The young London-based alternative video compression vendor V-Nova has found its feet as an enhancer of existing established codecs like HEVC rather than as a replacement for them and is now gaining some traction at both ends of the quality spectrum. Its Perseus technology was launched on April Fool’s day 2015 and in the ensuing four years we have all been trying to work out whether the joke is on V-Nova or critics, which come mostly from the legacy camp. The answer is somewhere in between since Perseus has failed to dislodge the incumbents except sometimes in video contribution but has carved out a significant niche by reducing bit rates achieved at a given level of video quality by up to 50%.
V-Nova has been taken more seriously than other upstart codec vendors because it had Sky Italia as its launch partner, which adopted Perseus technology firstly for contribution and then for IPTV distribution to extend reach of its service by distributing HD video to customers on slower connections, cutting bit rates from 8 Mbps to 4 Mbps by enhancing Harmonic ViBE VS7000 encoders. Then at the other end of the quality spectrum, Perseus was deployed by Indian movie distributor FastFilmz to deliver ultra-low-bandwidth SD video to Android-based mobile devices.
In the latest move, V-Nova has ported its Perseus Plus software to FPGA (Field Programmable Gate Array) cards from Xilinx, based in San Jose, California, to improve scale for large deployments and encoding efficiency for real time services, especially where some form of ultra HD is involved. There are two versions, Perseus XSA to accelerate encoding for existing live or VoD pipelines where codecs have already been deployed and Perseus XDE, which incorporates both the V-Nova software and HEVC codec on the same FPGA card for new services.
In the case of XSA, just the Perseus software runs on an FPGA to enhance an existing HEVC codec already deployed on other hardware. The latter is aimed at the many service providers that have already invested in transcoding and want to enhance their performance without having to upgrade to a next generation codec.
Development of the new Xilinx FPGA-based Perseus XSA and XDE products was motivated by the need of many large cloud service providers and large-scale streaming services to encode 4Kp60 content cost effectively in real time, according to Fabio Murra, V-Nova SVP, Product and Marketing. “The addition of Perseus Plus to FPGA means that the same chip can now increase by four times the number of encodes of a platform and provide live 4Kp60 within a single FPGA card, which is otherwise impossible at premium video qualities,” he claimed.
Murra was keen to emphasize that while V-Nova’s principle market, the target for the XSA version, is as an enhancement to existing codecs, Perseus is a codec in its own right rather than just an optimizer that either pre-processes the video or improves the quality of the encoded output. “The Perseus Plus layer is a compression technology creating a low complexity enhancement to improve the efficiency and performance of the underlying codecs,” Murra insisted.
This is certainly correct, although Perseus operates on different principles to traditional codecs from the MPEG/ITU family of which HEVC is the latest descendent. It is based on elegant algorithms more satisfying to the mathematical purist, which are actually independent from the video itself and can in principle be applied to any complex data set that can be represented numerically. It works inside frames on an iterative or hierarchical process which outputs at a variety of resolutions so that the one best suited to an application can be selected. It starts by stripping out redundancy just locally among the data and then progresses to greater depths and distances across the data set.
Apart from enabling tuning to different resolutions, this approach ensures that quality degrades gracefully as bit rate is cut for a given video source. Most other video codecs encode frames by dividing them into blocks of some form and extracting data from those to meet the target bit rate. This causes the blocky artifacts which Perseus avoids, which instead becomes generally blurry as bit rate reductions become more aggressive and exceed its quality threshold.
The downside of Perseus in its pure mathematical form is that it operates only intra-frame and cannot exploit temporal redundancy, which is why it was adopted initially for compression only. That was rectified in Perseus Plus by working in tandem with existing codecs, so there is a natural synergy there.
To start with, it worked with just H.264 but since then it has added an HEVC base layer and also enhancements that do take specific account of video’s peculiarities with techniques such as luminance and contrast masking enabling the algorithm to match more closely how humans see noise and detail in video.
Another necessary move was acquisition in February 2017 of video processing patents from Faroudja Enterprises, which improved Perseus performance at low bit rates after the associated techniques had been incorporated. These took account of video specifics such as noise that are best removed prior to compression and do not show up clearly in raw data.
Murra singled out eSports providers as a target market for the new FPGA versions of Perseus. He pointed out that Xilinx had recently announced deals with both large-scale cloud service providers and what he called hyperscale streaming services, including several eSports players.
“For them the solution is particularly suitable as they deal with a lot of hard-to-encode, fast content at high-resolutions and frame rates, 1080p60 and 4Kp60, as well as being easily deployable as an add-on card,” said Murra.
The most prestigious such client is Twitch, which is the largest and one of the fastest growing live streaming video platforms in North America dedicated to gaming and eSports content. Its principal research engineer Yueshi Shen said that Twitch had selected the Xilinx FPGAs to enable the industry’s first broadcast-quality, live streaming platform using the VP9 codec.
Twitch had evaluated various options including server class CPUs and ASIC based approaches, but they fell way short of the 60 frames per second encoding requirement. Perseus can then slot in as an add on card to boost quality at the available bit rate while keeping to that tight schedule.
Perseus then may have failed to take the transcoding field by storm as it hoped four years ago but has carved out what looks like a heathy niche as an enhancement to existing codecs with prospects of growth as high-quality streaming driven by gaming, live and eSports proliferates.