Nokia, Ericsson, Fraunhofer Join Forces to Drive 6G-Era Video Coding Standardization
European connectivity leaders Nokia and Ericsson have partnered with Erlangen-based Fraunhofer Institute for Integrated Circuits (IIS) to shape and drive the next generation of video-coding standardization for better immersive media and mobile-video experiences in the 6G era.
The three partners combined their codec-research expertise for the first time, jointly demonstrating a new video codec with considerably higher compression efficiency than current standards (H.264/AVC, H.265/HEVC, and H.266/VVC) without significantly increasing complexity, while improving energy efficiency and scalability. Their proof-of-concept submission was positively evaluated by both the ITU-T Video Coding Experts Group and the ISO/IEC Moving Picture Experts Group (MPEG).
Streaming Media’s Jan Ozer spoke with Ville-Veikko Mattila, Head of Multimedia Technologies at Nokia, about the collaboration and its implications for the next decade of video compression. You can watch the interview above or read the lightly edited Q&A transcript below.
Q&A with Ville-Veikko Mattila, Nokia
Jan Ozer: Give me some background on this. We all know that H.267—is this a proposal that relates to H.267, something else, or something altogether different?
Ville-Veikko Mattila: This press release came out on Monday this week, and it is exactly about this joint work by Nokia, Ericsson, and Fraunhofer IIS. It’s about our response to this call for evidence when it comes to video-compression standardization that we could develop a new codec standard—something that is beyond VVC when it comes to, for example, its compression performance.
Jan Ozer: Can you give us any background on this? There were some things about European technologies and participation from Europe as a whole. Why is that important to people who may use the standard down the road?
Ville-Veikko Mattila: Yeah, I mean, we have a long background at Nokia developing video-codec standards—more than three decades. And if you think about moving from one standard to the next, the typical requirement is that you should achieve something like 50 percent better compression gain while still maintaining the same picture quality. So basically doubling the compression efficiency.
And as we know nowadays, something like 80 percent of all the traffic in communication networks is video. So therefore such an achievement—that suddenly you can double the compression efficiency—of course can have a very huge global impact on traffic and communication networks.
Therefore we are again exploring new technologies and a new standard for video coding. So we have this name H.267 even though that option is not yet decided—what the final name for the new standard will be—but we are in the exploration phase currently.
We developed the current standard, Versatile Video Coding (VVC or H.266), which we completed in 2020, and since then we’ve been exploring the next generation of video coding. And we do know that the amount of IP traffic is increasing quite rapidly—something like 18 to 28 percent every year—so therefore we also need better compression technologies to provide various kinds of experiences, like video-on-demand, social-media videos, or even video conferencing, for example what we’re doing now here.
Jan Ozer: Do video codecs need to be kind of network-aware? I mean, is the network knowledge that you bring to the table in Ericsson helpful when it comes to video-codec development, or is that just kind of extraneous?
Ville-Veikko Mattila: There are actually of course multiple standards that you basically then need at the end to really implement a service—whether it is video-on-demand where perhaps the latencies are measured in seconds, or whether it is for video communication where the latencies are much shorter. So of course the video codec as such produces a compressed bitstream, and in that way it is kind of agnostic when it comes to the network technologies themselves. But for example these codecs like VVC are widely applied to mobile communication as well as to core VOD and social-media services.
Jan Ozer: Your announcement spoke about a proof of concept. What is that? Is that a working codec implementation or an early research model that was submitted for evidence?
Ville-Veikko Mattila: Yeah, the proof of concept is a fully functional codec implementation. It’s not just a research model, and it is developed as a part of this joint call for evidence on video compression beyond VVC or beyond the latest standard. And it demonstrates that higher compression efficiency is possible—and it’s possible also with lower encoder complexity and with improved energy efficiency. And this was created by us at Nokia, Ericsson, and Fraunhofer.
We just had a meeting earlier this month where we evaluated—and here we meaning this joint video-expert team of ISO/IEC and ITU-T. So this team evaluated all the responses, and our response was really evaluated to provide great performance and a very balanced complexity-compression performance. So keeping the complexity low but still being able to provide very nice compression gain.
