Codec innovator V-Nova will go to NAB next week with a new version of its Perseus P.Link contribution encoder, which drives down contribution feeds by another 40% to 50%. The key ingredient in the new version are machine learning algorithms.
V.Nova has made it clear in the past that its proprietary approach to codecs is to express intraframe video in a hierarchy – so this is the art of compressing the image for each frame – it produces layer after layer which are each more complex than the previous one. But it only sends the differences from the prior layer. This is seriously important in contribution codecs as they must be lossless, whereas Interframe compression is about sending change information for distribution and is sometimes called temporal compression and that’s all about describing the next frame by referring to changes from the last frame.
Some of V-Nova’s implementations break the video into 12 or 14 layers, and the machine learning uses rules extracted from looking at millions of hours of video to decide precisely where the boundaries between different layers should be for the lowest bandwidth use.
“We already had a codec that in sending intraframes had 3 times the compression of other lossless codecs like JPEG 2000. Now we can reduce the delivery of intraframes by a further 40% to 50%,” explained Fabio Murra, SVP, Product & Marketing at V-Nova.
The hard numbers are that a contribution UHD stream which using JPEG 2000 might take up 800 Mbps to 1000 Mbps of bandwidth, and that used to compress down to 240 Mbps using the previous generation of P.Link. Now this gets closer to 140 Mbps. The 240 Mbps number came from V-Nova sending the Euros soccer championships coverage to Rai in Italy last year.
V-Nova targets both contribution encoders – sending video from live and remote events to editing suites – as well as distribution encoders for delivery of video to homes or mobile devices – but found its first breaks came in contribution, operating its first such deal at Sky Italia. One reason is that each operator controls both ends of a contribution feed and there are few obstacles to embracing a new system; another is that it is easy to quantify how much savings an operator will make on cost of bandwidth and compare it with the cost of implementing a new contribution system. This is starting to make the P.Link product something of an easy sell.
P.Link now incorporates machine learning in the form of Convolutional Neural Networks. The new version of Perseus Pro has been trained to identify edges and curves and shares this across its various layers, reducing the residual data requirement for reconstruction.
The Perseus technology works in 2 ways: Perseus Pro is an Intra codec for mathematically lossless and visually lossless professional production, and Perseus Plus is a temporal codec designed to enhance a base codec such as H.264 or HEVC, adding additional layers of detail and upscaling for video distribution.
V-Nova also announced this week that uCast, which claims it runs one of the largest Multi-Channel Networks in Asia has integrated the Perseus Plus codec into its platform for Internet TV services. uCast powers services such as Qello Concerts, Stadium, and Acorn TV.