The debate between flexibility and cost is always a changing one, and broadcast equipment makers are facing a change in the way they put systems together as system on chip solutions are starting to displace the FPGA makers.
The traditional approach for broadcast encoders sees designs being implemented in large, expensive FPGAs as a way of encapsulating all the knowledge and expertise in image quality and system management.
Now that the latest video encoding standard, MPEG4part10 (also called H.264 and the Advanced Video Codec – AVC) has settled down and is now being implemented in encoders for high definition satellite TV and IPTV over the phone lines, SoC providers are now looking to provide this key technology almost as a commodity item for Tier One encoder makers such as Harmonic.
Strategic shift
“There is a shift in technology and the way Harmonic designs encoders and that depends at the end of last year,” said Arnaud Perrier, senior product marketing manager for Advanced and HD Encoding. “We realised that the technology hadn’t been ideal and we were getting feedback from customers on the first MPEG4 deployments.
“We took reference models from codec vendors and we used those internally in a ‘shoot out’. Some were ASICs, some were FPGA, some DSP, and we spent a lot of time doing that. One technology in particular, which was ASIC-based, blew the others out of the water. We were really surprised. It gave us better density, with 4HD channels in 1RU rack unit.
“It’s a different approach. We have some functions that are hardwired and that frees up a lot of cycles for the other tools that do matter, such as motion estimation and search frames. We have an innovative adaptive and hierarchical search algorithm that means we can process a whole frame at once. We still have DSP and FPGA for certain processes, and there is no one technology that can address everything, but we are taking a really holistic approach.
“We could not just take the technology and put it in a box and make it work in a broadcast environment – it’s just not possible. We have a team that’s optimising the microcode and replacing it with our own. It’s not just a plug and play approach.”
A chip is not enough
US chip startup Telairity identified the same trend and has developed a single chip which combines five independent very long instruction word (VLIW) processing cores, a video controller, and a DRAM controller supporting an I/O bandwidth up to 5.3Gbit/s in the SoC. However, equipment makers weren’t interested in just a chip.
“We went to all the customers and they said they needed a complete solution,” said Shubha Tuljapurkar, vice president of marketing at Telairity.
The company launched an encoder reference platform that uses eight of these chips with Telairity’s own software to encode one channel of AVC in real time, including key tools to improve the quality of the image such as CABAC, replacing up to 24FPGAs and DSPs.
“It means you can programme in C like a gentleman rather than Verilog,” said Howard Sachs, CEO of Telairity. “We are providing the fundamental video encoder hardware and software design, so each OEM can focus on its particular value-add while getting to market much more quickly,” he said.
Similarly Thomson has developed an SoC for encoding that samples this month. The 90nm Mustang chip has two ARM processor with six programmable DSPs and dedicated hardware for video-specific tasks like motion estimation and entropy encoding, as well as a 10Gbit/s PCI Express bus. It will be used in the ViBE encoders shipping at the start of next year from Thomson subsidiary Grass Valley.
“Grass Valley is in a unique position to develop this extremely complex single-chip solution,” said Marc Valentin, president of Grass Valley. “The result is that the ViBE encoder based on the new chip will offer incredible performance and flexibility and we will be able to build the chip into other products across the range from format converters to servers and production switchers.”
However, this commodity view has been dismissed by other encoder makers, as the implementation of the algorithm is still key to quality of the video encoding.
Added value
“With any encoder design it’s about the encoding algorithm, not about the implementation,” said Simon Bigg, senior vice president of engineering for compression systems at Tandberg Television.
“Although we are here to make the best encoders on the planet we have to provide what the market needs, so we are very, very agnostic about what technology we use. If someone came out with a DCT chip and it was cheaper for us to use that then implement it in a Xilinx FPGA then we would but the thing we will never give way in is the flexibility and the ability to put all the blocks together,” he said.
“Even if it looks like encoding is becoming a commodity, I don’t agree,” said Ovadia Cohen, vice president of marketing at Israeli equipment maker Scopus. “You can find encoding engines coming from companies like LSI Logic or IBM or even from Japanese manufacturers and I do believe that a lot of effort has been invested in that silicon but it is not enough. There’s a huge effort needed around it, not just the audio, multiplexing and transport, but statistical multiplexing (to allow channels to share bandwidth), and handling the hidden data such as VBI. There’s huge amounts of work that you have to do even if you are using ready-designed silicon or your own design.
“I think the high end encoders from the top five companies are definitely differentiated from others using the low cost chips,” he said. “The quality of the video is the number one factor.”
Differentiation will be in the system software, he says.
“Now you will find the silicon vendors are developing encoding engines and companies will use those engines. The question is how they will use it. It’s like buying a CPU from Intel – does that mean you have a PC or a laptop? No. The more you invest in pre-processing, statistical multiplexing and many other issues like that, the more you get.”
FPGA fights back
But the FPGA makers are not too worried. Altera has been focussing on the broadcast market for several years, and points to PC-based encoder maker Ateme. “Three years ago their solution was DSP based and now it is on a single FPGA for an encoder with broadcast quality,” said Herve Mer, Market Development Manager for Broadcast and Consumer at Altera Europe.
“Today there’s several ASSP vendors with chip for the consumer marker and they design it first to make some money in the broadcast market,” he said. “That’s a dead end solution because you fix it in hardware and it doesn’t develop.”
Similarly, Xilinx sees its role as providing more of the system hardware and interfaces, with the move to 65nm driving up performance and driving down the cost per function.
“I think the wide word, multiple processor technologies are valuable but there’s always the issue of connecting to memory or to the SDI data ports.” said Gregg Hawkes, principal engineer for video applications in the Advanced Products Division at Xilinx. “One view is with the geometries being driven down we can throw software at this problem but we always tend to end up with a DSP and FPGA. H.264 isn’t the last compression technology we will have to deal with.”
He also sees the FPGAs as able to mop up other functions such as SDI interfaces and the audio compression. “For example, we can integrated the audio through a reference design for the FPGA, for as many channels as you need, rather than having to add an extra board with perhaps 8 extra chips,” he said. “Before it was video, now it’s audio, what will be it be in 07?”
Xilinx is shipping samples of its Virtex-5 LXT FPGAs, which are the first FPGA with a hard-coded PCI Express Endpoint and Tri-mode Ethernet Media Access Controller (MAC) blocks for broadcast systems.
Xilinx has been working with several customers such as Agilent in its early access program since June of last year, and while the parts are sampling now they will not going into production for another year, in 2008.
“What started in the communication segment to improve bandwidth, cost and scalability is now becoming an industry-wide migration from parallel to serial interfaces across many applications within wired/wireless, video, storage, servers and consumer. A one-size-fits-all FPGA approach is no longer sufficient,” said Steve Douglass, vice president of Product Development at Xilinx. “The Virtex-5 LXT platform is the first of several high-speed serial platforms to be offered in the Virtex-5 family. The LXT platform is aimed at the large number of serial connectivity applications ranging from 100Mbps to 3.2Gbps.”
And the FPGAs are widely used across broadcast. For example, Canadian equipment maker Miranda is using the latest Xilinx FPGAs for its multi-room, multi-image display plus HD/SD routing in a single, expandable chassis that can display up to 96 separate streams of video on 8 different screens.
Similarly, designers of video compression equipment have used banks of FPGAs and DSPs to implement the latest standards.
The latest process technology opens up new opportunities for SoC devices in high quality broadcast applications systems, and the systems will be more of a combination of different devices and technologies. This gives designers even more opportunity to get the best performance out of innovative system design.