5G analytics systems and the predictive network
Before we dive into the technical specifics, let’s first understand why predictive networks are relevant for operators.
Business benefits of predictive networks
The predictive network enables a host of benefits to service providers, including:
- Ensure Quality of Service (QoS), Quality of Experience (QoE), and Service Level Agreements (SLAs) for 5G services by intelligently tuning to network conditions
- Prevent performance issues, predicting them before they occur and taking anticipatory corrective measures
- Lower downtime by minimizing network disruptions
- Support advanced next-gen use cases
- Maximize return on network-capacity investments
- Continuously tune the network and balance network load to prevent over-engineering
Operator success will be defined by ensuring that their predictive systems can manage their networks more efficiently while ensuring a superior customer experience.
What exactly is a predictive network?
Traditionally, proactive systems, popularly known as Self-Organizing Networks (SON), have analyzed real-time network data to detect anomalies either when an event occurred, or the system was impacted. Their algorithms would then suggest a resolution or take corrective actions that involve human intervention.
In contrast, a predictive network uses months of historical data to predict the recurrence of network events. It essentially treats each event as a statistical problem that can be solved using Artificial Intelligence (AI) and Machine Learning (ML) techniques, learning from the past and predicting what it should anticipate.
The role of predictive analytics in the 5G ecosystem
5G acts as a catalyst for predictive networks because it introduces dedicated data analytics network functions (NFs). Standards bodies have defined Network Data Analytics Function (NWDAF) and Network Exposure Function (NEF) to provide a centralized predictive analytics platform for the 5G core network. These NFs collect and expose network data in real-time to machine learning applications deployed at the network edge.
3GPP has also defined standardization guidelines for data collection, predefined analytics insights, and data exposure interfaces for customers. Accordingly, the NWDAF collects data from multiple sources like user equipment, network functions, network edge, data plane, operation, administration, and maintenance (OAM) systems, and more.
The data gathered by NWDAF can be fed into an analytics engine to provide insights and take necessary actions. It is designed to defragment proprietary network analytics solutions and standardize the way mobile network data is produced and consumed.
The 5G NWDAF, combined with AI and ML, empowers proactive closed-loop network operations, ensuring the network can analyze historical data and learn from it.
The ubiquitous architecture of 5G includes an edge NWDAF co-located with core network functions and a central NWDAF. The edge NWDAF serves low and ultralow latency use cases, while the central NWDAF supports use cases that do not have real-time requirements. It also includes functions such as the data and ML models repository that help ensure the AI/ML models and continuously trained.
Along with NWDAF, 5G also introduces data analytics functions at the following layers:
- Big data, management, and orchestration (Big Data/MDAF)
- Application function level (AFDAF)
- User equipment/RAN (DAF)
- Data network (DN-DAF)
With these critical 5G functions and an edge platform, the network can meet the performance needs of more complex next-gen use cases. It will develop a system to capture network data from all functions and understand the network; measure and predict service performance; and proactively ensure high QoE, QoS, and network availability round the clock. Automating this system using AI/ML will enable operators to maximize return on investment.
Predictive analytics use cases
Some of the many use cases that can be powered by NWDAF and AI/ML include:
|Load Analysis||Network Performance||Service Assurance||Device Behavior Analysis|
|Load level of network slice instance||Congestion information of user data in a specific location||Network performance predictions by analyzing traffic changes at the cell or area level||Behavior analytics like communication patterns for individual or groups of UEs|
|Load analytics information for specific NFs||Network load performance in an area of interest||Density changes in important alarms based on historical data||Abnormal behavior and anomaly detection for individual or groups of UEs|
Analytics systems today and into the future
The analytics journey can be classified into four stages:
- Context-sensitive or diagnostic analytics gathers and visualizes data, identifies patterns in historical data, and detects why the event occurred.
- Predictive analytics analyzes large volumes of data and forecasts probable future events, using machine learning techniques.
- Prescriptive analytics provides insights and options to optimize the network.
- Cognitive analytics takes optimized decisions to rectify problems without human intervention.
Broadly speaking, the industry is currently at the predictive analytics stage. With the advent of 5G analytics functions and maturity in AI/ML algorithms, we will soon see cognitive systems taking intelligent decisions without human intervention.
The future predictive network will analyze large datasets from multiple channels and identify complex network patterns, thus making near-accurate predictions. To achieve this, the operator’s monitoring and maintenance system must rely on 5G NWDAF and advanced predictive algorithms, both of which have equally innovative core functionalities.
Growing data is a reality of modern networks, and analyzing this data is the key to business success. It’s therefore essential for operators to devise and optimize their analytics system and continuously measure its maturity, especially as cloud technology permeates.
5G data analytics functions and AI/ML are set to disrupt the way we design and operate our networks. High compute and 5G network speeds make it easier to analyze huge amounts of data and transmit the results in real-time.
AI and ML technologies also enable the development of more sophisticated data analytics systems that can do more than analyze data and relay information. These intelligent systems can perform self-assessments, auto-adjust, and perform complex tasks on their own, without needing human intervention. As we see more widespread adoption, they will be transformative for the industry.
Director – R&D (5G)