The 5G Future Will Be Powered By AI
5G networks will demand AI because they are far more complex than previous-generation networks. AI-fueled insights will deliver higher QoE and better services.
Scan the tech headlines circa 2019, and you might think the robots really are taking over. Across every vertical and all the hottest market segments—cloud, Internet of Things (IoT), FinTech, Big Data—the big money is flowing towards artificial intelligence (AI).
In most segments so far, only one of the basic AI promises has been used. That is machine automation, which is the ability to perform tasks that historically required human beings—faster, more accurately, and at a much lower cost. Some technologies, however, will be so dependent on machine intelligence, they won’t really be able to function without it, and therefore AI will be exploited to its full strength. That’s true for perhaps the hottest emerging technology out there: 5G.
Machine learning will play a central role in the way operators plan, deploy, monitor, and operate their 5G networks. You could even make the case that, in the next decade, the key differentiator among mobile networks and operators worldwide will be the quality of their AI. This will be also true for network infrastructure vendors as long as 3GPP will provide just guidance regarding the performance of the AI to be embedded in the network elements.
Why is AI such a big part of the 5G story? And whay will separate the companies using maching intelligence effectively from those that do not? Let’s take a closer look.
The biggest reason 5G networks demand AI is that they’re far more complex than previous-generation networks. Start with the radio technology itself, which operates at higher frequencies, introduces more complex antenna configurations, and employs more sophisticated connectivity mechanisms like beamforming. There is a lot more going on behind every connection than there used to be.
At the level of network and service design, 5G also adds new layers of complexity. In previous networks, everything was tuned towards the same basic goal: high-quality voice and data experiences. 5G networks must support multiple use cases (enhanced mobile broadband, ultra-low-latency applications, machine communications), each serving different verticals with very different requirements.
Unlike previous-generation infrastructures, 5G networks are also far more dynamic. Network resources can now scale up or down—even at the level of individual network slices—in real time, in response to changing conditions. They’re also non-deterministic, with performance fluctuating based on location, time, device, application, and other factors.
Enabling New Experiences
Many aspects of 5G applications will also require analytical capabilities and responsiveness beyond what human beings can provide. For example, take one of the fundamental innovations of 5G: context-aware service delivery. If operators want to optimize every connection for every device and application, they need to shift their network planning, and monitoring tools from a network-centric to a user-centric view. That requires a continuous, end-to-end view of real-time network behavior. Which, in turn, demands the ability to correlate vast amounts of network testing and statistical data towards an accurate picture of the quality of experience (QoE).
Along the same lines, delivering on the promised value of new 5G verticals (connected cars, industrial IoT, and others) will demand literally superhuman analytical capabilities. Assuring the performance of these diverse applications requires continuous network monitoring, troubleshooting, and optimization based on an accurate end-to-end view of network behavior. That means collecting and processing data from a dizzying variety of sources—simultaneously, in real time.
Rise of the Machines
Bottom line, a 5G network will produce vast amounts of data that operators will need to be able to understand and act on. Managing, optimizing, and operating this “network of networks” will demand automated correlation of data across multiple sources, including network planning data and ongoing testing and telemetry statistics. Operators will also need to track a broader, more diverse set of key performance indicators (KPIs) than in previous-generation networks, as well as track how those KPIs interact with each other in real time.
There is simply no way to do all this using human-centric approaches—at least not cost-effectively or on the timescales needed to remain competitive. Which is why every vendor selling 5G technology, and every operator deploying it, is working towards closed-loop automation.
Ultimately, 5G networks will be able to identify when something is wrong (service or network failures, load outages, coverage issues), diagnose the cause, and fix it—automatically, without human intervention. The AI inside 5G networks should grow smarter over time by observing network behavior. Eventually, it will even predict problems before they occur and proactively optimize network resources to deliver the best performance at the lowest cost.
Inside 5G AI
Clearly, if 5G networks are going to be as performant and responsive as operators envision—and at a cost that’s not exorbitant—machine intelligence will have to do much of the heavy lifting. But not all AI is created equal. Even the best machine learning algorithm is still just an algorithm. Its effectiveness in achieving a desired outcome depends on other factors: the quality of the data it’s training on and the domain expertise of the people guiding that training.
When evaluating 5G tools, operators should look for those with access to the most diverse and highest-quality data. For example, you may have an excellent tool for analyzing drive-test data or geodata in 3G and 4G networks. But, if that tool only works with a few narrow data sources, it’s only looking at one piece of the puzzle. More effective 5G machine intelligences will pull from all those sources simultaneously, as well as planning data, device-based measurements, and other real-time and historical data sources, to capture a holistic view.
Of course, it’s not just the number of data sources you feed your AI. An effective machine intelligence needs to know which data is actually meaningful across that ocean of information, and how to effectively prioritize and act on it. The training models an AI uses make all the difference—and the quality of those models depends directly on the domain expertise of the people developing them.
Stay Focused on the Big Picture
5G networks really will deliver outstanding new experiences, and AI will play a key role in making it happen. But just because a given tool uses machine learning doesn’t mean it’s using it effectively. Recognizing the importance of machine intelligence to so many areas of 5G, operators should look for solutions with broader, more holistic data capabilities whenever possible. They should seek out tools that analyze data from the most diverse sources, and that are designed by organizations with broad and deep experience in mobile networks and RF performance. Last, but not least, operators need 5G testing tools that enable machine learning techniques to detect and evaluate in real time the performance of the AI embedded in the network; which is expected to be network vendor specific and based only on high-level guidance from 3GPP.
With the right data and models, AI can help operators achieve unprecedented levels of performance and automation in their 5G networks. For the companies that do it right, AI-fueled insights and capabilities will translate to higher QoE, lower costs, reduced risk, and a significant competitive edge.
Keep pace with 5G developments by attending this year’s Interop19 conference in May.