Video streaming traffic over cellular will continue to grow
Today 66% of the worldwide population has access to mobile connectivity, which is expected to grow to 71% in 2023 (1). And 63% of global mobile internet traffic is video. Forecasts indicate that will rise to 76% by 2025 (2). Video traffic is even higher today in some mobile networks.
This growth is due to multiple factors, including new consumption habits of end-users (i.e., multiscreen, user-generated content, social media), the emergence of wireless/cellular smart TV screens, the multiplication of Fixed Wireless Access (FWA) deployments, multistream events, larger screens, higher frame rates, and higher bit rate formats (i.e., full HD for mobile devices, UHD/4K for FWA usages).
Source: Ericsson June 2020 Mobility Report Source: Cisco Annual Internet Report (2018-2023)
The average mobile network connection speed is constantly increasing as well: from 13 Mbps in 2018, it is predicted to be 44 Mbps by 2023 (1). This offers a great perspective for mobile video streaming services.
Yet the video streaming experience over cellular is not satisfying
Watching live TV, AVOD and SVOD content over a cellular network is largely a hit-and-miss experience today. Consumers very often suffer from rebuffering, poor video quality, slow start time, and high latency on live content.
Therefore, they increasingly value having a smooth video experience when watching a popular TV show, a premium sports event, a movie or the latest TV series on their smartphone, via a cellular network.
Going forward, it is likely that the end-user perceived quality of video delivery will become a key differentiating factor. Network operators will not only compete on price, raw throughput and gigabytes of data, but also on video quality of experience (QoE). A poor user experience is the third most important reason why subscribers terminate a streaming service, according to a recent Digital TV Europe survey (3).
Not surprisingly, regulators and independent companies assessing the quality of mobile networks have set video streaming quality as one of the most important criteria (4).
5G will likely not solve the problem
5G offers a huge radio capacity, up to 1000 times bigger than 4G depending on the amount of aggregated bandwidth. While this will de facto bring a big breath of fresh air to the capacity of cellular networks and to the average video experience of mobile users, it will be some time before 5G is available everywhere.
Moreover, there is little doubt that congestion will resurface sooner or later in some locations. The improved video experience will certainly boost usage, and the throughput available per user will progressively shrink again. Also, video applications and content will continue to be ever more sophisticated and bandwidth-intensive (i.e., 4K/8K, stereoscopic video 360, future 6DoF volumetric streaming, etc.), which will fill the new 5G pipe even faster and generate congestion and QoE issues, again.
Before describing S4Streaming, a disruptive innovation from Broadpeak that significantly improves video streaming QoE in mobile networks, let’s revisit the reasons why video delivery is so challenging in these networks.
The roots of video delivery challenges in mobile networks
The impossible bandwidth measurement
In video delivery over cellular networks, contrary to wireline access distribution modes, the last mile (i.e. the radio cell) and to a lesser extend the mobile backhaul crystallizes most of the congestion challenges (5).
There are two main reasons for that. First, the air interface is a limited and shared medium, with multiple users competing for the same bandwidth. The number of these users and the amount of individual traffic load they generate are constantly changing. Second, the medium is wireless; the radio propagation conditions are also constantly changing, even faster than the load, because users not only move but their environment is also changing independently.
As a result, the bandwidth available for each user in mobile networks is highly volatile and unpredictable, which makes its measurement – the responsibility of the video players – extremely challenging (much more challenging than in a wireline network).
It just so happens that adaptive bit rate (ABR) streaming essentially relies on the player’s bandwidth estimation. If this measurement is inaccurate or slightly delayed compared with the actual highly fluctuating available bandwidth, inappropriate bit rates will be selected by the player.
This has two immediate consequences. First, the QoE is damaged, either due to increasing rebuffering occurrences (measured bandwidth value higher than actual) or because of poor video profiles served (measured value lower than actual). The usual solution for the player is to increase its buffer, but then this affects latency, another important QoE criteria for live video quality. Second, the precious spectrum resource is not optimally used, and the ratio between QoE and cost is consequently reduced.
In conclusion, the unreliability of the mobile bandwidth estimations performed by the players is the essential root of the QoE problems that occur when video is delivered over cellular networks.
