The desire to optimize the mobile network is a no-brainer. Who could possibly not want to extract the best performance, the most revenues out of the existing, already deployed mobile network infrastructure? Indeed, mobile operators have been optimizing their networks from the beginning, trying to pack as many voice calls as possible. And then trying to pack as many bits – either voice, video or data – as possible.
This approach is no longer good enough. Maximizing throughput – making most use of the existing network capacity – has been the goal of network optimization for a long time. It is a goal that at first sight appears to be eminently reasonable, if not plain obvious. Networks are built to transport traffic, and the more traffic they transport the more value the operator can extract from them.
But this is not always the case, and maximizing throughput is a strategy that can backfire. It gives false comfort because it is easy to measure and control, but it does not necessary lead to the best subscriber quality of experience (QoE). Network capacity remains a crucial foundation to enable good QoE, but it is how this capacity is used that determines the level of QoE.
QoE is not just a set of metrics to gauge subscriber happiness. It is – or should be – the goal of network optimization. Mobile networks are built to provide a service to subscribers and their performance should be assessed on their ability to do so, and this is what QoE measures.
Mobile operators have started to explicitly shift their focus from throughput to QoE in network optimization, as they realize that the two are not linearly correlated. In truth, mobile operators have always tried to capture QoE and improve it, but for a long time they lacked the tools and capabilities to do so directly. Maximizing throughput has been – and most of the time still is – the best proxy they had to maximizing QoE.
With LTE and more intensive data – and especially video – use, the discrepancy between throughput and QoE has come to forefront. In turn, this has pushed mobile operators to look for ways to use QoE metrics to improve network performance, and vendors to develop solutions to do so.
Problem solved? Not so fast. As operators try to measure QoE and use QoE metrics to optimize network performance, they realize that it is not trivial to measure QoE, and to map QoE metrics onto network key performance indicators (KPIs). Not only are QoE metrics inherently more subjective that KPIs as they depend on the subscriber perceptions and behaviors, their links to the performance of elements within the network is difficult to establish. A poor video experience that may be due to high latency in the subscriber device may be caused by problems in the device, by congestion in the radio access network (RAN), by insufficient backhaul capacity, by signaling overflow, by bottlenecks in content delivery network. Regardless of what the cause is, the subscriber will most likely assign blame to the mobile operator.
QoE metrics do not directly tell us where the problem originates, but they provide crucial information that, when correlated with network KPIs and other performance metrics, can lead operators to identify causes and to address the issues. Establishing how these correlations emerge – when and how they are triggered, how they can be used to solve or prevent network performance issues – is a major learning effort for the entire wireless ecosystem. Each network element, as well as the interaction among network elements, have to be included in the assessment, because they all contribute to network performance.
Not only QoE metrics have to be mapped into network elements and into end-to-end network performance. They also need to be collected and analyzed at the application level. Averaging QoE and network metrics over time and space is necessary to avoid optimizing the network for each subscriber, but it can also be uninformative or misleading when done across either time or space or both.
For instance, a relatively high average network latency may not affect or even improve QoE. A network in which voice and video latency is low, but data latency is high may result in better performance than a network with an average lower latency, because the latency is the same for all traffic flows. Distinguishing between these two situations provides operators with the understanding of network dynamics they need to resolve performance issues. But analyzing latency – or other metrics – at the application or traffic flow level adds complexity to the optimization efforts, and requires skills and solutions that are new in the wireless industry.
To further complicate this efforts, QoE is not only subjective, but it is deeply embedded in time and space. What matters is not the experience of a subscriber through a month, day, or hour, but the experience at a given time, in a given location – and how it relates to the experience of other subscribers in the same time and location, and to network conditions in the same spatio-temporal coordinates. Network performance varies continuously as a function of subscriber usage behavior, location and interaction with other subscribers.
This causes another shift in network optimization. No longer data is collected and analyzed offline, and used to tune the network future performance. Instead, data is collected to optimize network performance in real-time or near-real-time, all the way from the RAN to the mobile core. The move to real-time optimization forces another increase in the complexity that permeates our efforts in optimizing mobile networks.
The move towards a QoE-driven network optimization that is done in real time and at the application level requires a major learning effort and a much more active traffic management that affects deeply how mobile operators manage their networks. But it is a transition that is needed to finally be able to leverage QoE data from subscribers, and, in turn, to make the use of network resources more efficient and profitable.
Read more about RAN optimization in Senza Fili’s latest report “The smart RAN. Trends in the optimization of spectrum and network resource utilization.” The next report will be on traffic management and core optimization. If you would like to contribute to the report, contact Monica Paolini at [email protected]
