By Chris Pearson, President, 5G Americas
You hear a lot lately about how AI is going to automate jobs, by some estimates up to 29% of jobs by the year 2030. Scary thought. But you may also be aware machine learning (ML) and artificial intelligence (AI) could also provide some amazing capabilities that otherwise might take entire teams of humans many years to accomplish – but in less time and with more accuracy. That kind of capability opens up new opportunities, which might not ordinarily exist.
Take for instance, 5G networks. Today’s wireless networks are facing an ever-increasing demand for mobile data, up to 77 exabytes of data per month by 2022 according to Cisco, which is driving the need for network densification (more cell sites!). This network densification requires a tremendous amount of labor and capital to accomplish because filling in wireless data coverage ‘gaps’ involves teams of experts to fan out and analyze challenging real-world conditions affecting where 5G network radios should be sited.
What do I mean by that?
A wireless network provides coverage in a series of ‘cells’ which can encompass a wide area, usually provided by a ‘macro cell’ – the kind of which you might see on a cell tower, or a small cell which can be backpack-sized and small enough to fit indoors – or something in between. By increasing the density of these cells, you can increase the amount of data that a network can manage.
Small cells are compact, low-power and lower cost than macro towers. This means they need to be located closer to demand hot spots where consumers and businesses are using their mobile devices to effectively supply data and deliver a good return on investment for network operators. Small cells should be placed as close as possible to demand peaks, so a best practice (outlined in our 5G Americas and Small Cell Forum white paper) for small cell siting is to place it within 20-40 meters of the highest demand peak.
Here’s the challenge: how do you determine where a ‘demand peak’ is located?
It turns out that determining the location of demand peaks is challenging and requires an ultra-precise planning process for deployment. This is because demands peaks are impacted by a huge variety of factors, which can include signal-to-interference ratio, spectral efficiency, line of sight, network traffic estimates, weather patterns, height maps of buildings and topography, directionality and position of mobile devices, and overlapping cell coverage.
In addition, there’s also the issue of supplying the small cell with all the right ingredients to allow it to work, including power and backhaul availability, spectrum availability, as well as license agreements with site owners and other numerous considerations. Therefore, network planning must be examined holistically to address the wide variety of factors and inputs.
While humans can assess these factors to a high degree of certainty, AI takes it to another level. AI and ML can take into consideration all of these impacts at once and provide a very good prediction of how they’ll interact together. The result? A very solid forecast of the location of that demand peak.
AI and ML can do amazing things, but the iterative nature of machine learning requires algorithms to be repeatedly tested against large amounts of data. The more data, the more accurate the result. Aggregation of very large data sets into ‘data lakes’ are important to provide algorithms with enough test data to inform results.
AI and ML can be also be used to augment existing user device location techniques, such as GPS tracking or Observed Time Difference of Arrival (OTDOA) to predict device travel patterns across a cell coverage area. These enhancements can greatly improve the accuracy and increase throughput of a small cell. Mapped against 3D visualization, which is superior to 2D visualization, AI and ML can begin to make predictions as precise as traffic variations between different floors of a building.
Today, network operators are just beginning the early stages of incorporating AI and ML into their small cell siting process. The results are impressive and indicate a considerable opportunity to save network operators time, labor and costs when deploying their networks. In one test case, an operator was able to provide the same level of network throughput and coverage in lower Manhattan with 40% fewer sites – an incredible result.
We’re likely still a ways off from a full-scale adoption of these AI techniques across 5G networks. After all, 5G is in the first inning of a nine-inning baseball game. But if history is any indication, once a technology is successfully applied, it can quickly be replicated by an operator and provide a positive outcome for their network.
The quality and availability of underlying local real-world, real-time information will have long-lasting impacts on the success of 5G networks. The fact that operators are already thinking about this today may be an indicator of how fast change will come.