@Device[dover]
@Style[Font=timesroman11]
@Make[Article]

@Section[Introduction]
Computers are increasingly common in the home and small businesses; many
applications require access to external computer systems and services.
Consequently, the communication capacity of the home or business must also
increase.  Larger organizations can afford to install local communications
with LANs and then connect these LANs to the rest of the world via
telephone, microwave, optic and satellite links.  For a small business or
home, with only a few computers, high speed metropolitan area communications
are out of reach.  The user is forced to use a phone line at 4800 baud or
less.  Compounding this problem is the new generation of personal computers
arriving on market.  These machines have faster processors, more use of
graphics and bitmap displays, and larger memory sizes.  These increases in
data manipulation will increase the need for high speed data communication.

The telephone system is used for communication because it is already
installed.  Fiber optic cable may eventually replace copper twisted pair
phone lines, bringing more communication bandwidth to the home or office;
but fiber is currently expensive and widespread use is probably five to ten
years away for large business customers and twenty years away for the home.
But already a high bandwidth communication link runs to many homes in the
form of a coaxial cable for the local CATV system.  Unused bandwidth on a
CATV system could bring high speed (1-10 MBPS), low cost (less than $1000
per interface), packet switched communication facilities into the home or
office.

This paper proposes a preliminary design for a packet switched communication
system installable on a CATV network to provide digital data communication
over a metropolitan area.

@Section[System Design]
The network design takes advantage of existing commercial CATV systems, so
understanding CATV systems helps in understanding the network design.

@Subsection[Cable Television]
A cable TV network is a coaxial cable transmission system that is
frequency-split into 6 MHz. bands.  In the most common, single-cable design
used for residential areas, most of these bands are @i[downstream] channels
occupying the spectrum from 54 Mhz. to 300 MHz., 450 MHz., or higher
depending on the particular system and converter technology employed.
Downstream channels carry signals, typically entertainment video signals,
from the cable TV headend to the subscriber. The @i[upstream] channels
occupy the 5.75 - 29.75 MHz. range and carry signals from the subscriber to
the cable headend.  Frequencies from 29.75 MHz. to 54 MHz.  form a @i[guard
band] and no signals are transmitted at these frequencies.

Sources of noise on a CATV system include ingress of off-the-air signals
from short-wave radio stations, insertion noise from equipment connected to
the cable, white noise from the amplifiers on the system, loose connections
at the drop, signal reflections because of impedance mismatch and beat
products from rectification by corroded connectors.@Cite[SIRAZI]  The
downstream channels of a cable system have low noise because the system is
primarily designed for distributing video signals from the headend to the
subscriber with good picture quality.  The noise upstream is high because a
CATV system has a converging tree topology; the noise arriving at the
headend is the amplified sum of all the noise over the entire system.  In
addition, the frequencies used in the upstream direction are routinely used
for short-wave communication and the CATV distribution plant make a very
effective antenna at these frequencies.  Network design must take this high
level of noise into account.

@Subsection[Design Goals]
The network consists of stations that transmit data to the headend on an
upstream channel.  The headend broadcasts the received data from the
upstream channel to all stations via the downstream channel.  A design goal
for data communication over a cable TV network is to change the existing
cable system as little as possible.  Ideally, the only additions would be a
network controller at the headend and modems at the subscriber end.  A
related goal is complete compatibility with the pre-existing cable
equipment, television sets and cable channel allocation.  In addition,
expensive and complex equipment for data communication should be placed at
the headend in exchange for cheaper and simpler equipment at the subscriber
nodes.  To ensure a high upstream bitrate, the upstream link to headend
should be via the same cable as the downstream link; other types of upstream
channels such as telephone lines are undesirable. The network should support
multipoint communication instead of point-to-point communication, because
any station may communicate with any other station without prior
arrangement.  A system must exist to allocate bandwidth on the network to
allow communication and to support addressing of the multiple points on the
net.

In summary, the problem of designing a CATV based packet switching network
can be broken into four parts:
@Begin[Enumerate]
Transmission method from headend to subscriber

Transmission method from subscriber to headend

Subscriber control logic

Access scheme to determine bandwidth utilization

@End[Enumerate]

Two of these problems also appear in other data communication systems and
only slight redesign is necessary.  The two interesting problems are those
of transmission from subscriber to headend and the access scheme used for
bandwidth allocation.

