Automation is one of the hottest topics in telecoms, spurred in part by an expected jump in network complexity with the arrival of 5G, which will make them larger, more dense in terms of frequency bands and channels, and more diverse in terms of base station types and scenarios. Not only will this mean more O&M complexity, it will actually drive down utilization efficiency.
Network services are diversifying rapidly from traditional voice and data to mobile IoT. Service integrity has become a prerequisite, even with network slicing making real-time resource orchestration for multiple services more difficult. With tomorrow’s mobile networks set to be 100 times more complex than those in use today, the old ways simply cannot cope with the service development needs of 5G – AI is the answer.
So What Will it Do?
AI is set to bring three major benefits to the wireless domain: enhanced performance, simplified O&M, and the enablement of new businesses. Its introduction into just three scenarios (Massive MIMO, multi-band/multi-system, wireless fingerprint positioning) has already been shown to enhance O&M efficiency tenfold, while improving experience by 20% and increasing positioning accuracy by three to ten times.
Massive MIMO 5G base stations will provide high-capacity coverage and vast improvements in spectral efficiency. With 4.5G, there are nearly 300 broadcast beam combinations that improve cell coverage; there will be over 10,000 with 5G. Manual dynamic adaptation to this many traffic scenarios and models will be impossible, but enhanced learning-based automated optimization can do the job, with these dynamic models facilitating automated batch fast-lock optimized patterns, at double the O&M efficiency, meaning better overall cell performance － 20% increases in cell throughput and number of connected users have already been demonstrated.
As Narrowband IoT (NB-IoT) applications proliferate, for example, bike-sharing services, demand for positioning capabilities will only increase. GPS-based positioning offers accuracy to the meter, but chip costs and energy consumption make it prohibitive. Wireless AI can use network data for a positioning accuracy of approximately 30 meters, which would be sufficient for bike sharing, but sub-meter positioning is in the pipeline, which would greatly expand the reach of IoT.
What Does It Need?
But Wireless AI won’t develop overnight, or in a vacuum. It’ll depend on other capabilities in big data, intelligent algorithms, and architecture.
More Big Data: Mobile networks already generate huge quantities of data, but much of it is “read and burn,” with structured storage lacking. Wireless AI will change the game in three regards: real-time to historical, single-dimension to multi-dimension, and fragmented to structured. Mobile base stations will become mobile big data sensory systems, with data generated internally and externally, which can be analyzed for the accurate profiling of grids, cells, and users – thus enabling full network simulation.
Dedicated Machine Learning (AutoML): Wireless network systems have their own unique characteristics. Huawei has built a library of AI/ML algorithms for wireless networks based on its experience in mobile communications. This library integrates various industry- and Huawei-developed algorithmic model data, enough for 90% of mobile scenarios. We’re also partnering with universities to research new algorithms. Huawei has developed the industry’s first AutoML platform for mobile communications, and reduced the data modeling cycle from months to days – thus accelerating Wireless AI incubation.
In the future, we see the continual interaction, understanding, and integration of Wireless AI as powering mobile networks that understand you, are dedicated to you, and work for you.