1975 , Volume , Issue May-1975

1975 , Volume , Issue May-1975
MAT 1975
HEWLETT-PACKARDJOURNAL
© Copr. 1949-1998 Hewlett-Packard Co.
An Understandable Test Set for Making
Basic Measurements on Telephone Lines
This new portable test set uses a digital processor to pre
sent direct-reading, autoranged measurements of level and
frequency, substantially reducing operator errors. Other
conveniences simplify set-up and operation.
by Michael B. Aken and David K. Deaver
CONTINUING EXPANSION OF THE world's tele
phone network intensifies the need for means of
testing telephone lines expediently. Now more than
ever, with the growth of data communications, oper
ating companies and end users need test equipment
that can check telephone lines quickly without re
quiring a lot of personnel training.
The HP Models 3551A (Fig. 1) and 3552A (Fig. 2)
are dedicated test sets designed to fill this need.
These instruments make basic measurements on
voice-grade lines according to North American
(Model 3551A) and CCITT (Model 3552A) standards.
They measure tone level, noise level, and frequency,
from which they obtain measurements of loss, atten
uation distortion, message-circuit noise, noise with
tone, noise to ground, and single frequency inter
ference.
Both test sets include an oscillator, a frequency
meter, a level meter, and the various filters required
for voice-channel measurements. They can send a
test signal while simultaneously measuring it in a
loop-back set-up, or two can be used as a pair for one
way measurements. Combining the send and receive
functions in one box makes the test set easier to carry
and it also speeds measurements by making it pos
sible to switch between send and receive without re
connecting anything and while maintaining the
telephone line in an "off-hook" condition.
Design goals achieved with the realization of these
test sets were:
Reduction of operator set-up time because of an
easily-understood front panel;
Reduction of operator errors because of the autoranged, direct reading digital display;
Increase in accuracy because of the low-distortion
test signal with settability within 1 Hz;
Increase in convenience because of the compact
ness, light weight (13 Ibs), and choice of battery or
line power.
Front Panel Clarity
A careful look at the front panel (Fig. 1) discloses
how the telephone craftsman can tell the status of the
instrument at a glance, and modify that status quickly
with a minimum likelihood of making a wrong move.
The right portion of the panel has the controls for
the send unit. Frequency is selectable within a 40-Hz-
Cover: Telephone channels
need to be tested quickly,
especially when restoring
service on thousands of lines
following a major disruption.
Our thanks to the New York
Telephone Company for the
background photo of crafts
men cleaning 16,000,000
relay contacts contaminated by the recent fire.
The instrument is Model 3551 A Telephone Test
Set, a number of which were used in restoring
service. Also described in this issue is Model
5453 A, a programmable, computerized tele
phone test system that uses digital signal analysis.
In this Issue:
An Understandable Test Set for
Making Basic Measurements on Tele
phone Lines, by Michael B. Aken and
David K. Deaver
— page 2
A Computer System for Analog
Measurements on Voiceband Data
Channels, by Stephan G. Cline, Robert
H. Perdriau, and Roger F. Rauskolb . page 10
A Precision Spectrum Analyzer for the
10-Hz-to-1 3-MHz Range, by Jerry W.
Daniels and Robert L. Atchley page 18
e Hewlett-Packard Company. 1975
Printed in U.S. A
© Copr. 1949-1998 Hewlett-Packard Co.
Fig. 1. Model 3551 A Transmis
sion Test Set makes basic mea
surements on telephone lines
according to North American
standards. The front panel is
arranged to clarify operation of the
instrument. The digital display is
autoranged and automatically
compensated for the selected
impedance.
to-60-kHz range and the amplitude is adjustable from
-60 to +10 dBm. None of the controls is calibrated
because the built-in measuring circuits enable read
out of frequency and level with better resolution and
accuracy than dial markings could provide. The fre
quency range switch also includes a position for
quick selection of the fixed holding tone used in
noise-with-tone measurements.
The central portion of the front panel is concerned
with the type of measurement to be made on the re
ceived signal. One switch selects the measurement
to be made (tone, noise with tone, message circuit
noise, noise to ground) while the other switch selects
one of four weighting networks. Measurements can
be made on tones that have amplitudes ranging from
— 70 dBm to +15 dBm or on noise that ranges from
0 to +125 dBrn.
The left portion is concerned with the physical con
nection to the telephone lines. Two sets of terminals
are provided to permit simultaneous send/receive
measurements, each set consisting of a standard
phone jack in parallel with a pair of 5-way binding
posts. Either set may be used for measurements on a
single pair. The FUNCTION switch selects the role of
each set of terminals (send or receive) and it enables
the roles to be interchanged without requiring any
disconnecting and reconnecting of the lines. It also
determines whether the receive line is to be bridged
or terminated. A concentric switch selects the load
impedance for a terminated line (and also establishes
the send source impedance).
A set of clip posts for connecting a lineman's hand
set can be switched in parallel with the left set of ter
minals for dialing up a connection. A holding circuit
is provided so a connection to a "wet" line can be
held in an off-hook condition while the line is used
for either send or receive measurements.
The readout is an LED display that gives four-digit
resolution in frequency, three-digit resolution in tone
level, and two digit resolution in noise level measurerhents. The quantity displayed (receive level or
frequency, or send level or frequency) is selected by
pushbuttons. Both frequency and level measure
ments are autoranging and automatically compen
sated for the impedance selected to give fast, direct
readout of the measured quantity. No mental calcula
tions are required on the part of the operator.
A monitor loudspeaker helps the craftsman iden
tify single-frequency interference and, by the char
acter of the sound, the source of other types of inter
ference.
Battery and AC Power
Each of the instruments has internal rechargable
Fig. 2. Model 3552/4 Transmission Test Set is fundamentally
identical to the Model 3551 A but has connectors, filters, and
impedance levels that conform to CCITT standards.
© Copr. 1949-1998 Hewlett-Packard Co.
