T1 Clear Channel CSU/DSU
        A. Expository on T1 Service

                        T1 for Computer Networks

The Bell System's Digital Signal Hierachy
        To improve signal/noise ratio on multi-line phone trunks, Bell
began converting some frequency division multiplexing (FDM) lines to time
division multiplexing (TDM) back in the 1960's.

        The digitization technique chosen was pulse code modulation (PCM),
taking 8000 samples/second of the analog waveform and quantizing it to 8 bit
precision with an analog to digital (A/D) converter.  When the bits are
serially shifted out, the signal source is called a "DS0" by the phone company.

        Including several DS0 channels in one TDM bit stream requires the
addition of framing bits, so the individual channels can be identified
on recovery.  A "DS1" is composed of 24 byte-wise interleaved 8-bit samples
(from 24 different DS0's) and one framing bit.  The total bit rate is:

total rate = 8000 samples/sec * [(8 bits/sample * 24 samples) + 1 frame bit]

           = 1.544 Mbps

   a1 a2 a3 a4 a5 a6 a7 a8 b1 b2 b3 b4 b5 b6 b7 b8 ... x6 x7 x8 f0
  |                       |                       |            |  |
    sample from 1st DS0      sample from 2nd DS0    .....     frame
                           Sample Bit Frame

        Four DS1's can be combined into a DS2; 7 DS2's compose a DS3.  There
are also DS4's and DS5's, used for long-distance trunks often running on
optical fiber.

T1 Framing Patterns
        To synchronize with the bit stream, the receiver picks a random
bit, and then examines every 193rd bit for the presence of the special
framing pattern.  If too many received bits differ from the pattern, it
delays one bit and begins the search again.  At 1.544 Mbps, it does not
take long to synchronize.

        The Bell company first used a "D1" framing pattern when T1 began.
The pattern was so simple that putting a 1 KHz tone (the standard Bell test
frequency) on one of the DS0 channels would cause the circuitry to synchronize
on the wrong bit.  They changed to 1004 Hz test tones, and later changed to D2
framing patterns.

        D4 framing is the current most common framing pattern, but recent
advances in signal processing make it slightly more redundant than necessary.
"ESF" refers to a framing pattern in which three of every four framing bits
(8000 frame bits/s * 0.75 = 6 kbps) are used for control and error-checking

Modulating the 1.544 Mbps bit stream
        While framing patterns facilitate timing recovery at the receiver,
special encoding techniques must be used for operation with T1 line
"repeaters", the T1 signal amplifiers and conditioners located about
every 6000 feet in Bell intracity wiring.

        Zeros or spaces in the bit stream correspond to periods of zero
volts, while ones or marks are converted to 2.7 to 3.3 volts.  Using
alternate mark inversion (AMI), all adjacent marks are of opposite polarity.
When adjacent marks have the same polarity, a "bipolar violation" (BPV)
has occurred.  Some telephone repeaters can tolerate these.

        Repeaters also require a "minimum ones density", a pulse at least
every 8 bits, in order to recover timing information.  For voice channels,
forcing one bit in each 8 bit sample to a mark does not seriously degrade
quality.  Thus, most voice channels actually occupy 56 kbps.  This is also
why digital dataphone services (DDS) offered by the phone company comes in
56 Kbps chunks.

1 DS0 = ((8 bits - 1 bit) * 8000 samples/s ) = 56,000 "bits per sec" = 56 Kbps

Connecting to a T1 line
        As private branch exchanges (PBX) and computer networking became
more popular, phone companies began offering end-to-end digital lines.
Equipment connected to these lines which ensures proper signal levels,
protects against surges, and cleans up BPV's are called "channel service
units" (CSU's).  Together with PBX's, the equipment is sometimes called
"customer premises equipment" (CPE).

        The specific CSU's intended for T1 lines are referred to as "T1 CSU's",
but may be abbreviated as just "CSU".  The actual wires bringing the T1
service onto the customer premises may provide a DC current source (60 or
140 mA) for powering the CSU.  Such a T1 line is called "wet"; lines not
providing this "span power" are called "dry".  Though most CSU's allow for
use with or without span power, be careful when touching T1 lines, or
servicing CSU equipment even when powered down; the constant current source
may provide several hundred DC volts without a load.

