April 1, 2008

Bullpen: One Network or Two?

By Michael Adams

In my last column, I talked about evolution of the real-time, two-way signaling network in cable systems and some important properties of networks built on the principles of the seven-layer Open Systems Interconnection (OSI) model. One of the most important ideas is link layer abstraction, allowing the network layer - Internet protocol (IP) - and all the layers above to care "precisely not at all" about which particular link layer is being used.

See Figure 1 for a "Martini Glass" view of the OSI model. I've tried to show how the layers spread out in terms of the number of alternative protocols above and below the network.

Figure 1: The OSI model

We also mentioned that traditional video transmission systems do not follow the seven-layer OSI model because they can be cost-optimized by blending some of the layer functions together and leaving out functions that are not required altogether. They are also optimized for one-way transmission because that is what's needed to deliver entertainment video services.

Layers

Many video engineers might look at Figure 1 and say, "I understand the need for physical, data-link and network layers, but why do you need all those layers on top?" In fact, a favorite question of mine is, "What's the difference between the transport layer and session layer, and why do I need both of them?"

What is more, video transmission really doesn't have much use for the re-transmission functions in layers 2 and 4. If a packet containing video is dropped by the network, by the time the loss is detected and a retransmission requested and received, the video has already been decoded and displayed. That is why forward error correction (FEC) algorithms are a must - so that most transmission errors can be corrected without the need for retransmission. If the FEC algorithm cannot reconstruct the bit stream correctly after a serious transmission error, then the viewer will see a glitch in the video, but that should happen rarely in a well-designed system.

Cable systems were originally built as one-way video transmission systems for analog video. Subsequently, digital video channels were added to carry compressed MPEG-2 digital video, and a two-way signaling network was added to support on-demand services.

In contrast, the Internet uses one network for everything, signaling and delivery. This isn't optimal for either one, but a single-network architecture is very simple to operate, and that's the beauty of the approach. Moreover, the Internet has been rapidly catching up in terms of throughput by following a trend to lower transmission costs. And it has done this by swapping out the link layer and physical layer technologies each time a new one came along. We have had rapid migration from T-1 circuits (1.536 Mbps) to DS-3 (45 Mbps) to synchronous optical network (SONET) (155, 622, 2,400 Mbps) to optical Ethernet (100, 1,000 Mbps, 10 Gbps). In 20 years, this ability to swap the link-layer has allowed the Internet to go from being a tortoise to a rocket-propelled hare.

While cable systems follow a two-network environment with in-band (video transmission) and out-of-band (signaling) networks, we now have a new entrant called IPTV that follows a single-network, Internet-style model. So which is more cost-effective, one network or two?

Let's compare costs and look for significant trends.

Cost per bit

A quick Web search brought up a retail price of $1,300 per port for a 10-Gigabit network adaptor, a cost of $0.13 per megabit. Compare this with a quadrature amplitude modulation (QAM) modulator at about $400 per port, a cost of $10.30 per megabit, making the cost-per-bit of the QAM modulator an astounding 80 times higher! Why the disparity? Optical systems can operate at much higher throughputs (in this case 10 Gbps, or 263 times faster than a 38.8 Mbps QAM modulator). This is not an entirely fair comparison because the edge QAM modulator is actually performing some significant gateway and control functions.

However, the 10 Gbps network adaptor only supports a point-to-point connection over fiber, so each set-top would also need a $1,300 network interface. Suddenly this doesn't seem like such a good idea. But equally, 10 Gbps is probably more bandwidth than even the most bandwidth-hungry subscriber needs.

This is why passive optical network (PON) technology is being used for residential fiber deployments; typically, a single optical line termination (OLT) can serve up to 32 optical network units (ONUs). For example, the ITU G.984 GPON standard is fast gaining favor and uses a 2.488 Gbps downstream rate. G.984 uses time division multiplexing (TDM) to share the 100 Mbps among subscribers and relies on the fact that statistically not all of them are going to be actively viewing at the same time.

A good ballpark price for a GPON OLT is $1,800, working out to $1.38 per Mbps, or about 13 percent that of a QAM modulator. However, this isn't quite fair to GPON because it provides equivalent network termination to a bank of QAM modulators and a (shared) laser transmitter (at a cost of about $1,000). Moreover, there are standards for 1 Gbps (GEPON) and 10 Gbps (10GEPON) waiting in the wings, which promise to dramatically reduce cost per bit.

Cable options

Meanwhile, the standard for high-speed data delivery over cable (DOCSIS) also follows the single-network model. In other words, a single channel carries both payload and signaling information. Of course, DOCSIS is built on the 7-layer OSI model, so is able to use any physical channel.

However, the 6 MHz RF channel limitation means each channel can support only 38.8 Mbps, which is why DOCSIS 3.0 bonds multiple RF channels to achieve higher downstream and upstream rates (rather than wider RF channels). An alternative proposal to use PON as the physical layer for DOCSIS (DPON) would bypass the need for RF modulation altogether, but, of course, would only work over a fiber-only network. Finally, RF over glass (RFOG) would retain modulation, but eliminate optical-to-RF translation at the node.

We are clearly in the middle of an optical transition. Low-cost optical devices are being introduced that can support voice, data and video services over a single optical carrier. Fortunately, cable operators have several options from which to choose, including the following:

• Move to a single-network environment by using DOCSIS 3.0 to deliver all services (voice, video and data) and to provide signaling
• Migrate to all-fiber networks, but stay with a broadband network, such as RFOG
• Migrate to all-fiber, baseband networks, employing standards such as GPON

Michael Adams is vice president, Systems Architecture, for Tandberg Television. Reach him at MAdams2@tandbergtv.com.