THE BELL SYSTEM
TECHNICAL JOURNAL
volumexlv September 1966 number7
Copyright © 1966, American Telephone and Telegraph Company
An Experimental 224 Mb/s Digital
Repeatered Line
By I. DORROS, J. M. S1PRESS, and F. D. WALDHAUER
(Manuscript received May 12, 1966)
An experimental digital repeatered line has been developed which trans-
mits information at a rate of 224 Mb/s as part of an experimental high-
speed digital transmission system. The PCM terminals and time division
multiplex portions were described by J. S. Mayo and others in the No-
vember 1965 issue of the Bell System Technical Journal. The repeatered
line is described in this paper. The performance of this line is shown to be
suitable for coast-to-coast operation.
The line utilizes 0.270-inch copper coaxial transmission lines and re-
generative repeaters at one-mile intervals. Ten repeaters have been operated
in tandem to form ten miles of repeatered line. Each repeater uses 25 tran-
sistors, most of them a new germanium, design with a cutoff frequency, f t ,
of /f. GHz. Esaki diodes provide the decision thresholds for the regeneration.
Power to the repeaters is supplied by dc over the center coaxial conductor.
The pulse transmission code is paired selected ternary (PST).
I. INTRODUCTION
Digital transmission of information is becoming increasingly attrac-
tive in the telephone plant because it competes favorably both in per-
formance and in cost for telephone service and it allows all kinds of
other services such as data and television to share the transmission
media with virtually no interaction among the various signals. The
digital transmission process 1 is based on (i) pulse code modulation of the
analog signals to be transmitted, (it) time interleaving of the resultant
pulse streams to form a composite pulse stream, and (in) regeneration
993
994 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 19CG
in the repeatered line to nullify the effects of noise and distortion en-
countered in transmission.
The Tl carrier system, 2 - 3 extremely successful since its introduction
in the Bell System in 1962, provided the first wide use of the digital
transmission concept. In Tl, 24 exchange area voice signals are pulse
code modulated, time division multiplexed, and transmitted over cable
pairs at a rate of 1.544 megabauds. Tl repeatered lines are now also
in use for short-haul data services, 4 since these lines are virtually insensi-
tive to the source of the pulse streams transmitted.
There has also been interest in making use of the features of digital
transmission for long-haul transmission. Recently, at Bell Telephone
Laboratories, a 224 megabits/second (Mb/s) experimental system was
constructed to demonstrate the feasibility of a coast-to-coast system.
The PCM terminals and time division multiplex portions of this system
were described by J. S. Mayo and others in the November 1965 Bell
System Technical Journal. 5,Bi7 It was there indicated that the 224
Mb/s information rate would serve either four coded 600 channel mas-
tergroups for a total of 2400 voice channels, 144 Tl signals representing
24 channels each for a total of 3456 voice channels, two coded network
TV signals, or some combination of these and other digital signals. In
this paper, we describe the experimental repeatered line. The significant
accomplishment is the realization of repeater circuits with suitable oper-
ating margins that detect and regenerate pulses at a 224 megabaud rate
(a baud interval of 4.5 nanoseconds) when each pulse is dispersed by the
transmission medium into more than 30 baud intervals and also atten-
uated to a level limited by thermal noise considerations.
In Section II we give a general description of the line and a summary
of performance. In Section III we describe the design of the overall line
leading to the requirements on each of the repeaters. The design con-
siderations include the pulse transmission code, the equalization for the
loss-frequency characteristic of the coaxial transmission medium and
the control of the accumulation of jitter in a chain of repeaters. In Sec-
tion IV, the design of the repeater is described with emphasis on critical
circuits such as the Esaki diode regenerator. Section V reports on per-
formance under laboratory conditions.
II. GENERAL DESCRIPTION
2.1 Brief Description of the Experimental Repeatered Line
The configuration of the experimental repeatered line is shown in
Fig. 1. The overall performance of such a line is characterized by the
EXI'EKIMENTAL DIGITAL REPEATERED LINE
995
rate of errors introduced into the transmitted stream of information and
by the amount of pulse jitter introduced into this stream.
The line operates at error rates below 10 -10 through ten repeatered
links. A coast-to-coast system of about 4000 repeatered links requires
an overall error rate below 10~ 6 , or below 2.5 X 10 -10 per link. Hence,
-4000 mile error performance has been achieved under laboratory con-
ditions.
The pulse jitter measured through ten repeatered links under labora-
tory conditions was 13 degrees rms. From the model for the accumula-
tion of jitter in a chain of repeaters, 8 this 13 degrees implies that the
significant component of jitter introduced per repeater is about 3 degrees
rms. As will be discussed in Section 3.3.2, 3 degrees is an entirely ac-
ceptable performance level in a coast-to-coast system.
An arbitrary stream of binary unipolar pulses from the multiplex at a
rate of 223.880 Mb/s is introduced into the line at a binary-to-PST
translator as shown in Fig. 1. This translator converts the 2-level
pulses into a selected ternary code, called paired selected ternary (PST),
for transmission. The details of this code are described in Section 3.1 and
also in an earlier paper. 9 Briefly, the ternary transmission of the binary
information provides sufficient redundancy (i) to allow the transmission
of unrestricted binary sequences while still providing timing information
to the repeaters, (ii) to eliminate dc components from the transmitted
spectrum, thus permitting ac coupling in the repeaters and powering at
dc, and (Hi) to allow in-service error monitoring for maintenance pur-
poses.
The ternary signal from the translator has its positive and negative
pulses represented as positive pulses on separate leads. These are applied
to the transmitting repeater which adjusts levels, combines the two
streams, and adds dc power to the signal for serial powering of the re-
peaters.
PST
(+.0,-)
BINARY
(1,0)
BINARY
TO PST
TRANS-
LATOR
TRANSMITTING
-' REPEATER
RECEIVING
REPEATER
B- >0
PST
(+,o,-)
REPEATER
REPEATER POWER
PST TO
BINARY
TRANS-
LATOR
PST
VIOLA-
TION
MONITOR
BINARY
(l.O)
Fig. 1 — Experimental repeatered line.
996 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1966
This system utilizes essentially one mile of an experimental cable
with eighteen 0.270-inch coaxials, designed at Bell Laboratories and
fabricated at the Western Electric Company's Baltimore Works. The
cable design will be reported on separately. Each repeatered link uses
one of the 18 coaxials in the cable. Electrically, the coaxial line inserts
a loss that is proportional in dB to the square root of frequency (\/f)
over the frequency range of interest. At a nominal temperature, the loss
of a mile of 0.270-inch coaxial line is 57 dB at 112 MHz, one-half the
baud rate. The propagation characteristics of the experimental cable
are essentially equivalent to those of standard Bell System coaxial
cables used for L-carrier transmission. 10 Hence, the spacing of the
repeaters on the 0.375-inch standard coaxials now generally being in-
stalled would be that of the experimental system modified by the ratio
of the diameters [(1.0 miles) X (0.375/0.270) = 1.4 miles].
The regenerative repeater, of the forward acting complete retiming
variety, 11 performs the three R's — reshaping, retiming, and regenera-
tion. To reshape, an equalizer compensates for the y/f propagation
characteristic of the transmission medium in such a manner as to
compromise among the intersymbol interference, the noise entering at
the repeater preamplifier, and other important degrading effects. The
repeater spacing is controlled by this compromise. To retime, a sine wave
with average frequency equal to the baud rate is extracted from the
pulse train by nonlinear means. This sine wave in turn generates regu-
larly spaced pulses of short duration for sampling the equalized wave-
form. To regenerate, a 3-level decision is made at each sampling instant
to determine whether a +, 0, or — pulse is to be emitted in each baud
interval. A detailed description of the repeater is given in Section IV.
In the experimental system, 10 such repeatered links have been oper-
ated in tandem, looping through 10 coaxials of the 18-coaxial cable to
form 10 miles of line.
In the receiving repeater, the dc powering circuit is completed, the
levels are adjusted, and the positive and negative pulses are separated
into two unipolar streams for application to the PST-to-binary trans-
lator. Finally, a PST violation monitor makes use of a redundant prop-
erty of the PST code to monitor errors as described in Section 3.1.4. 8
2.2 Main Stations
An actual long-haul repeatered line would have main stations spaced
at appropriate intervals to house equipment for (i) powering the line
repeaters through the transmitting and receiving repeaters, (ii) isolating
and automatically switching out a section of line for maintenance pur-
EXPERIMENTAL DIGITAL REPEATERED LINE 997
poses, (in) controlling the accumulation of pulse jitter in a long re-
peatered line, 5 ' 7 and (iv) dropping and adding information-bearing
pulse streams (multiplexing).
The main station spacing appears to be determined primarily by the
multiplexing considerations. The limitation due to powering is based on
the power required by the repeaters and the maximum practical voltage
which can be applied to the coaxials and the repeaters. The limitation
due to section isolation considerations depends on the combined relia-
bility of the cable and the repeaters. We will not deal with the power
and section isolation questions in this paper, but we will show in Section
3.3 that jitter control is required only at very distant spacings, and
hence is not a consideration in the spacing of main stations.
III. DESIGN OF THE LINE
The overall design of the experimental 224 Mb/s digital repeatered
line is described in this section; the repeater itself is described in Section
IV. We discuss the pulse transmission code in Section 3.1, the equaliza-
tion of a link of repeatered line in Section 3.2, and the retiming of such
a link in Section 3.3. The derivation of performance objectives for a
single repeater from the overall objectives for a 4000-mile system is
included; we treat error rate in Section 3.2 and jitter in Section 3.3.