        The reason for powering CSU's is to ensure a "keep alive" signal
of all marks sent on the T1 line, even if main electrical power at the
customer site is removed.  Without ones density, a repeater can oscillate,
affecting communications on adjacent T1 lines.

        Formally, the phone company is to be notified whenever a T1 CSU
is connected or disconnected, but recent advances in T1 repeaters make
this "not always necessary" (I didn't say it.)  Official T1 line specifications
are available in Bell publication 62411.

        The repeater nearest to the CSU is guaranteed to be within 3000
feet.  The CSU provides enough drive to operate that far, and often includes
a switchable attenuator known as "line build-out" (LBO) in case it is much
closer.  The optimum setting of this switch should be provided to the
customer by the phone company.

Using T1's to carry computer data
        Clearly, the 56 kbps and T1 data rates and formats were not chosen
with computer data in mind.  But if we don't violate the specifications for
our applications, the phone company does not care about the type of information
source we use.

        Interfacing computers to T1 lines requires a special formatter
to clock serial serial data from a computer (on an RS-422 or V.35 interface,
for example) at one rate, insert framing patterns and ones density bits as
needed, and then shift out the data at another (possibly different) rate.
Also needed is a CSU to properly interface to the line.

        Of course, the reverse operations need be done at the receiving
end.  An integrated piece of electronics to perform both these functions
is called a "clear channel CSU".

        If we are using T1 modems "in house", over our own wires in a
building less than 6000 feet apart, we can run them at full 1.544 Mbps;
no framing bits are needed.

        It's possible for the phone company to provide a T1 line between
two locations in the same city.  If told so by the phone company, then only
the repeater requirements need be met; the framing bits are irrelevent.
However, in most cases the framing bits are included by the equipment,
anyway.  For intercity T1 lines, framing bits must almost always be added.

        Adding the framing bits is straight forward.  The formatter, or
the clear channel CSU, inserts them into the bit stream.  Note that right
away, available data bandwidth is reduced to 1.536 Mbps:

effective rate = 8000 samples/sec * [(8 bits/sample * 24 samples)]

               = 1.536 Mbps

        Meeting repeater requirements of one's density are more difficult,
and several approaches are available.  They include:
1) B8ZS
        Standing for "bit 8 zero substitution", this technique transmits
        data at 1.536 Mbps by inserting the pattern 00011011, with BPV's
        in the fourth and seventh positions, wherever ones density
        requirements are not met by the unmodified data.  It requires
        the CSU to not remove BPV's, and works only where the phone
        company equipment can tolerate them.
2) Clever encoding
        If we know enough about the format or information content of our
        bit stream, we could perform some clever conversion to suppress
        strings of eight consecutive zeros.  Such techniques rely on the
        actual information rate being less than 1.536 Mbps, even though
        that is the final clocking rate of bits onto the line.
        Three possible specific applications include:
        a) Run Length Encoding
                By looking for all consecutive strings of eight or more
                zeros, and encoding them in a special way within the data
                stream, ones density can be met.  Such an approach is
                often used to encode image data (often with long stretches
                of zeros or ones), and is very similar to...
        b) ZBTSI
                "Zero Byte Time Slot Insertion" is a proprietary technique
                used by Verilink in their 551VCC/U clear channel CSU, where
                long strings of zeros are encoded, and the decoding information
                is inserted within the framing pattern.  (Remember how ESF
                makes available 6 Kbps for special functions).  It offers
                the most generalized scheme of increasing throughput, at the
                correspondingly highest price.
        c) HDLC/SDLC
                If we understand the protocol enough to know where ones
                must be, we can scramble the bits and spread them out
                evenly, satisfying ones density.  The Digital Link DL551
                offers this approach, and eventually Proteon gateways are
                to use HDLC or SDLC.
3) Ones Insertion
        Just as the phone company sacrifices one bit in eight for each
        DS0, so can we force every eighth bit to a mark, and reduce computer
        link bandwidth to 1.344 Mbps.  Whether such a reduction is tolerable
        depends on the specific application being considered.

      effective rate = 56 Kbps * 24 = DS0 rate * 24 = 1.344 Mbps

T1 Testing
        The phone company often guarantees service performance in terms of
"percentage error free seconds per month", though actually measuring that
quantity is difficult.  In order of increasing thoroughness, some techniques
for testing include:

1) Loop Up/Loop Down
        The most primitive indication of line operation is to attempt to
        "loop up" the remote CSU, by sending the standard remote analog
        loopback pattern of "10000".  The remote end should return the signal
        within five seconds of application.  Looping down with "100" pattern
        may take slightly less time.  This test, often built into CSU's,
        takes the line out of service, but is usually only done to determine
        whether complete link outages are due to the line or computer.
2) Passive monitoring
        By using the MON jacks available on some CSU's, you can watch the
        incoming bit stream and check for proper D4 framing bits.  If any
        of these are in error, you can assume a line error occurred, and
        multiply the frequency of framing bit errors by 193 to estimate
        total line errors.  The FIREBERD bit error rate tester does this.
        It can test continuously, with no interruption to service.
3) ESF
        The extended superframe officially divides the 6 Kbps bandwidth
        scavenged from the D4 framing pattern into a supervisory channel
        of 4 kbps, (for interogating remote equipment, for example), and
        2 kbps for a cyclic redundancy check (CRC).  The standard specifies
        that this be computed using all bits, including data, so it has
        a better statistical chance of catching line errors than examination
        of framing bits only.  Some companies offer conversion equipment
        which takes D4 framed signals and adds ESF functions to them.  It
        provides continuous testing while the line is in service.
4) Bit Error Rate Tester
        For suspected line quality problems, a bit error rate tester (BERT)
        is usually put on one end of the line, with the other end looped
        back.  Whether it provides a useful measure may depend on: whether
        gapped clocks are used (as with the DL551V), and whether loopback is
        analog or digital.  You should ask the manufacturer under what
        conditions this technique is appropriate with a given CSU.  It
        requires taking down the link, and is therefore usually only done
        when quality is so poor as to significantly impede link utilization.

Appendix 1: What's a Gapped Clock?
        A clear channel CSU, or CSU-formatter combination, usually provides
the serial data clock to the computer equipment.  Depending on the type of
encoding, the clock may be 1.544, 1.536, or 1.344 MHz.  However, that's
given in clock transition rate; they are not necessarily evenly spaced.
For example, the Digital Link DL551 clear channel CSU provides a 1.344 MHz
clock like:

          _   _   _       _   _   _   _   _   _   _       _
       |_| |_| |_| |_____| |_| |_| |_| |_| |_| |_| |_____| |_|
                      ^                               ^
                   missing                         missing
                 transition                      transition

        If the transitions had been included, the total rate would be
1.544 MHz.  But everywhere the CSU inserted a ones density or framing bit,
it simply gapped the clock to the computer.  This clock is incompatible with
some BERTs.

Appendix 2: Digital vs. Analog Loopback
        Remote loopback of CSU's is an analog loopback, as it basically
sends the same incoming voltage back out the line.  However, you can also
provide digital loopback, either by placing a loopback connector on the
digital signal interface to the computer, or sometimes by configuring the
computer interface a certain way.

        Either digital approach may indicate BERT errors even with a
good line.  The reason is that each modem may independently determine
the rate at which it sends data out on the T1 line.  For example:

  ----------             -------              -------             ----------
  |        |-----TxD---->|     |\/\/\/\/\/\/\/|     |-----RxD---->|        |
  |Computer|<----TxC-----|     |/\/\/\/\/\/\/\|     |-----RxC---->|Computer|
  |   #1   |             | CSU |              | CSU |             |   #2   |
  |        |<----RxD-----|     |\/\/\/\/\/\/\/|     |<----TxD-----|        |
  |        |<----RxC-----|     |/\/\/\/\/\/\/\|     |-----TxC---->|        |
  ----------             -------              -------             ----------

        TxD refers to transmitted data, and TxC is the clock for this data;
similarly for received data.  Note how the CSU provides each clock to its
associated computer.

        As is usally the case, each CSU determines the rate at which it
transmits data from an internal osciallator.  It must be 1.544 MHz, +/- 75 Hz.
The rate at which it clocks in the received data is of course equal to the
rate of the other CSU's transmission.  Thus, if the clocks are the slightest
bit off (and they usually are), digital loopback produces a skewed return
signal, producing bit errors at a rate related to the beat frequency of
the two oscillators.

