September 6, 2013 – For pay TV providers trying to navigate an ever more complex matrix of encoding challenges surrounding past, present and future generations of service the good news is an abundance of new solutions has emerged to offer direct escape routes out of the maze.
The bad news is the more options there are to choose from, the harder it becomes to make a choice. New generations of off-the-shelf microprocessors used in conjunction with software-based transcoding system offered by firms like ATEME, Elemental, Envivio and Digital Rapids are strengthening the case on the software side of the argument as traditional headend suppliers double down on claims that their hardware systems have opened an insurmountable future-proof lead against the likes of Intel.
Two years ago, pay TV operators looked on the software-based systems as solutions for their IP TV Everywhere services. Today the question is whether to cap the legacy systems by moving to a converged headend environment where the software-based systems can be used to handle increasing volumes of legacy VOD content and to bring MPEG-4 AVC (Advanced Video Coding) into play with higher volumes of live content.
“As the pace of change of formats, requirements and new business models continues to quicken, product form factors are drifting away from dedicated hardware appliances toward agile software solutions,” notes Avni Rambhia, senior industry analyst for Frost & Sullivan. “This trend is particularly pronounced in segments such as production and multiscreen transcoders, and is creeping into traditional hardware domains such as linear pay TV encoding.”
Actually, ATEME makes the case for both software and hardware systems, depending on the applications scenario. As explained by ATEME chief strategy officer Benoit Fouchard, the firm’s Titan software transcoding platform uses Intel processors running on COTS (commodity off-the-shelf) servers at densities as high as 18 blades per seven rack-unit (RU) chassis to support pay TV distribution from operators’ headends. Its modular Kyrion hardware system, utilizing purpose-build FPGAs (Field Programmable Gate Arrays), is meant to be used for contribution applications, such as DTH satellite or field encoding of sports and other high-value broadcast events for distribution to broadcast facilities at superior quality and extremely low latency.
DirecTV, for example, is using the Kyrion platform to deliver HD VOD service in the U.S. and has found other uses for the technology as well. In 2010, the DirecTV Latin America deployed the platform to bring back feeds of the World Cup Soccer games from South Africa for broadcast across its footprint.
Meanwhile, the Titan transcoding system has been widely deployed in global markets. For example, U.K. OTT pay TV provider TalkTalk last year implemented the ATEME transcoding platform in conjunction with its transition to delivering pay TV via the YouView connected TV set-top. In Korea digital broadcaster CJ Powercast is using the software platform to simplify operations for its VOD and Catch-up TV services. India’s Digicable, the country’s second largest cable operator, is putting the technology to use also to support rapidly expanding transcoding requirements for VOD services.
In June, in a demonstration of how both the hardware and software platforms can be used in tandem to support next-generation TV technology, ATEME teamed with distributors for demonstrations involving live broadcasts of the French Tennis Open. Working with the terrestrial broadcaster France Televisions and satellite broadcaster Eutelsat, ATEME supported distribution of matches in the ultra-high HD format 4K. At the same time, in a trial conducted by the 4EVER consortium, a group devoted to promoting HEVC (High Efficiency Video Coding) and 4K, ATEME and Orange teamed to deliver HD HEVC multiscreen service coverage of the French Open over the Orange fiber network using the Titan transcoding platform
“Our company is about compression technology that can used with software or hardware platforms to deliver superior performance based on what the situation requires,” Fouchard says. “In contribution applications, we’re not in the density game; it’s all about superior reliability and quality with no latency. The density play is at the headend with our Titan product line. There’s no better density than what we offer with Titan.”
In other words, ATEME as a software transcoding supplier is claiming high density as a strength even as hardware-based competitors claim one of the reasons to use their solutions is the need to overcome the presumed low density inefficiencies of software-based systems. And ATEME is saying that, to achieve the highest possible performance for contribution encoding by a hardware system, it is using FPGAs as the best option for a hardware solution without worrying about what has to be sacrificed in terms of density.
Fouchard acknowledges “you can do high density with ASICs (application specific integrated circuits)” but adds, “you can’t achieve what’s required for high-performance contribution applications with that technology. And, when it comes to transcoding on ASIC platforms, don’t have the flexibility that pay TV operators need to stay competitive in this environment.”
With the Titan platform the operator gets the best combination, he says: very high density with the flexibility to evolve with encoding technology and new needs that can only be accomplished on a software-based platform. Fouchard notes the firm’s density advantage with the Titan platform rests in part on an alliance with European blade server supplier Bull, which is known for providing COTS blades for space exploration, medical radiology and other high-end applications. “Our platform operates on 18 blades in a 7 RU (rack unit) space,” he says. “By comparison, if you’re using HP you can do 16 blades in 10 RUs.”
Currently the Titan platform is designed to run on the Intel Xeon E5-2600 family of server chips, which until now represented the cutting edge of Intel’s commercially available server products. The combination of Xeon chips and Bull blades with Titan algorithms produces capacity to process up to 90 HD or up to 360 SD inputs per 7 RU configuration.
