In my last column, I argued that one network is better than two for video delivery. In this column, I want to talk about the mode of delivery of video services.
Of course, the original delivery mode was broadcast, and cable networks can trace their roots to coaxial-based video distribution systems. However, with HFC upgrades and real-time, two-way signaling, cable systems have become sophisticated two-way communications networks that can support a number of different delivery modes.
For video in particular, several modes now exist: broadcast, on-demand, and switched digital video (SDV). Here, by "on-demand," I mean both movies on demand and network personal video recorder (nPVR) services (such as Time Warner Cable’s "Start Over" service). Overall, the trend seems to be toward "everything on demand," driven by our busy schedules and by the Web alternatives (such as hulu.com, iTunes, etc.). But how realistic is everything on demand from a capital cost and operational point of view?
Let’s start with broadcast, the most efficient mode if everyone is watching a limited range of programming. For example, to broadcast 500 standard definition (SD) channels, you would need 250 MHz of spectrum using MPEG-2 compression at 12:1.
But if you have to add 500 HD channels, you would need 1,500 MHz using MPEG-2 compression at 2:1. Even if you can reclaim spectrum currently used for analog video, this approach doesn’t look promising.
What is more, the choices that customers are demanding are exploding. SDV and on-demand turn the equation around; as more than one analyst has observed, "I don’t need 500 channels; I need just one channel of my choosing."
SDV was developed to address this problem by only transmitting channels in a service group that are actually being watched. Technically, SDV is considered a multicast delivery mechanism because if two or more subscribers in a service group are watching the same channel – CNN, for example – the channel is sent only once, and all subscribers get the same channel.
However, SDV is now being recast (pun intended) as a unicast service because as programming becomes more diverse, the probability of more than one person watching a specific program in a small service group (500 tuners, for example) is relatively small. In addition, there is considerable interest in targeted advertising, where ads are addressed to an individual subscriber. Moreover, to achieve a seamless transition to time-shifted viewing (nPVR), you have to size network bandwidth for the worst case scenario of 100 percent unicast delivery. Finally, to support telescoping in a linear stream to longer-form material, nPVR is required to bring you back to where you left off.
These additional requirements could be seen as the final nail in the SDV coffin, or alternatively we could just define "Next Generation SDV" as a unicast service that supports targeted advertising, nPVR and telescoping. This becomes more interesting when we look under the covers at the different implementation models for unicast programming (regardless of what it is called).
There are two main ways to skin the unicast cat. In the "SDV implementation model," the edge quadrature amplitude modulator (QAM) provides replication of an input stream to the set-top boxes. In the "nPVR implementation model," a video server provides the replication function. SDV model 1. The input program, typically received from satellite, is first conditioned to be a constant bit rate (CBR) feed. The rate is commonly 3.75 Mbps for SD channels, but more recently higher rates have been added for programming where quality would be degraded by conversion to 3.75 Mbps, such as high-action video programming such as sports. For HD channels, the rate is commonly 15 Mbps (with the same caveat).
2. The CBR version of the feed is distributed using IP multicast to all edge QAM devices via Gigabit Ethernet (GigE) networking.
3. Each edge QAM replicates the program to each output QAM channel where it has been "tuned" by a set-top (by means of the SDV channel request protocol). nPVR model 1. The first step is identical to the SDV implementation model because current video servers are limited to CBR streams. Future video servers may be able to ingest in native variable bit rate (VBR) formats, but don’t hold your breath – we’ve been talking about VBR-capable servers for nearly 10 years.
2. The feed is ingested by a video server (or by multiple servers) in real time and written into storage.
3. The video server replicates the program to each set-top that "tunes" to it. In "live streaming mode," the server plays the stream out immediately after ingest with almost no delay, about 1-2 seconds in practice. (Under the covers, standard VOD signaling protocols are used to request the program instantaneously in much the same way as SDV).
The advantage of the SDV model is that it is less expensive to deploy because the replication is done by the edge QAMs, essentially for free. However, the nPVR implementation model is a single, integrated solution that supports targeted advertising, nPVR and telescoping.
Let’s examine how the nPVR model supports these additional functions. Targeted advertising Targeted advertising can be supported by play-listing in the video server. When the session is created, a call is made to an ad decision system (ADS), which returns a play-list of all the ads that should be sent to this individual subscriber based on rules set by the campaign manager. (This architecture has been defined in SCTE 130 by Working Group 5 of the Digital Video Standards sub-committee.)
Because live content contains embedded SCTE 35 messages and equivalent markings can be added to the on-demand metadata, the ad decision manager (ADM) knows where it can insert or replace ads to address the individual subscriber.
nPVR: nPVR functionality, for example TWC’s "Start Over" service, is also supported. When the set-top "tunes" to the program, a local graphic is displayed: "Press Select to watch this program from the beginning." If the subscriber presses the select key, the program is played from its start point in time-shifted mode. When the program finishes, the subscriber is given the choice of watching the next program from the beginning.
Telescoping: Telescoping can make advertising interactive. An enhanced TV (ETV) application is embedded into the ad (using two additional PIDs) and an ETV-enabled set-top captures and executes the application to display a graphic: "Press Select for more information on this product." If the subscriber presses the select key, the long-form infomercial is played. After it completes, the subscriber is automatically returned to the point where the programming left off. Complexity and cost The nPVR implementation model is significantly more complex simply because it supports a greater range of functions. There are a number of initiatives to build implementations based on well-defined interfaces between well-defined components (such as SCTE 130 for advertising components) that promise to help to address this additional complexity.
The SDV implementation model is much cheaper than nPVR model because it does not require video servers. However, to target ads using the SDV implementation model would require a splicer for each subscriber. This could be done by adding a hardware splicing function into the edge QAM, but suppliers have resisted this because the edge QAM market is highly price-competitive.
The nPVR model lets you splice ads into an individual stream at almost no incremental cost over pure nPVR because a server that supports play-listing costs no more than one that doesn’t.
Finally, if video server prices continue to fall (from about $400 per stream in 1998 to less than $25 per stream today), then the nPVR implementation model will soon become cost effective. It also brings us to everything on demand delivery of all services – as just one channel tailored to you. Michael Adams is vice president, Systems Architecture, for Tandberg Television. Reach him at firstname.lastname@example.org.