It would be so much cooler to be an RF engineer today than it was back in early 2000 when I was out in the field conducting hundreds of miles of drive tests in remote locations after every software upgrade and climbing 50 m high towers to adjust antenna tilts to keep the radio access network running at the best humanly possible quality.
Today it’s another story. Now we can use sophisticated software to conduct remote drive tests without generating truck rolls of fuel emissions, and we send fly drones to inspect high rise sites. And that’s just the beginning of how such a ‘physical, field level, operational task’ involving millions of man hours and of sweat, is being transformed by software, big data analytics and machine learning.
Looking ahead to the anticipated demand, which will lead to some 10-100x more cells in the network by 2020, certainly erases any twinge of nostalgia for those tower climbing days! Imagine the situation this creates for network operations teams. In terms of incidents per hour, this would lead to a 25x increase from 400 incidents per hour to as many as 10,000 incidents per hour for a network serving 10 million users – an impossible task to handle manually. The solution lies in automation to manage the scale and complexity, reduce cost, reduce resolution time and ensure quality.
Self-Optimizing Network (SON) technology is already available to simplify the management of base stations through automated configuration, automated optimization of parameters like antenna-tilt and automated recovery procedures in case of failures. However as complexity grows, SON alone is not sufficient. We need to add the power of machine learning to SON automation in order to teach networks to be self-aware.
Machine learning algorithms learn from unstructured big data and events to provide signals, patterns and models that help in predicting network conditions and making the right improvement decisions. For example machine learning techniques can find patterns in large, incomplete, unstructured and noisy data sets. Knowledge representation schemes provide techniques for describing and storing the network’s knowledge base. And reasoning techniques utilize the knowledge base to propose decisions even with uncertain and incomplete information. Ultimately, we believe that big data analytics and machine learning will help to evolve SON into what we call a “Self-Aware Network”, one that is able to handle complex end-to-end optimization tasks autonomously and in real time.
Sense -> Analyze -> Decide -> Act
There are four distinct stages of applying analytics and machine learning to create Self-Aware Networks: Sense -> Analyze -> Decide -> Act:
* Sensing imports data not just from the network but also from other touch-points such as sensors, social media and historic data.
* In the ‘Analyze’ phase, the algorithms ascertain what is happening, where, why (root cause), and how, and predicts expected impacts.
* In the ‘Decision’ phase, learning from the past is applied in anticipation of the future and decisions are proposed.
* Finally in the action phase, respective triggers, corrections, and settings are automated.
At the base level, SON forms the foundation by automating daily local network operations based on the principles of self-configuration, self-optimization and self-healing. SON capabilities are now expanding beyond SON management, coordination and verification to address newer domains such as HetNet, Cloud, 5G, Customer Experience and becoming more end-to-end covering Radio, backhaul and Multi-vendor SON.
However, SON does not allow early identification of network incidents “hidden” in the generated network data and root cause analysis, which is very important to ensure timely counter actions. Therefore, second level ‘preventive optimization’ provides automated anomaly detection of relevant KPIs. Nokia’s Predictive Operations is a great example here. It is the world’s first managed service for predicting mobile broadband service degradations, so they can be fixed before subscribers notice quality issues.
Even then, different SON and preventive troubleshooting functions may be in conflict. A resolution layer, network level orchestration, is required to provide fluent coordination of the underlying functions. A great example of this is Nokia’s SON for Mobile Backhaul innovation which solves the challenge of how to adjust the backhaul to traffic demand dynamically and automatically. The solution is based on SON-MBH agents, which are the ‘Sensing’ and “Acting” agents located at various points in the backhaul to measure and analyze Quality of Experience continuously, detect and localize anomalies degradation and act when advised by the SON MBH manager.
Finally, the top layer introduces decision support systems. It correlates insights from all the layers as well as takes into account business objectives to provide guidance on how networks can be planned over the long term. The best example here is Nokia’s Dynamic Experience Management innovation, an automated end-to-end experience management solution proven in 100s of use-cases on both ‘Personalizing Network Experience’ and ‘Teaching networks to be self aware’.
Teaching Networks to be self-aware is a fundamental pillar of Nokia’s Technology Vision 2020, which provides guidance on how to enable mobile broadband networks to deliver Gigabytes of personalized data per user per day profitably and securely by 2020, and lays the foundation for cognitive and cloud optimized networks of the 5G era.
Take a deeper dive into ‘Teaching networks to be self-aware’ in our Technology Vision 2020: Self-Aware Networks White Paper.
Please share your thoughts on this topic by replying below – and join the Twitter discussion with @nokianetworks using #NetworksPerform#FutureWorks.