Inside Google's master plan for faster, sharper streaming video
MOUNTAIN VIEW, California — We’ve all been there: After clicking or tapping a YouTube link, we’re greeted with a long wait, then video marred by blurry details and distracting blocky patterns.
Google has been trying to improve the situation with technology called VP9 that compresses the video data so they can move across networks faster. And with a successor called VP10 due in a couple of years, it’s promising an even bigger boost to image quality — not just sharper images, but also richer color and a better ability to span from bright highlights to dark shadows.
Google’s free-to-use VP technology is the proverbial underdog to the established standard for data compression, known as H.264. But it is getting new attention because of new patent fee problems afflicting its major rival for next-gen video, a technology called HEVC or H.265. What’s shaping up is a potential battle for how video gets compressed and distributed.
Compression standards and video patents are arcane matters, but they matter to anyone who watches video — which is to say just about everybody. Video compression is crucial to the shift from DVDs and Blu-ray discs to online video; to the arrival of video services like HBO Now that let people cancel expensive cable TV subscriptions; and to the livelihoods of Michelle Phan and other YouTube stars that younger viewers watch on smartphones and Web browsers. Figuring out the right standard that everyone can agree on is a critical step in ensuring that “House of Cards” streams to your television crisply and clearly.
Compared to H.264, VP9 roughly halves the network capacity needed to send video of a certain quality. And in an interview here at Google headquarters, engineering product manager James Bankoski revealed that Google expects a similar improvement with the forthcoming VP10. “We are trying to cut it in half again,” Bankoski said.
VP9 and VP10 are no shoo-ins. HEVC still has major momentum, especially when it comes to broad support extending to cameras, smartphone processors, Blu-ray discs and other domains beyond streaming video. And significant players including network giant Cisco Systems and Firefox maker Mozilla chose to launch their own HEVC/H.265 alternatives rather than rely on Google.
But Google is trying to move fast with its compression technology. Already this month, Google engineers began adding the first VP10 changes to the VPx software project.
“We’re hoping to hit the performance target by the end of next year,” Bankoski said. It’ll take some time after that for Google’s hardware and software partners to bring their VP10 support to market after that.
New video features, new data demands
New compression is key to both traditional TV and new-era online video. At the top of the list is the industry’s effort to introduce high “4K” resolution, which in principle offers more detailed imagery by at least quadrupling the number of pixels from today’s 1,920×1,080 to 3,840×2,160. With doubled efficiency, VP9 and HEVC/H.265 mean 4K video requires the network capacity to increase by a factor of two rather than four.
Beyond that are other features, each with a new appetite for data. One is a wider gamut of colors for richer, more realistic scenery. Another is high dynamic range (HDR), which means bright clouds and sunlit faces don’t wash out into blazing white patches and black-suited Batman can still be seen amid the shadows.
Better compression also can deliver more video frames per second, which is helpful when showing fast-twitch video games. If 3D video ever catches on, it, too, places new demands on networks.
Google has more power than most over video technology because it controls both YouTube, which sends gargantuan quantities of video, and Chrome and Android, which consume it. That means it can introduce and improve technology without having to wait for lots of other industry players to cooperate.
Developing the technology is only the first step in making a compression “codec” — which is essentially the translator for coded video signals. To be effective, a codec must be built deep into software and Web browsers and baked into processors so watching video doesn’t sap precious smartphone battery power or overheat TVs and set-top boxes.
There’s a tradeoff: Saving network capacity requires compression schemes that use more processing power. PCs are powerful enough to decode video using general-purpose processors like Intel’s Core family, but for mobile phones showing high-definition video, the compression technology needs to be built directly into the processor.
Moving to VP10 will increase the processing demands even further, but Google believes it shouldn’t be more than 40 percent more work to decode VP10 video than VP9.
The hardware support is coming, most notably in the Samsung Exynos 7420 processor used in the Galaxy S6 family of smartphones. Other chips with VP9 support are coming from MediaTek, STMicroelectronics, Sigma Designs, Nvidia and Broadcom.
“Almost all the 4K TVs that came to market this year already have VP9 hardware decoding support for 4K,” said Jani Huoponen, the Google product manager in charge of VP9 hardware support.
YouTube compatibility is a big incentive: The site began streaming VP9 video in late 2013. But VP9 and its VP8 predecessor still aren’t built into Microsoft’s Windows for PCs or into Apple’s chips in iPads and iPhones, meaning that online video companies can’t count on it the way they can rely on rival standard H.264.