@Subsection[Philosophy of Design]
Both the headend and the subscriber nodes should be modular in design.  In
the headend, the control logic can be attached to any transmit and receive
modems desirable.  The upstream and downstream environments are different,
thus different modulation techniques may be appropriate.  Modulation and
signalling techniques can be experimented with by switching modems.

The subscriber end is also modular, thereby lowering the cost per node of
future changes by reusing most of the interface components.  The four
modules for the subscriber node are the receiving modem, the transmitting
modem, the subscriber logic, and the bus interface.

Modularity allows the system to be put together in a building-block manner.
The bus interface is easily changed as new computer systems are added and
the modems can be changed during the initial testing to determine which
transmission techniques work best.

@Section[Tested Network Components]
Transmission from the headend to the subscriber and the subscriber logic are
parts of the network design already tested in other data communication
systems.  Only slight redesign is necessary to use these ideas for the
two-way data communications network.

@Subsection[Transmission from Headend to the Subscriber]
The headend transmitter and the subscriber receiver use the same modulation
system.  Transmission from the headend to the subscriber is the easier of
the two transmission paths to engineer because CATV systems are designed for
video signal distribution from the headend to the subscriber with a good
signal-to-noise ratio.  Since the downstream data transmission channel will
be a 6MHz. television channel, surrounded by video channels on a system
designed for video transmission, it makes sense to have the downstream data
method be compatible with video transmission as much as possible.  CATV
amplifiers expect certain signal levels, amplifier triple-beat
characteristics are optimized for video signals,  and television sets have
filters for eliminating adjacent channel interference that expect carriers
to be at certain frequencies.  Research at the Sony Corporation has already
demonstrated a data communication method compatible with CATV.

The Sony system has the right characteristics needed for the packet
switching network:
@Begin[Enumerate]
High speed

Low error rate

Easily introduced into existing CATV networks

TV signals not adversely influenced

Low receiver cost
@End[Enumerate]

The Sony system transmits data at 7.4 MBPS using one channel of 6MHz.
Transmission tests on North American and Japanese CATV networks have proven
the system to be workable.  The system uses VSB (Vestigial Side Band)
transmission, the same as that of a television signal, modulated with a
bandwidth-restricted, two-level baseband signal.  The subscriber receiver is
inexpensive because it uses mass-produced TV components.  The system has a
raw error rate of about 10@+[-7] even in the worst carrier-to-noise ratios
seen on CATV networks.@Cite[SONY]

A second possibility for the downstream channel is spread-spectrum
modulation.  In some areas, certain channels on a CATV network cannot be
used because radiation of signals from the cable system might interfere with
aircraft navigation.@Cite[ESTRIN]  Spread-spectrum signals can make the
energy across a channel uniformly low so that radiation from the cable might
be acceptable.  Spread-spectrum receivers are more expensive than VSB
receivers, but in an area where cable channels are at a premium, or in the
future when demand rises, it may be worth the extra receiver cost to reclaim
the unusable channels.


@Subsection[Subscriber Logic]
The subscriber logic should be similar to the logic used in local area
network interfaces, such as the Proteon proNET.  The network interface
should be simple to operate, intelligent enough to do several operations at
once, and have multiple transmit and receive buffers to avoid missing
packets.  The maximum packet size should be about one or two kilobytes to
keep transmission delay small.

@Subsection[Bus Interface]
The bus interface will initially be for an IBM PC, since the PC is a popular
computer for home and business use.  The IBM PC bus is easy to interface to
and the bus interface will be connected to the subscriber logic via a
standardized interface.

@Section[New Techniques]
Two new pieces of the design of the CATV network are the upstream
transmission technique and the access scheme for bandwidth allocation.  Most
of the research will be focused in these two areas.

