o that you can send to anyone, anyw
here, over a ubiquitous digital network will be the confirmation that convergence has arrived.
What's the payoff for being able to send video as easily as a page of text? For one thing, videoconferencing ceases to be nine parts setup and one part collaboration. And tele-presence becomes real, so you can call an expert on another continent to help you overcome a technical problem on a complex piece of machinery or aid in a difficult medical procedure. Sales presentations and training sessions can take place without anyone having to hop a plane.
But don't cancel your airline tickets just yet. The requisite high-bandwidth pipes aren't fully in place, and technologies for transmitting video globally aren't fully formed. The existing telephone system handles point-to-point connections and can establish these connections quickly. But bandwidth limits mean that companies have had to maintain separate data and video networks. The main alternative is the cable TV (CATV) network, which can
deftly handle video streams but was built to broadcast video in a one-to-many model; point-to-point, temporary, and symmetrical (i.e., equivalent sending and receiving transmission speeds) connections are stumbling blocks for CATV. These aren't the only possibilities. Some analysts see wireless technology as a powerful candidate. Unfortunately, all these approaches have weaknesses that keep them from offering a total solution for creating an anywhere-to-anywhere video network.
The high demands of video mean the final solution will touch on all the different layers of the communications infrastructure. This includes the actual, physical medium that carries video (i.e., copper phone wire, coaxial cable, optical fiber, or wireless mechanisms) and the transport standards, like asynchronous transfer mode (ATM), that manage the flow of data. The final component will be the actual applications that implement such things as telepresence, videoconferencing, or Internet-based virtual reality.
Gr
eat Expectations
For MPEG-1 video quality, a network needs to handle about 120 to 140 Kbps. That translates into the equivalent picture quality of a VHS tape (352 by 288 pixels, at 8 bits per color and 30 frames per second). MPEG-2 quality, which would be necessary for a telepresence application that ships detailed technical information in real time, requires about 500 Kbps, which translates into the sharpness of S-VHS.
For video to truly become a ubiquitous data type, tomorrow's networks will need to transmit video in both directions at the same speed (i.e., full duplex). For this you'd need 4 to 6 Mbps. (You would also need real-time encoding to compress raw video data on-the-fly.) Video data also has to arrive in a continuous stream and in real time: It is unacceptable for any individual frame to be delayed or for any associated audio track to arrive out of sync with the picture. The result would look like a poorly dubbed foreign movie.
Shared connections are often unsuitable for r
eal-time video applications. If network traffic is low, your video communication will have enough bandwidth as long as you're using one of the evolving broadband technologies like cable modems. But if the cable alternative becomes a popular way for people to set up video connections, the increased demand could mean many people are vying for the same pipe, and there is no guarantee that the portion available to every user will be adequate. That's one advantage of a dedicated connection, like a T1 link, to connect your company's offices: You have exclusive use of that connection. However, you also pay thousands of dollars a month for the service, and your communications abilities are anything but global. A leased line doesn't connect you to the outside world, including that potential customer who might be sold on the flashy sales presentation your corporate communications department recently created.
To make matters worse, an MPEG data stream is by no means constant. It varies greatly, depending on the ty
pe of video you're encoding. The compression ratio improves when images in consecutive frames don't change much. Sequences with lots of motion afford less compression and therefore need much more bandwidth. Compression ratios may vary between 1:50 and 1:200 within the same video file. As a result, a video network should guarantee a minimum throughput level.
To resolve this situation, some protocols, namely ATM, let you reserve the right amount of bandwidth for the transmission by assigning a particular "quality of service" level when you first make the connection. (For more on ATM's service levels, see "Virtually Well Connected," August BYTE.)
How well does the Internet fare in meeting these demands? Today, the Net isn't reliable enough to deliver real-time video. Too often, bandwidth on the Net is like a magic act: Now you see it, now you don't. The variations depend on the time of day you're logging on as well as the topology between the points you want to connect. Add to that the reality that the
Internet isn't a commercial entity, so there's no central authority that can guarantee you get the resources you need. This situation may change when the Net's data paths become wide enough to absorb video (see this month's cover story, "Breaking the Bandwidth Barrier").
Video's demanding requirements have kept it a specialty data type that needs expensive and dedicated networks. The technologies to let us send and receive video at the same transmission speeds using connections that we set up and tear down as needed aren't mature, but some are getting close.
Video Dial Tone
Plain old telephone service (POTS) may be plagued by having lower bandwidth capabilities than the cable TV network, but there's nothing in the physical infrastructure of the telephone network to make it inadequate for video. The telephone system's designers assigned the lowest 4 KHz of the available spectrum on twisted-pair wiring for voice calls. New, higher-bandwidth services using the same wires must eith
er coexist with traditional services by using higher frequencies, or telephone calls must be converted to digital and be interleaved with other data. Much of the current phone system is already digital; the final connections to end users are the remaining analog holdouts.