Patent troubles open a door
Hamstringing HEVC/H.265, however, are disputes over how much to pay to use the technology. Those problems stem from clashing business and technology priorities among technology powers like Sony, Philips, Apple, Dolby and Samsung. That’s raised the profile of Google’s VP technology as a possible alternative.
In today’s video world, Blu-ray discs, smartphones, websites and TVs all rely on H.264. The same industry group that made H.264 — MPEG (Moving Picture Experts Group) — finished work on HEVC/H.265 months ago, and it was poised to become just as dominant as H.264 is today.
But new patent royalty requirements from a group called HEVC Advance have cast a shadow on HEVC/H.265. With supporters Philips, Dolby, Technicolor and General Electric, HEVC Advance charges a higher fee to use HEVC in a device and unlike MPEG LA, demands payments for streaming video — 0.5 percent of all revenue derived from HEVC-encoded video.
“I think the HEVC Advance announcement has made companies look at a lot of options,” said Matt Frost, head of partnerships for Google’s Chrome Media team. “HEVC Advance adds another factor to be considered,” an additional cost to be weighed against HEVC/H.265’s performance improvement.
HEVC Advance, for its part, believes a shake-up is necessary.
“There are many many device manufacturers out there that believe placing the burden solely on device manufacturers is no longer a viable methodology, and we need to expand the content piece,” said Pete Moller, formerly with GE, but now HEVC Advance’s chief executive.
Where MPEG LA charges 20 cents per device, HEVC Advance charges 80 cents for mobile devices and $1.50 for TVs. That’s just for the basic version of HEVC without some bells and whistles.
Add it up
The fees add up. A company selling 10 million HEVC-equipped smartphones or smartphone chips would have to pay $2 million to MPEG LA and $8 million to HEVC Advance. For comparison, Apple sold 47.5 million iPhones just in the second quarter of the year, although only iPhone 6 and 6 Plus models have HEVC support so far. Netflix, which charges $12 per month for a subscription including 4K content and uses HEVC/H.264, would pay HEVC Advance 6 cents a month per subscriber. A $10 4K movie rental at UltraFlix would mean a fee of 5 cents for HEVC Advance.
The licensing fees likely won’t end with HEVC Advance. Companies including Sony, Panasonic, Sharp, Toshiba, LG Electronics and Microsoft all have significant video patent portfolios that aren’t part of MPEG LA or HEVC Advance, and another group may form to extract even more licensing fees.
Avoiding this patent royalty burden is precisely why Google bought the VPx technology with its $123 million acquisition of On2 Technologies in 2010.
“We needed a high-quality, free alternative so the we could implement technology without having to be immediately on the hook for very high license fees,” Frost said. Traditional compression technology wasn’t advancing fast enough, he added: “We were hampered by old slow-to-evolve technology that came out of the cable industry and the optical-disc industry. There was desire to really pick up the pace of innovation, to start to evolve these technologies at the speed of the Internet.”
In 2014, YouTube delivered more than 25 billion hours of VP9 video, with billions of hours that were shown in high-definition resolution that H.264 only could have delivered in less detailed low-resolution standard-definition video, the company said.
Not for everyone
There are a few other players in the mix. Cisco’s Thor and Mozilla’s Daala also were developed to avoid patent royalty problems — and Google’s control over VPx. Both technologies were designed to leapfrog the existing standards.
Cisco’s interest in compression stems from its videoconferencing products, which range from its WebEx and Spark software running on PCs and phones to much larger room-size systems. For the latter, HEVC/H.265 is appropriate, but for WebEx and Spark, it’s a nonstarter, said Jonathan Rosenberg, chief technology officer for the collaboration products. Its Thor project already was under way because of MPEG LA’s license fees for HEVC/H.265, but “the final hammer fell with HEVC Advance,” he said.
“The worst case is every time somebody hits the Web page, we might owe money,” Rosenberg said. That’s no way to run a free service, even one with paid premium options, especially because HEVC Advance has no payment caps.
Cisco and Mozilla aren’t trying to go it alone. Instead, they’re contributing their compression technology to an effort to create a royalty-free streaming video standard called NetVC. Rosenberg said he believes its alternative is better than having one company control a standard compression technology.
Mozilla, too, shies away from Google’s control over VP9. “Because NetVC is being created via an open, multi-party standards process rather than by a single company, we believe it has the best chance of wide acceptance,” said David Bryant, Mozilla’s interim chief technology officer and vice president of platform engineering.
NetVC is still a distant option, though, with a planned delivery in 2017. Compared to the simpler days of H.264, there will be plenty of video technology uncertainty and confusion. “It’s up in the air again,” Rosenberg said.