Executive Interview with Dr. Alex Eleftheriadis, Chief Scientist and co-founder of Vidyo, who explains how Vidyo technology can make visual communications available to millions of users – Part 1

29 March 2009. Vidyo Inc., a start-up company based in New Jersey, USA, says it has invented the next generation of videoconferencing which on the desktop it calls personal telepresence.

Alex Eleftheriadis is Chief Scientist and a co-founder of Vidyo. Alex drives the technical vision and direction for Vidyo and also represents the company on standardization committees and technical advisory boards. He is an award-winning researcher, bringing over 19 years of research experience in video compression and communications to his role at Vidyo.

Prior to Vidyo he was an Associate Professor of Electrical Engineering at Columbia University. Alex has more than 100 publications, holds 15 patents, has served as the Editor of the MPEG-4 Systems specification at the ISO and Co-Editor of the H.264 SVC Conformance specification at the ITU, and is currently Co-Editor for the IETF’s RTP Payload Format for SVC.

Vidyo has developed a next-generation architecture for HD videoconferencing that leverages the H.264 SVC standard and does away with the MCU. It has the potential to be a disruptive technology. Read the interview to decide for yourself.
*  *  *

Editor: Unlike competitors such as Polycom, TANDBERG and others who use H.264 AVC, Vidyo chose H.264 SVC. What is the difference between AVC and SVC and why did Vidyo choose SVC?

Alex: H.264 AVC came first. H.264 SVC came along in 2007 as an extension of AVC, in order to support scalability. Since AVC is a subset of SVC, it carries forward all its benefits to SVC. If the network has very few packet losses, and the endpoints are the same or similar, then AVC can work well. But AVC generates a single-layer bit stream and hence the video degrades very quickly if packets are lost. As a result AVC needs provisioning in order to ensure a high quality network, with very few packet losses.  

In today’s environment, people want to videoconference over a variety of IP networks with different endpoints, including desktop PCs, laptops and maybe even mobile phones. People want to use video over the public Internet which is subject to packet loss of up to one packet in ten and in bursts one in five. SVC can be used to overcome this packet loss. So for an optimum visual communications experience today, you need to use H.264 SVC.

Editor: How does H.264 SVC work and why is it good for use on the public Internet?

Alex: Scalable video coding (SVC) is a technique that enables a video stream to be broken into
multiple levels of picture resolution, different quality levels, and multiple frame rates.  To address variable conditions, Vidyo divides the video stream into multiple sub-streams or layers: a base layer, and one or more enhancement layers. So if packet loss occurs or the available bandwidth is reduced, the video signal at the base layer still gets through 100 per cent, without picture degradation. (See picture)

You can see a demonstration of the difference which Vidyo technology makes on the Vidyo web site; a Vidyo SVC stream of video and an AVC stream of video are compared when they are subject to packet loss of 2, 5, 10 and 20 per cent. There is no degradation on the Vidyo SVC stream.

Editor: Does Vidyo technology overcome picture degradation caused by packet loss better than other competing packet loss technologies?

Alex: Yes. Error resilience and error concealment technologies are proprietary and there has been no attempt at ITU to create an international standard. Like other vendors, we use proprietary error correction and concealment algorithms.  What’s different in our case is that we discovered that the SVC approach allows for tremendous robustness. This robustness is a necessity in order to be able to use the Internet.

Using the VidyoRouter there is minimal latency to the bit stream because all encoding and decoding is done at the endpoints. We find that Vidyo’s technology can reduce latency at the server to a very low 10 to 20 milliseconds (ms), resulting in a total delay well below the limit of 180 ms. This allows for natural, interactive person-to-person conversation and is the only way to achieve interactivity in multi-person conversations. And if one endpoint has a link experiencing congestion, you can arrange for the video to change instantaneously to a lower spatial resolution to accommodate this, or change from 30 fps down to 15 fps if the congestion persists. When these adaptations are short-lived, they are not visible by the user.

Editor: An enormous advantage of Vidyo technology is that it enables a multipoint conference through the VidyoRouter and without an MCU. How does it achieve this?

Alex: Using SVC we change the way the video signal is represented. Vidyo represents the video signal in a number of layers, for example three in a typical case. So if you use the Vidyo SVC implementation, the VidyoRouter’s intelligence can decide which layers will be transmitted to each site, at what frame rate and over what bandwidth.