In the early 1960s, engineers at Bell Labs designed multiplexing systems that digitized voice into 64-Kbps data lines and then sent several of them in a framed stream. This first structured signal was called DS1, which spawned T1 and E1 lines (at 1.54 or 2.04 Kbps, respectively). However, these lines need repeaters every 6000 feet and require 1.5 MHz of bandwidth, which is high by today's standards. Interference is another disadvantage: Only one T1 line can operate problem-free in a 50-pair cable. This is why a variety of
digital subscriber line
(DSL) technologies may be the long-term answer.
DSL can turn a standard pair of copper wires into a sophisticated digital highway when you add digital modems at both
ends. DSL variations come in a torrent of acronyms:
--ADSL (asymmetric digital subscriber line)
is the first DSL variation ready for high-quality video data, but with a catch. As the name implies, it works asymmetrically, so much more information can be transmitted to the subscriber than what comes back. ADSL's downstream rate is affected by distance, so at 9000 feet, ADSL achieves a speed of 8.4 Mbps; at 18,000 feet, the rate drops to 1.54 Mbps. The upstream rate is between 16 and 640 Kbps. The main reason for this limitation is interference caused by capacitive couplers within the phone system. This means ADSL is better suited to distribution-type services, such as video on demand, than any-to-any connections. (It is also suitable for Web browsing.) Furthermore, ADSL lets you carry on conversations on the same line where video is streaming, and your phone service works even if the ADSL modem fails.
--HDSL (high-data-rate digital subscriber line),
using more advanc
ed modulation, is less demanding on spectrum and needs no repeaters. It also delivers 1.54 to 2.04 Mbps. HDSL can go up to 12,000 feet, but it requires two twisted-pair lines for T1 or three for E1.
--SDSL (single-line digital subscriber line)
delivers the same speed as HDSL over a single line, at up to 10,000 feet.
--VDSL (very-high-data-rate digital subscriber line)
is, for the time being, also asymmetric, but at even higher data rates. Downstream, it offers speeds between 12.9 Mbps at 4500 feet and 51.8 Mbps at 1000 feet. Upstream rates are from 1.6 to 2.3 Mbps. VDSL is intended for ATM networks. Both ADSL and VDSL provide error correction.
DSL is taking the world a step closer to high-end, point-to-point video, and the bandwidths it offers are enticing. But as long as the technologies are asymmetric, DSL is not a complete solution for transforming the existing phone infrastructure into a powerful, multimedia data highway.
Cable Guy to the Res
cue
Just as DSL provides new capabilities for standard POTS networks, the CATV network is being enhanced to deliver fast data-access services. Cable modems are currently being used in trials throughout the United States and Europe. These projects aim to find a way to make the high-capacity cable network suitable for point-to-point connections. Video data still goes to all the cable subscribers connected to the loop, but the modem filters out the information addressed to a particular subscriber. This establishes a virtual point-to-point connection, at least in one direction.
But much of the existing cable plant still can't accommodate two-way traffic. To overcome this limitation, cable companies need to upgrade their infrastructure by adding routers. This refurbishing project could cost an estimated $500 per subscriber, not including the price of the cable modem, which could double that amount.
But the promise of broadband speeds over an existing network can make some people gloss over
the costs. The theoretical limits of cable
connection speeds
may be as high as 30 Mbps. But realities quickly bring this theoretical rate to more modest levels, perhaps to speeds that are two-thirds slower, or roughly equivalent to what ADSL modems offer. Performance penalties are due in part to the fact that cable subscribers share available bandwidth, much like clients on a LAN.
One glimpse into how cable companies might offer CATV networks as a video link to corporate customers is
Motorola's CableComm
, a turnkey solution for implementing video CATV services. The package includes modems and subscriber access software. The necessary cable router that installs at the head-end interfaces with a hybrid fiber-coax (HFC) distribution network to local or remote IP networks. In the CableComm solution, a channel occupies a 6-MHz slice of downstream bandwidth, in the range of 65 to 750 MHz. The cable router manages all the modems in the system and prompts them to shift
to alternate channels. Each downstream channel provides a 30-Mbps raw data rate, with 768 Kbps upstream. Since all traffic passes all the premises that are connected to the network segment, the system encrypts messages with shared private keys unique to each modem.
