Harmonic has finally confirmed to Faultline what we unofficially knew about the ongoing work with V-Nova to integrate LCEVC (low complexity enhancement video codec) into a Harmonic encoder.
Slowly but surely, we are scratching below the surface of the first real-world test for LCEVC, following the revelation that Harmonic’s Thierry Fautier, VP of Business Development and Innovation, announced to around 300 people at an SMPTE event in Italy last week that the vendor is working on a European project with V-Nova.
It feels like only yesterday that Harmonic and V-Nova were considered direct rivals, so having the company talk about embracing the standardized MPEG technology with open arms is a huge development that should not go unrecognized.
Could this have anything to do with the TIM/DAZN assertions that Faultline explored last week? While Harmonic could not reveal any additional details, the fact that Harmonic’s LCEVC plans have only been revealed to Italian audiences suggests we are pretty close to the money.
It was surprising then that Fautier, speaking during the Harmonic Live Summit this week, was more focused on encouraging the industry to focus on existing video codecs than drumming up excitement for the research going into next-generation codecs, including that of his own company.
His rationale is a reasonable one that we have echoed many times at Faultline. Existing codecs account for billions of devices, while devices supporting next-gen codecs are like gold dust. Harmonic’s own numbers show approximately 10 billion H.264 (AVC) devices in the field today, while 2 billion support HEVC, and 4 billion are currently streaming VP9. As for AV1, VVC, EVC, LCEVC, as well as China’s SARFT (the Chinese alternative to VVC), Fautier described the current presence as insignificant.
Based on these figures, Harmonic projects next-gen codec device penetration reaching 5% in the first year, 10% by year two, and onto 20% after three years, while remembering that internet traffic is increasing at around 40% each year, while existing codecs are supposedly improving by 10% each year. Fautier’s point is that the most efficient bandwidth savings on unicast HTTP networks today comes from existing codecs – and it will take years for that to change.
“People tend to forget legacy – 16 billion devices versus nothing,” he implored.
Of course, he is bundling LCEVC into the same bracket as standalone codecs, when the technology is capable of enhancing any base codec – established or next-gen.
Looking at the elephant in the room, licensing, Fautier said he has seen an early version of LCEVC’s licensing terms, adding that we should receive finalized licensing information “very soon”.
Fautier is not alone in having strong opinions on video codec licensing, describing the
scene as still “murky at best.”
“We think the industry has to work harder on the commercial aspect of licensing, rather than promoting the technical benefits. We don’t need better compression than HEVC, we need better licensing,” declared Fautier.
Judging by the Q&A session that followed Fautier’s presentation, the vast majority of webinar attendees had turned up to hear about licensing terms rather than learn more Harmonic’s products, which somewhat took the shine off Fautier’s core message, which focused on how the greatest compression improvements today are being driven by ML and AI, while in the past it was all about core codec improvements.
ML and AI research is enabling better frame rate selection, better resolution selection, better encoding settings, and better video quality prediction – which in turn creates a lower CPU footprint, enhanced QoE, and bitrate savings. “When you combine all three, you get really impressive compression efficiency,” said Fautier.
In Harmonic’s case, the ML/AI-based compression process starts with a huge library of test files that Harmonic has put together, to which it applies in-house video quality measurement. It then encodes all content offline into multiple modes, multiple resolutions, multiple frame rates, multiple control modes, before entering content into a big machine learning system equipped with a prediction model. This is done using man-based supervision, but requires a lot of CPU resources, so Harmonic is tapping into all the cloud resources it can to estimate all the different modes, Fautier assured.
Spookily, he then referenced Sky Italia as a commercially deployed customer case study for Harmonic’s AI/ML-based EyeQ technology, keeping in mind that Sky Italia is V-Nova’s flagship customer. There are few coincidences in the Faultline ecosystem.