3.1 Pulse Transmission Code
3.1.1 Purposes of a Transmission Code
The binary information from the multiplex must be coded into a se-
quence of signal symbols that is readily transmitted over a practical
line. The transmission code adds redundancy to the binary information
to permit the three specific functions described below.
Conceptually, the simplest pulse transmission code is unipolar in which
the binary marks and spaces are coded for transmission as presence and
absence of pulses. There are three significant practical problems asso-
ciated with this unipolar format :
(i) Timing information must be extracted from the pulse train at each
repeater to determine when the pulse, no-pulse decisions should be
made and to retime the regenerated pulses emitted by each repeater.
Long sequences of binary spaces result in long periods without pulses
and hence, without liming information. This in turn yields poor timing
performance, leading to increased error rate and pulse position jitter.
(ii) Since the repeaters are serially powered by means of dc trans-
998 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1966
mitted on the center conductors of the coaxials, the signal path in the
repeaters cannot be dc coupled to the cable medium. Consequently,
the transmission of varying densities of marks results in dc wander of
the pulse stream, thereby reducing drastically the margins in the de-
tection process. DC restoration circuits which could eliminate this
wander appeared unfeasible for the experimental repeater because of the
high baud rate.
(in) Some method of determining error performance without the
interruption of service is essential for maintenance purposes. In-service
monitoring of the line error rate with the unipolar format as described
above is impossible because each unipolar binary symbol carries exactly
1 bit of information with no redundancy.
All three of these problems can be eliminated by introducing redun-
dancy into the coding process. In the experimental line, the required
redundancy is obtained by employing 3-level transmission (+,0, — ) at
a line baud rate equal to that of the binary information rate. This redun-
dancy is log 2 3 — 1 = 0.59 bits/symbol.
Another means of achieving the necessary redundancy is to utilize
polar binary transmission at a line baud rate higher than the information
rate. This appears to be an unattractive alternative for a coaxial me-
dium because the higher baud rate, resulting from a compromise satisfy-
ing the above three constraints with reasonable coding complexity,
more than offsets the potential signal-to-noise advantage of detecting
2-level signals rather than 3-level signals.
3.1.2 The Paired Selected Ternary (PST) Code
The pulse transmission code is paired selected ternary (PST). 9 In this
code, the binary sequence to be transmitted is framed into pairs and
translated into a ternary format according to Table I. There are two
modes in the PST code and the mode is changed after each occurrence
of either a 10 or a 01 binary pair. As an example of PST coding, consider
the following binary sequence and the corresponding PST sequence.
BINARY 10 1
01 11 01
PST +0 0-
- + 0+ +- 0-
Note the change of mode after each occurrence of a 10 or a 01 pair in
the binary sequence.
Six of the nine possible ternary symbol pairs are used to represent
the four possible binary symbol pairs. The remaining three unused
ternary symbol pairs are used for framing the pairs at the receiver.
EXPERIMENTAL DIGITAL REPEATERED LINE
Table I — The Paired Selected Ternary Code
999
PST
+ Mode
— Mode
11
10
01
00
+ -
+ o
+
-+
+ -
-0
0-
-+
Change mode after each 10 or 01
The alternation of modes produces a null in the power spectrum at dc
when the positive and negative pulses are balanced (identical except for
sign). This enables dc powering of the repeaters. The PST power spec-
trum, W(f), normalized to the baud interval, T, is presented in Fig. 2 for
the case of equally likely marks and spaces in the original binary se-
quence [p(l) = p(0)], cosine-squared pulses one baud interval in dura-
tion at the base, and balanced positive and negative pulses. For other
than this idealized situation, see Rcf. 9.
The coding of the 00 binary pair into the — |- PST pair eliminates the
timing problems associated with the transmission of long sequences of
0's. An additional feature of the PST code is that timing information
1 . p (1 ) = p (0) IN BINARY TRAIN
2. COSINE -SQUARED PULSES
ONE BAUD INTERVAL, T, IN
DURATION AT THE BASE
3. BALANCED POSITIVE AND
NEGATIVE PULSES
100 150 200 250 300 350
FREQUENCY IN MHZ
Fig. 2 — PST power spectrum for a random binary sequence.
1000 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 19GC
H
CO
.a
•a
o
h5
bO
EXPERIMENTAL DIGITAL REPEATERED LINE 1001
at the repeaters can be extracted by nonlinear means (rectification).
This avoids the harmonic and phase distortion problems associated
with other schemes involving the linear extraction of timing information
from the low-level components received at the baud frequency.
A logical diagram of the binary-to-PST translator is shown in Fig. 3.
3.1.3 Translation of PST Sequences Back Into Binary
At the receiving end of the line, the original binary sequence is re-
covered from the selected ternary sequence in the PST-to-binary trans-
lator.
Framing is essential to associate the symbols which are part of the
same PST pair. The effect of incorrect framing is demonstrated below.
Binary 10 01 00 01 11 01
PST +0 0- -+ 0+ +- 0-
Incorrectly framed PST + 00 -- +0 ++ -0 -
Incorrect binary ? ? 10 ? 10
The unused ++> , and 00 ternary pairs are detected to indicate an
out-of-frame condition.
The PST-to-binary translator is shown in Fig. 4. Out-of-frame indica-
tions are applied to a flywheel circuit which prevents randomly occurring
line errors from initiating a change in frame. The flywheel requires that
three out-of-frame indications occur within 19 consecutive pairs (38
baud intervals or 170 nanoseconds) to initiate a change in frame. The
cumulative probability of obtaining three out-of-frame indications
within J consecutive pairs after going out of frame is given by the N = 3
curve in Fig. 5 for the case of equally likely marks and spaces in the
original binary sequence. The mean time (in PST pairs) to occurrence of
N out-of-frame indications is shown as M.
To reduce the effect of erroneous frame shifts due to line errors, the
flywheel is disabled for a period of 19 pairs after a change in frame.
During this period, only one out-of-frame indication is required to ini-
tiate a second frame shift pulse, thereby reestablishing the original fram-
ing condition. The cumulative probability of obtaining one out-of-frame
indication within J pairs is indicated by the N = 1 curve in Fig. 5.
The mean time to an incorrect initiation of a change of frame due to
randomly occurring line errors is shown in Fig. G as a function of error
rate, assuming p(l) = p(0) = ]/2- For the required error rates of 10 -6 ,
such false misframes are extremely unlikely.
1002 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 19GG
s?
1=
ID O
'
r
6 C
\i
- u
o
3"
It
A
1
/
/
,*
^\
1
K
FIRST
FRAME
SHIFT
PULSE
o
h
/ ^-^
\_f "}ul
v,
£
/i
' (\
r "
c
■r
r
1
1
1
ID
z
1
, lu \ /
FLYWHEEL
170 nS PULSE STRETCHERS
1 LU
1 -J
-^
1 LU
1 5
I-
u
m
— o
|
1 a -
°
h-
1 U
LU
^
1 | <
1 a
1 LL.
i
t; r
5 ^K
Lt
O
LU(-
(/1<
ja
3 LU
Q. Z
LU
O
LU LU
2ui
< -I
a 3
Li
+
i
c
■n
\
^y
iN [
» 1
J
i~ i
r
IP
I u
LU
h-
1 UJ
1 Q
LU
1 1
1 <
1 LL
. ? 1
h
D
0-
Z
H
\
OSS /
J. < <
m
\
l
LU
1-
LU
Q
+
+
A
UJ
111
o
1
A
c
c
o<
OL
1
L
u
U
k
J 1
II
i
O
c 1
fa
c
in
LU
Lt
LU
V
1 >"
J_i
0.
H LU
too:
»
.
NEGATIVE
STREAM
r
=3
tfj
* —I
EXPERIMENTAL DIGITAL REPEATERED LINE
1003
M = MEAN TIME
TO OCCURENCE ;
P(1) = P (0) IN
BINARY SEQUENCE
N = 3^
/
^=1
r*M ~
/m
) 5 10 20 40 60 80 90 95 99 99.9 99.99
CUMMULATIVE PROBABILITY OF OBTAINING N OR MORE PST
OUT -OF -FRAME INDICATIONS IN J PAIRS
Fig. 5 — Probability of obtaining N or more PST out-of-frame indications in
J pairs.
a
z
S! 4n9
-CENTURY
</>
RAMES IN
o
-YEAR
-WEEK
UI
-DAY
5
Z
UI *
ui (0 3
t-
UI
-HOUR
-MINUTE
Hi
2
z
<
UI
2
10-3
ERROR RATE
Fig. 6 — Misframes due to random line errors.
1004 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER I960
3.1.4 In-Service Error Performance Monitoring
A method of monitoring line error performance without interrupting
service is a necessary maintenance feature of a long-haul digital trans-
mission system. PST sequences contain certain properties which are
violated only when errors occur. Violations of these properties can be
monitored to determine the error performance on an in-service basis.
Four different properties whose violation provide means of error
monitoring are given in Ref. 9. One of these, the bipolar property of PST,
is employed in the experimental line.
To error monitor with the bipolar property, we first remove all of the
+ — and — + ternary pairs from the pulse stream. The remaining pulses,
which come from coding the binary 10 and 01 pairs, alternate in polarity.
