This brings back so many memories of covering new compression technologies over
the years. Many of the claims sound so familiar!
The one thing that strikes me in this is the disconnect between capabilities
and applications. V-Nova claims it will be a break thru for mobile and for UHD
(4K). The whole notion of 4k is still in flux - if there is a market it surely
will be for large home entertainment screens, not smartphones and tablets. And
the massively parallel processing somehow seems an oxymoron when we are talking
about the power efficient processors in these devices.
But there are two aspects of this new codec that do seem relevant.
- Hierarchical coding
- Efficient use of parallel processing for encoding on mobile devices
If the company has developed a truly new transform that can deal with images in
a resolution hierarchy, it may have a chance. Many of us thought wavelets would
succeed because of its native resolution hierarchy, but it never panned out.
For mobile networks, pulling in image information in resolution layers makes
good sense. A doctor looking at an X-Ray or other medical scan can start with
an overview to locate areas of interest, then pull the detail data for the
regions of interest. JPEG 2000 offers some of this capability, but V-Nova
claims to have major advantages for this application.
It will be interesting to see where this leads...
Regards
Craig
Regards
Craig
On Apr 10, 2015, at 5:57 PM, Manfredi, Albert E
<albert.e.manfredi@xxxxxxxxxx> wrote:
Not much detail here, but sounds intriguing. I guess it's to be demoed at NAB.
Bert
----------------------------------------
http://www.eetimes.com/document.asp?doc_id=1326323&;
Parallel Processing Spawns Non-MPEG Codec
Junko Yoshida
4/10/2015 10:08 AM EDT
V-Nova's Perseus encodes UHD quality at HD bitrates
MADISON, Wis. - What if video experts were to design a brand new video codec
fit for 2015?
We live in the moment of 4K Ultra High Definition (UHD) TV, with a growing
number of consumers viewing video sent over unreliable and
bandwidth-constrained mobile networks, while CPUs and GPUs inside their
mobile devices come with multiple cores, massively parallel processing
capabilities and a very large cache.
And yet, we all know that no current video codec can comfortably handle 4K
distribution to UHD TV at home, let alone deliver HD video to mobile devices.
The new video compression algorithm, if it emerges in 2015, probably, won't
be based on MPEG, or even the brand-new H.265.
A London-based startup V-Nova believes that the new video codec everyone
should consider is its brainchild, called Perseus. It is designed to do
hierarchical and scalable video encoding by leveraging massive parallel
processing, while sidestepping the complexity of block-based compression
algorithms.
V-Nova will next week come to Las Vegas for the National Association of
Broadcasters (NAB) convention, to showcase Perseus, which the company claims
has achieved "2x - 3x average compression gains," compared with legacy video
codecs. V-Nova promises to make 4K transmission commercially viable, while
enabling HD on 3G or 4G mobile network by using less power.
Perseus can, in short, encode and transmit UHD quality at HD bitrates, HD at
SD bitrates, and SD video at audio bitrates, explained Guido Meardi, CEO &
Founder, V-Nova in a phone interview with EE Times. More importantly, with
Perseus, "Service providers can upgrade their systems in software, without
touching their existing delivering infrastructure or digital assets," he
added.
Just as human brains function in a hierarchical manner, by first visually
recognizing a human being, then, a face, then a goatee, and then, gray hairs
mixed into the goatee, Perseus will provide additional details of data
representation to the elementary stream of MPEG-2 or H.264, explained Meardi.
V-Nova said it's been developing and testing Perseus for more than five
years, suggesting that the technology has long passed its slideware stage.
The first transmission testing of Perseus was carried out by Sky Italia back
in 2012. Hitachi provided Sky Italia with the first Perseus-based UHD video
gateways.
V-Nova has reportedly collected more than 20 global industry leaders - all
working with V-Nova - in an "Open Innovation" consortium. The consortium
members publicly listed in V-Nova's press release include Broadcom, European
Broadcasting Union (EBU), Hitachi Data Systems (HDS), Intel, and Sky Italia.
Broadcom's spokesperson, reached by EE Times, said, "We support V-Nova with
our reference design and software environment." Perseus implementations are
being "driven by V-Nova through a software implementation of their
algorithm." Meanwhile, Broadcom's set-top chip, currently field-deployed, can
support Perseus by using its CPU power.
Gary Arlen, president of Arlen Communications, a Bethesda, Maryland-based
research and consulting firm, said, "The bandwidth compression [Perseus
claims] is terrific, and its credentials - big names listed in their
consortium - are impressive. But I haven't seen their demo yet. All I can say
at this point is that it looks promising."