@Subsection[Upstream Transmission]
The subscriber transmitter and the headend receiver have the same modulation
method.  Upstream transmission is harder than downstream transmission
because of the high noise level that accumulates on the upstream channels.
Low signal-to-noise ratios on the upstream channels suggest three modulation
techniques: coherent PSK, phase-continuous FSK and spread spectrum.
Coherent PSK and phase-continuous PSK have a low energy-to-noise ratio for a
given error rate@Cite[OETTING], suiting them for high noise channels.  But,
the type of noise found on the upstream channel of a CATV system would
likely cause these two simple modulation techniques to perform poorly.  The
noise on a CATV system is probably closer to jamming than to gaussian noise
distribution because of strong interference by short wave radio stations
entering the upstream channel by ingress.

Spread-spectrum could be the modulation technique needed to overcome noise
on the upstream channel.  The military uses spread-spectrum techniques to
produce systems that work even in the presence of intentional signal
jamming.

Upstream spread-spectrum can be implemented in two ways.  The first is to
allocate a 6 MHz. slot in the upstream band exclusively for the network
data.  An advantage of this method is that it fits with the modularity of
the CATV system - to add more networks, allocate more upstream channels.

The second method is to use spread-spectrum across the entire upstream band
(24 MHz.) at a low enough signal level so that it can sit "underneath"
conventional services occupying the same bandwidth.  The advantage is that
the network upstream channel can exist independently of other services using
the upstream channels.  The disadvantage is the possible effect on
signal-to-noise ratio on the upstream channels by the spread-spectrum
modulation and vice-versa.

Digital regenerators could replace some of the upstream linear amplifiers on
the appropriate channels.  Not all of the linear amplifiers would be
replaced - only enough to reduce the noise to an acceptable level.  The
advantages are that noise is reduced and a simpler modulation system can be
used.  The disadvantage is the need to change some upstream amplifiers,
making the system more complex and expensive.

@Subsection[Access Scheme]
The access scheme for a MAN is a very important and difficult problem.
Because the network serves the home or small business, the overall packet
distribution will be bursty.  Bursty traffic is difficult to schedule on a
network with a large propagation delay and static allocation of time or
frequency slots would make network utilization unacceptably low

The downstream channel will probably run at a higher data rate than the
upstream channel because of the difference in signal-to-noise ratio between
them.  Therefore, the downstream channel can be broken into two parts:
downstream data repeated from the upstream channel and access control
information.  The headend then gives permission to send on the access
control part of the downstream channel.

The upstream channel could be split into two channels, a data communication
channel and an access channel for queueing.  The separate access channel
allows stations to enqueue while data is transmitted by some other station.
When the transmitting station is finished, the headend notifies the next
station in the queue to start transmission.  The downstream channel could be
split into a data communication channel and a enqueue acknowledge channel.
The headend sends acknowledgments as the requests for transmission are
received so the next station can begin data transmission without delay.

The stations must somehow make the desire to transmit known.  CSMA/CD may
work efficiently on a small network, but as size and speed increase,
efficiency decreases.  On a geographically large system, the efficiency
would degrade to that of an Aloha system.  RF collision detection is very
hard to do; because the system is broadband, detection of conflicting
signals on the upstream channel is uncertain, so the only way to know that
the network is in use is to watch the downstream channel.  Collision
detection on the downstream channel only increases the delay of collision
detection.  Uniform station distribution implies that the number of stations
at a distance r from the headend is proportional to r@+[2] because a CATV
network covers area and is not linear as is an Ethernet.  The average
station distance is R@+[2]/2, where R is the radius of the system.  The
large average distance between colliding stations also slows down collision
detection.@Cite[LNNE]  Inefficient collision detection increases the loss of
bandwidth because of collisions.

Polling is an access scheme frequently used on cable systems.  The headend
can poll every station to allow those with data to send to transmit.  But,
since most stations would not be ready to send traffic, polling is slow and
thus not suited to the interactive environment that this cable network would
support.  Polling a large network (100,000 customers) can take six seconds,
causing low throughput.@Cite[ESTRIN]  Adaptive polling polls active stations
more often than inactive ones, so the polling is more efficient, but still
not fast enough.

Multiple polling systems poll all stations at once by giving each station an
orthogonal channel for reply to the headend.  Frequency division or code
division (for spread-spectrum)  orthogonal channels allow stations to reply
within a short amount of time.  The headend determines the stations with
traffic and assigns a transmission order.  The headend polls again when the
queue is empty.