Jan Ozer: What were the compression gains that you quantified?
Ville-Veikko Mattila: We of course have the previous standard as a kind of reference, and so compared against VTM, which is the test model for VVC, we can achieve almost 30 percent compression gain as part of this response—and also achieving that with encoder complexities that are not higher than what we have with VTM.
Jan Ozer: It seems like the bottleneck for implementation on most of these codecs recently is playback. So what are the playback resources that you’re targeting for this new model that you’re proposing?
Ville-Veikko Mattila: If you think about playback resources—what kind of resource it would need—I think it’s quite early to specify exact requirements because we are still in the exploration phase, and the actual collaboration phase regarding H.267 should only start in early 2027. But of course we want the standard to be designed for a wide range of devices and architectures, and we have great examples.
If you think about VVC, Fraunhofer has an open-source software encoder and decoder available, and the decoder can run in software on general-purpose CPUs. So it’s good to note that we do not necessarily need a hardware implementation for the decoder anymore.
But the new standard may also be unique in the sense that it is the first time we integrate neural-network technologies as part of the codec. If you think about the decoder and inferencing these technologies as part of the decoder, that may require NPUs going forward—but it’s quite early to say anything because we don’t yet have the codec in place. We are still exploring these technologies.
Jan Ozer: Of the model that you proposed, how big a component was AI versus some of the traditional technologies that you pioneered 30 years ago?
Ville-Veikko Mattila: The response jointly developed by Nokia, Ericsson, and Fraunhofer had two versions. We had one version which is more about conventional coding, and in the other response we added a neural-network-based loop filter as part of the response. So these were the two responses we submitted.
Jan Ozer: Are these alternative responses, or are you expecting a hybrid-type codec?
Ville-Veikko Mattila: Definitely a hybrid codec could make sense. Neural-network technologies are quite interesting and they’ve shown great performance for certain test cases which otherwise can be challenging for conventional coding—for example running water or grass. It’s quite difficult to do inter-prediction for those, and we’ve learned that network-based technologies can be quite handy to provide texture which may otherwise be missing. Whether the texture is really authentic is another question, but it can definitely look very good.
So these technologies can be quite handy to improve the subjective quality of video.
Jan Ozer: So just to make sure we’re saying the same thing—are you hoping your technology will be able to decode in software on mobile and computers?
Ville-Veikko Mattila: I think it’s an exciting opportunity if you could really run the decoder in software, because that would also mean a much faster entry to market. You wouldn’t need to wait for hardware implementations before bringing the codec to the market. Especially considering use cases like video-on-demand, where the content is encoded once but decoded by many, the decoder could, if running in software, even be part of the streaming application.
Jan Ozer: What happens on smart TVs and similar devices that typically don’t have robust CPUs?
Ville-Veikko Mattila: Of course there will be many alternatives and definitely hardware implementations for encoders and decoders, but in some cases—for example if you’re consuming video on your mobile—you could also run the decoder in software. Many people nowadays use mobiles to watch media; that’s very common.
Jan Ozer: One of the things that accelerated HEVC adoption was the fact that it opened up the HDR 4K market. What do you see as being possible for this new codec in terms of new markets that could accelerate its adoption?
Ville-Veikko Mattila: We should complete the codec around 2030, then it will still take perhaps a couple of years before it’s integrated into hardware—so something like 2032. By then immersive experiences will matter even more. There’s a lot of discussion about augmented reality and immersive video.
In MPEG there’s also ongoing work on Gaussian Splat coding for 3D video—you can even walk into the video—but currently the bandwidth requirements are several hundreds of megabits per second. Those immersive application codecs are implemented on top of core codecs like VVC and the upcoming H.267. Together they can enable volumetric video in the future in a scalable way so that bandwidth requirements are low enough for many to consume such immersive experiences.
Cloud gaming is another interesting area—turning every playback device into a gaming device. And there are many industrial uses: quality control, surveillance, automation, low-latency machine vision.
Jan Ozer: Feature Coding for Machines—do you see this codec as potentially competing with those as well?
