Harmonic Claims 50% OTT Performance Improvements with EyeQ
Harmonic has announced a real-time video compression optimization solution that it says does the near-impossible trick of not only delivering a superior viewing experience on net-connected devices but reducing bitrates by 50 percent.
The product of at least eight years of research and several filed patents, EyeQ is available as an option for the software-based Harmonic PURE Compression Engine. It is in the labs of a dozen customers and will be released in Q4.
According to the company, the codec "directly improves the bottom line for video content and service providers." It does this through reduced network delivery and storage costs, "an increased ability to reach more consumers over congested mobile networks, and by enabling a more consistent viewing experience with enhanced video quality and less buffering."
"Video quality matters in the living room and is increasingly important on mobile. Buffering is not an option. To make matters even more challenging, accelerating consumption rates have driven up storage and delivery costs," says Tom Lattie, VP of market development. "Larger screens are driving higher resolutions and bitrates of greater than 3.5Mbps with conventional encoding. When viewers spend more time in front of the screen, there’s more opportunity for content monetization. By lowering CDN and storage costs by half, EyeQ delivers dramatic CAPEX and OPEX savings, and increased profitability, for operators."
"Back in the mid-90s compression engineering was straightforward," Lattie says. "You would hypothesise an algorithm, model it, and test it on a span of content. You only had one format in MPEG-2. Wind forward to the early 2000s and we had MPEG-2 and AVC layered with ABR. Since then, devising compression solutions have become even more complex.
"As long as eight years ago we began a process of trying to teach machines to be much smarter in the way they look at content, to mimic more of the human experience. In addition we have gotten better at optimising software effcient techniques, plus Intel continues to give us more cycles per processor.
"At the same time we started to migrate from ASIC-based to software-based compression cores," he continues. "All of these developments smashed together have brought this breakthrough."
EyeQ is embedded after the encoding process. "Once a codec has assessed frames and made those frames available, the EyeQ engine performs analysis based on quality thresholds," Lattie says. "It is assessing in real time whether the codec did a good job or a bad job, was it as efficient as it could be or was it too aggressive and more bits needs adding. I like to say it's like having a Harmonic algorithm engineer inside the encoder approving or rejecting based on automated settings and then sending that information back to the encoder to do it again. All of this has zero latency."
Harmonic says the solution delivers bandwidth savings using existing AVC-based codecs with no requirement to upgrade existing client devices.
What Lattie terms "in-loop artificial intelligence" guides the codec to focus bits where and when they matter. "This leverages the mechanics of the human visual system to continuously assess video quality in real time and focus bits where and when they matter most for the viewer," he says. "Unnecessary bits can be eliminated, allowing content providers to consistently meet video quality targets at the lowest possible bandwidth."
Importantly, Harmonic has moved from a constant bit rate (CBR) to variable bit rate (VBR). "We are able to much more precise about how many bits are needed to achieve a prescribed video quality and to free ourselves from CBR," he says. "We are confident of the 50% figure. Obviously this depends on content but we have customers delivering cinema content where EyeQ could achieve 60-65% gains. 50% is the averge saving over a typical channel lineup."
This technique is said to average a 50% bitrate reduction across content types versus CBR encoding. The encoder output is plain vanilla H.264 AVC; there is no stream syntax manipulation.
EyeQ has been internally tested and it's been running on a public CDN for some time "working with all sorts of devices worldwide," says Lattie.
Can EyeQ be compared to a capped VBR technique? "Absolutely not," says Lattie. "This is a complete encoding paradigm shift. Traditional state-of-the-art capped VBR techniques are leveraging picture and scene level quantizations to feed a pseudo-linear bit allocation model. It is an inherently limited method that can ony bring 10-15% impovements without degading the picture quality. EyeQ is using a full active visual qualiy measurement and analysis leveraging the mechanics of the human visual system—it brings a complete new dimension to bit allocations."
Harmonic is demonstrating EyeQ at IBC stand 1.B20