An error causes a violation of this alternation or bipolar property, as in
the example below.
PST pulse stream
4-0 + 0+ + - 0-
PST with +- and - +
removed
+0 0-
0+ 0-
PST with error
r
+0 + -
+0
-+ 0+ +- 0-
PST with error and H — and
— H removed
r
0+ 0-
Error
Violation
The violation can occur some time after the error, but all singly occurring
errors* are detected.
A logical diagram of the PST violation monitor is shown in Fig. 7.
Since this form of violation monitoring requires framing common to
PST-to-binary translation, these two functions have been combined in
one circuit in the experimental system.
3.1.5 Circuit Techniques for PST Code Translation
The laboratory realization of the PST code translation equipment has
made extensive use of Schottky barrier diodes and emitter-coupled
current-routing pairs using transistors having an f T of 3 GHz. The diodes
perform the logic and the cm-rent routing pairs provide isolation, am-
plification, amplitude regeneration, and, when appropriately connected,
logical inversion. Fig. 8 shows one of the four binaiy pair detectors in the
* For the low error rates we are interested in, the rate of singly occurring errors
is essentially the total error rate.
EXPERIMENTAL DIGITAL REPEATERED LINE
1005
NEGATIVE
PST --+■•
STREAM
PST FRAME
CLOCK {FROM
PST-TO- — -
BINARY
TRANSLATOR)
POSITIVE
PST
STREAM
VIOLATION
OUTPUT
Fig. 7 — Logical diagram of the PST violation monitor.
FRAME
CLOCK "
INPUTS
X*
t
^
^A/V
Fig. 8 — Binary pair detector at the input of the PST-to-binary translator —
typical of the logic gates in the code conversion equipment.
1006 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 196(i
binary-to-PST translator, a circuit typical of the logic gates used
throughout the translation equipment. The input signals to the gates are
at least one-half volt to insure complete switching of the emitter current
from one transistor to the other.
Another circuit used extensively in the code translation equipment is
the Goto pair binary counter, shown in Fig. 9, which employs two Esaki
diodes.
— nwy^
Fig. 9 — Goto pair binary counter using two Esaki diodes.
3.2 Equalization of a Link of the Repeatered Line
3.2.1 The Error Rate Objective
The preliminary overall error rate objective for a 4000-mile digital
repeatered line is a maximum of 10~ 6 , based upon subjective testing and
computation of noise power in the terminal signals due to line errors.
Such an error rate will introduce negligible degradation into analog
signals arising from services such as message, television, Picturephone*
and voiceband data, and will yield high performance for signals already
in digital form such as computer outputs. 12 If we assume an approximate
repeater spacing of one mile, we obtain a maximum error rate objective
of 2.5 X 10 -10 per repeatered link, or approximately 10~ 10 . This calcu-
lation assumes uniform operation of all repeaters. In a field situation,
we would expect to permit somewhat poorer performance in several re-
peaters since the bulk of the repeaters normally would be operating at
substantially better than this level. To place an objective on this, how-
ever, requires more knowledge of uniformity of repeater operation than
is presently available.
Service mark of the Bell System.
EXPERIMENTAL DIGITAL REPEATERED LINE
1007
3.2.2 Model of the Signal Path
A model of the signal path in a link of a repeatered line is shown in
Fig. 10. The signal path begins at the regenerator output in a repeater
and ends at the regenerator input in the next repeater. The overall
transmission of the signal path from S N (f) to Ritif) is denoted T(i).
The loss of a coaxial tube in a cable, C(f), is primarily dependent on
the skin effect in the conductors and therefore this loss in dB is essentially
proportional to the square root of frequency in the range of interest to
us, and linearly proportional to the cable length. A single pulse through
C(f) alone is highly attenuated and severely dispersed and therefore gain
and equalization are required to amplify and shape the pulse stream
prior to detection in the regenerator. In the model, there are two equaliz-
ing blocks, Ei(f) and E 2 (J), and two amplifying blocks, Ai(f) and A 2 (f).
The details of this separation will be discussed later, but it can briefly
be stated here that the amplifier split is to sectionalize the required high
gain, and the equalizer split is a compromise to limit the noise band-
width, to prevent overload of the preamplifier, and to provide surge
protection for the repeater output. The power separation filters at the
input and output of the repeater are included in Ai(f) and Ei(f).
Our analyses have been carried out with cosine-squared regenerator
output pulses one-half baud interval in duration at half-amplitude.
This is a good approximation for the experimental system. The detailed
shape of these pulses is not important because the spectrum of such a
narrow pulse is nearly flat over the important frequencies in T(j), which
only go up to about three-fourths of the baud frequency. Hence, it is
Fig. 10 — Signal path model.
1008 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1966
only the total area of these pulses that is important. The pulse duration
used was chosen for convenience in implementation.
The outer conductor of the coaxial is 0.005-inch thick copper with a
soldered seam. The conductor thickness is sufficient to essentially isolate
the signals inside the coaxial from the outside world. Hence, the con-
trolling disturbance to the signal is the sum of the thermal noise of
the cable and the noise generated in the preamplifier. This is shown in the
model as an equivalent noise source at the preamplifier input, N(f).
The key measure of the performance of a repeater is the error rate.
This rate, in turn, is dependent on the ability of the regenerator to decide
correctly whether +, 0, or — pulses were transmitted in each of the
baud intervals. We are, therefore, highly concerned about the details
of the waveform at the regenerator input, r N {t). Equalization Ei(f)E 2 (f)
compensates for the shape of C(f) and the imperfections in the trans-
mission characteristic of Ai(f)A 2 (f). Also, the bandwidth of the noise
entering the regenerator is primarily determined by E 2 (f). The optimum
equalization maximizes the repeater spacing for a given error rate ob-
jective by means of a compromise between intersymbol interference and
noise, while taking into account variations in coaxial loss due to temper-
ature variations, sampling amplitude threshold offset, sampling timing
misalignment, and other practical degradations.
An analytical solution for the optimum equalization in the face of all
of the practical degradations in the repeatered link is not tractable.
The general problem of optimum equalization for pulse detection has
been treated in the literature, 1314 but nowhere, including in analyses of
our own, is there a solution to an applicable model. Consequently, we
simulated the repeatered link on a digital computer in such a manner
as to allow examination of the effects of the following:
(i) variations in the transmission characteristics of the coaxial,
C(J), the amplifiers, Ai(f) and A 2 (f), and the equalizers, Ei(f) and E 2 (f);
(ii) Gaussian noise with arbitrary power spectrum ;
(Hi) static and dynamic sampling time misalignment;
(iv) variations in detection threshold;
(v) variations in the statistics of the input binary sequence;
(vi) variations in the regenerator output pulse shape.
3.2.3 Signal-to-Noise Ratio and Error Rate
The theoretical error rate in the detection of a signal in the presence
of Gaussian noise follows the well-known curve shown in Fig. 11."
Note that we have plotted peak signal-to-rms-noise. It is the peak sig-
nal that is of interest to us since the power limitation is in the repeater
EXPERIMENTAL DIGITAL KEPEATEREI) LINE
1001)
10- 5
)0" 6
tr
|io-'
LU
LL
O
>- (0- S
_l
m
a.
a.
io-'°
to-"
IO-«2
\
\
s
\
\
\
\
^
\
\
1
\
16 18 20 22 24
PEAK SIGNAL -TO -RMS -NOISE IN DECIBELS
Fig. 11 — Theoretical error rate as a function of peak signal-to-rms-noise ratio.
peak output capability, not in its average signal power. Also, because
of the shielded nature of the coaxial medium, transmitted signals do not
interfere with any other signals.
The interpretation of Fig. 11 is that the average error rate indicated
by the ordinate will prevail if the ratio of the peak of an isolated pulse
to rms Gaussian noise, with no interfering signals or other degrading
effects at the time of detection, is the quantity in dB indicated by the
abscissa. Under practical conditions with a high baud rate there will
be intersymbol interference and other degradations. Hence, at the de-
cision instants the signal will depart from the ideal case, where there is
either a full peak or no signal at all, and where the decision time instant
and threshold arc perfect. Fig. 11 must, therefore, be interpreted ap-
propriately.
To avoid the difficulty of dealing with statistical intersymbol inter-
ference and other degradations, we conservatively design for the worst
case. Hence, Fig. 1 1 may be used to determine the upper bound on the
error rate if the peak signal is interpreted as the difference at the sam-
pling instant between the minimum amplitude of all received pulse
signals and the maximum amplitude of all received no-pulse signals.
1010 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 19G6
The worst case is the combination of the worst intersymbol inter-
ference resulting principally from cable temperature changes and cir-
cuit imperfections, and the other degradations such as nonideal sampling
where both the time instant and the threshold are imperfect. The lump-
ing of the other degradations with intersymbol interference is illustrated
in Fig. 12 using the ternary eye. This eye represents the inner boundaries
of all possible pulse sequences superimposed and synchronized with the
baud rate; in addition, the boundaries have been displaced inward verti-
cally and horizontally to account for the other degradations. Resulting
eye heights H+ and //_ now represent the peak signal for use on Fig. 11,
and the time and amplitude crosshairs now are of zero width. In what
follows, H is taken to represent either H+ or //_ .
