The pros are whispering, "AI is exciting, but most of it isn't ready to put to use." Caroline Gabriel, one of the very best analysts in wireless, surveyed many major operators and discovered a huge gap between what is being delivered by automation and the many brave words we hear so often. She writes,
Plans to implement advanced machine learning-based automation in 5G networks are grinding to a halt, with operators slowing investment and reducing their ambitions. ... most plans had gone backwards. Globally, only 18% expect to have more than 40% of functions automated by the end of this year.
Last year 21% of the same operators said they would have more than 60% of processes automated* by the end of 2020. Those hitting this 60% mark fell to 14% in this year’s study, suggesting respondents had been over-optimistic previously or that projects are snagged, or harder than they seemed.
[Without effective automation, the] business case for 5G in the period to 2024 will be severely compromised – resulting in higher costs, less network flexibility, lower ROI on 5G RAN and on virtualization."
There is no doubt machine learning can do some remarkable things. Every day, I read articles on four Chinese news sites in Google translation. I can almost always understand the articles. Google translate is the culmination of a decade of work by a tribe of researchers worldwide over a decade. When I first tested the Chinese to English Google a few years ago, it was godawful.
She asks, "Is it a distant dream?" So far, the evidence is that most forms of RAN automation are not ready for prime time. What I've seen so far includes some brilliant ideas and exciting trade show demonstrations. I'm sure some of them will work well in a few years. Huawei is spending hundreds of millions on AI or maybe more. I expect some interesting results coming fairly rapidly.
Caroline knows the problem is, "A vast complexity in the network with a huge number of new elements. This means modern networks are becoming hard for humans to understand, predict or manage." If I read her correctly, she believes many of the solutions will be quickly found if enough money and energy were applied. I don't know that's it's been proven that more money will yield quicker solutions.**
Gabriel believes high frequencies and densification are essential for 5G. That will be true at the point at which traffic demand outstrips what the carriers can sensibly deliver with mmWave and more density. Ted Rappaport, who virtually invented millimetre wave 5G, believes that point is coming soon. The engineers building 22 of the 24 5G networks currently deploying disagree. They are building 5G in mid-band with only modest densification. They believe that will be enough for practical purposes for a number of years.
Traffic growth has been falling rapidly in the developed world. In the UK it is down to 25% and in Germany under 37%. Cisco predicts the U.S. will fall to 31% in 2021. Last year, I thought most telcos would profit from the extra capacity of mmWave.
We will see.
*The functions the operators are referring to are mostly much simpler than the full RAN automation
** See The Mythical Man Month