Your browser is not supported. Please update it.

10 November 2022

StreamTV surfaces Xperi spin-off Adeia, live streaming scale problems

The final session of the StreamTV Show was at least not painful to listen to. It was the first time Faultline got a peak behind the Adeia curtain, the licensing firm that was spun-out from Xperi recently, with the topic of live streaming at scale being the task at hand.

For a bit of context, Özyeğin University Professor of Computer Science, Ali Begen, set the stage for the discussion, noting that in 1969, the moon landing had an audience of 650 million viewers. Using its one antenna, the lander sent its signal some 250,000 miles back to Earth, in a trip that took less than three seconds. Pointing to 2019’s Super Bowl LIII, which was captured by some 115 cameras, the reported broadcast latency clocked in at ten seconds. OTT latency was 45 seconds.

In the three years since, Begen – who is also a technical consultant in Comcast’s Advanced Technology and Standards Group – said significant progress has been made on the latency front. This is thanks to the arrival of Common Media Application Format (CMAF), in 2018, and Low Latency MPEG-DASH (LL-DASH) and Low Latency HLS (LL-HLS) in 2019, which have since gained traction in OTT workflows.

Begen’s delivery had just a hint of mourn, when recalling the (LHLS) format introduced by Twitter’s Periscope in 2018, which was later adapted by Amazon’s Twitch in 2019 but then abandoned. A world where Twitter was responsible for a transformational low latency protocol seems quite farfetched a concept, given recent happenings at the company.

Continuing, Begen warned that “running adaptive bitrate blindly can introduce visible artefacts, which are not ideal. To this end, we introduced a context-aware algorithm, to ensure that artefacts are minimal.” Özyeğin University has published this code, and Begen was keen for those with artefact problems to reach out to collaborate.

This set the stage for a discussion between the Streaming Video Technology Alliance (SVTA) head Jason Thibeault and Serhad Doken, the CTO of Adeia. The session suffered at least one last-minute absence, but Doken kicked things off by saying that Adeia has around $400 million of revenue, currently.

The main topic was challenges for operator streaming at scale, and Doken immediately dove into the point that small problems in a workflow can scale exponentially once audience numbers begin to climb. To that end, proper planning is prudent.

Doken quickly added that this discussion often ignores last-mile and in-home issues, as well as the mess of different settings and features that are not shared by the plethora of video devices within homes. Doken sounded particularly exasperated when pointing out that there are multiple different current definitions of ‘5G.’

The fix for those problems is bigger and better pipes, said Doken, who noted that “fixed-line operators that have launched mobile offerings are in a strong position to be able to deploy some clever caching techniques,” thanks to being able to observe user movements throughout both wired and wireless networks.

But perhaps more interesting was the sidelink features added to the 5G standard in the 3GPP’s Release 17 document – a way to pull media packets from nearby devices, rather than reach out to the access and core network. “Device-to-device will have a comeback,” said Doken, “both for transmission and caching.” As Release 17 has only been available for about six months, there has been little public progress on video implementations.

Doken pointed to CloudFlare’s recent announcement of WebRTC support as promising news, and Thibeault jumped in, to discuss how operators need to consider such technologies. In a sports stream, delivered via HLS, Thibeault said how a user that wants to bet on the game can be moved into a WebRTC version of the stream, so that they are on the same page as other betters – with latency measured in perhaps a single second, rather than perhaps 45.

Here, the default of using scalable HLS allows the less scalable WebRTC to not burden the distribution infrastructure. However, Thibeault warns that the production and distribution workflow is now much more complicated. This prompted Doken to recall recent unnamed operator sports streaming failures, although to the best of our knowledge, the likes of TIM and Viaplay were not running parallel WebRTC streams to support betting.

Doken warned that in such decisions, everything is a tradeoff, where new features will increase costs. Lessons can be learned from the gaming industry, he said, although the point about games supporting millions of real-time users should have some major caveats, we feel.

Massively Multiplayer Online (MMO) games split the total player count into multiple realms or worlds, and even ‘single server’ games like EVE Online use clustering to separate those players strategically. More mainstream titles like Call of Duty or Fortnite range from 50 to 100 players per instance.

There was time for a brief SVTA pitch, when Thibeault addressed institutional expertise at these operators. “They have to encapsulate heaps of knowledge that would normally have been a specialist pursuit.” Doken added that a lot of education and ecosystem work is needed to further this collective progress.

As for future work, Thibeault said, “We’re seeing some great innovation, but not at scale. Metaverse, XR, AR; those are being tested by operators, but I think stadiums and arenas are going to be the proving grounds. With 5G Ultrawideband and Multi-access Edge Computing (MEC), these immersive experiences are available, but not yet at scale. So, more edge compute, network slicing, and guaranteed Quality of Service (QoS) are needed.”

In a separate session, a very painful lesson on the importance of low latency was present, when a panel’s moderator was around two seconds out of sync with the other panelists. This led to many painful instances of people wanting to comment on a point being cut off by the moderator’s attempt to switch the discussion to the next planned talking point – in what felt like a very scripted affair.

This panel, on the potential of operator aggregation, also did not introduce the speakers, which led to the audience scrambling to pick up on the potential nuance being imparted into the points made. Very little of interest was said, despite the bounty of speakers, and no audience questions were taken, but it was all perhaps worth it to spot the Comcast rep squirm when one of the other panelists described set tops as a legacy platform.