From Fig. 11, we see that in order to maintain the per repeater 10~ 10
error rate objective indicated in Section 3.2.1, under extreme conditions
(due to pulse sequences, cable temperature, parameter drifts, and
sampling time and threshold displacements), the ratio of H to the rms
noise must be 22 dB. It follows then that the ratio of the peak of the
pulse, P, to the rms noise is greater than the theoretical value of 22 dB
by an impairment denned as
/ = 20 log (P/H).
The impairment, /, is the excess peak-signal-to-rms-noise ratio in dB
required to compensate for the worst case degraded eye due to the total
of the intersymbol interference and other degradations. Due to the rapid
increase with frequency of the loss of the coaxial, a small increase in
equalized bandwidth for the purpose of reducing intersymbol inter-
ference produces a large increase in noise. The compromise between
intersymbol interference and noise, therefore, is balanced in favor of
reduced bandwidth. Our specific compromise allows enough intersymbol
interference and other degradations to reduce H to 0.2 P, which corre-
sponds to an / of 14 dB.
It should be pointed out that the eye opening, H, is an artificial one
with idealized crosshairs and, therefore, is not what is seen on an os-
cilloscope at the regenerator input. What is seen on an oscilloscope is
the eye with all degradations up to the input to the practical detector.
The detector, however, has its own degradations which are included in
the artificial eye.
3.2.4 Signal Path Parameters for the Experimental Line
The digital computer simulation and actual circuit performance
achievements have led to the signal path parameters of Table II for a
EXPERIMENTAL DIGITAL REPEATERED LINE
1011
OUTLINES OF
DEGRADED
EYES
OUTLINES
OF IDEAL
EYES
Fin. 12 — Ideal and degraded eyes.
repeatered link. Those parameters have been realized, but under labora-
tory degraded conditions rather than worse case conditions.
Table III gives an example of the extreme conditions simultaneously
permitted by the design implied by Table II.
The signal path parameters of Table II impose the requirements on
the design of the repeater itself, to be discussed in detail in Section IV.
For the present, we have specified, (i) the loss of the cable at the half
Table II — Signal Path Parameters of a Repeatered Link
Worst case error rate
lO" 10
Repeater peak output power
+ 14 dBm
Preamplifier noise figure
6dB
Theoretical peak signitl-to-rins-noi.se ratio for 10 -10 er-
22 dB
ror rate
Impairment, I, for worst case degradations
14 dB
Ratio of peak equalized single pulse to rms noise under
30 dB
nominal conditions
Loss of coaxial between repeaters
57 dB at 112 MHz
(Corresponding length of 0.270-inch coaxial at nominal
, 55 T)
Equalizer singularities Zeros:
(5320 ft)
11 MHz
15 MHz
70 MHz
Poles:
94 MHz
161 MHz / ± ii7°
196 MHz /±H4°
1012 THE HELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 19GG
Table III — Example of Extreme Conditions
Simultaneously Permitted
Cable temperature rise
Equivalent echo at the sampling instant
Sampling amplitude threshold offset
Static sampling timing misalignment*
Dynamic sampling timing misalignment distribution'
Noise figure
21 °F (maximum expected)
20% of the pulse peak
10% of the pulse peak
20°
cos 4 with G0° base width
7dB
* 360° is one baud interval.
baud rate which bears on the gain in the repeater, f (ii) the required
output power and noise figure of the repeater, (Hi) the equalizer singu-
larities, and (iv) the allowed impairment. The repeater design will take
up from there.
As an indication of the significance of these signal path parameters,
we refer to Figs. 13 through 17. First, the loss at 55 °F of a coaxial be-
100
80
40
4 6 8 10 20 40 60 100
FREQUENCY IN MHZ
200
400
Fig. 13 — Loss of 5320 feet of 0.270-inch coaxial at 55°F on the usual log /
scale.
tween repeaters is shown in Fig. 13 plotted on the usual log frequency
scale in accordance with conventional characterization of transmission
media. This curve (loss in dB) follows essentially the square root of
frequency and goes through 57 dB at 112 MHz.
The nominal losses of the cable, the equalization including the effect
of the amplifiers, and the overall channel with 50 dB net effective flat
gain, are shown in Fig. 14.
t The loss of the peak of a pulse through the signal path is given approximately
by the loss of this path at a frequency corresponding to half the baud rate."
(00
90
70
60
30
10
CABLE/
/
/ OVERALL CHANNE
WITH 50 dB NET
/
\ /
\EC
UALIZ/
VTION>
i
50 (00 (50 200 250 300
FREQUENCY IN MHZ
Fig. 14 — Cable, equalization, and overall channel loss characteristics.
l.o
0.9
0.8
Q
l "
5
< 0.5
a
LU
N o.4
_J
<
I 0.3
o
\
\
\
\
\
■
0.2
0.1
-0.1
10 20 30 40 50 60 70 60
TIME IN BAUD INTERVALS
Fig. 15 — Nominal isolated pulse without equalization.
1013
1014 THE BELL SYSTEM TECHNICAL JOURXAL, SEPTEMBER 1966
Due to the great variation of the cable loss over the frequency band
of interest, the nominal isolated pulse without benefit of equalization is
severely dispersed as shown, normalized to its own peak, in Fig. 15. It
has a rise time of about 3 baud intervals and a fall time of about 25
baud intervals. The undershoot is due to the lack of dc transmission in
the power separation filters and is of little significance since there are
balanced numbers of positive and negative pulses in the PST code. As a
result of the dispersion, the peak amplitude of all possible unequalized
pulse sequences varies over a 28-dB range. The problem of dealing with
the resulting large dynamic range in the repeater is discussed further in
Section 4.1.
The nominal single equalized pulse is shown in Tig. 16. It is apparent
that the amplitudes at -1.0, +1.0, +2.0, and +3.0 baud intervals are
small. Under degraded conditions, however, these amplitudes are sub-
stantial. The low amplitude negative tail of such an isolated pulse,
which obviously must be present in the response of a channel without
dc transmission, is off the chart. The inner boundary of the superposition
of this single pulse in all possible PST sequences forms the nominal eye
shown in Fig. 17. The uppermost curve in this figure is the envelope of
the maximum values of all pulse sequences. The slow-acting automatic
1.0
0.9
0.7
0.6
O.b
t
0.2
0.1
\
\
\
0.1
-2.0 -1.5 -1.0 -0.5 0.5 1.0 1.5 2.0 2.5 3.0
TIME IN BAUD INTERVALS
Fig. 16 — Nominal single equalized pulse.
EXPERIMENTAL DIGITAL REPEATERED LINE
101.'
i.5 -0.4 -0.3 -0.2 -0.1 0.1 0.2
TIME IN BAUD INTERVALS
0.3 0.4 0.5
Fig. 17 — Nominal positive eye at the regenerator — inner boundary of the
superposition of all possible PST sequences using the nominal single equalized
pulse shown in Fig. 1G.
gain control (AGC) in the repeater operates substantially on the peak of
this curve. It should be emphasized that this nominal eye is the one seen
on an oscilloscope and docs not include the effects of the finite crosshair
widths.
a. 3 Retiming in the Repeater ed Line
3.3.1 Timing Extraction at Each Repeater
The function of the repeater timing path is to extract timing informa-
tion from the equalized ternary pulse stream. As discussed previously,
this timing information is used to determine when the pulse, no-pulse
decisions should be made as well as to retime the pulses transmitted
from the repeater.
The repeater uses forward acting complete retiming 11 with nonlinear
extraction of the liming component from the PST sequence. The equal-
ized pulse stream is full-wave rectified and the upper portion of the
1016 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER I960
resulting unipolar stream is amplified and applied to a simple tuned
circuit. The output of this is amplified and limited to produce a constant
amplitude sinusoid which, in turn, drives a timing pulse generator.
Each repeater introduces phase variation, or timing noise, called jitter,
into the extracted timing information. This jitter contains a systematic
component which is a function of the transmitted pulse pattern and
accumulates in a chain of repeaters considerably faster than the pattern
independent, nonsystematic component.
The amounts of jitter introduced in and transmitted through each
repeater can be kept low by a small effective bandwidth in the timing
path. The smaller this bandwidth, however, the larger the static sampling
timing misalignment for a given and inevitable mistuning of the timing
extraction circuit.
The jitter accumulation and the objective for the jitter introduced in
each repeater are discussed next.
3.3.2 Control of Jitter Accumulation and the Requirements on a
Single Repeater
The preliminary objective for band-limited jitter at the end of a
4000-mile line is as shown in Fig. 18. This was derived on the basis of
coded mastergroup message service in which the top frequency is 2.788
MHz, such as in an L600 mastergroup or a U600 mastergroup shifted
down in frequency for more efficient sampling.* Based upon preliminary
analysis and subjective testing, however, this objective is probably
controlling for all digital and analog type services including data and
network television. These two latter services, however, require further
study. 12 The effect of low-frequency jitter within the objective shown is
a signal-to-distortion ratio in the message channels in the frequency
multiplexed mastergroup of greater than 30 dB; the effect of high-
frequency jitter is crosstalk between channels less than theoretical
9-digit quantizing noise.
In order to meet this objective with an economical repeater design,
appropriately spaced high Q jitter reducers 7 are required along the line.
A jitter reducer includes an automatic phase control (APC) loop to
smooth the jittered timing wave, and buffer storage for the information
pulses. The information pulses are read into the store by the jittered
timing wave and read out by the smoothed wave.
The jitter performance required of each repeater is a function of the
* In Ref. 5, this objective was given for a coded U600 mastergroup in which
the top frequency is 3.084 MHz.