Fiber-optic lines combined with CATV's coaxial networks may play an important role in the development of anywhere-to-anywhere video networks. A single fiber can transmit up to 30 terabits per second, or the equivalent of 450 million simultaneous phone conversations with digital quality. HFC is a combination of fiber-optic trunk lines, which bring the data over long distances, and coaxial cable networks, which provide the local-loop links.
On paper, the combination of high bandwidth with an existing infrastructure makes HFC an ideal candidate for video applications. The main drawback of this approach, however, is implementation expense. Large-scale HFC deployment requires new cabling or at least refurbishing existing coaxial cables. In
cities, that means a lot of digging; in less populated areas, amortization seems impossible. Estimates on cost per subscriber start at $700.
Wireless Wonders
One way to get around the technical problems inherent in today's telephone and cable TV networks is to bypass those infrastructures altogether and go wireless. Advantages of wireless solutions, like satellite networks, include flexible and high-speed bandwidths, high reliability, and the ability to reach any location with near-zero setup time.
Some Europeans are already familiar with the advantages of satellite communications. For example, the Astra family of direct broadcast satellites now delivers hundreds of TV and radio channels to about 10 million customers. But wireless broadband networks have trade-offs of their own. Two examples illustrate this point. The first
is DirecPC,
by Hughes Network Systems. DirecPC transmits data from a geosynchronous communications satellite directly to a dish antenna
that you attach to your PC. Direct satellite broadcasting minimizes industry investment in new ground-based infrastructure, largely because users buy most of the equipment themselves. The antenna, receiver, ISA interface card, and software cost about $1000. Professional installation costs about $500 more unless you install the equipment and align the dish yourself -- a nontrivial task.
Service charges for Internet access via DirecPC depend on usage, which is measured in megabytes, not minutes. You pay 80 cents per MB during peak hours and 60 cents during off-peak times. You can buy monthly packages that range from $15.95 for 30 MB a month to $39.95 for 130 MB a month. You also need a separate dial-up account with an Internet service provider (about $20 to $30 a month).
Why a dial-up account? Although DirecPC can broadcast data downstream to your compact dish antenna, you can't transmit data upstream to the satellite without turning your PC into a NASA tracking station. Instead, the return path is
an ordinary analog modem that dials up your local Internet provider over a standard SLIP/PPP connection. The result is a highly asymmetric network: 400 Kbps downstream and 28.8 Kbps (or whatever the speed of your modem is) upstream. The downstream path is about 14 times faster than a 28.8-Kbps modem but well short of the megabit-per-second speeds that cable modems and ADSL potentially deliver.
Actually, DirecPC has enough bandwidth for a maximum rate of 11.79 Mbps downstream. However, that bandwidth is shared by all users on a single satellite transponder. Hughes says it will add more transponders if there's enough demand. Still, the lack of a broadband return path will limit DirecPC to applications that tend to be asymmetric, such as Web browsing and software distribution. This isn't the solution for Web publishing, videoconferencing, or fast-lane telecommuting.
Another broadband wireless alternative is local multipoint distribution service (LMDS), a two-way digital broadcasting system that esche
ws satellites in favor of ground-based transceivers. LMDS needs a portion of RF spectrum, which the U.S. government plans to auction off this fall. Spectrum buyers will then have to build a new infrastructure to launch LMDS.
LMDS will use a 1-GHz-wide chunk of spectrum that starts at the extremely high frequency of 28 GHz (27 GHz in Canada). Due to this very high frequency, the user's dish antenna can be even smaller than DirecPC's (probably 9 to 12 inches in diameter). The antenna points at a neighborhood hub station that's mounted on a high roof or pole. The hub station houses a transponder that communicates with the network's central office. The central office is analogous to a cable TV system's head-end; it handles all routing and switching and also bridges to the Internet. Although LMDS is not a fully switched network, it can establish virtual point-to-point circuits.
Like most other broadband solutions, LMDS is asymmetric. It divides the 1 GHz of airspace into an 850-MHz downstream path and
a 150-MHz upstream path. QPSK (Quadrature phase shift keying) modulation yields 1.3 Gbps downstream and 240 Mbps upstream. That's aggregate bandwidth shared by all users on a hub. Depending on the network design, LMDS will assign multiple users to separate channels that are 20 or 40 MHz wide. Each channel can deliver 32 or 64 Mbps of raw bandwidth, which drops to 25 or 50 Mbps after error correction and overhead.
Users need a special LMDS modem that attaches to an Ethernet port, which limits the maximum bandwidth to 10 Mbps. As with cable modems, actual throughput depends on the amount of traffic on the same channel and the computer's I/O efficiency.
Proponents hope to deploy large-scale LMDS networks for about $1000 per user. That includes the cost of the modems and hub stations but not the RF spectrum. Assuming that LMDS networks come close to that price target, users might pay about the same access fees as they would for cable modems. But LMDS companies may be able to build their networks faste
r than cablecos can upgrade their infrastructures for broadband service.