The showstopper on low-latency streaming
Mobile usage makes users particularly sensitive to lag time and in demand of low latency. That’s because viewers are more likely to be close to the broadcast event itself (e.g., stadium) or close to a zero-latency streaming source (e.g., traditional TV broadcast in a bar or on the main screen at home) when they are watching live events on their devices.
Low latency vs. live has become an important video QoE criteria, according to 86% of the users interviewed in a recent Digital TV Europe survey (6).
For low-latency streaming (e.g., using CMAF/CTE), where every video segment is divided in smaller sized chunks, one key challenge is fast bandwidth variations. The player faces the problem that the time of download – of the multiple chunks of a segment – is no longer constrained by the available bit rate but by the time it takes to create the segment.
Bandwidth measurement issue in low-latency mode
The result is that the bandwidth calculation of the player becomes flawed in live low-latency environments, and the player constantly makes wrong decisions when choosing the video quality to request from the CDN. And this time, it cannot play with its buffer as it would lead to the reverse effect.
In the best case, the player falls backs to a mode where it requests a bit rate that is safely lower than the available bandwidth, thus decreasing video quality. In the worst case scenario, the player asks for a much higher bit rate than the bandwidth actually available, leading to very poor experiences with recurring freezes and incremental latency being added.
In conclusion, without a reliable and accurate bandwidth measurement mechanism, low-latency streaming can hardly work in practice in ABR delivery networks, and it is even worse in mobile networks.
The total absence of operator control
There is another fundamental problem in the way ABR streaming works today. The players not only perform the bandwidth estimations, they also make the important decisions with regards to the video profiles to request, the depth and use of their buffer, etc.
This mode prevents operators from having any control over the streaming, unless they’ve implemented a certain degree of customization with specific players. This lack of control is even more critical when the delivery is done over cellular, because as previously stated, the players are unable to cope with the broad bandwidth variations of mobile networks. They make wrong decisions that significantly affect QoE.
Operators cannot apply common rules for a group of mobile users, a type of mobile devices or a type of content. The devices and their players behave autonomously, in an uncontrolled way, without any possible central command from the operator.
Some players have, for instance, a behavior that is detrimental to the global QoE. They unfairly ask high video profiles for VOD content even though they don’t have the necessary bandwidth. They use their buffers to avoid switching to lower bit rates when the network conditions would push them to do so. This obviously impacts all the other users watching live content.
This negative effect is highly augmented by the particularly wide range of mobile devices and players that usually co-exist among subscribers of the video service.
The successive generations of operator mobile TV apps downloaded by users and the diversity of operating systems on handheld devices has created a heterogeneous park of players. There is no possible consistency between their behavior, leading to a poor use of the network bandwidth, and a heterogeneous QoE among users.
This lack of control from operators is a major weakness when they want to ensure a homogeneous experience among their subscribers in a highly unpredictable mobile environment and fragmented park of players. It’s also an issue when they decide, temporarily or not, to solve serious congestion problems on the mobile network with radical measures like the general reduction of the video bit rates (as was observed during the COVID-19 global health crisis when major OTT providers had to reduce the maximum resolution of their content).
Why and how Broadpeak S4Streaming solves the problems
Unreliable bandwidth assessment, tough low-latency streaming constraints, problematic player heterogeneity, absence of operator’s control over the streaming experience: all these issues have an additional criticality in a mobile network context.
Broadpeak S4Streaming solves all of these challenges, thanks to a simple but disruptive principle: ABR bandwidth estimation and video quality selection can now be piloted by the CDN instead of the players.
S4Streaming is compatible with the vast majority of video players, as it is compliant with the HLS/DASH specifications.
Reliable bandwidth measurement
With Broadpeak S4Streaming, the bandwidth measurement is done on the server side, and uses congestion control algorithms at the transport layer’s level, like BBR (Bottleneck Bandwidth and Round-trip).
This has three main advantages:
- the server is aware of all the ongoing sessions, contrary to the player that is only considering its individual connection,
- the algorithms at the transport layer’s level provide instantaneous congestion and bandwidth information,
- the bandwidth measurement is both more accurate and more dynamic than the HTTP-based player’s estimation.
