In order to maintain perceived quality consistent with that of standard-definition TV, IPTV must be designed so that frames are not dropped.
By Ed Ellebracht, Ample Communications
Streaming video over Ethernet networks is especially problematic for equipment providers. Today's equipment is geared for broadband data access and is able to accommodate only the relatively modest bandwidth needs of voice-over-Internet-Protocol (VoIP). The high-bandwidth demands of high-definition TV (HDTV) and DVD-quality video streams, their intolerance for lost frames, and the low-latency demands of channel surfers can quickly overwhelm the capabilities of much of today's enterprise and metro access devices.
IPTV is the mechanism used by the large wireline carriers to provide a video broadcast that can compete with the cable industry's strong video service offering. Wireline carriers are offering broadband Internet access using either legacy DSL or, more recently, passive optical networking (PON) or an enhanced DSL, such as ADSL2 or VDSL. Each carrier also offers (with varying degrees of success) VoIP service, marketed as a replacement for legacy POTS. While POTS service is carried over the traditional PSTN, the VoIP offering may be carried over a mix of PSTN and the Internet or solely over the Internet.
These three services, when combined, form the widely touted "triple play." They allow the service provider--wireline or cable--to maximize revenue per customer and minimize customer turnover. This latter goal is achieved via the triple play because customers are especially reluctant to simultaneously change their telephone, TV, and Internet service agreements.
Wireline carriers have allowed the cable industry a significant headstart, as cable has offered triple-play services for years. While the wireline carriers have long sold broadband data services and POTS service, they only recently have begun to offer broadcast video. However, IPTV subscriber growth is forecast to grow from about 7.5 million subscribers in 2006 to over 25 million subscribers in 20081. Moreover, the IPTV market is projected to grow to $10.9 billion by 20092, and revenue will experience a compound annual growth rate (CAGR) exceeding 100%.
IPTV technology
As often occurs in markets based on rapidly evolving technologies, IPTV may be able to leapfrog the standard cable offering in certain areas. However, because more bandwidth is inherently available in data streams sent over coax than those that use twisted-pair wiring, the IPTV industry faces significant challenges.
Bandwidth needs
The majority of IPTV transmission today uses MPEG2 encoding, which is a "lossy" compression algorithm that encodes standard video (SDTV) into roughly 4-Mbit/sec streams and HDTV into roughly 20-Mbit/sec streams. The size of HDTV streams is problematic for wireline providers, as only the enhanced DSL technologies can transfer a stream exceeding 10 Mbits/sec, and they suffer from limited reach and/or high cost. This is pressing the IPTV providers to change their encoding to MPEG4, which roughly doubles the compression ratio of MPEG2.
Using the combination of enhanced DSL and MPEG4 technologies, wireline carriers can transmit a single HDTV stream with some bandwidth available for voice and data. However, subscribers demand more than one HDTV channel, forcing wireline carriers to press forward with aggressive, optics-based transmission schemes. The most successful appears to be PON, an Ethernet-oriented scheme that provides roughly 100 Mbits/sec per subscriber and uses passive optical splitters. PON-based systems can deliver several HDTV streams, broadband data, and several VoIP streams to each subscriber. Wireline carriers can only be truly competitive with cable when these networks are in place.
Whatever the technology used, the increase in bandwidth of IP traffic is tremendous. Just a few households receiving HDTV channels have the same bandwidth demand as a small city of DSL users--which profoundly affects existing equipment.
Loss sensitivity
MPEG4, typically encoded to comply with H.264, uses an extremely complex distributed Discrete Cosine Transform (DCT) to remove high-frequency components. It also takes advantage of the limited color space that the human eye can perceive to further remove information content. In addition, it uses motion prediction to estimate what the forward frame will look like and only sends the differences between the actual frame and the predicted frame forward. Since the only information that is sent is the difference between the predicted frame and the actual frame, required bandwidth is reduced.
All of these are combined in a stream of key frames (I-frames) containing complete images, each followed by one or more P-frames, each of which contains the predicted frame data. Interspersed between the I and P frames are a short sequence of B frames, which contain the differences between the I and P frames at different time points. Because the complete image is refreshed at the I-frame rate, which typically occurs at two frames per second, error magnification can be significant. Errors in the P and B frames will persist for up to half a second, and I frame errors can cause errors for up to a second.