AutoAttenuator
15 kHz Flat
Fig. 3. processing block diagram of the 3551A/3552A Transmission Test Sets. Signal processing
is analog up to the detector, then processing becomes digital. The hold circuit is a current source
that appears as a broadband high impedance to the telephone line while supplying the current
necessary to hold central office relays. The diode bridge, protected against high line transients
by the gas-discharge tube, functions as an automatic polarity switch for the hold circuit.
batteries that can power the instrument for four to
six hours on one charge. The instruments can also
operate on ac lines of 100, 120, 220, or 240 volts.
In the new test sets, a level-sensing circuit moni
tors the battery voltage and shuts off the instrument
whenever the voltage falls below a useful level. This
prevents erroneous readings and it also prevents
cell reversal from deep discharge.
This arrangement is considered preferable to
meter monitoring because the NiCad batteries used
maintain a fairly constant voltage during use and
then lose voltage rather rapidly as they approach to
tal discharge. With the usual meter monitoring, the
operator would have to check the battery voltage
quite frequently to avoid overlooking the onset of the
rapid voltage fall.
Internal Details
The two instruments are fundamentally the same
except for certain characteristics that conform to the
telephone measurement standards where the instru
ments are to be used. The Model 3551A has imped
ances, weighting filters, and a hold tone that conform
to standards established by the telephone industry in
North America. In most of the rest of the world, stan
dards are set by the International Telegraph and Tele
phone Consultative Committee (CCITT) and the
Model 3552A conforms to these standards.
The block diagram shown in Fig. 3 applies to both
instruments. The received signal is filtered and de
tected in conventional analog fashion. The output of
© Copr. 1949-1998 Hewlett-Packard Co.
the detector, however, is converted to a proportionate
time interval in the logarithmic converter. The digital
circuits measure this time interval to get a digital
indication of the input signal level. From here on, the
measurement information is manipulated digitally,
with appropriate factors added so the number dis
played gives the measurement in the desired units.
In a frequency measurement, the signal frequency
is multiplied in a phase-lock loop so the counting
circuits can accumulate enough counts in 50 millise
conds for 4-digit resolution. This enables a 10-per-second sampling rate, even at low frequencies.
One of the significant differences between these in
struments and other telephone test sets is the use of a
function generator as the send unit, rather than the
traditional RC oscillator. The use of a function gener
ator achieves significant cost and space savings. The
basic function generator circuit was modified, how
ever, to obtain a sine wave with very low distortion.
Triangles First
A schematic representation of the send unit's func
tion generator is shown in the box on page 6. As in
other function generators, a triangular waveform is
generated by using constant currents to alternately
charge and discharge a capacitor, shown as range ca
pacitor CR in the diagram. The triangular wave is
shaped into a sine wave by a nonlinear network.
Diodes in this network are biased at progressively
higher levels so more and more attenuation is switched
in as the triangular waveform moves towards its posi
tive or negative limit. By suitable choice of the attenu
ation switched in by each diode, the triangle is
rounded off to a sine wave.
Maintaining high sine-wave purity under field
conditions required some modifications to the basic
waveform-generating mechanism. To begin with,
the sine shaper is compensated for temperature
changes by diodes that adjust the voltage on the bias
ing networks to compensate for the change in diode
forward voltage drop caused by temperature changes.
For optimum operation of the sine shaper, the up
slope of the triangle wave must be exactly equal to
the down slope and the waveform must be centered
on the zero level (equal positive and negative excur
sions). Circuits for maintaining these conditions are
designed into the function generator. These are de
scribed in the box on page 6.
As a result of these measures, total harmonic distor
tion in the send unit output is more than 50 dB below
the fundamental within the telephone voice band
and at least 40 dB below outside that band. These
specifications are held over an operating tempera
ture range of 0°C to 55°C.
Level Detection
As noted in the block diagram of Fig. 3, the re
ceived signal is appropriately filtered and adjusted
Logarithmic
Converter
V, = dc Proportional to
Signal Level
Fig. 4. quasi-rms combines the outputs of peak and average detectors to derive a quasi-rms
indication. The detector output is converted to a proportionate time interval by the logarithmic
converter.
© Copr. 1949-1998 Hewlett-Packard Co.
A Function Generator with a
Well-Defined Output
The triangular waveform is shaped into a sine wave by the
nonlinear network described in the text preceding. To assure
proper operation of the sine shaper, the triangular waveform
must be symmetrical. Equal positive and negative waveform ex
cursions (x-axis symmetry) are assured by integrating the trian
gular waveform and using the resulting dc level to modify the
lower steady-state level in the bistable switch. The waveform's
peak negative excursion is thus adjusted to equal the magni
tude of the peak positive excursion.
Maintaining equal up and down slopes (y-axis symmetry) is
accomplished by sensing the symmetry of the square wave. Re
ferring to the y-axis symmetry circuit in the diagram, current lb
flows into the adjacent integrator when the square wave is at
the upper level, reverse-biasing diode D1. When the square
wave is at the lower level, the reverse bias is removed from
diode D1, allowing the lower current source to draw a current
equal to 2IS through diode D1 , with half of the current (ls) being
drawn from the integrator. Thus, as long as the waveform is sym
metrical, the average integrator output is zero. If the waveform
were not symmetrical, a net charge would remain on the inte
grator output, which would add to or subtract from the charging
current in the triangle generator. The up slope of the triangle
is thereby altered to make it match the down slope.
Definite space-saving and cost advantages result from using
a function generator as the send unit's oscillator in the Model
3551 A/3552A Transmission Test Set. Certain modifications, how
ever, had to be made to the basic circuit to assure the wave
form purity required for telephone tests.
A simplified diagram of the function generator is shown be
low. Positive and negative currents are switched by transistors
Q1 and Q2 to alternately charge and discharge capacitor CR,
thereby generating a triangular waveform. To minimize switch
ing transients, steady current flow is maintained by operating
transistors Q1 and Q2 in a bridge configuration with Q3 and
Q4. This arrangement sinks one current to ground while
the other is charging the capacitor and vice-versa.
The transistors are turned on and off by a bistable switch that
changes states when a comparator indicates that the triangle
waveform has reached the same level as that being held by the
switch. The amplitude of the triangular waveform is thus deter
mined by the bistable switch, a precision circuit that maintains
a well-defined level in either of its two states.