EXPERIMENTAL DIGITAL REPEATERED LINE
1017
? o
1.8
SI 1,0
|g 0.8
ncg
hi/)
of
Q z 0.4
a.
A MASTERGROUP CODER
(TOP FREQUENCY 2.788 MHz)
-
-
\
s
;;
0.26-
0.3
0.01 0.02 0.05
0.1 0.2 0.5 (.0 2 5
JITTER BANDWIDTH IN kHz
Fig. 18 — Objective for the maximum overall band-limited jitter due to random
pulse sequences in a 4000-mile digital repeatered line.
overall jitter objective, the number of jitter reducers in the overall line,
the effective Q of each of the jitter reducers, and the effective Q of the
timing circuits in the repeaters.
The accumulation of jitter in a chain of repeaters with uniformly
spaced jitter reducers can be analyzed by an extension of the technique
developed for the analysis of jitter accumulation in the Tl system. 817
The analysis is based on a model in which the following assumptions are
made:
(i) The significant jitter at the end of a chain of repeaters arises from
the addition of the systematic, or pulse pattern dependent, jitter intro-
duced in each repeater.
(it) At the end of a long chain of repeaters, the accumulated jitter
due to a random pattern is gaussian.
(m) With respect to the transmission of accumulated jitter, the
repeaters and jitter reducers may be replaced by the low-pass equiva-
lents of their timing paths.
(iv) In each repeater, the effect of all sources of jitter due to random
patterns is equivalent to the effect of a single, band-limited, white
timing-noise source at the input to the equivalent low pass network.
(y) The timing noise introduced in the jitter reducers is negligible.
The model has successfully predicted experimental results for the Tl
system, 8 and there is every indication that the same will be true for a
high-speed repeatered line with jitter reducers. The model for the high-
speed line is shown in Fig. 19. F R (s) and Fj(s) are the low-pass equiva-
1018 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 196(5
N R REPEATERS
*R *R
L
,1
FrCSJ
Fr(s)
"~ 1
Fr(S)
Nj JITTER REDUCERS
N R REPEATERS
4>a * D
■*l
'R
JITTER
REDUCER (-— -j F R (S)
FAS)
F R (S) F R (s) —
JITTER
REDUCER
Fj(S)
FROM
TRANSMITTING
TERMINAL
TO
RECEIVING
TERMINAL
Fig. 19 — Model of a chain of repeaters and equally spaced jitter reducers for
the analysis of the accumulation of jitter.
lents of the repeater and jitter reducer timing paths, and <£ B is the power
spectral density of the equivalent band-limited white timing noise
source.
The results of an analysis are presented in Fig. 20 for a transcontinental
system containing 3G0O repeaters each with an effective Q of 80, and
uniformly spaced jitter reducers each with an effective Q of 10 6 . In this
zz
< Q.
Si£o.4
ct a-
a. 5
a Q
r
w"0.2
si
uzo.
la
NUMBER OF REPEATERS = 3600
Q OF REPEATERS =80
EFFECTIVE Q OF JITTER REDUCER =IO e
so.
DOB
I 4 10 20 40 60 80 100
NUMBER OF EQUALLY SPACED JITTER REDUCERS
IN THE LINE
Fig. 20 — Objective for the systematic component of jitter (rms) contributed
by a single repeater versus number of equally spaced jitter reducers.
EXPERIMENTAL DIGITAL REPEATERED LINE
1019
figure, the requirement is given for the systematic component of jitter
(rms) contributed by a single repeater due to random pulse patterns as a
function of the number of equally spaced jitter reducers. If no jitter
reducers were to be used, the repeater objective would be 0.008 ns rms
or 0.6° mis, which when properly scaled is more stringent than the
approximately 1° measured in Tl. 8 Because of the higher-speed tech-
nology here, a practical requirement should be more liberal than the
reported Tl performance. Based on our laboratory experience, a reason-
able objective for the contribution of a single repeater to the systematic
jitter due to random patterns is 0.1 ns rms or 8° rms. This leads to a
need for four jitter reducers, each with a Q of 10 6 , in 4000 miles, or one
jitter reducer every 1000 miles. In an actual system, jitter reduction is
required at all multiplex points. These occur, on the average, at inter-
vals substantially smaller than 1000 miles, so that jitter reduction does
not influence the main station spacing. Further, jitter reducers with
Q's of less than 10 6 would be used.
IV. DESIGN OF THE REPEATER
A block diagram of the repeater is shown in Fig. 21. As briefly de-
scribed in Section 2.1, the regenerative repeater perforins the functions
known as the three R's — reshaping, retiming, and regeneration. The
PRE-
AMPLIFIER
AMPLIFIER
WITH AGC
POWER
SEPA-
RATION
FILTER
rv
EQUAL-
IZER
^
n
EQUAL-
IZER
I
POWER
SEPA-
RATION
FILTER
AMPLIFIER
+ 8 4V -64V
ISOLATION
FILTER
— I REPEATER
-±=- GROUND
POWER
SUPPLY
ISOLATION
FILTER
rn EARTH
///GROUND
Fig. 21 — Functional block diagram of the repeater.
1020 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 196(3
linear signal path (Fig. 10), comprising Equalizer I in the previous
repeater, the preamplifier, Equalizer II, and the amplifier, reshapes and
amplifies the pulse in preparation for detection. The timing path extracts
a periodic timing wave from the signal train and generates a train of
short sampling pulses occurring near the centers of the baud intervals,
where the eye has its maximum opening. The regenerator provides
crosshair pulse detection; at a time established by the sampling pulses,
the signal pulse is compared with a threshold voltage, and if the thresh-
old is exceeded, a new pulse is generated and transmitted to the next
repeater. Since this is a ternary repeater, two such thresholds are pro-
vided. A second set of short periodic pulses controls the output signal
pulse duration by turning off the regenerator after one-half baud in-
terval.
The circuits shown on the block diagram of Fig. 21 were constructed
on printed wiring boards and interconnected by sections of shielded
transmission line on a printed interconnecting board. The development
of the individual circuits was carried out independently between re-
sistive terminations equal to the characteristic impedance of the trans-
mission line used. Good cascading behavior was insured by controlling
the forward transmission and the input impedance with load.
The complete repeater uses 25 transistors; 14 of these are a Bell System
pnp germanium planar design with a cutoff frequency, f T , of 4 GHz;
9 are a Bell System npn silicon design with an f T of 1 GHz; and the
remaining two are lower-frequency transistors of standard codes. A pair
of gallium arsenide Esaki diodes provide the decision thresholds; a pair
of charge storage diodes are used to generate the short sampling and
turn-off pulses. Schottky barrier diodes are used in many circuits where
high speed and low capacitance are required.
The description of the repeater circuits is organized into four sections:
reshaping by the linear signal path, retiming by the timing path, re-
generation, and secondary features such as powering and surge protection.
4.1 Reshaping: The Linear Signal Path
4.1.1 Equalization
The overall shape of the equalization is specified by the singularities
indicated in Table II. Since the peak output power of the repeater is
limited, minimum noise reaches the decision point in the regenerator if
all of the passive equalization is placed beyond the preamplifier, a source
of the noise. It is advantageous, however, to sacrifice a small amount
of noise performance by placing a portion of the low-frequency attenu-
EXPERIMENTAL DIGITAL REPEATERED LINE
1021
ating equalization ahead of the preamplifier to prevent its overload by
certain pulse sequences strong in low -frequency content. With the PST
code, the peak amplitude of the unequalized sequence varies over a 28-
dB range, as stated earlier.
Once placed in front of the preamplifier, there is further advantage in
placing this low frequency attenuating portion of the equalizer at the
output of the previous repeater. This aids in protecting the regenerator
against lightning and power surges, both rich in low frequencies rela-
tive to our band of interest.
The response characteristics and circuit configurations of Equalizers
I and II are shown in Figs. 22 and 23. Equalizer I consists of a single
, — *
1
1
1
1
-
1
1
1
18.6 dB
1
1
1
1
1
1
1
-
11.1 94.0
FREQUENCY IN MHZ (LOG SCALE)
AAA/
BALANCED
REGENERATOR
OUTPUT
1
EQUALIZER I
UNBALANCED
COAXIAL
LINE
POWER
SEPARATION
FILTER
Fig. 22 — Equalizer I — attenuation characteristic and circuit configuration.
1022 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER I960
* 2(3
15 161
FREQUENCY IN MHZ (LOG SCALE)
70 196
FREQUENCY IN MHZ (LOG SCALE)
j wv
INPUT
^Mr
MV
AAAr
Fig. 23 — Equalizer II — attentuation characteristic and circuit configuration.
doublet with the zero at 11 MHz and the pole at 94 MHz. Notice the
balanced nature of Equalizer I, which couples the balanced regenerator
outputs to the unbalanced coaxial line through a balun and the un-
balanced power separation filter.
Equalizer II comprises two constant R sections, each with a real zero
and a complex pole pair. The first section has a zero at 1 5 MHz and poles
at 161 MHz /±117° and the second section has a zero at 70 MHz and
poles at 196 MHz /±114°.