Ville-Veikko Mattila: There are other standardization tracks ongoing in MPEG developing video-coding technologies for machines, like Video Coding for Machines (VCM) and Feature Coding for Machines (FCM). These are quite different technologies. However, they may also use core codecs like VVC as the inner codec. So these core codecs become the fundamental part even of those other standards. They rely on the core codecs but develop something on top—for example, FCM transmits features, not video as such. We apply video coding to feature maps for machine-vision tasks using neural networks designed for those tasks.
Jan Ozer: What about the encoding side—complexity and machine requirements? Is it ASIC-based?
Ville-Veikko Mattila: When we’re discussing H.267 it’s very early to say anything because we don’t really have the standard in place yet. But typically for encoders, hardware implementations are needed. For use cases like user-generated content, encoder hardware matters a lot—being able to record, compress, and share quickly.
Jan Ozer: You’re talking about 2030 and 2032 for implementations. What are the risks of a totally different, perhaps AI-based, technology gaining acceptance before then?
Ville-Veikko Mattila: We’ve already discussed hybrid coding—enhancing conventional coding with neural-network-based technology. That could be one solution going forward, taking benefits from both sides, conventional and neural. One challenge for neural-network-based coding is that inference at the decoder side has high complexity.
With pure end-to-end learned codecs there’s a lot of research, but reproducibility across hardware can be a problem. If different companies develop encoders and decoders separately and run them on different hardware, we’re not sure how feasible that is in practice because the calculations may differ. Even hybrid systems can become unstable if those differences aren’t handled carefully.
Jan Ozer: One issue I’ve been grappling with myself is the inter-compatibility of NPUs. If Apple has an NPU and Nvidia has an NPU, can you develop one codec that plays on both, or is it not that simple?
Ville-Veikko Mattila: It’s a very good question, and it’s something we’re addressing in standardization. In MPEG under the MPEG AI umbrella we’re exploring exactly these topics—reproducibility and guaranteeing the same results across hardware. It’s a very interesting area for new research and new methods to ensure interoperability of neural-network-based encoders and decoders.
Jan Ozer: When does a typical publisher—Netflix or Paramount—need to start thinking about H.267? Walk me through the best-case timeline.
Ville-Veikko Mattila: We completed VVC in 2020 and since then have been exploring technologies for H.267. We had the joint call for evidence evaluated in October, and the next phase is the call for proposals. Organizations, companies, and universities can submit proposals to JVET; we’ll review them in January 2027. Then comes the collaboration phase, lasting about three years, which takes us to January 2030 for completion of the new standard.
Jan Ozer: So best case, software playback could appear around 2030, and silicon two or three years later?
Ville-Veikko Mattila: That’s consistent with what has happened before.
Jan Ozer: What’s your sense of where VVC is right now?
Ville-Veikko Mattila: VVC is by far the best-performing codec in the market. It’s called Versatile Video Coding because it provides good performance for many different content types—ultra-high-definition, HDR, 360-degree, low-latency, and screen-content coding. It’s truly versatile and provides the best compression performance today.
Jan Ozer: What about implementation?
Ville-Veikko Mattila: It’s coming to market. VVC is integrated into broadcast standards in Europe and North America. Pretty much all TVs may have the hardware already; sometimes firmware or player support is missing. You can already find apps that play VVC content. It usually takes a few years after standardization for adoption, and that process is happening now.
Jan Ozer: Thanks for taking the time today, and congratulations on the collaboration with Ericsson and Fraunhofer.
Ville-Veikko Mattila: Thanks for having me. It was really nice to have a conversation with you.
Related Articles
Candidates for H.267 already significantly outperform VVC as the hunt for a new video compression standard gets underway
17 Oct 2025
HEVC rode the 4K/HDR wave to success, but AV1, VVC, and LCEVC lack an equivalent killer app. With CDNs racing to the bottom on pricing and new royalty pools threatening to increase costs, codec adoption is no longer just about technical merit—it's about survival. These are the issues I'll explore in this article.
28 Mar 2025
H.267 should be finalized between July and October 2028. If history holds, this means H.267 won't see meaningful deployment until 2034-2036, long after I hang up my keyboard. Here's a brief description of what the standard is designed to deliver and my free and totally unsolicited advice to the committees and technology developers that will create it.
04 Feb 2025