Elisa automation use case: 4G LTE load balancing
Our mobile customers in Finland are among the most data-happy in the world. In Q3 2017, the average Elisa subscription consumed close to 14 GB per month. This high usage makes us one of the mobile operators that carry the most data traffic in Europe, even though Finland has a population of only 5.5 million.
More frequency bands: A good start – and a problem
Since close to 90% of that data traffic is carried on 4G LTE, it’s fair to say that we know a thing or two about how to deal with high 4G LTE load. The general solution is to add more frequency bands. By combining a number of carriers – for instance in the 800, the 1800 and the 2600 MHz band – the total capacity of a 4G LTE site and of a network can be increased. It’s a good start, but not more than that.
The problem is that each frequency band has its own propagation characteristics. Lower bands such as 800 MHz reach longer whereas higher bands such as 2600 MHz propagate much shorter distances. So, when the carriers are co-sited, the lowest band “wins” when it comes to signal strength. This means that the mobile devices of the customers will prefer to camp on the lowest frequency band leaving the higher frequency band unoccupied. And that’s not good when the whole point of adding new carriers was to increase the capacity.
“It’s not beneficial to deploy more 4G LTE carriers without any load balancing” explains Dr. Jarno Niemelä, the technical lead of Elisa SON. “The added higher frequency bands are simply not taking enough traffic unless an operator is building new dedicated sites for the higher frequency bands – something which is very costly.”
Network-wide prioritisation: Not a solution
To address the issue of higher frequency bands not taking traffic, operators can assign different priorities to different bands. By e.g. giving higher priority to the 1800 MHz band most mobile devices could be forced to camp on 1800 MHz even if the 800 MHz signal is stronger. But a network-wide prioritisation is not ideal. Every cell and cell environment is unique. Due to changes in customer behaviour, traffic patterns aren’t static. Weekdays are for instance different than weekends. Spectrum allocations are different between operators and are not uniform throughout the country. It demands manpower and it is difficult to copy across networks.
By now, we have listed a lot of problems. It’s time to identify the solution.
What if the load balancing between the bands could be done adaptively based on the actual daily performance? If the parameters were set in a closed-loop fashion without human interaction? How much higher could the capacity of the network be?
Our developers were excited about a potential breakthrough in 4G LTE capacity, and embarked on a project to code, test and gradually refine an automated algorithm in Elisa SON which automatically tunes every cell in the network for maximum performance – every single day.
Dr. Jarno Niemelä: “The logic was coded and implemented in a time-period of a couple of weeks, piloted to prove the concept and then productised in April 2016. We benchmarked the algorithm against vendor-specific load balancing algorithms. Already at this point, our approach gave a better result. Between January 2016 and October 2017 we developed and tested five different versions of this use case in Elisa SON, each time raising the bar of what could be achieved.”
The optimum balance of how many mobile users should be kept on each band depends on the spectrum mix of the operator. During the refinement of the algorithm, we adjusted the targeted balance between the bands multiple times to achieve equilibrium with regards to end-user throughput.
Early results: Improving the customer experience
We achieved astonishing results already in the first version of the use case in balancing the load between 800 and 1800 MHz bands (as illustrated below)
When our engineers evaluated the outcome of the automatic optimisation they realised that it would have been impossible to find one optimal network-wide parameter value. The automatic optimisation had given the parameters a wide range of values. It could have been done manually – an experienced network optimisation engineer could realistically do it for one cell in five minutes – but with thousands of cells, a daily tuning would call for at least 50 full time employees – doing nothing else. That was in 2016. Since then the number of cells has increased a lot – and then we haven’t even spoken about what 5G will bring.
“Automated tuning is mandatory”, says Dr. Jarno Niemelä. “We don’t have that manpower – and we most certainly don’t want to carry the added cost”.
Learning and development
Energised by these first results, we refined our algorithms in Elisa SON. The timeline below shows the main steps taken to date.
We saved money too
Thanks to the algorithm, the number of cells with an average user throughput below 5 Mbit/s decreased with more than 40% within two months. In Elisa this threshold has an immediate CAPEX implication since it triggers new base station investments. By letting the algorithm work in a closed-loop fashion, we managed to postpone investments in the network.
“In 2017, we were able to improve our CAPEX efficiency by 3%”, says Dr. Jarno Niemelä. “Postponing investments would also lead to OPEX savings in e.g. site rental, transmission and electricity. In practice though, we use Elisa SON load balancing to steer capacity investments to the locations where it’s most needed.”
Our plans for the future
Although we have refined this algorithm several times already and made it multi-vendor capable, the development of it continues. Our plan is to implement a yet more accurate and efficient load balancing algorithm. We believe that a shorter optimisation cycle – perhaps down to just one hour – will improve the performance even more. This is also what we expect from a possibility to tune the algorithm based on an estimated traffic increase. We want to be able to use end-user throughput as a balancing criterion. To improve even more, we also plan to include inter-site 4G LTE load balancing.
This Elisa SON use case – 4G LTE load balancing – resulted in the following longer-term benefits:
- Improved average end-user throughput with up to 30-40%
- 20% reduction in network-related customer complains
- Improvement in NPS1
- 3% improvement in base station CAPEX efficiency triggered by a decrease of more than 40% in the number of cells that fell below the investment threshold
- 2% reduction in energy OPEX
- Reductions in both transmission and site rental OPEX
We’ve done it. Now it’s your turn.