4.1.2 Amplification
It is the function of the preamplifier and amplifier to provide essen-
tially flat gain while maintaining the desired equalized pulse shape. To
accomplish this without complex delay equalization requires a band-
width exceeding 300 MHz. The phase characteristic of this gain was
EXPERIMENTAL DIGITAL REPEATEREI) LINE
1023
taken into account in the computer simulation that led to the equaliza-
tion choice. A calculation of the gain required in the preamplifier and
amplifier at 1 12 MHz is as follows:
Cable loss
Equalization loss at 112 MHz relative to minimum loss
Minimum loss of equalization
Matching padding at amplifier input to achieve ade-
quate return loss
Nominal loss of variolosser (in amplifier for A(!C)
Effective ratio of amplifier to regenerator peak power
outputs
Total gain required 79 dB
57 dB
(i dB
9dB
(i dB
5dB
-4 dB
This gain is split with 26 dB in the preamplifier, and 53 dB in the am-
plifier.
The preamplifier curcuit, shown in Fig. 24, uses three transistors,
each having a 4-GHz/ r , in common emitter configurations with collec-
tor to base feedback. The series diode gate at the input provides surge
protection for the input transistor, and in the process loses 1 dB in
noise figure and in gain. The overall circuit shown has a 6-dB noise
figure at the frequencies of interest and a gain of 2G dB.
The amplifier functions (i) to terminate Equalizer II accurately in 50
ohms, (ii) to amplify the signal 48 dB nominally, with an AGC range of
±5 dB from nominal, (in) to provide balanced outputs of 1 volt across
each of the two regenerator 50 ohm inputs, and (iv) to rectify the bal-
IFROM
! POWER
SEPARATION
FILTER
r-VvV-W- -VW-Jk-W-AMr- -)|-f-Wv- I N Wv
rWVi
MSL
TO
EQUALIZER
n
Fig. 24 — Preamplifier circuit.
1024 THE HELL SYSTEM TECHNICAL JOUKNAL, SEPTEMBER 1900
I
60
■AAA If—) 1
EXPERIMENTAL DIGITAL HEPEATERED LINE 1025
aneed output to provide signals for the timing path. A circuit diagram
of the amplifier is shown in Fig. 25. The basic gain stages are two-tran-
sistor doublets. 18 All transistors in the forward signal path are the 4-
GHz pnp type, except the final transistor which is the 1-GHz npn type.
The two transistors in the dc amplifier for the AGC are standard codes
as'noted.
The AGC diode compares the positive peak equalized signal amplitude
with a reference derived from the power supply. When the amplitude is
too large (small) the difference is amplified and the PIN variolosser
diode current is increased (reduced). The advantage of a PIN diode over
a pn-j unction diode is that variolosser action is not obtained by the
nonlinear conductance of a junction, but rather by the conductivity of
the intrinsic region, determined by the dc control current. In this manner,
currents of tenths of milliamperes are used to control signal power up to
1 milliwatt. The AGC loop gain is 40 dB at midrange.
4.2 Retiming
The timing path generates two trains of periodic subnanosecond
pulses from information extracted from the equalized pulse stream. One
train is for timing the decisions in the regenerator; the other, of opposite
polarity, is for control of the duration of the regenerator output pulses.
The relative phase between the sampling pulses and the information
pulses at the regenerator input is determined by the timing path.
The timing path begins with the full-wave rectifier at the output of
the amplifier (Fig. 25). The diodes are biased to transmit the upper 65
per cent of the rectified signals. This clipping level was shown to give
best jitter performance both in an analog computer simulation of the
timing path and in tests on the actual circuits. As indicated in Fig. 21,
the clipped signal drives the timing preamplifier, which in turn drives a
resonant tank tuned to the baud frequency. The output has pulse pattern
dependent amplitude variations of approximately 14 dB; the
+0—0+0— • • • sequence gives minimum amplitude and the
H 1 1 — • • • sequence gives maximum amplitude. This signal is
amplified and limited to obtain a uniform high-amplitude sine wave,
which in turn is coupled to a pair of oppositely poled charge storage
diodes to generate the two required subnanosecond pulse trains.
4.2.1 The Resonant Tank
The tank is a loaded cavity with inductive loops for input and output
coupling. The loaded Q is 80. An exploded view in Fig. 26(a) shows its
1026 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1066
(a)
^wr-
^OQOr
FROM TIMING
PREAMPLIFIER
^WP^
TO TIMING
AMPLIFIER
(b) "
Fig. 26 — The resonant tank; (a) exploded view; (b) equivalent circuit.
construction, and the equivalent circuit in Fig. 26(b) shows its opera-
tion. The resonant frequency is stabilized with respect to temperature
by the use of invar for the center post. This frequency can be adjusted
over a narrow range by trimming the capacitance of the tank by ad-
justing the position of the disc attached to the trimming screw shown at
the left end of Fig. 26(a).
We allow a maximum of ±0.1 radian (5.7°) of static timing mis-
alignment due to tank mistiming with age (20 years), and temperature
(a range of 40°F). The phase shift, <p, of a high Q resonant circuit is
EXPERIMENTAL DIGITAL REPEATERED LINE
1027
given for small values by
<
where / is the resonant frequency and A/' is the mistuning. Hence, for a
Q of 80, the tolerance on the resonant frequency is
4f
±0.1
1G0
or ±0.0625 %
This level of performance is attained with the machined structure of
Fig. 20, provided that it is operated between well-controlled impedances.
4.2.2 Amplification in the Timing Path
For the pulse sequence with maximum timing energy, the baud
frequency component of the rectified timing signal has a peak amplitude
of 150 mV. For the sequence with minimum energy, this component is
30 mV, or 14 dB lower. The peak amplitude of the required timing am-
plifier output is 3 volts, or 40 dB above 30 mV. In addition, as will be
shown in Section 4.2.3, a minimum of 8 dB of limiting for the lowest
level signal is required for good performance. Further, the tank has a
loss of 0.5 dB at the baud frequency. Thus, linear gain of 40 + 8 +
0.5 = 48.5 dB is required in the timing path.
The 3-stage timing preamplifier, shown in Fig. 27, provides 22.5 dB
of gain, and the 5-stage timing amplifier, shown in Fig. 28, provides 26
(^Mr
TO
TIMING
Fig. 27 — Timing preamplifier circuit.
1028 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1966
ft
to
.9
to
EXPERIMENTAL DIGITAL REPEATERED LINE
1029
dB of gain at levels low enough not to produce limiting. Each stage has
about 8 dB of gain, but the amplifier includes a phase adjustment circuit
with 13 dB of loss, as described in Section 4.2.4. Both amplifiers employ
common base stages coupled by bifilar-wound transmission line auto-
transformers to provide current gain. All eight transistors are the 1-GHz
npn silicon type.
As indicated in the composite frequency response of Fig. 29, the band-
widths of the timing path amplifiers are quite broad. This broadband
design reduces the sensitivity of the timing path phase response to
variations in amplifier reactive elements with age and temperature.
4.2.3 Limiting
A series gate employing Schottky barrier diodes at the output of the
second stage of the amplifier is used to perform limiting, as shown on
Fig. 28. Transistor limiting was avoided in order to keep amplitude-to-
phase conversion at a minimum. The main cause of amplitude-to-phase
conversion in this series type of limiter comes from diode shunt capaci-
tance. With very large signals, significant reactive current flows through
this capacitance and advances the phase of the output wave. This effect
is kept small by the use of the shunt diodes to reduce the voltages
reaching the series diodes. The amplitude-to-phase conversion of the
limiter for the 7-mA diode current used is shown in Fig. 30(a).
Since the phases of the pulse trains generated by the charge storage
diodes are heavily dependent upon the sine wave amplitude, the flatness
of limiting shown in Fig. 30(b) is also important. Notice that an input
of 0.35 volts at the lower end of the 17 -dB range results in an output of
150
200
225 250 275
FREQUENCY IN MHZ
350
Fig. 29 — Composite frequency response of the timing path amplifiers.
1030 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1966
HI
If)
< I/)
1 w
Q. UJ
O a.
,/) Q
^Z
2 ~
(0 1-
Z U-
$ =
20
\
(a)
v_
- 0. 15
_l*
ill <
> UJ
UJ Q.
h I-
3 -I
X o
H >
0.0 5
(b)
r*
17 DECIBELS
RANGE
i
0.35 0.5 1.0
INPUT LEVEL IN VOLTS PEAK
2.0 2.5 3.0
Fig. 30 — Limiter performance; (a) transmission phase shift vs input signal;
(I)) output signal vs input signal.
0.14 volts. This corresponds to the 8 dB of minimum limiting referred to
earlier. The 17 -dB range is the sum of 14 dB due to timing wave am-
plitude variations and 3 dB due to loss variations in the phase adjust-
ment circuit to be described next.
4.2.4 Phase Adjustment Circuit
In order to set the sampling pulse at the center of the eye, a phase
adjustment is required in the timing path. The circuit at the input to
the timing amplifier (Fig. 28) was designed to permit a ±45° adjust-
ment range on the phase of the timing wave. The coaxial transmission
line provides 90° of phase shift between the input and the upper end of
the potentiometer. Due to the balun, there is 180° phase difference
between the upper and lower ends. Two currents are summed at the
emitter of the first amplifier stage. One current, I R , at reference phase
in the vector diagram of Fig. 31, comes directly from the input; a quad-
rature current, I Q , comes from the movable tap. By moving the tap
upward in the diagram, more phase lag is introduced and vice-versa.
Over the extreme range of the potentiometer, for which / Q is indicated
by the dashed lines, 3 dB of amplitude variation is introduced, an amount
which the limiter removes.
EXPERIMENTAL DIGITAL REPEATERED LINE
1031
-Ir
Fig. 31 — Vector diagram for phase adjustment circuit.