The frequency is determined by the rate at which the capaci
tor charges and discharges. The vernier frequency control
changes the charge and discharge currents. Ranges are
changed by switching in capacitors of different values.
X-Axis Symmetry
3.3V
3.3V
capacitor to the negative peak of the waveform. The
average detector integrates negative half-cycles of
one waveform to get an average value. The summing
resistors at the input to the integrator determine the
ratio of peak to average in the combined result. Dur
ing tone measurements, only the average detector is
used.
in gain before being applied to the detector. For noise
measurements, the detector functions as a quasi-rms
type that derives the equivalent of an rms measure
ment by combining the outputs of peak and average
detectors in accordance with Bell System Technical
Reference PUB 41009.
The peak detector (Fig. 4) is a diode that charges a
6
© Copr. 1949-1998 Hewlett-Packard Co.
To
Digital
Counter
From
Attenuator
Amplifier
Fig. 5. Frequency multiplier
generates an output two times the
input for inputs of W kHz and
higher, and 20 times higher for
inputs below 10 kHz.
The detected dc level is then applied to the log con
verter. This compares the detectors' dc level to the
voltage on a discharging capacitor, generating a
pulse when the two are equal. Ten times per second
during tone level measurements, switch Si (Fig. 4)
closes long enough to charge capacitor C1 to a fixed
level. When Si opens, Ct discharges and when it
falls to the level of the detector voltage, a pulse is
generated. The time interval between the opening of
Si and the comparator pulse is thus inversely propor
tional to the absolute value of the input voltage and,
since the C^ discharge curve is exponential, to the
logarithm of the input voltage. The digital circuits
measure this time interval to derive the number for
display. The 100-kHz clock frequency gives a mea
surement resolution of 0.02 dB, which is truncated
to give a display resolution of 0.1 dB.
During noise measurements, an additional lowpass filter is switched in at the input to the log con
verter to provide more averaging for the noise sig
nal. Switch Si is then activated only two times per
second.
The digital circuits also use the output of the log
converter to sense when the attenuator range should
be changed. If the time between the opening of switch
Si and the comparator pulse is less than 5 ms (sig
nal too large), the attenuator is up-ranged. If the time
is greater than 20 ms (signal too low), the attenuator
is down-ranged. The 15-ms interval between these
points is equivalent to 15 dB, giving a comfortable
overlap of the 10-dB range on each attenuator step.
The attenuator consists of resistive dividers with
taps switched by an eight-channel analog multi
plexer under control of the digital system. The con
trol sequence is such that an amplitude measurement
and range correction is always made before a fre
quency measurement is made. The frequency mea
suring circuits are thus assured of a suitable signal
level.
Fast-Responding Frequency Measurements
To get 1-Hz resolution in a conventional frequencycounter measurement of an audio frequency, say
4 kHz, a one-second counting time is required. A
sampling rate of 10 per second was desired for the
Models 3551 A/3552A so the results of adjustments to
the telephone line are immediately apparent to the
craftsman. The counting time was thus made 50 milli
seconds, which allows time for an amplitude mea
surement within each 100-millisecond measurement
interval. To achieve 1-Hz resolution in this time in
terval, input frequencies lower than 10 kHz are mul
tiplied by a factor of 20 before counting. Input fre
quencies of 10 kHz and higher are multiplied by 2,
giving 10-Hz resolution with the four-digit counting
circuits.
The frequency-multiplication circuit is shown in
Fig. 5. A commercially-available integrated-circuit
phase-lock loop is at the core of the circuit. It has a
voltage-controlled oscillator controlled by a phase
detector that compares the input signal to a divideddown version of the VCO output frequency. The VCQ
is thus locked to a multiple of the input frequency.
Digital Control
The operating simplicity of these instruments re
sults from the use of a digital processor to manipu
late the raw measurement information. The proces
sor monitors the signal level and frequency and the
front-panel control settings, and uses this informa
tion to derive control signals for the measurement
routines and for the display.
The digital processor (Fig. 6) is an algorithmic
state machine (ASM) that uses MSICs (medium-scale
integrated circuits) and 8000 bits of memory in lowcost ROMs. It is divided into two parts, a control sec
tion and a display section, although functions over
lap somewhat.
A four-digit counter that is part of the display section
performs measurements of both level and frequency.
It counts 100-kHz pulses for the duration of the log
converter's measurement interval to derive a digital
number proportional to signal level. During the
50-ms frequency measuring interval, it counts the
output of the frequency-multiplier circuit (Fig. 5).
At the end of a frequency measurement interval,
the counter's contents are latched into a register and
held there for direct application to the display. The
contents of the latch register are scanned digit-by-digit and applied to a seven-segment decoder that
drives the display LEDs one at a time. Scans occur
2500 times per second.
© Copr. 1949-1998 Hewlett-Packard Co.
Clock
Range
Up Down
Counter
•^^
From Log Converter
Input Qualifier from
Display and Receive
Switch Controller
Output
Qualifier and
Output Select Lines
Digit Activation Lines
V7
Measurement
Interval
Programmed
for Digit
Scanning
Time Base
and Clock
Circuits
Frequency
Multiplier
Frequency
Underrange
to Controller
Qualifier
Select Block
From
Frequency
Multiplier
• Range Block to Automatic Amplitude Ranging Circuit and Qualifier Select Block
Display Control Lines
Fig. 6. Simplified block diagram of Test Set's digital circuits.
the qualifier select block, it monitors timing signals
and determines the proper frequency range and the
points at which autoranging is initiated. It partitions
the 100-ms sampling interval into the 25-ms signal
amplitude measurement interval and the 50-ms fre
quency measurement interval. It issues steady-state
command voltages for opening and closing
switches, and pulsed commands for initiating timerelated functions.
At the end of an amplitude measurement, the regis
ter contents are used as an address to find a new
number in the display ROM according to the mea
surement being made. For example, in tone measure
ments, a 0-dBm input signal results in a log converter
time interval of 10 ms (1000 counts). This is dis
played as 0 dBm. A higher count (longer time inter
val, lower amplitude) is interpreted by the display
ROM as a larger number (dB below 0 dBm) and it
adds a minus sign to the display. A lower count
(shorter time interval, higher amplitude) is also dis
played as a larger number (dB above 0 dBm) but
with a plus sign. Although the counter is unidirec
tional, it appears to behave as a reversible counter be
cause of the ROM processing.