        Sometimes CSU's can be configured to adjust their transmit clock to
match the rate of the receiver clock, or even to lock transmission rate to
an external clock.  This may make remote digital loopback work for a BERT,
but has the disadvantage of requiring different hardware configurations for
each end of the link.

Appendix 3: Foreign Standards
        In case it comes up in conversation, European phone networks space
repeaters somewhat closer than every 6000 feet, allowing them to use a
2.048 Mbps stream for their equivalent "T1" trunks.  Some vendors produce
multiplexing equipment capable of connecting countries of different

.Date: Tuesday, 16 February 1988 01:28:34 ESTFrom: Eugene.Hastings@morgul.psc.edu

Thanks to Marty Schulman and Eugene Hastings for the excellent introduction
to T1 signaling technology and CSU equipment.  Their writeup removes much
of the "black magic" surrounding T1.  At BARRNet we have built our regional
network based entirely on T1 circuits and hence have gained a wealth of
experience with T1 equipment and common carrier providers during the past
1.5 years.  With this preface, I'd like to share some T1 knowledge and correct
a few errors in Marty's and Eugene's writeup.

The T1 Carrier standard specified in Bell Pub 62411 specifies minimum ones
density in two ways (both minimums must be met) (a) an average ones
density of not less than 12.5%, and (b) no more than 15 consecutive zeros
between one bits.  In North America, a bit is "robbed" in each DS0
subchannel every sixth frame to carry circuit signaling information (e.g.
on-hook and off-hook indications).  Thus a bit is NOT stolen from every
byte in a DS0 signal, but rather only from every 6th byte of any particular
channel.  However, since data terminal equipment (DTE) has no easy way of
determining which byte will have have a bit robbed from it, it is simplier
just to have a 56 kbps clock (7/8 * 64 kbps) for all DS0 data circuits.

If a T1 formatted bit stream does not represent 24 DS0 channels, then there
is no need to do anything funny with one of the bits from each channel every
sixth frame.  In other words, it is NOT necessary to force every 8th bit of
user data to be a one.  This just needlessly decreases usable bandwidth
(more on this below).  The reason it is done so often in clear channel CSUs
is because it is easy to implement and clearly meets (actually far exceeds)
the ones density requirements listed above.  More sophiscated CSUs
(such as the Verilink 551VCC) do not treat the user's bit stream as 24 7-bit
bytes, but rather operate on a larger group of bytes in a more intellegent
manner (hence their higher cost).

The ones density requirements, as Marty says, is to keep the T1 line
repeaters operating properly.  Note that the 62411 standard was developed
when analog repeaters were the only ones available.  Today's digital
repeaters can operate on a much lower ones density; some military spec
repeaters operate with up to 50 or 60 consecutive zeros.  Unfortunately
you have no way of knowing what kind of repeaters are in any particular T1
circuit and hence all commerical CSUs are built to the 62411 standard.

Other minor discrepancies in the writeup:

        1) the "I" in ZBTSI stands for interchange, not insertion.
           The algorithm exchanges the "time slot" occupied by a byte of
           all zeros with another non-zero byte.  The position of the zero
           byte in the data stream is indicated by a 7 bit (inherently
           non-zero) index value with the 8th bit indicating if there are
           additional zero bytes present.  Either framing bits or data
           bits (see #5 below) are used to flag the fact that the data stream
           has been encoded.

        2) ZBTSI is now a Bell standard and is not proprietary to Verilink.
           The Verilink encoding scheme is actually slightly different from
           the ZBTSI standard.

        3) ZBTSI has nothing to do with extended superframe format (ESF).
           Both ZBTSI and Verilink's proprietary clear channel technique
           work independently of ESF.

        4) CSUs should not be configured to generate their own clock, rather
           they should always recover the clock from the network.  Common
           carriers have gone to GREAT lengths to ensure synchronized
           clocking.  In the USA, there is a nominal USA-wide master clock
           generated from an atomic time source located (I think) in
           Atlanta.  Obviously the phase of this clock varies from location
           to location across the USA, but the frequency should always be
           1,544,000 Hertz EXACTLY.

        5) Because of the different encoding schemes, there are actually
           several options for getting the highest effective user bandwidth
           on a T1 channel.  The data rates that commonly come up are:

           1.544 Mbps - The total bit stream including both user data and
                        framing bits.  The standard framing bit format today
                        is D4 and includes both "T" (terminal) framing bits
                        and "S" (multiframe alignment) framing bits.

