proof-of-Concept Generative Diffusion inside 5G System
Finally, our solution is deployed on the NWDAF, a 3GPP compliant Network function in the 5G Core for real-time analytics. We implement our solution for seamless integration, deployed on the MLOPS layer of the NWDAF without any modifications in the data-ingestion services.