Noise measurements are processed in a similar
manner so the correct number is displayed. In the
case of measurements of noise-to-ground, the ROM
adds 40 dB to the measurement number to correct the
reading for the signal attenuation caused by the
noise-to-ground input configuration (see Fig. 3). The
craftsman is never required to add this correction
mentally.
Using firmware to process the numbers, rather than
adders, registers, and other hardware, reduced the
parts count significantly and hence reduced the cost.
The control section of the digital processor uses a
4K ROM that contains all of the instructions for the
level, noise, and frequency measurements. Through
Test and Calibration
The control section ROM also contains instruc
tions for a progressive series of tests for troubleshoot
ing both the digital components and analog subsys
tems. These are internally-accessible tests, which
means that they cannot be accessed by the operator
from the front panel but they can be used as a trouble
shooting tool by maintenance personnel, either in
the shop or in the field. Tests may be conducted in
either a fixed informational display mode, which
indicates proper operation of the test routines and
thus the instrument, or a dynamic response mode
using an oscilloscope, which allows individual IC's,
circuits, and subsystems to be examined at normal
operating repetition rates. Some of the tests that can
be performed are:
• High-speed exercising of all transfers, resets, com
mand outputs, and flip-flops to assure their proper
operation.
8
© Copr. 1949-1998 Hewlett-Packard Co.
points than would have been possible manually
while at the same time reducing test time. The sys
tem automatically sets a digital voltmeter and a
synthesizer to the proper ranges, checks the frontpanel control settings of the instrument under test,
tells the technician the adjustment to make, and
checks that the calibration is within prescribed limits.
• Monitoring the display ROM outputs to check
proper operation of the ROM and its associated cir
cuitry.
• Fast or slow repetitive operation of the analog
autoranging circuitry to check proper gains and tran
sient responses.
One of the cost-saving features of these instru
ments is the inclusion of two test sockets that allow
connection to a calculator-controlled test system.
The inclusion of a digital controller in the Models
3551A/3552A made it relatively simple to make most
of the pertinent information available to the HP Inter
face Bus.1 It was thus possible to design a calculatorcontrolled test system that assures that the necessary
calibrations are performed. It also checks more
Acknowledgments
The authors would like to acknowledge the contri
butions of Ray Hanson for the design of the send oscil
lator and as the group leader of the 3551A. Dick Huff
man provided the mechanical design. S
Reference
1. D.W. Ricci and P.S. Stone, "Putting Together Instru
mentation Systems at Minimum Cost," Hewlett-Packard
Journal, January 1975.
David K. Deaver
Dave Deaver joined HP in 1969
after earning his BSEE degree at
Washington State University.
Since being with HP, he con
tributed to the designs of the
Models 3575A Gain/Phase Meter
and the 3570A Network Analyzer
be?ori? jninina the 3551A project.
He is working towards his MSEE
degree at Colorado State Univer
sity in the HP Honors Co-op pro
gram. Dave plays basketball in
the Loveland, Colorado, com
munity intramural program and
carries this activity over into
community youth work, refereeing Saturday morning basket
ball games. He and his wife have one daughter, 2.
Michael Aken
With HP since 1966, Mike Aken
worked on voltmeters (3450A and
3480A) and the 3570A Network
Analyzer before joining the 3551 A
project. A graduate of the Univer
sity of Wisconsin (BSEE), he
earned his MSEE at Colorado
State University in the HP Honors
Co-op program. Married, and
with three children, 10, 8 and 5,
Mike enjoys gardening, both
outdoor and indoor with fluores
cent lights, and he plays some
tennis.
SPECIFICATIONS
Model 3551 A 3552 A Transmission Test Sal
Receive Section
Noae-Mtn-tone -SOdBmlo 5 dBm (60011. 900Ã1)
Noae-to-growna - SO dBm 10 * 35 dBm'
Resolution 1 OB
Sample rite 2 second
Detecto) type RMS responding
ACCURACY
Message orcmt na»e -1 dB (-70 dBm 10 -5 dBm). -2 dB (-90 dBm to
- 70 dBm)
None-mttvtone ±idB( -70 dBm to -5 dBm} *2dB( -SOdBmto 70 dBm)
Nose-to-grotnd • ' dB ( - 30 dBm to * 35 dBm) - 2 dB I - 50 dBm 10 - 30
dBm)
WEIGHTING FILTERS Teiephon.(CClTTP»ophometnc) ShHziat iSXHifiai
programme (CCITT)
FREQUENCY MEASUREMENTS
FREQUENCY RANGE 40 Hi to 00 kHz
DYNAMIC RANGE - 1 5 dBm to - 70 dBm
RESOLUTION 1 Hz (40 Hz to 10 kHz}. 10 Hz ( 10 kHz to 60 kHz)
SAMPLE RATE 10/eecond
ACCURACY - 1 count
LEVEL MEASUREMENTS
FREQUENCY RANGE 40 Hi to 60 kHz
DYNAMIC RANGE - 1 5 dBm lo - 70 dBm
RESOLUTION 0 1 08
SAMPLE RATE 10>MCOnd
DETECTOR TYPE •vO'·g· rMfxxxttig
ACCURACY
FREQUENCY
*OH1 100H2
u j
10KMJ 20KH2 80 KMl
- 1 5
ï -10
200 Hz
13511 IMPEDANCE NOT SPECIFIED BELOW 200 Hz
NOtSE MEASUREMENTS (3S51A)
DYNAMIC RANGE
Nowe-Mth-tone lOdBmio -85 dBm ,80011 9001 It
Nowe-to-oround -40 dBm W • 125 dBm
RESOLUTION 1 dB
SAMPLE RATE ¿Mcond
DETECTOR TYPE Quasi RMS
ACCURACY
Meeaage orcuri none ±1 dB i -20 dBm to -85 dBm). -2 dB |0 dBm w
-20 dBm)
Nowe-wttvtone ri dB ('20 dBm to -85 dBm) -2 dB (10 d6m to
• as dBm)
NCtte-tO-ground :1 dB i-8C dBm lo -125 dBm]. ±2 dB I -40 dBm to
• 80 dBm}
WEIGHTING FLTERS C-manaoa. 3 kHz ftat. 15 kHz tat. program
NOISE MEASUREMENTS (35S2A)
DYNAMIC RANGE
Message a'ciÃ-t nas» -90 dBm lo -5 dBm
S*ndSsction
FREQUENCY RANGE: 40 Hz K> BO kHz
RANGES 40 HZ to 600 Hz 200 Hz to 6 kHz 2 kHz lo 60 kHz. 1004 Hz teed
(3551A) or BOO HZ |3552A) Ovwr frequencies avariaole for 3552A
HARMONIC OtSTOfmON: - -50dBTHD(100Hzio4kHz) > -40dBTHD(40Hz
» -00 Hz and 4 kHz to 20 kHz) --55 dB (al hermorao 100 Hz K 4 kHz),
• -60 dB THO ( 1004 Hz fiied)
ACCURACY: -1 oount
LEVEL RANGE: - *0 dBm lo -60 dBm (40 Hz to 00 kHz) -6 dBm to -60 dBm
(1004 Hz fixed)
RESOLUTION: 0 1 dB
9
© Copr. 1949-1998 Hewlett-Packard Co.