Almost all of the video sent assumes that the previous frames were correct, which is very different from SDTV in which the screen is refreshed more than 12 times per second. In order to maintain perceived quality consistent with that of SDTV, IPTV must be designed so that frames are not dropped.
Protocol encoding for IPTV
The roundtrip delay of today's networks does not allow time for the retransmission of errored or dropped frames. The video stream typically is encapsulated using real-time transport protocol (RTP), encoded as described in the IETF standard, RFC 2250 or "RTP Payload Format for MPEG1/MPEG2 Video." This encapsulation places key information, such as temporal references and frame-type information in the header, where it can be used to recover from errors and frame losses even before the video decoding is performed. The RTP-encoded frames are transmitted using user datagram protocol (UDP) encapsulation. UDP is extremely lightweight and does not provide any correction for errors or losses. This differs from Internet traffic, which uses transmission control protocol (TCP)/IP. The UDP/IP frames are transferred over the wireline providers' networks using Ethernet equipment.
By using the same protocol stack as the Internet, wireline providers reap the benefit of lower-cost datacom equipment. This benefit is increased through the use of multicast IP protocols. By encoding a video stream as a multicast Ethernet frame, many subscribers can share the same stream. The selection of which streams to send to which subscribers now can be made using multicast control protocols, such as Internet group management protocol (IGMP) at the edges and protocol independent multicast (PIM) at the core. This has the tremendous benefit of allowing the demands of the users to dictate where the video streams flow.
Network structure
Switched versus linear networks
The installed cable network uses a linear distribution scheme in which all channels are sent to all subscribers simultaneously3. This wastes a great deal of bandwidth and is currently impractical for IPTV. It also limits the number of channels that can be transmitted from a cable headend. To conserve bandwidth, IPTV uses a switched scheme where subscribers only receive the channel they are currently viewing. This allows the network to support a much greater number of channels compared with cable linear networks.
IPTV network structure
A simplified diagram of an IPTV network is shown in Figure 1. Because of the massive bandwidth involved with sourcing the many video streams, several congestion points occur. The use of multicast group protocols works to reduce the number of bottlenecks but cannot eliminate them. As shown in the figure, DSL and PON can happily coexist within a single network. Because of the reduced bandwidth capabilities of DSL technology, an intermixed network will require additional video streams, adding to the bandwidth crunch.The IPTV-aware line card
Ethernet equipment currently uses media access controller (MAC) devices that are "streaming-unaware." Systems that make use of network processing units (NPUs), which include nearly all equipment with Layer 3 capabilities, find this structure especially problematic, as they must combine the video streams for each port individually. The conservation of NPU cycles is especially important to equipment providers; the NPUs are the workhorses, providing the more advanced and, therefore, higher revenue features.
To this end, next-generation MAC devices with integral multicast and broadcast functionality are becoming available. Figure 2 depicts a block diagram illustrating how this functionality is integrated into the devices. By providing an external hardware solution to the problem of intermixing the video streams with other data destined for subscriber ports, the NPU is significantly offloaded. This allows the equipment provider to increase the number of ports serviced by each NPU or to use a lower-cost NPU, resulting in reduced equipment costs and improved performance.
Emerging IPTV bandwidth needs will greatly affect current wireline carrier networks. In addition to the exponential growth in bandwidth, the sensitivity to packet loss and latency directly affect the customer's quality-of-service (QoS) perception. The introduction of next-generation MAC devices with integral multicast support is designed to address this issue, reducing system cost while enhancing performance and the subscriber experience.
References:
1 www.iptvarticles.com/IPTVMagazine_2005_07_market_update.htm
2www.mercurynews.com/mld/mercurynews/business/technology/14905473.htm
3 Cable providers have begun to use switched networks; see also www.lightreading.com/document.asp?doc_id=97828&print=true
Ed Ellebracht is chief architect at Ample Communications. He may be reached via the company's Web site at www.amplecomm.com.