           1.536 Mbps - The total bit stream rate less the framing bits.
                        I.E. the maximum usable user bandwidth on a T1
                        channel.  It is this bit stream that is usually
                        modified to meet the ones density requirements
                        (this is because the framing bits must conform to
                        the D4 standard and hence cannot be modified,
                        although one of the Verilink 551VCC products does
                        modify the framing bits).

           1.528 Mbps - A DACS-compatable (digital access and cross-connect
                        system) clear channel bit stream.  Some telco
                        central offices contain DACS equipment which
                        strips the framing bits off from T1 bit streams, then
                        reframes the stream later on.  Since some encoding
                        methods (e.g. Verilink VCC) purposely inject errors
                        into the "T" framing bits on a T1 signal, these bit
                        streams are not compatable with DACS equipment.  To
                        make these encoding schemes compatable with DACS, an
                        8 kbps "channel" is used for the encoding control

           1.344 Mbps - (= 24 * 56 kbps).  This is the user data rate obtained
                        when using the brute force method of insuring minimum
                        ones density.  As Marty and Eugene point out, the
                        method may be so crude as to clock the user data in
                        "gapped" form, essentially stalling the DTE data
                        clock while the CSU inserts its own bits for ones
                        density and framing.  CSUs which operate in this mode
                        essentially "throw away" 192 kbps of user bandwidth
                        by robbing every 8th bit position in the user's data

Not so much a discrepancy, but something that should be pointed out is the
fact that very little of today's installed telco equipment (~1%) is capable
of working with B8ZS (bit 8 zero substitution) signal format.  I'm told by
telco personnel that within 10 years 90% of all T1 equipment will be B8ZS
compatable.  What this means to you, the user, is that it is unlikely that
any B8ZS CSUs you buy today will work today.  If you buy a clear channel
CSU that works via B8ZS encoding be sure you test it with the telco T1
circuit before you commit your dollars.  If your T1 channel is multiplexed
and demux'ed by the common carrier, B8ZS can't work until all the mux'ing
equipment is upgraded to understand receiving intentional bipolar violations
and regenerating them at the far end.

ZBTSI, on the otherhand, will work with all of today's equipment.  As VLSI
circuits are developed to implement the ZBTSI algorithm (it requires
buffering 96 bytes of data and encoding these as a unit), more manufactures
will offer ZBTSI equipment.  The only manufacturer I know of currently
offering a ZBTSI clear channel product is Verilink.

--tom ferrin

From cisco@spot.colorado.edu Tue Feb 16 21:24:12 1988
I stand corrected on a couple of points in my recent email message:

        1) AT&T's USA-wide master T1 clock is located in Hillsboro,
        Missouri, not Atlanta.  Hillsboro was chosen because it is the
        "geographic center" of the country.  (Does this mean if there
        was a giant H-bomb it would be dropped there?  Never mind...)
        There are several backup master clocks arranged in a hierarchal
        fashion in case of failures in the primary synchronization system. 
        An article about this recently appeared in Data Communications.
        I'm told it is not necessarily easy to slave CSUs to the master
        clock.  The clock is used by telco COs, but it may not be easy for
        you to get at it.

        2) The ZBTSI ANSI standard is part of T1X1 committee and will
        be balloted on shortly. Several telco's are already using the
        current ZBTSI document as a defacto standard.  The standard
        does, in fact, require ESF.  It uses 2 kbps of the 4 kbps ESF
        data channel for transmitting "Z" control bits used to flag
        encode control information.  The properitary Verilink 551VCC
        product does not require ESF.

        Other differences between ZBTSI and Verilink VCC are more
        substantial than I orignally implied.  They include: (a) 500
        microsecond delay on xmit and recv for Verilink, 500 microsecond
        delay on xmit only with ZBTSI, (b) no modification of the T1 bit
        stream if it already meets density requirements for Verilink,
        channel 96 time slot always exchanged with channel 1 time slot
        for ZBTSI, (c) no bit scrambling with Verilink, 5-bit scrambler
        added to ZBTSI data stream to minimize error multiplication.

From cisco@spot.colorado.edu Wed Feb 17 16:52:59 1988

Press your browser's BACK button to return
goto Telecommunications Tidbits page