ACCURACY
40 Hz
••Ml
MONITOR cu'i-in speaker montón received ex tranarmned *gn*
BALANCED IMPEDANCES: '3511 (3551A|. 15OI1 (3552AJ. 60011 90011
BRIDGING LOSS OlOB
RETURN LOSS: • 30 dB
LONGITUDINAL BALANCE: -60 dB at 6 kHz
HOLD r 24 T»»amp« concur* current >50 kfl impedance, r
ha» protection
INPUT OUTPUT PROTECTION Mocfcs 300 V dc maximum tongrtudneJ vot
lag* 200 V rm»
BATTERY SUPPLY: 4-6 hours contvuous operafcon on «*em*J rechargeable
battene* ai25'C Battery dram n aulomaucaiy lurr^ off «rnen oHcharged beto*
prop»' Operating kevel Complete '«charo* m 12 hours
ACUNE: 100 120 200 240 V ±10% 48 Hz to 440 Hz. 14 VA
TEM«RATU« RANG*
OPERATING 0*C U M'C (32"F lo 130*F)
STORAGE 20-C to 65'C ( -4'F to 148*F)
RELATIVE HUMnfTY: 0»95%(<100*F >40*C)
WEIGHT N« 66Hg|l3b)
DIMENSIONS: 133 mm M • 343 mm W * 254 mm D with front-pane» cow (5 25
PRICES IN U.S.A.: 35S1A $1750 35S2A $2000
MANUFACTURING DIVISION: LOVELAND INSTRUMENT DIVISION
P O Bo» 301
815 Fourteenth Street. S W
LoveÃ-and Colorado 00537
A Computer System for Analog
Measurements on Voiceband
Data Channels
Besides making nine data-channel performance tests
automatically in less than two minutes, this new Trans
mission Parameter Analyzer is capable of a much broader
range of measurements.
by Stephan G. Cline, Robert H. Perdriau, and Roger F. Rauskolb
sentative list of measurements appears in Table I.
There are two categories, one consisting of general
measurements that may be useful in a broad range of
applications, and the other consisting of nine charac
teristics of voiceband data channels that are common
ly measured in North America. The 5468A Trans-
MANKIND TODAY POSSESSES a truly stag
gering ability to disseminate information. Much
of this ability depends upon networks that transmit in
formation in the form of electrical signals to the far
corners of the earth and deep into space. Over the past
decade, it has become necessary to transmit a grow
ing volume of information at higher speeds and with
greatly increased accuracy and reliability. This has
led to the imposition of increasingly stringent per
formance requirements on the transmission networks
and the components that comprise them.
The 5453A Transmission Parameter Analyzer
(TPA), Fig. 1, has been developed to aid in the design,
manufacture, installation, and maintenance of today's
high-performance voice-grade communication chan
nels and components. Transmission parameters are the
properties of an electrical path that must be suitably
controlled if information is to be successfully transmit
ted over the path. The path may be a simple amplifier,
for which frequency response, noise, and distortion
are the primary parameters of interest, or it may be as
complex as a long-haul telephone channel that must
be optimized for high-speed data transmission.
The 5453A TPA performs digital signal analysis,
measuring signal properties using computational
techniques rather than analog circuitry. No hardware
detectors, demodulators, or signal generators of the
type normally found in analog instrumentation are
incorporated in the 5453A. New measurements may
be implemented, or existing measurements modified,
purely through software. Programs can be written in
either FORTRAN or BASIC together with a simple
calculator-like language. The 5453A also offers sig
nificant advantages in terms of speed and accuracy
when compared to equivalent analog instrumentation.
The system is capable of both stimulus-response
(network analyzer) or response-only (spectrum ana
lyzer, power meter, counter) measurements. A repre
Fig. 1 Model 5453 A Transmission Parameter Analyzer is
both a general-purpose stimulus-response test instrument
and a special-purpose analyzer for performance tests on
voiceband data channels and components. Table I lists some
of its capabilities.
10
© Copr. 1949-1998 Hewlett-Packard Co.
Stimulus Generation
ponder (see box, page 16) is used with the TPA for
many of the data-circuit measurements.
A simple block diagram of the system appears in
Fig. 2. A digital-to-analog converter (DAC), together
with suitable filtering and impedance matching,
generates the desired stimulus. On the response side
of the device being tested, an analog-to-digital con
verter (ADC) samples the incoming waveform and
prepares it for digital processing. The mass memory
provides storage for measurement and analysis pro
grams, test stimuli, test results, and other data that
may be required for a measurement, such as digital
filters. A CRT terminal serves as the sole operator
interface. From it, programs can be prepared and
executed and tabular results displayed. No manual
controls are present or necessary. An optional CRT
display provides graphical output.
