More Big Codec News
A couple of weeks ago, we wrote about the development of new codecs and looked at the pros and cons of different technologies in particular at the new AV2 codec. Just after we went to press, we saw a report of another development that is expected to go beyond VVC and AV2.
The announcement was that Ericsson had been working with the Fraunhofer HHI and Nokia to develop a new codec that is said to have ‘considerably higher’ compression efficiency with ‘no significant increase in complexity. Ericsson said that this is the first time that the three partners, all representing significant experience in the compression field, have worked together on a new codec.
No Boost in Complexity is Big
As we saw in the previous article, no increase in complexity marks a change as the development of previous generations of MPEG codecs have tended to trade complexity against efficiency. With energy efficiency high on the agenda and Moore’s Law, while not over, being increasingly under pressure, a move away from higher complexity is a real advantage.
Avoiding an increase in the computing power needed for a new codec would also have advantages at both the compression and decompression stages. In compression, as Spin Digital has demonstrated, VVC compression in real time can be performed on industry standard CPUs and GPUs using software. The same should be true of a new algorithm. Second, at the client end, mobile device power continues to increase, so newer devices are more likely to be able to support a new codec in software, enabling quicker adoption and less of a ‘chicken and egg’ challenge.
New Codec in ‘Proof of Concept’
The new codec is currently at the ‘Proof of concept’ stage, but details have been submitted to the ITU-T Video Coding Experts Group (VCEG) and ISO/IEC Moving Picture Experts Group (MPEG) and the announcement said that the idea was ‘positively evaluated’ and may become a formal technical standard around 2029/2030. It is reported to be a working codec. As we have seen with other codecs, market adoption is likely to be significantly later and Nokia said that this could be around 2032 or 2033 when supporting hardware has been developed.
The partners have not released any information about the technology in this initiative, but there’s an interview with Nokia here that gives some more flavour, without revealing much about the the technology.
There has been some speculation that the codec might represent a move away from the AVC-HEVC-VVC block-based approach to something based around AI-based technology and the interview confirms that the new proposal has a neural-network-based loop filter and could be considered a ‘hybrid’ codec. However, others see the new proposal as “H.267” and the next step in the evolution that we have discussed.
AI in Codecs
As was pointed out on a recent panel that we attended at CEDIA 2025, no technical discussion is considered complete these days without a reference to AI. Codec discussions are even more likely to involve AI as much of the breakthrough work on AI models was enabled by the release of a huge database of images that allowed developers to work on the concept in 2012. Further, a lot of work has already been done on image-based processing for such critical image work as mammography analysis, where AI assistance is already seen as being better than un-assisted human workers.
Google , Meta , Tencent and the Fraunhofer HHI (among many others) have shown AI-based codec technologies that seem to be more efficient in compression than, for example, VVC. A key advantage of AI-based coding is that the codec can be more ‘content aware’ – so can adopt different techniques or parameters for different scenes or even for different content within a scene. For example, AI could detect faces or animations or sports. Each might be able to use a better algorithm. (Spin Digital already does some of this in its VVC implementation).
At the client end, the increasing availability on devices of tensor cores or Neural Processing Units (NPUs) could mean that existing hardware could mean hardware assistance without so much new hardware.
AI Upscaling is Already Here
The writer doesn’t claim expertise in this field but the approaches being used to develop AI-based codecs looks similar to the approach described to us in an interview with Samsung’s specialist in AI-based upscaling, Dr. Tien Bau.
Of course, everything that is described as AI-based is not always truly or fully using the technology. As Dr Bau pointed out to us, upscalers from companies other than Samsung that he has tested use AI to decide how to upscale, but then use traditional techniques to perform the process. In the same way, Spin Digital’s VVC implementation exploits machine learning to optimize the parameters used in its compression process. Codec specialist, Jan Ozer, in a presentation last year pointed out the need to ensure you understand what the real measurements are of any new technology. The presentation in his link has a lot of detail after he surveyed a range of companies in the segment and is worth a look if you are interested in this topic.
The interest in AI codecs is currently huge and just before we went to press we heard that Interdigital, a well established video technology company, had agreed to acquire Deep Render, a company that had very big claims for its AI coding efficiency when we reported last year. Jan Ozer reports that this is the first time that a ‘traditional codec IP powerhouse has invested in AI-native compression’ (and has looked at the codec) So the move might represent a technology inflection point (although it could also mean that Interdigital is acquiring it for ‘defensive’ purposes!).