For a network with R meter spokes (2R meters in diameter) and a propagation
velocity of V meters/second, the one-way propagation time is R/V seconds
(amplifier delay is negligible).@Cite[LNNE]  A worst case polling cycle
includes a request out (R/V seconds), delay time at modem (D@-[m]), arrival
time of the last possible carrier (R/V seconds), carrier frequency
determination (F), permission to send packet out (R/V seconds), delay at
modem (D@-[m]), and time until the first message arrives (R/V seconds).  The
minimum polling time is 4R/V + 2D@-[m] + F seconds.  If there are N stations
on the network, each assigned a unique poll-response frequency in a B hertz
bandwidth slot, then each has a slot bandwidth of B/N.  A rough
approximation of bitrate is B/3 bits/second for a bandwidth B, so each of
the N slots would have a bitrate B/3N bits per second.  Only one bit per
station (request to send) is needed, so the time per station is 3N/B seconds
per bit.  Since all stations transmit carriers simultaneously, the total
time to receive the one bit for each of the N stations is 3N/B seconds, so F
= 3N/B.  The maximum polling time is (4R/V + 2D@-[m] + 3N/B).  If R = 30km,
V = .87c, D@-[m] = 1@g[m]s., B = 6 MHz., and N =1024, then the maximum
polling time is 974 microseconds. Compare this with N(2R/V + D@-[m] + 3/B)
for conventional polling, which is 28.7 milliseconds.

@Section[Other Considerations]
@Subsection[Error Correction Codes]
Packet error checking should be done end-to-end, but a lower-level check
might be used to determine the location of defective cable, amplifiers or
sources of jamming.  If the raw bit error rate is low enough, then a parity
bit should work; but a CRC will probably be necessary.  CRC calculation on a
chip, as on Ethernet cards, is a simple way to add this.

@Subsection[Security and Availability]
For data security, users can implement end-to-end encryption.  Using
end-to-end encryption allows the user to use any type of data encryption
desired (including none at all) and thus set his own level of security.
Encryption will not stop traffic analysis, but for the type of traffic this
network will carry, traffic analysis should be of minimal concern.

Interference is a more difficult issue to deal with.  Interference on the
downstream channel would cause local problems for other users on the same
cable segment.  If the upstream channel is still open, the interfaces being
interfered with could complain to the headend.  Knowing the network topology
and the station numbers of the stations complaining should make isolation of
the interfering station fast.

Interference with the upstream channel is more difficult to handle.  One way
to find the source of the interference would be to use a binary search with
intelligent bridger amplifiers.  The headend can remotely switch the
amplifiers on and off to find and isolate the offending unit, leaving most
of the system intact.@Cite[ESTRIN]

@Section[Test Bed]
Newton, Massachusetts has a cable system that is an ideal test site - a 400
MHz. hub-shaped system with upstream amplifiers.  The Newton cable system is
near MIT and the city of Newton has many potential test subjects - MIT
faculty member that already have IBM PCs.

Bell Communications Research has a CATV system built inside a lab for
experimental use.  This system is smaller, has less noise and is not for
commercial use.  Running preliminary tests at Bell would determine whether
the network equipment works and insure that there is no interference with
other channels on the cable before testing on an operational commercial
system.

Testing on campus can be done on the MIT cable system because the system is
small and near by for easy access.  Probably most of the system testing will
be done on the MIT cable system for convenience.

@Section[Conclusion]
The need for inexpensive, high speed data communication within a community
is growing.  Because the physical plant of CATV is already installed, a MAN
built on a CATV system allows a communication system to be installed quickly
and economically.

Initially, downstream transmissions will use the Sony VSB system and the
upstream transmissions will use spread-spectrum.  The subscriber control
logic will be similar to the proNET interfaces, and the access scheme will
be multiple polling.  During system testing, other alternative systems will
also be tried.

The CATV MAN project explores three important areas: transmission over a
broadband cable, spread-spectrum communication and access schemes for MANs.
All these areas are important to data communication in the future.

@Subheading[Acknowledgments]
My thanks to Jerry Saltzer and Deborah Estrin for their ideas and comments
on this paper.  Their groundwork on CATV data networks really helped to get
this project going.

@Heading[References]
@Use[Bibliography="CATV.Bib"]
@Bibliography