Virtually any desired stimulus may be quickly and
easily designed from the CRT terminal. The stimulus
may be as simple as a sine wave or as complex as
pseudorandom noise. It may be described as a wave
form in the time domain or as a spectrum in the fre
quency domain. In the latter case, the user has com
plete control of both the amplitude and the phase of
each spectral component.
Spectra and the corresponding waveforms can be
designed in minutes, stored in the mass memory, and
retrieved either on command from the operator or
automatically by the measurement program. Before
output, the waveform is represented in the computer
memory by a block of 16-bit words, each word repre
senting one time sample of the waveform. To convert
this to an analog signal, the 5453 A has a 13-bit DAC
followed by programmable gain circuitry. A com
bination of block scaling and gain setting is used to
achieve the desired peak or rms power output. Sig
nals can be generated over a range of 0 to —40 dBm
with distortion products down 60 dB.
The number of words per block is variable in
powers of 2 from a minimum of 64 to a maximum of
4096. The number is selected along with the scan
ning rate to achieve the frequency range and resolu
tion required in a given application. Generation of
the test signal is accomplished by causing the pro
gram to read the block out through one of the com
puter's high-speed direct memory access (DMA)
channels to the DAC. This process may be contin
ued for as long as desired, resulting in a periodic sig
nal being applied to the test device. The frequency
range is limited by the settling time of the DAC and
the DMA rate of the computer. Frequencies up to 10
kHz may be generated by the 5453A.
Table I
Representative Measurement Capabilities of the 5453A
Gain/Loss
Phase
Envelope Delay
Frequency
Power
Noise
Nonlinear Distortion
Impedance
Frequency Modulation
Phase Modulation
Amplitude Modulation
Conversion Loss
Characteristics of Four-Wire Data Circuits
Measured by the 5453A/5468A
Attenuation Distortion
Envelope Delay Distortion
Level
Message Circuit Noise
C-Notched Noise
Phase Jitter
Intermodulation Distortion
Frequency Shift
Single-Frequency Interference
5478A System Interface
Graphic
Display
(Optional)
Digital-toAnalog
Converter
Channel A
Analog-toDigital
Converter
CRT
Terminal
Channel B
Analog-toDigital
Converter
Direct Digital-to-Analog
1 Converter Output
Anti-Aliasing
Filter, Impedance
Selection
Voiceband
Digital-to-Analog
Converter Output
Direct Channel A
Input
Anti-Aliasing
Filter. Impedance
Selection
Direct Channel B
Input (Optional)
11
© Copr. 1949-1998 Hewlett-Packard Co.
Voiceband
Analog-to-Oigital
Converter Input
Fig. 2. Transmission Parameter
Analyzer block diagram. A digitalto-analog converter provides
stimulus signals for the network
being tested. Analog-to-digital
converters sample the network's
response. Parameters of interest
are then computed digitally.
Digital Processing
tion can be executed and the results displayed im
mediately upon entry. In this mode the 5453A is used
very much like a general-purpose scientific calculator.
Engineering personnel can learn to operate the sys
tem in one or two hours.
Once the desired keyboard program has been
written, it can be executed directly or, more typical
ly, it can be stored in the disc memory and retrieved
and executed using a CALL statement from a FOR
TRAN or BASIC controlling program. Both the con
trolling and keyboard programs have access to any
data blocks stored in the disc memory. The disc mem
ory is also used to transfer data from one program
to another. In this manner, measurement data result
ing from the execution of a keyboard program is avail
able to the controlling program for further computa
tion, formatting and output, or decision making. In
addition, data may be synthesized by the controlling
program and passed to a keyboard program for use in
a specific measurement. The disc memory is capable
of storing a large number of keyboard programs,
any one of which can be executed at will by the con
trolling program. The result is that powerful digital
signal analysis capabilities are now available in the
context of standard engineering-oriented computer
languages.
The number of possibilities for digital processing
of the raw data is large and they cannot all be dis
cussed in this article. We will discuss two examples
that illustrate processing of steady-state and random
signals and then look at an example of using the
5453A to generate or simulate a desired impairment.
On the response side of the 5453A, the incoming
waveform from the test device is sampled by the ADC
and converted to a block of 16-bit words representing
successive time values of the input. The incoming
waveform may have been generated by the DAC and
distorted by the test device, or it may be an external
signal. From this point, digital processing in either
the time domain or the frequency domain is used to
extract the pertinent information.
Because sampling is used as the means of gathering
the raw data, we must, of course, be aware of the con
straints imposed by aliasing, leakage, and quantiz
ing noise.1 The balanced, voice-frequency ports of
the 5453A are provided with seven-pole elliptical
filters that keep aliased products down at least
50 dB. For applications outside of the filter frequency
range, the direct DAC output and ADC inputs must
be provided with suitable anti-aliasing filters. In
put frequencies greater than 100 kHz can be accom
modated. Leakage is reduced by using Manning or
other appropriate windows on the data or, in some
cases, by measuring the amount of leakage and ac
counting for it. Dynamic range greater than 70 dB is
obtained and the system noise floor over the voice fre
quency band is approximately —90 dBm.
Digital signal analysis, by its very nature, involves
operations on or between blocks of data words.
These operations include block arithmetic, forward
and inverse Fourier and Hubert Transforms, power
spectrum, convolution, correlation, integration, and
so on.1'2 In most practical situations, an ordered se
quence of such operations must be performed on the
raw input data. Programs to accomplish this may be
written in FORTRAN. However, an alternative soft
ware approach, developed for the 5453A, provides a
simple keyboard language that may be used to call
for any desired sequence of block operations. The
name "keyboard" derives from the fact that any
block operation may be programmed by pressing at
most two keys on the CRT terminal. Each block opera
Measuring Insertion Loss and Phase
As a first example, suppose it is desired to measure
the insertion loss and phase of a two-port network.
Insertion loss and phase are defined as follows:
Insertion Loss = 20 Log [V0(f)/VN(f)] (1)
Insertion Phase = [<f>0(f) - </>N(f)] (2)
Fig. 3. spectrum typical insertion loss measurement on a two-port network showing the spectrum (a) of
the test (c) generated by the TPA, the test signal itself (b), and the network's insertion loss (c)
computed by the TPA.