4.2.5 The Short-Pulse Generator
The circuit to generate the subnanosecond sampling and turn-off
pulses is shown in Fig. 32. A sine wave of current from the timing am-
plifiers flows through the diode in the forward direction, storing charge.
During this portion of the operation, the rather small forward voltage
drop appears across the diode. When the sinusoidal current reverses
polarity, the stored charge permits reverse conduction until the charge
is depleted, whereupon the diode current abruptly falls to zero. 1920 The
feed inductor current is abruptly switched from the diode to the load to
produce a rapid rise of current. The decay transient determining the
short pulse duration is established by the coupling network with the
diode open -circuited.
i — OKKT^
JUVJL
FROM
TIMING
AMPLIFIER
^\AA —
—^5W^
^WP-
-WV
SAMPLING PULSES
TURN-OFF PULSES
TTT
Fig. 32 — Short-pulse generator circuit.
1032 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER I960
4.3 Regeneration
In the regenerator, the ternary signal requires two amplitude thresh-
olds, which are obtained by providing two identical decision circuits
driven with oppositely phased signals from the balanced amplifier out-
put. Each decision circuit incorporates a two-input AND gate and an
Esaki diode, shown as part of Fig. 33. The signal is applied to one input
of the gate and subnanosecond sampling pulses to the other. With a
positive signal pulse present, the sampling pulse diverts the AND gate
current from the sampling pulse diode to the AND gate output, where
the Esaki diode is triggered to its high-voltage state. With either a zero
or negative input signal, the gate current flows toward the signal source,
and the Esaki diode remains untriggered. A subnanosecond negative
turn-off pulse, one-half baud interval after the sampling pulse, returns
the Esaki diode to its low- voltage state, establishing the repeater output
pulse duration of 2.2 nanoseconds.
The Esaki diode voltages of the two decision circuits are amplified in a
pair of current routing output stages, employing the 4-GHz germanium
transistors. The resulting balanced outputs of these stages are combined
to give the appropriate signal polarities at the repeater output.
> -8.4V
Fig. 33 — Regenerator circuit.
EXPERIMENTAL DIGITAL REPEATERED LINE 1033
4.3.1 Input Networks
The inputs to the regenerator are ac coupled. The shunt capacitors
(Fig. 33) reduce timing energy coupling onto the input signal leads.
Since the timing path begins at the amplifier output, such coupling feeds
energy back into the timing path, thereby increasing pattern dependent
jitter. The other elements of the input networks provide good 50-ohm
terminations for the amplifier signals.
4.3.2 The Threshold Circuits .
A biasing circuit fixes the dc level of the signal relative to the thresh-
old voltage, which is established by the Esaki diode. This threshold
voltage and the signal bias are related through the back-to-back diodes
of the AND gate to provide temperature tracking. The signal bias is
chosen to place the Esaki diode threshold voltage at the center of the eye.
A discussion of the operation of the threshold circuit follows. At the
threshold of triggering the Esaki diode, with the sampling pulse present,
equal current flows through diodes Di and D 2 of the AND gate. It can
be shown that this condition corresponds to maximum signal trans-
mission, which provides maximum regenerator sensitivity.
An applicable model of the Esaki diode and its sources, shown in Fig.
34(a), is the parallel combination of a signal current source, I s , a bias
current source, I B , a linear source resistance, a capacitance, and a non-
linear resistance whose static characteristic is shown in Fig. 34(b). The
difference between the load line current and the static characteristic
current is capacitive current I c of our model. Threshold voltage V A is
the voltage at point A, the unstable intersection of the load line and
the diode static characteristic. Initially, the voltage is V c , the low
voltage state. The subnanosecond sampled signal pulse charges the
capacitance, raising the voltage. If after this pulse has passed, the re-
sulting voltage is greater than V A , excess current is available to further
charge the capacitance, stable point B will be reached, and the decision
will be that a pulse was present. If the voltage is less than V A , the
capacitance will discharge, operation wall return to point C, and the
decision will be that no pulse was present.
For good dynamic performance — that is, high circuit speed at the
threshold level — the load line should intersect the diode negative re-
sistance at a steep part. For high circuit gain, on the other hand, the
load line should be raised toward the peak so that smaller AND gate
current can be used (the amplifier must supply a peak-to-peak signal
current equal to the gate current) and larger voltage can be obtained to
1034 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER I960
— o
fills Ieltl Ir|| Ic|
SOURCE ESAKI DIODE
EQUIVALENTS EQUIVALENT
(a)
Cf* — A
L — ~»»^ ■it
\ *~ ^^^
\ ^"^^-i
\ J c 7
\ ■ /
\ i /
\ i /
\ i /
\w /
1 \ /
1 \ /
1 \ /
J D \^_^
l
1
»
/
(b)
Fig. 34 — Esaki diode model; (a) equivalent circuit; (b) volt-ampere charac-
teristic of nonlinear resistance.
drive the current routing stages. Further, for stability against changes in
diode peak current with age and temperature, biasing current I B should
be small and the gate current should be large. As a suitable compromise
among these factors, a bias current of 8 mA and a gate current of 4.5
mA were chosen in conjunction with a gallium arsenide Esaki diode
having a peak current, I P , of 10 mA.
The input-output dynamic regenerator characteristic was calculated
using the equivalent circuit of Fig. 34(a) and a piece-wise linear approxi-
mation to the diode characteristic of Fig. 34(b). The performance was
measured for six regenerators in the experimental setup of Fig. 35(a).
A comparison of the limits of the measured characteristics and the calcu-
lated characteristic is shown in Fig. 35(b). It was indicated in Section
3.2.2 that the total area and not the amplitude of the regenerator output
pulse primarily controls the amplitude of the pulse arriving at the sub-
EXPERIMENTAL DIGITAL REPEATERED LINE
1035
sequent regenerator after transmission and equalization. Hence, the
use of the simulated equalized line for T{f) permits the proper com-
parison of equalized amplitudes, V out .
4.3.3 Output Amplifiers
The emitter-coupled current routing pair of Fig. 33, which amplifies
the Esaki diode voltage, has five important features. First, the circuit is
fast because of the prevention of saturation. Second, it performs the
inversion required for the negative pulse decision circuit. Third, its
nonlinear forward transmission characteristic provides additional am-
plitude regeneration. Fourth, it has good dc temperature stability due
£
h\
y\ &<
REGEN-
ERATOR
SIMULATED
EQUALIZED
LINE T(f)
v ouy A
Vm
>
t ^
(a)
MEASURED /
'-"
■^^
^ =S *"^
0.9
LIMITS OF SIX^-
i
REGENERATORS
"V£>= CALCULATED FROM
r
"*- ESAKI DIODE
0.8
0.7
o
Ul
^ 0.6
•!
2
S 05
Z
t
t
J"
0.3
0?
G 1
I
I
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
V, N (normalized)
(b)
Fig. 35 — Input-output dynamic regenerator characteristic; (a) experimental
setup; (b) measured and calculated characteristic.
1036 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER I960
to the oppositely poled emitter junctions between the signal and the
reference inputs. (The Esaki diode signal voltage is a unipolar pulse
stream which includes a dc component.) Fifth, to a first approximation,
signal currents flow equally and oppositely in the two output leads and
thus no current from the output flows through the ground system of the
repeater. Many kinds of repeaters powered serially over the trans-
mission line suffer from feedback problems due to such currents.
The output collectors of the two current-routing circuits are paralleled
into balanced 75-ohm loads, to which they deliver peak output voltages
of 1.5 volts of each polarity.
4.4 Secondary Features
4.4.1 Power Arrangement
Repeaters are powered serially by dc over the center conductor of the
coaxial. The line current is 450 mA. Power supply voltages in the re-
peater are obtained by passing about 50 mA of this current through a
series pair of 8.4-volt Zener diodes as shown in Fig. 36. Thus, each re-
peater consumes 7.5 watts.
In Fig. 36, we show only the circuit elements required to separate the
dc power from the signal. Feedback from output to input is attenuated
REPEATER
3 b
+L
REPEATER
CIRCUITS
+8.4V
■w-
T
-e
REPEATER
OUTPUT
*h
X
ISOLATION FILTER
X
X X
ISOLATION FILTER
Fig. 36 — Repeater powering circuit.
EXPERIMENTAL DIGITAL REPEATERED LINE 1037
in the power circuits by greater than 130 dB over the signal frequency
range.
It is difficult to prevent spurious signals from appearing between earth
ground and local repeater ground. In a long system, these two grounds
may differ by as much as 1000 volts dc and capacitors with adequate
voltage rating have appreciable impedance at frequencies of interest.
Filtering inductors are required to prevent these ground-to-ground
voltages from affecting the sensitive repeater circuits. The philosophy
here is to isolate the repeater circuits from earth ground as much as
possible.
4.4.2 Surge Protection
Partial surge protection has been provided at both the input and the
output of the repeater. At the input, a series diode gate (Fig. 24), with 4
mA of current through each diode, limits the surge at the base of the
first transistor. At the output, Equalizer I reduces the low frequency
power which could harm the transistor collectors. These surge protection
features arc laboratory precautions only. Complete protection against
lightning and power surges has not been accomplished in this experi-
mental repeater.