Ericsson, with its new AVC 4000 encoding processor, is claiming performance well beyond the most advanced Xeon processors. As previously reported, the company has made the new chips software programmable to enabled introduction of new codecs like HEVC without requiring new hardware, which is one of the advantages cited by software-based systems. Giles Wilson, head of the video compression business at Ericsson, says the new AVC 4000 chipset handles three trillion operations per second. The Xeon E5-2600 does about 300 billion per second.
But Intel is moving well beyond these levels, having already introduced the new Phi coprocessor platform performing at one-trillion operations per second. These chips are designed to work in standalone mode or as boosters to Xeon processors, meaning that any application designed to run on a Xeon processor can run seamlessly across both types of processors.
Of course, there’s a lot more to transcoding than processing speed, but the argument that purpose-built video processing chips or proprietary hardwired applications of DSPs (digital signal processors) are better at handling video-related tasks than general-purpose Xeon chips may be losing its steam. As Erin Jiang, segment marketing manager at Intel notes, the need for video processing support has become so pervasive in server applications, the firm’s development path necessarily dictates that its chips be purpose built for video.
“This has been core to our approach for a long time,” Jiang says. “For us the goal is to introduce advances that continually improve video processing performance.”
With the launch of the E5-2600 “Sandy Bridge” series of Xeon chips Intel integrated the capabilities of its GPU (graphics processing unit) chips into the Xeon core, bringing into play what it calls Quick Sync Video to utilize the on-processor graphics capabilities with greater power and density efficiency then can be done when the GPU and CPU are separate. While experts argue over whether the Intel GPU capabilities measure up to those of competitors like NVIDIA and AMD, it’s instructive that Elemental Technologies, which has long employed NVIDIA GPUs with Intel CPUs to deliver a highly successful software-based transcoding platform, now gives customers the option of running its system on the unified E5-2600 platform.
With the next version of its server chip, Intel is taking the Quick Sync Video capabilities to another level, Jiang says. “With our Haswell architecture, the embedded GPU can be used to handle all the video processing requirements, not just graphics, which frees up the CPU-related processing for other things,” she explains. Developers using the Intel Media Software Development Kit can assign video encoding, decoding and transcoding to the on-chip Intel Core processor HD graphics while allocating CPU resources to things like local ad insertion or support for advanced features, she adds.
These new capabilities are incorporated into the just-released Xeon E5-2600 V2 family of chips for two-socket systems. In the fourth quarter Intel plans to introduce high-end Xeon E7-8800/4800/2800 v2 processors for four- to eight-socket servers. As quoted by eWeek, Xeon product line manager Dylan Larson asserts the E5-2600 V2 offers a ten X performance improvement in H.264 transcoding compared to the previous generation.
One factor contributing to this improvement is the fact that the Haswell generation of microarchitecture employs 22-nanometer circuitry, a significant gain over the 32-nm circuitry used with Sandy Bridge. This translates into an 18 percent gain in power efficiency, according to Larson. In 2014 Intel says it will begin moving to 14 nm. circuitry on some of its microprocessors.
The closer COTS-based video processing gets to performance levels touted by proprietary hardware encoders, the more the issues of flexibility and costs of operations become the determining factors in operators’ choices. And as these considerations gain ascendancy, the true differentiators will have to do with the performance capabilities of the software architectures in utilizing processing power to maximum advantage.
Where ATEME is concerned, Fouchard argues the company’s software knowhow, starting with construction of standards-compliant codec algorithms from scratch, should be the determining difference in pay TV providers’ decision-making process. Last year, the company introduced EAVC4, entailing a top-to-bottom redesign of its encoder employing a patented technique called “Multi-Screen by Design.” The process employs “intelligent paralleization” to perform common processes only one time for all bitrates and screen formats, resulting in higher performance for multiple outputs from a single content stream, Fouchard explains.
The EAVC4 video encoding architecture exploits a combination of multi-processor and multi-core 64-bit architecture with multi-threading from slice level up through the macro-block level to achieve its efficiency gains, he says. The result is bandwidth efficiency improvements of up to 20 percent based on objective measures and higher gains based on visual perception, he notes.
Demonstrating the benefits in side-by-side clips of a basketball game encoded in HEVC, Fouchard says ATEME’s HEVC performance can be seen as the benchmark for expectations from the new standard for the next couple of years. Objective measurements show 35 percent to 50 percent improvements in bitrate efficiency compared to AVC while subjective feedback puts the gains at 40 to 60 percent, he says.
“In our tests we’ve set sequences in MPEG-4 at the quality levels required by Freeview (the UK DTT broadcaster), which are higher than the norm for U.S. cable,” Fouchard says. “Then we do multiple encodes in HEVC, cutting back in increments from 7 megabits per second, to 6, 5, 4 and so on until the viewer says the quality is the same.”
How big a performance improvement in terms of bandwidth efficiency is achievable over HEVC depends on frame rates and resolution, with higher levels showing better improvement ratios. “There are a lot of tricks used with HEVC that you can’t take advantage of at lower frame rates,” he says. “For example, at 30 frames per second you have less ability to find duplications between frames than at 60 fps.”
A big leap in resolution, as represented by super HD or 4K, is where the difference really shows, he adds. “When you multiply the resolution by two you only multiply the bitrate by the square root of 2,” he says.