12
© Copr. 1949-1998 Hewlett-Packard Co.
where the "o" subscript refers to conditions at the
load with the source directly connected, and the
"N" subscript refers to conditions at the load with the
network inserted. V(f) and <¿>(f) are voltage and phase
expressed as a function of frequency. Further assume
that the measurement is to be made at the frequencies
contained in the spectrum of Fig. 3a.
A simple program to accomplish this measure
ment might first instruct the operator to bypass the
network and connect the source directly to the load.
It would then generate the waveform of Fig. 3b, cor
responding to the spectrum of Fig. 3a. Next it would
sample the DAC output, deriving a 512-word block
of data (representing, in this case, a 64-ms time rec
ord), and compute the complex spectrum. In practice,
several such records would be sampled and averaged
to reduce the effect of external noise. The resulting
averaged spectrum would be saved, the operator in
structed to insert the network, and the process
repeated. We now have two complex spectrums,
VolO/foolf) and VN(f)/<ftN(f). Performing the calculations
indicated in equations 1 and 2 yields the desired re
sult. Accuracies of ±0.1 dB and ±0.2 degree are ob
tainable, and the measurement can be accomplished
in only a few seconds. Fig. 3c illustrates the results
of an insertion-loss measurement on a simulated
communication channel.
Noise Measurements
The measurement of noise is perhaps one of the
most common maintenance activities in telecom
munications, and the 5453A offers several capabilities
in this area. A conventional measurement of noise
might be modeled as shown in Fig. 4. In the case of a
telecommunication channel, the input signal spec
trum Sxxif) may be zero (input terminated), or it may
represent a holding tone intended to bias compandored facilities to their normal operating points for
continuous signals. In the latter case, the tone is re
moved by including a notch at the appropriate fre
quency in the transfer function of the weighting net
work. In either event, the weighted spectrum Syy(f) is
indicated on an rms-responding power meter.
S,x(f) .
Weighting
Network
Rms
Responding
Power Meter
Fig. 4. Model illustrating the measurement of noise. The
S,,(f) represent complex voltage spectra as functions of
frequency. The transfer function of the weighting network is
H(f).
From the model we write:
Syy(f)=H(f)*[Sxx(f)+Snn(f)]
(3)
or, for Sxxif) = 0
Syy(f) = H(f)*Snn(f)
4]
Since we are interested in power, we multiply each
factor in equation 4 by its complex conjugate and
write:
GÃ-f)= |H(f)|2*GÃ-f)
(5)
In words, the weighted noise power spectrum is equal
to the unweighted power spectrum multiplied by
the squared magnitude of the transfer function of the
weighting network.
The measurement is implemented with the 5453A
TPA by first causing the TPA to gather a record (data
block) representing the waveform associated with
Snn(f). From this raw data, the complex spectrum
Snn(f) is computed, followed by computation of the
power spectrum Gnn(f). In practice, this process is
repeated several times and the computed Gnn(f) are
averaged to obtain a reliable estimate of the noise
power spectrum. With an estimate of Gnn(f) available,
the desired weighting is applied in accordance with
equation 5. The weighting function, |H(f)|2, may be
obtained by actual measurement of a physical net
work or it may be computed from the ratio of polyno
mials that describe the network. In either case, it is
most often stored in the mass memory and used as
needed. The resulting data block, representing Gyy(f),
is then integrated to obtain the total (mean square)
weighted noise power. Specific frequencies, such as
60 Hz and its significant harmonics, may be elimina
ted by excluding them from the limits of integration.
The resultant data block is then passed, via the mass
memory, to the controlling program, where it is con
verted to the appropriate units (dBm, dBrnC, etc.) for
output.
Each of the steps is called for by simple keystrokes
and, once programmed, may be repeated as desired.
It is possible to apply any number of weightings with
out repeating the measurement. The power spectrum
is available and may be scanned by the controlling
program to determine the frequency and level of any
interfering tones that may be present. A typical flatweighted noise power spectrum with an interfering
tone at 1 kHz and -53 dBm is illustrated in Fig. 5.
Finally, if Sxx(f) is non-zero — representing, for ex
ample, the output of a data set — it is possible to check
for proper operation as regards both the frequencies
and power level transmitted.
13
© Copr. 1949-1998 Hewlett-Packard Co.
e¡ (t) = Acosat + Bcos/3t
SOdBm
171
Substituting (7) into (6), applying the appropriate
identities, and assuming A = B for simplicity, we
obtain:
+ a1 A(cosat+cos/Jt)
(cos2at + cos2/3t)
+ a2A[cos(a-/3) t + cos(a+/3) t ]
(8)
The output contains a dc component, linear terms
at a and /3, second harmonics, and sum and differ
ence frequencies. Second-order distortion is the ratio
of the power at the sum and difference frequencies to
the power at the fundamentals. By selecting a^ and
A, it is possible to compute a2 for any desired amount
of second-order distortion.
The input spectrum can now be entered into a data
block, transformed to the time domain, and the cal
culations indicated by equation 6 performed. The
result is a test signal that can be used directly to
evaluate system performance. Fig. 6b illustrates a
test signal of this type as it appears on a swept spec
trum analyzer while being output by the DAC.
This example is also an excellent illustration of the
potential speed advantage of digital signal analysis.
The spectrum in Fig. 6b was taken with an analog
spectrum analyzer over a 2-kHz sweep width with
3-Hz resolution. Approximately one-half hour was
required for a single sweep. The same result, over a
wider bandwidth, with equivalent resolution and
dynamic range can be computed by the 5453A in
approximately twelve seconds.
4 kHz
Fig. 5. Typical flat-weighted noise power spectrum com
puted by the TPA. An interfering tone is apparent.
Simulating Impairments
Digital signal analysis can also be used to simulate
known impairments. Fig. 6a shows the power spec
trum of a clean test signal containing energy at 703
and 1172 Hz. Such a signal might be used as a stimu
lus when performing measurements of intermodu
lation distortion.