Eventually, one broadband technology, whether it's wired or wireless, will probably emerge as the dominant solution. But it'll be many years before any broadband network achieves universal coverage. In the meantime, there'll be room for alternative solutions that can reach users who aren't covered by other networks.
Package Deals
The pipes -- or airwaves -- we dedicate to moving video from point to point are just one piece of the video-network puzzle. The protocols and standards that package and transport video files are equally important and, in some cases, similarly a work in progress.
ATM integrates video, data, and voice. ATM packages cells at fixed lengths to achieve fast switching. Cell size is 53 bytes, including a 5-byte header containing code for error control, address information, and priority control. The other 48 bytes carry the payload. ATM is connection-oriented, so, like a telephone ca
ll, an ATM transmission first registers with all switches along the way. Each cell is then guided to the next node by each switch. The data stream itself need not be concerned with routing, a degree of transparency that makes ATM flexible. Thus, computers, TVs, phones, and fax machines may eventually all be equipped with ATM switches. The costs of an ATM connection will depend on the amount of data that's being transferred, independent of the distance.
ATM speeds are scalable from 25 Mbps to more than 1 Gbps. In addition, ATM switches can transparently buffer and thus adapt data rates between slow and fast devices. Because ATM can guarantee a minimum bandwidth for a connection, it is ideally suited for point-to-point video.
The main drawback to ATM so far is its slow rollout. But that is changing: Finland and Germany are among the first countries that have commercially available switched ATM networks in place. (For more information about ATM's evolution, see "Is ATM Ready to Catch Fire?," August B
YTE.)
In the meantime, other protocols exist for video applications. Distributed Queue Dual Bus (DQDB) works at speeds of 34 and 140 Mbps, duplex mode. DQDB uses two separate directional buses to which all nodes are attached. The bus's head-end generates empty cells that pass other nodes and acquire information payload. This payload data then travels onto the recipient node, as if it's in a container.
DQDB is connectionless: Each cell contains the necessary routing information, and end points communicate with each other at different speeds. A cell's size is 53 bytes, 5 of which are address information. This is the same size and format as ATM cells, which makes the schemes compatible. The double-bus structure adds redundancy, which adds to DQDB's reliability: Even if one bus line fails completely, the system can reconfigure itself in a matter of seconds.
Broadband ISDN (BISDN) is a set of services that define how to transport video and other types of data at speeds starting at about 150 Mbps
. The transport backbone for BISDN will be an optical time-division-multiplexed network. In the U.S., BISDN will be based on synchronous optical network (SONET) technology; in Europe, on synchronous digital hierarchy (SDH) technology. Because it's fiber-based, BISDN provides higher data rates and lower error rates than the current DSL-based ISDN. It offers both synchronous and asynchronous transfers. ATM will be the underlying transport technology within the BISDN protocol stacks.
Wait for the Sequel
What will be the components of tomorrow's switched video network? In a perfect world, we'll send video over high-bandwidth fiber using ATM to transport real-time images smoothly and efficiently. But to reach that happy ending, we have to wade through a messy world of copper wiring, coaxial cables, and incomplete solutions.
Meanwhile, pioneering switched video networks will convert phone lines into video corridors or tap the potential of CATV. Wherever the two are not availabl
e and fiber is too expensive, satellite solutions, complemented with wireless systems, may shed their science-fiction image and become more commonplace data transport mechanisms.
Name
Data Rate
HDSL (high-data-rate digital 1.544 Mbps (requires two twisted-pair lines)
subscriber line) 2.048 Mbps (requires three twisted-pair lines)
SDSL (single-line digital 1.544 Mpbs (one line)
subscriber line) 2.048 Mbps (one line)
ADSL (asymmetric digital 1.5 to 9 Mbps downstream
subscriber line) 16 to 640 Kbps upstream
VDSL (very-high-data-rate 13 to 52 Mbps downstream
digital subscriber line) 1.5 to 2.3 Mbps upstream
illustration_link (25 Kb
ytes)
Motorola's CableComm provides security by encrypting cabled messages with private keys unique to each modem.
illustration_link (26 Kbytes)
Two ways to broadcast: DirecPC zaps data fro a geosynchronous satellite to a dish attached to your PC; LMDS uses ground-based transceivers and avoids analog modems.
illustration_link (9 Kbytes)
Until phone lines go totally digital, cable has the speed advantage.
Udo Flohr is a BYTE contributing editor who lives in Hannover, Germany. You can reach him at
flohr@dfn.de
.
Global Video Village
