As a mobile network operator, competing demands can feel overwhelming.
The realities of RAN as you gear up for 5G only add to this, with networks becoming so much more complex to deploy and operate.
And it’s not just complexity. It’s the sheer magnitude of necessary infrastructure: analysts recently estimated that in the UK alone 400,000 new ‘super masts’ would be needed to cater for the adoption of this new generation.
Could artificial intelligence (AI) play a part in helping you through these challenges or, more specifically, those of network planning? In this short blog post we set out to answer that question by:
- Outlining the challenges;
- Showing how AI will improve RAN planning; and
- Providing a bit of background about how it works.
To meet your challenges, 5G networks must be deployed faster than ever before. This calls for simpler, more efficient planning tools that generate reliable results from Day One. Traditional planning methods won’t cut it.
Consider this range of new factors:
- Increased vendor numbers, all of them with necessitating different planning scenarios;
- New frequencies to plan for, including dynamic spectrum sharing;
- Dealing with massive MIMO;
- The accelerating use of millimetre waves;
- The rise of open RAN;
- Developments in virtualization and cloud-based technology; and
- Initiatives in network slicing.
Each of these deserve several blog posts of their own.
In short, there’s lots to build into 5G planning.
How AI will improve RAN planning
We hope you’ll excuse a product plug, because we’re the first in the market to do this, so it’s hard to talk in more general terms, but this is where AIM comes in.
AIM is Planet’s Artificial Intelligence Model.
Launched this month as part of the Planet 7.5 release, AIM is the world’s first carrier-grade planning solution driven by advanced machine learning.
By pre-calibrating real-world measurement data, its machine learning is continually evolving to deliver:
- The best-fit predictive algorithms to match propagation reality;
- Versatile RF modeling that can automatically adapt to multiple deployment environments; and
- Fast 3D ray-tracing.
All of this is needed for effective 5G deployment.
The benefits of our AI in planning are clear:
- Simplification and acceleration of radio network modeling;
- Drastic reduction in costs associated with prediction model tuning;
- Improved coverage prediction reliability, which helps retain customers and attract new ones; and
- Better control of radio network investments.
These assertions are backed up by our research, which suggest that’s that:
- Cell coverage prediction is up to 3 times as fast as that of traditional methods;
- There’s a drastic reduction in measurement surveys;
- Network plans are more accurate and reliable; and
- CAPEX savings in radio sites are in the region of 10 to 20%.
In short, with AI your team can plan your 5G network faster, more easily and more reliably. This means a quicker time to market, scaling without advanced skills, better QoS, fewer radio measurements and less optimization costs in future.
But how does it work?
How it works: before and after
Engineers have traditionally used a mixture of statistical and physical propagation models to simulate the different environments needed to plan their networks, ensuring that they deliver the right levels of performance, capacity and QoS.
These methods do work – you no doubt currently rely on them – but without AI they’re hugely time-consuming for a number of reasons. The two most important are:
- They necessitate labour-intensive and repetitive calibration and parameter manipulation so that simulated scenarios can be matched with the propagation reality on the ground to reduce discrepancies; and
- If you make mistakes that you don’t detect until the deployment phase, you face the costly prospect of have to rectify your errors through intensive re-planning and/or site optimization activities.
The increased number of masts needed for 5G only magnifies this problem.
However, planning tools that use AI to automate the calibration process will significantly reduce the first problem and mitigate against the second.
An out-of-the-box solution like Planet 7.5 achieves this by drawing from millions of anonymized drive test measurements gathered from users of testing solutions in the field over the years, taking in numerous different environments, frequency ranges and types of site deployment around the world.
This data feeds into machine learning processes that produce insight that means customers can dispense with most of the manual activities associated with calibration and parameter manipulation, creating an automated experience that’s vastly faster and more accurate than those associated with its manual forebear.
Pro-active rather than reactive, such solutions aren’t limited to propagation; they also allow users to spot potential problems before they happen, meaning they can fix them before customers are exposed to them.
In the future, live crowdsourced user data could be brought into the mix to further refine the intelligence offered in these solutions and enable more planning automation. But that discussion is for another day.