A multipoint control unit is far more complex. An MCU has to decode the video signals it receives, composite them into a new frame and then make it into a new signal that is encoded at the chosen frame rate and the chosen resolution. Transcoding in the MCU causes latency and lowers picture quality. An MCU may add 200 msec of delay due to its internal processing. So using the VidyoRouter ensures much lower latency and better picture quality than using an MCU.

Editor: So what are the significant features of the VidyoRouter and how are they superior to the use of an MCU?

Alex: Vidyo’s distributed processing approach enables a single VidyoRouter (a single rack unit appliance) to support 100 simultaneous HD video connections and scales easily to a deployment of thousands of endpoints by using additional VidyoRouters.

The VidyoRouter sees each participant in a multipoint call as an individual video stream; this means that each endpoint has the capability of signaling the VidyoRouter how it wants to receive each stream. Only the layers necessary for displaying a particular user are sent to a particular endpoint. For example, if you see a user in a small window, there is no need for the router to send you layers that correspond to high spatial resolution. Rate matching and the layout of continuous presence are simply routing decisions that are taken instantaneously by the VidyoRouter, and cause no latency.

For each endpoint, the MCU has to perform signal processing for both rate matching and continuous presence (how to place multiple windows in a frame). In most MCUs today, the picture of the speaker is transcoded and then re-transmitted at the resolution of the lowest resolution endpoint. This means the other sites only get the lowest common denominator in picture quality.

So by using the VidyoRouter instead of an MCU for a multi-site conference, you obtain much lower latency, a better picture quality, and scalability up to hundreds or even thousands of endpoints.

In a VidyoRouter multi-site call, each endpoint gets the best picture resolution that each sender, the network, and each endpoint can support. An endpoint with less processing power and network bandwidth might receive and decode fewer layers, whereas other more capable endpoints may receive all the layers.

Editor: Critics say the trouble with Vidyo technology is that it cannot easily interoperate with the two million legacy AVC videoconferencing systems already out there. Are they correct?

Alex: They are not correct. We offer the Vidyo Gateway to bridge the worlds of AVC videoconferencing systems and our Vidyo SVC videoconferencing systems. If you connect the AVC cloud through the Vidyo Gateway you will get the best picture quality that is possible from AVC systems. If there is a good quality connection you will get good picture quality; but if the connection is poor the picture quality will be poor.
Editor: You worked with the ITU Standards Committee for H.264 SVC Conformance. What further technology improvements can we expect?

Alex: The SVC standard was finished and published in November 2007. So far there has been little testing of interoperability of H.264 SVC systems of different vendors, since until recently Vidyo was the only vendor with an SVC product. Vidyo is committed to standards and interoperability, and plans to be as active in interoperability testing as it has been in standardization.

In addition to my serving as the Co-editor of the H.264 SVC Conformance specification, we have donated test video sequences as well as a large portion of the conformance bitstreams for SVC to the ITU. I am also Co-editor of the IETF’s RTP Payload Format for SVC, which defines how SVC is to be transported over the Internet.  That specification is practically complete. The recent announcements by several vendors that they will support SVC in their products means that very soon we will have to start the interop testing process.

Editor: If H.264 SVC has so many advantages, why do TANDBERG and Polycom not implement it?

Alex: Actually Polycom has announced that they will deliver an SVC solution in 2010. There are two ways to look at this. First these two vendors have been using an architecture centered on the MCU for many many years now. It is conceptually and organizationally very difficult to make such a fundamental change in direction, to eliminate such a big part of what they have been building.

The second way is to ask what kind of equipment we need to offer in order to make Telepresence and videoconferencing become ubiquitous. Vidyo’s answer is that we need a new next-generation architecture that eliminates the MCU and replaces it with a distributed architecture using an application-layer router that is scaleable to thousands of endpoints. Vidyo’s system is simple to use and maintain; it offers a 10 x increase in capacity and it is inexpensive. All these are requirements for mass adoption. TANDBERG’s and Polycom’s MCU-based systems have not led us to this type of mass adoption, even though they’ve been around for many years.  

Editor: Is H.264 SVC the way the industry will go over the next five years?

Alex: Absolutely. In the last analysis the market will decide how fast and how extensive the adoption of H.264 SVC is going to be. The much-improved video quality that SVC and the VidyoRouter deliver over the Internet suggest that their potential impact can be huge. The MCU has made videoconferencing possible. But you need SVC and the VidyoRouter to make it truly ubiquitous – available on the Internet, using a boardroom system, an office desktop computer, or a laptop. We designed Vidyo’s offering with that kind of market in mind.

Editor: Alex, thank you very much for these valuable insights.

*  *  *