Exploring the potential of harnessing the power of Google Cloud for Ericsson's Network Planning and Optimization Solutions
Regardless of the network architecture or generation, each radio access network (RAN) presents unique challenges in network planning and optimization. Ericsson’s network planning and optimization solutions, Cognitive Software, utilizes an innovative AI-based approach to optimize network performance and is exploring Google Cloud to rapidly adapt to dynamic use cases and faster time to market.
With hundreds of thousands of cells in a network to assess, identifying and resolving network issues is a daunting task. However, the strong legacy of AI-driven network optimization with Cognitive Software delivers intelligent solutions to detect and resolve network anomalies.
During technology exploration, Google Cloud and Ericsson Cognitive Software have collaborated to integrate advanced hyperscale cloud services, such as Vertex AI and BigQuery. This integration has led to a demonstration of the Cell Anomaly Detector use case with Google Cloud's Vertex AI, first launched at Mobile World Congress 2024 in Barcelona.
The demo showcases the capabilities of using Ericsson's Cognitive software AI model deployed on Google Cloud's Vertex AI to detect anomalies in cellular networks, which represents an exciting advancement at the intersection of cloud computing and telecommunications technology.
By utilizing AI technologies, the Google Cloud and Ericsson exploration further seeks to enhance network planning, optimization, and operation while providing telcos with dynamic, scalable solutions that significantly reduce time to market.
The Cell Anomaly Detector: A game-changer in network performance management
The Cell Anomaly Detector from Ericsson's Cognitive Software was developed to proactively identify, classify, and address cell issues in radio access networks (RAN).
This use case is a trailblazer in the telecommunications field, running a multi-dimensional analysis on over 200 KPIs to unearth hidden patterns and recognize abnormalities quickly and accurately. The tool classifies abnormal cells into several issue classes with an impressive accuracy of 98% — arguably higher than human skill level.
The results are then displayed on a web user interface, offering detailed insights into the issues and providing APIs to connect with other applications already in use by communication service providers. This approach has delivered significant improvements in network KPIs, reduced customer complaints, and minimized operational expenditure (OPEX) for over 60 network operators worldwide.
Cell Anomaly Detector
Diagnostics for Al powered cell anomaly classification
Al powered automatic analytics for the entire RAN every day
- Automatic detection of >50 different problem allowing the optimization team to significantly resolve more issues, reducing the resolution time.
- Directly boosts the network performance and minimizes user-perceived performance degradations
Multi-vendor and multi-technology support
Benefits & differentiators
- Al-driven quantitative causal analysis
- Unique functionality in the market
- Issue detection and classification is 75 percent faster1 compared to traditional methods
- Detection of 50 percent more performance-impacting issues2
- Field-verified accuracy of 92-98 percent2,3 in issue detection and classification
30%
An increase in capacity per optimization FTE1
40%
Average reduction in bad quality cells4
The value of hyperscale cloud provider in expanding Ericsson's Cognitive Software Capabilities
In today's rapidly evolving tech landscape, domain expertise is a critical foundation for success. At Ericsson, we offer our Cognitive Software by leveraging our industry-leading RAN domain expertise, integrated and packaged with advanced AI technologies designed to unleash the full potential of next-generation networks. We also explore with Google Cloud the benefits of server-less services to allow communication service providers (CSPs) to optimize their total cost of ownership (TCO).
This is where the role of a hyperscale cloud provider (HCP) framework comes into play. By leveraging services from a provider, like the Google Cloud, we can address many of these costs, maximize the value to the customer, and accelerate innovation.
The role of MLOps in augmenting Ericsson's Cognitive Software
Machine Learning Operations, or MLOps, provides a standardized set of processes and tech capabilities for building, deploying, and operationalizing machine learning systems rapidly and reliably. This approach is essentially an extension of DevOps to machine learning and data science.
With MLOps, we can improve our solutions' efficiency, scalability, and reliability. We can also enhance and reduce costs by automating parts of the machine learning process. For the Cell Anomaly Detector, the input from RAN performance management data is ingested and aggregated to KPIs stored in Google Cloud’s BigQuery. This data is then processed by the VertexAI MLOps platform, and the inferences are then sent to cloud storage.
Leveraging serverless Software as a Service facilities of Google Cloud products like BigQuery and Vertex AI can optimize the TCO compared to an Infrastructure as a Service (IaaS) version of the same solution, since the SaaS model allows for a you pay-as-you-go model when consuming the service.
The Demo implementation of Cell Anomaly Detector
Conclusion
The technical exploration with Google Cloud has demonstrated the potential of implementing HCP and MLOps together with Ericsson's Cognitive Software. The full automation of ML model lifecycle management enabled by the Vertex AI framework ensures robust, scalable, and flexible operations, streamlining ML model maintenance, detecting any deviations to accuracy, accelerating time to market, and, most importantly, reducing TCO with HCP’s pay-as-you-go consumption model.
RELATED CONTENT
Like what you’re reading? Please sign up for email updates on your favorite topics.
Subscribe nowAt the Ericsson Blog, we provide insight to make complex ideas on technology, innovation and business simple.