As a result, the bandwidth measurement with S4Streaming is very accurate, despite the high variability of the mobile network, even in low-latency streaming.
Full control given back to operator
Broadpeak S4Streaming lets the CDN decide which video quality to stream from the manifest (7). The server can make these decisions based upon a multitude of criteria and rules, which can be customized by the operator.
The operator can, for instance, decide to:
- decrease the bit rate when a major TV event starts (e.g., a popular live sports broadcast), to avoid congestion,
- downgrade the video layer when a given load limit is exceeded (8),
- adapt the bit rate to the mobile device capabilities to make efficient use of the bandwidth,
- select different video qualities per content,
- differentiate the bit rates served to individual mobile users, groups of mobile users, regions or even entire countries,
- ensure a fair repartition of the bandwidth among all users.
The resulting actions can be fully automated or triggered by the operator involvement.
Impact of S4Streaming on QoE
Our experiments show that S4Streaming has a tremendously positive impact on the video QoE in mobile networks.
When we activate S4Streaming on the CDN, we observe the following QoE improvements:
- the number of rebufferings in difficult network conditions goes down to nearly zero (provided that we use a rather small size of buffer to keep a reasonable delay vs. live, i.e., lower than 15 seconds); in the rare case when rebuffering occurs, it doesn’t last more than a couple of seconds,
- the average bit rate is higher,
- the latency vs. live is lower (as a large buffer is not needed anymore),
- the video quality is the same for all users (there are no “losers”).
If we upgrade the delivery chain with a low-latency protocol like CMAF/CTE, we observe that:
- low-latency streaming effectively works, reducing the delay vs. live to a few seconds, with a smooth experience for all users,
- video bit rates are, again, higher with S4Streaming (as there is no limitation anymore to multi-layer switching in low-latency streaming, so content is streamed with the highest possible quality).
Stream selection in low latency
Finally, in extreme overload conditions (e.g., in crowded places like train stations, during live TV sports events), the video quality of experience is kept under control thanks to the operator’s rules or intervention, and everyone can access the service.
Netflix has set the quality standard in our video streaming industry. They have the advantage of owning and controlling the video delivery chain end-to-end; for the last 10 years, they have been developing and implementing sophisticated mechanisms all along this chain that strengthen their control over the QoE.
This QoE improvement is what S4Streaming offers to pay-TV operators and content providers: in the highly unstable environment of cellular networks, they get a powerful variant of ABR that increases its reliability and performance, and gives them much more control over the QoE they deliver to their mobile and Fixed Wireless Access (FWA) users. At the end of the day, S4Streaming allows pay-TV operators and content providers to significantly increase the loyalty and the ARPU of their consumers connected over mobile and FWA.
It may now be the time to start evaluating the technology on your network to see how S4Streaming can improve your video service delivery over mobile networks.
(1) Cisco Annual Internet Report (2018-2023).
(2) Ericsson Mobility Report November 2019.
(3) 52% of the respondent find that” a poor user experience” is a “very relevant” factor leading “subscribers to terminate their streaming service subscription”, 38% “moderately relevant”.
(4) In its 2020 “mobile quality of service” measurement report, French regulator Arcep indicates as a key criteria the percentage of two-minute-long videos streamed with an excellent quality.
(5) Aggregation and backbone networks are also stressed but they are generally better equipped than radio access to absorb traffic overloads and peaks. In mature markets, most operators have started to deploy vast campaigns of end-to-end transport capacity upgrades, in anticipation of 5G. They are also increasingly merging their mobile and fixed transport networks, which allows to benefit from the statistical multiplexing effect and thus better absorb the overloads.
(6) 48% of the respondent find that “minimizing latency of the video stream is very important”, 38% “moderately important”, giving a total of 86%.
(7) The server can also use information from the player, like the depth of its buffer, in order to take an even wiser decision.
(8) The precision and granularity of applied video quality policies will depend on the level of integration with the mobile network and the availability of pieces of information like mobile user location, total cell load, quality of individual radio connections.