4.4.3 Repeater Equipment Design
The repeater circuits shown on Fig. 21 were constructed on printed
wiring boards which plug into a 3-layer printed wiring interconnecting
board. The plug-in boards are attached to aluminum backing plates
which provide support and an electrical ground plane for the circuits.
The plates slide into the grooved sides of pockets in an aluminum in-
vestment casting. Fig. 37 is a photograph of a repeater showing the
circuit boards in place in the casting, except for the amplifier board
which has been removed. Shielding covers have been removed and are
not shown.
Printed wiring carries shielded dc power to the individual circuit
boards from the power supply on the intercomiecting board (at the back
of the casting in the photograph). Signal interconnections are made by
miniature coaxial transmission lines to provide shielding.
Tantalum thin-film integrated techniques were considered for the
circuits of the repeater, and there appear to be no fundamental ob-
stacles to their use. A thin film version of the preamplifier was built and
performance was equal to or better than the printed wiring version.
The size of this repeater is approximately 4x5x11 inches for a volume
1038 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 190(1
p A
o O
13
n
h
Si
fe
EXPERIMENTAL DIGITAL REPEATERED LINE 1039
of 220 cubic inches. By design, there is access space for testing of the
experimental repeaters. It is anticipated that the use of integrated cir-
cuits and the elimination of excess space will result in a design occupying
about one-quarter of the volume.
V. EXPERIMENTAL PERFORMANCE
In this section, we report on the performance of the line under labora-
tory conditions. In general, the line has met all performance expectations
under these conditions. For volume manufacture, however, and for
operation under field conditions for many years, further development is
required. By the work reported on here, we have established the technical
feasibility upon which a design for service can be based.
5.1 Error Rate
The line operates at error rates below 10~~ 10 errors per baud through
ten repeatered links.
5.2 Jitter
According to our model, each repeatered link introduces a pattern
dependent, or systematic, component of jitter which is dominant. Pat-
tern independent jitter tends to be random at each repeater; hence, it
accumulates much more slowly and is negligible at the end of a long
chain.
Since pattern dependent and pattern independent jitter are indis-
tinguishable by measurement of a single link, we measure the jitter at
the end of the chain. To determine the systematic component for the
single link, we apply an equation derived from the model: 8
where 0i is the systematic rms jitter arising in each link, 6 N is the syste-
matic mis jitter at the end of N links, and P(N) is given by
P(tf)-?- (2N - 1)l
2 4"[(tf - l)!] 2
This function is tabulated in Ref. 8. For ten links, the above expression
becomes
0i = 0.247 dm .
1040 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 19G0
For only ten repeaters, the nonsystematic component may not be
negligible, so that this calculation gives only an approximation to the
systematic jitter contribution per link.
The total measured jitter at the end of our ten links is 13.3° rms. If
we assume this to be all systematic, we calculate a per-link systematic
jitter contribution of 3.3° rms. This laboratory measurement is well
within the 8° objective of Section 3.3.2.
Fig. 38 shows the measured accumulated jitter versus the number of
repeatered links. The intercept (N = 0) accounts for jitter introduced
in the transmitting and receiving repeaters at the ends of the line.
5.3 Waveforms
In Fig. 39 we show some of the key waveforms in the repeater. All the
waveforms are aligned in time for clarity. Fig. 39(a) shows the regenera-
tor output before Equalizer I; 39(b), the signal input to the positive
decision threshold of the regenerator; 39(c), the eye at this point when
the line is driven by a random binary sequence; and 39(d), the sub-
nanosecond sampling and turn-off pulses. For instructional purposes,
the eye diagram has been synchronized at an even submultiple of the
baud frequency to show the paired nature of the PST signal. The tim-
ing extractor causes the distortion in the negative eye. This distortion is
of no consequence since only the positive eye is used by the positive
threshold.
T 6
2
a. a
_i
<
5 2
1
^^\
5"^
/o
2 4 6
N, NUMBER OF REPEATERS
Fig. 38 — Measured accumulated jitter (rms) vs the number of repeatered links
in the experimental line.
EXPERIMENTAL DIGITAL REPEATERED LINE
+ - + +
1041
(a)0.5V/CM
!b) 0.25 v/cm
(C) 0.25 V/CM
ff w w w w
« 1 ft 1 Mi Ni
(d) i.o v/cm
2.5 ns/CM
Fig. 39 — Key waveforms in the repeater; (a) regenerator output before
Equalizer I; (b) signal input to the positive decision threshold; (c) the eye at the
positive decision threshold; (d) the subnanosecond sampling and turn-off pulses.
1042 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1966
VI. CONCLUSION
Ten 224-Mb/s experimental digital repeaters and associated code
translation equipment have been developed, constructed, and operated
over 10 miles of 0.270-inch coaxial line under laboratory conditions. The
resulting performance has indicated that such a line 4000 miles in length
is feasible for actual service and can be designed with existing techniques.
The transmission code is paired selected ternary (PST) which provides
the essential features for the repeater operation as well as for in-service
error monitoring.
Each repeater employs 25 transistors, most of them of a pnp ger-
manium planar epitaxial design with an f T of 4 GHz. The decision ele-
ments are gallium arsenide Esaki diodes.
The repeaters are serially powered by dc over the line with 450 mA
of current, and each repeater consumes 7.5 watts.
Good agreement among theory, simulation, and laboratory perform-
ance has been achieved throughout. The error rate per repeatered link
under laboratory conditions is less than 10 -11 and the systematic jitter
introduced in each link is about 3° rms.
VII. ACKNOWLEDGMENT
The work reported on herein was performed over a period of several
years primarily by the members of the PCM repeater department, under
supervision of the authors. Many individuals have made significant
contributions, but specific mention of these is not practical here. Prior
work under the supervision of R. V. Sperry is gratefully acknowledged.
The experimental cable was designed and manufactured through the
efforts of Bell Laboratories' Outside Plant Laboratory, and Western
Electric Company's Engineering Research Center and Baltimore Works.
Support is also appreciated from many other departments within Bell
Laboratories that have made contributions in areas such as systems en-
gineering, components, devices, networks, and power. The work is based
on earlier efforts over many years in the research department.
REFERENCES
1. Oliver, B. M., Pierce, J. R., and Shannon, C. E., The Philosophy of PCM,
Proc. IRE, 86, November, 1948, pp. 1324-1331.
2. Davis, C. G., An Experimental Pulse Code Modulation System for Short Haul
Trunks, B.S.T.J., 1,1, January, 1962, pp. 1-24.
3. Fultz, K. E. and Penick, D. B., The Tl Carrier System, B.S.T.J., U, September,
1965, pp. 1405-1451.
4. Travis, L. F. and Yaeger, R. E., Wideband Data on Tl Carrier, B.S.T.J., 44,
October, 1965, pp. 1567-1604.
EXPERIMENTAL DIGITAL REPEATERED LINE 1043
5. Mayo, J. S., Experimental 224 Mb/a PCM Terminals, B.S.T.J., 44, November,
1965, pp. 1813-1841.
6. Edson, J. O. and Henning, H. H., Broadband Codecs for an Experimental 224
Mb/s PCM Terminal, B.S.T.J., 44, November, 1965, pp. 1887-1940.
7. Witt, F. J., An Experimental 224 Mb/s Digital Multiplexer Using Pulse
Stuffing Synchronization, B.S.T.J., 44, November, 1965, pp. 1843-1886.
8. Byrne, C. J., Karafin, B. J., and Robinson, D. B., Jr., Systematic Jitter in a
Chain of Digital Regenerators, B.S.T.J., 42, November, 1963, pp. 2679-2714.
9. Sipress, J. M., A New Class of Selected Ternary Pulse Transmission Plans for
Digital Transmission Lines, IEEE Trans. Com. Tech., Com-13, September,
1965, pp. 366-372.
10. Elmendorf, C. H., et al., The L3 Coaxial System, B.S.T.J, 32, July, 1953, pp.
781-1005.
11. Aaron, M. R., PCM Transmission in the Exchange Plant, B.S.T.J., 41, Januarv,
1962, pp. 99-141.
12. Ross, W. L., Private Communication.
13. Aaron, M. R. and Tufts, D. W., Intersymbol Interference and Error Proba-
bility, IEEE Trans. Inform. Theor., IT -12, January, 1966, pp. 26-34.
14. Smith, J. W.. The Joint Optimization of Transmitted Signal and Receiving
Filter for Data Transmission Svstems, B.S.T.J., 44, December, 1965, pp.
2363-2392.
15. Bennett, W. R., Methods of Solving Noise Problems, Proc. IRE, 44, May,
1956, pp. 609-638.
10. Cravis, H. and Crater, T. V., Engineering of Tl Carrier System Repeatered
Lines, B.S.T.J., 42, March, 1963, p. 436.
17. Chapman, R. C, Private Communication.
18. Waldhauer, F. D., Wideband Feedback Amplifiers, IRE Trans. Circuit Theor.,
CT-4, September, 1957, pp. 178-190.
19. Goodall, W. M. and Dietrich, A. F., Fractional Millimicrosecond Electrical
Stroboscope, Proc. IRE, 48, September, 1960, pp. 1591-1594.
20 Moll, J. L., Krakauer, S., and Shen, R., P-N Junction Charge-Storage Diodes,
Proc. IRE, 50, January, 1962, pp. 43-53.
Live Music Archive
Librivox Free Audio
Metropolitan Museum
Cleveland Museum of Art
Internet Arcade
Console Living Room
Books to Borrow
Open Library
TV News
Understanding 9/11