To persuade U.S. operators like Comcast or AT&T to adopt Perseus, Arlen said
V-Nova still has to prove itself. But if what V-Nova is saying today holds
true, "this comes at a very interesting timing," he added, as service
operators are in the process of upgrading their infrastructure.
Massively parallel processing
So, how good and original is Perseus? And how do we know this isn't yet
another video codec said to be "ground-breaking to change the world forever"
but never quite pans out that way?
Perseus is neither a "trick" nor a tweak on the conventional MPEG standards,
Meardi said.
Rather, V-Nova went back to the drawing board to rethink how a video codec
should work.
Meardi noted that Perseus is the first video codec - designed from the ground
up - to compress video by leveraging "massively parallel processing."
In contrast, legacy MPEG standards are based on a fundamentally different
architecture - designed to do entropy encoding only sequentially, said Eric
Achtmann, V-Nova's executive chairman and co-founder.
The difference, for instance, will result in the Perseus video codec fully
running on GPU only, encoding lossless video at less than 2 milliseconds,
said Meardi. "Or take an example of 20 megapixel X-ray image. It would take
14 seconds to encode it (lossless) via JPEG2000, but with Perseus, it would
take only 10 milliseconds."
Since Perseus is designed to leverage multi-core CPU or GPU in a handset for
parallel processing, rather than sequential processing, it is much less
power-hungry compared with legacy MPEG-2 and H.264 technologies or even the
newer H.265 (HEVC), according to V-Nova.
By quoting data made available by a large, unnamed mobile carrier testing
Perseus, Meardi said Perseus uses 15 to 30 percent less power than H.264. If
a chip company like Broadcom decides to make a hardware accelerator for
Perseus, power consumption on mobile turns out even lower, he explained.
No more block-based coding
In the course of its "rethinking" of a video codec, the V-Nova team also
decided to eschew the block-based compression algorithms used in legacy
codecs. "The block-based codec would only add large complexity," said Meardi.
"As blocks create artifacts, we need to de-block them, and then we also need
to deal with variable block sizes."
With Perseus, V-Nova says it's bringing fundamental change to the codec
world, compared to the increments enabled by frequency-based Fourier
transformation or wavelet transformation.
By leveraging massively parallel processing power, the V-Nova team developed
for Perseus a new mathematical transformation that can deal with
"correlations [in each video image] properly," according to Meardi. With the
demand for high-quality images at high frame rates, "there are a lot more
repetitions in the frames we are sending," said Meardi. Without revealing
details, he said that with correct correlations, Perseus has potential to do
lossless video compression.
Other secret ingredients incorporated in Perseus include its hierarchical
nature and its ability for creating a single file that can be delivered to
devices in multiple scales.
Hierarchical structure is critical, because Perseus is designed to allocate
data progressively, adding more details as necessary. In the past, the MPEG
community also worked on scalable video coding and added that feature in
2007, Meardi explained. "But it was more of a patch bolted on top of a codec
that was not originally designed for."
Peeling away an onion
Perseus' multiple-scale nature can be best explained as "peeling away an
onion," as Meardi put it. Rather than encoding video tens of times in
different formats, "you can maintain a single file and stream it to different
devices with different capabilities. If an end-device can only decode MPEG-2,
it just decodes MPEG-2 while ignoring the rest," Meardi said.
This scalable feature becomes a huge advantage for inventory management,
Achtmann said. Imagine if you have to create, say 19 different files of one
single asset for multiple deliveries, and need to manage that in a library of
thousands of assets, he explained.
While the feature for no-need-to-save-multiple files means big savings in
server space, even bigger implications lie in the bandwidth use of network
infrastructure - whether satellite, cable or Content Delivery Networks for
the Internet. Rather than setting aside a special bandwidth for an HDTV
stream in addition to an SD stream, operators will be able to send a single
stream.
According to Cisco report, mobile video traffic exceeded 50 percent of total
mobile data traffic for the first time in 2012 and grew to 55 percent by the
end of 2014.
Predicting that 75 percent of mobile data traffic will soon be occupied by
video delivery, driven by only the 10 percent of subscribers who stream
video, Achtmann pointed out that easing the growing network infrastructure
strain is one of the biggest issues the telecommunication industry faces.
Achtmann concluded that Perseus is solving three problems previously
insoluble: 4K distribution, HD on mobile, and video delivery to a Third World
population who have neither 3G nor 4G. Meanwhile, Perseus is also making what
are already possible "more profitable" for operators, he added.
- Junko Yoshida, Chief International Correspondent, EE Times
----------------------------------------------------------------------
You can UNSUBSCRIBE from the OpenDTV list in two ways:
- Using the UNSUBSCRIBE command in your user configuration settings at
FreeLists.org
- By sending a message to: opendtv-request@xxxxxxxxxxxxx with the word
unsubscribe in the subject line.