Laying out more chairs. That’s the solution that springs to mind when thinking about data congestion. However, while more radio network capacity may seem like an obvious response to growing volumes of data traffic, it’s both time-consuming and costly. Chairs, or to leave analogy behind, radio network cells, are a straightforward but expensive way to tackle the issue of data congestion on our networks. Instead of simply catering to traffic volumes by expanding network capacity, perhaps we ought to look at how traffic itself is managed and optimized.
Let’s be clear: the insatiable demand for data should be a good problem to have for mobile operators. The average monthly data consumption globally was 15GB per user in 2022, predicted to triple by 2028. That presents a challenge for operators, but it also means that their subscriber base is fully engaged and ready to take advantage of whatever services come their way. In this respect, mobile operators have it good. They have an almost captive audience – who among us could give up our smartphone?
However, alongside this demand for data comes rising service level expectations. It’s a testament to how well operators have managed the rapid data increases across mobile networks that users simply don’t tolerate delay. According to findings from a recent survey, anything more than six seconds of buffering in a video stream, and most users will abandon the video altogether out of sheer frustration. What’s more, those same users will blame their network provider for the experience and are more likely to churn as a result. So while insatiable demand might be a “good” problem for operators to have, retaining subscribers and making that problem work for them is far from easy, and will only get harder as content becomes more bandwidth-intensive.
As 5G rollouts advance, the promise of more data-hungry apps and super-fast speeds will only drive user expectations higher. Even without these data-intensive use cases, there is an upward trend in data traffic. For example, the median size of web pages in the past year alone has jumped by 18% for mobiles. To keep up, operators either need to build more access (put out more chairs) or improve the current way they organize data traffic.
Addressing capacity, quality, and congestion
Upgrading and expanding network equipment is one obvious and perhaps inevitable step towards enhanced quality of experience (QoE). Network infrastructure overhaul is absolutely a necessary part of advancing 5G adoption, which is expected to make up over 50% of global mobile traffic by 2030, and more than 90% of traffic in regions such as North America, developed areas of Asia Pacific and the Gulf. However, throwing more radio signals at the congestion problem is costly and time-consuming and doesn’t address one of the most immediate underlying causes of poor QoE, congestion.
Network congestion occurs when many people in the same area are accessing data at the same time. When network operators need to build out their signal capacity, they typically look to install a radio network congestion monitoring component known as a Radio Congestion Awareness Function (RCAF) to their existing eNodeB units to detect and isolate congestion. These upgrades take time and resources to implement. They are useful as a long-term solution to a consistent capacity challenge rather than a fix for unexpected congestion.
Predicting congestion is key to enhancing quality
As network traffic management becomes software-driven, virtual, and agile, mobile network operators now have alternative tools at their disposal for improving the efficiency of their networks. Analytics, AI, and ML have an increasingly valuable role to play in detecting and optimizing traffic. By monitoring the vast volume of data on their networks, analytics tools can pick up on the data usage patterns that indicate congestion. For instance, a very large number of dropped packets in a protocol flow is usually an unmistakable sign of something having gone wrong downstream. However, congestion can be harder to detect when it builds gradually. In these cases, continuously monitoring parameters such as the slow increase in the number of dropped packets in a user’s flow, or protocol round trip time, or a decrease in throughput for a user are all ways of detecting usage patterns that will lead to problematic congestion.
AI models can now be built, and, over time, trained to detect these conditions without the need for overlay components and probes. Associating data congestion patterns with location is enough to model if a cell or eNodeB is tending towards a congested state in real-time. There are several advantages to using data analysis, modeling, and prediction tools to detect congested conditions. Over time, ML models can be trained using IP data traffic and standard radio data to avoid false negatives or feedback loops to check accuracy. These modeling tools rapidly become the first key step in detecting network hotspots without expensive overlay probes and additional components.
Taking action to improve QoE
Another advantage of using AI models in this way is that they provide real-time intelligence on the state of the network. Given that adding physical access takes time, pioneering MNOs in Europe are not only this network data to detect congestion, but also to mitigate it and improve the QoE for users in real-time. By moderating and optimizing the packet flows in real-time, operators can smooth the flow of traffic reaching the eNodeB, avoiding the unnecessary collisions that lead to congestion and result in a buffering video or dropped call. This technique has been a particularly valuable addition to operators’ toolkits because it can be applied to video streaming, the major contributor to data traffic and data traffic congestion, making up 60-70% of mobile data.
Improving RAN capacity and efficiency with AI
Even without additional physical radio infrastructure deployment, research has revealed that using AI tools to optimize data traffic can boost effective network capacity by 10% and the average quality of video streaming as perceived by users by 15%. This QoE improvement is achieved through selective action that network operators can decide on in advance. Instead of slowing all video streams down within a congested cell, as would happen with no video optimization in place, selective action can be taken on more data-heavy (for example, HD) video streams. These actions work in real-time when congestion is detected and can be tweaked to the operator’s preferred level, automatically disabling once the congestion has eased. This practical use of AI/ML solutions for selective intervention is capable of working on both 4G and 5G networks, prolonging the life of existing network investments and ensuring an efficient transition to 5G and beyond.