Suppose that we have devised a system for mea
suring the second-order intermodulation distortion
of a device using the signal in Fig. 6a as a stimulus,
and we wish to test this measurement system. To do
this we need a means of creating known and variable
degrees of second-order distortion. The design and
construction of a physical device to do this is expen
sive and time-consuming. An alternative is to use
the computational ability of the 5453A to generate
known distortions.
We begin by assuming a nonlinear device transfer
function given by:
e 0 ( t ) = a ^ i i f l + a z e f l t ) ( 6 )
Telephone Channel Measurements
The testing of telephone circuits used for the trans
mission of high-speed data is a difficult problem.
While many different types of voice-grade data chan
nels are available, we will limit our discussion to twopoint, private-line, four-wire circuits. Such a circuit
where a^ is the linear component of the transfer func
tion and a2 is related to the degree of second-order
nonlinearity. Next, we describe our test stimulus as:
Fig. 6. Spectrum of a typical
stimulus signal for intermodulation
distortion measurements (a), and
the spectrum of a typical response
(b). In this case the response
signal was generated by the
5453A and contains a known
amount of intermodulation dis
tortion. It could be used to test
a distortion-measuring system.
Other known impairments are
also easily generated.
14
© Copr. 1949-1998 Hewlett-Packard Co.
T™°n
Fig. 7. Two-point private fourwire data circuits like this one can
be measured in both directions
from the test center by the 5453A.
Circuits not passing through the
test center can also be tested.
Transmission
System
might be laid out as shown in Fig. 7. Several transmis
sion media may be used between intermediate points
along the circuit, including PCM, FDM cable, and mi
crowave systems. The user leases the circuit from a
common carrier and it is available to him on a fulltime basis.
Most transmission systems were originally de
signed to enable people to talk to people. As data traf
fic has increased in volume, speed, and importance,
a number of circuit characteristics, most of which
offer little or no degradation to voice traffic, have be
come significant. A list of the parameters presently
measured on data circuits by the 5453A appears in
Table I. The interested reader unfamiliar with the ter
minology will find additional information in refer
ences 3 and 4. The parameters encompass such fun
damental characteristics as insertion loss, power, fre
quency, noise, distortion, and incidental modula
tion.
A circuit like that of Fig. 7 may be tested from end
to end or between any two points at which voice-fre
quency access is available. Assume that a 5453A is lo
cated in the test center and that we wish to character
ize the portion of the circuit from that office to user lo
cation B. The transmit and receive sides of the circuit
could be connected together at the user location.
This would form a loop and the line could be treated
as a two-port network by the 5453A. However, this
approach does not make it possible to separate the
characteristics of the two sides of the circuit. Its major
usefulness lies in the ability to characterize a knowngood circuit on a looped basis and to save the result
in the mass memory. Subsequent troubles may then
be traced to either the circuit or the terminal equip
ment by repeating the measurement and detecting
changes from the benchmark. This can normally be
accomplished without dispatching a trained repair
man to the remote location.
The 5468A Transponder has been developed to
provide for two-way measurements between distant
locations. When connected to the circuit at the user
location, it can be commanded automatically from
the 5453A to generate the test signals required for
measurement of the receive line or to process signals
generated by the 5453A in a manner that allows the
transmit-line characteristics to be calculated (see
box, page 16).
With the equipment in place, the operator requests
any or all of the transmission parameter measure
ments listed in Table I. The 5453A will make all nine
measurements in both directions on the circuit in ap
proximately two minutes. Fig. 8 illustrates the data
output.
Other capabilities of the program include storage,
retrieval, and purging of test results in the mass me
mory. It is also possible to compare data to a bench
mark or to specifications the circuit must meet. Data
taken on two segments of a circuit may be combined
to yield the overall characteristics. Operator interac
tion with the program is purely conversational, al
lowing him to accomplish complex tasks rapidly
with a minimum amount of training.
Other Applications
The 5453A is not limited to testing installed com
munication channels. The same approach could be
applied equally well, for example, to end-to-end
checkout of a complete communications system on
the production floor. The speed of digital tech
niques makes it feasible to do 100% testing and have
complete records even for high-capacity systems.
Additional 5453A applications are to be found in
the design and testing of all types of communica
tions equipment, such as data sets, facsimile trans
ceivers, equalizers, multiplex-channel modems, tele
phone sets, and loop extenders.
Correction
In our April 1975 issue, page 10. it is stated that Model 5308A time-interval measurements are
guaranteed accurate within one nanosecond That sentence should have read. Measure
ments accurate guaranteed accurate within five nanoseconds and are typically accurate within
Fig. 8. 5453/4 TPA printout of data-circuit test results. Nine
tests are made in less than two minutes.
one nanosecond.' The editors apologize for losing a crucial line of type.
15
© Copr. 1949-1998 Hewlett-Packard Co.
Portable Transponder Allows
Two- Way Data Channel Measurements
When used with the 5453A Transmission Parameter Analyzer
(TPA), the HP 5468A Transponder, Fig. 1 , provides the capabi
lity to characterize four-wire voice-grade facilities automatically
in both directions of transmission. Control of the transponder is
by means of coded command tones generated by the TPA. The
transponder provides the test signals needed to characterize
the receive line and conditions test signals received from the
TPA so the system can compute the transmit line characteris
tics. The automatic feature can be overriden, allowing manual
measurements of received level and noise without tying up the
TPA.
Output
Fig. 2. Transversal filters are used in the transponder to
generate low-distortion test signals.
length) pseudorandom binary sequence. The final output
signal is then obtained by multiplication using the circuit shown
in Fig. 3.
For measurements on the transmit line command tones are
sent from the TPA to program the transponder into its signalconditioning modes of operation. Attenuation and envelope
delay distortion are measured by causing the transponder to
provide an equal-level loopback. The characteristics of the
transponder and the receive line are then subtracted from
the measurement of the entire loop.
Measurement of noise with tone on the transmit line is accom
plished by first passing the received signal through a 20-dB
notch en With the tone reduced in amplitude by 20 dB, the en
tire spectrum (noise plus tone) is given 20 dB of gain before be
ing looped back on the receive line. Thus noise on the receive
line has a negligible effect on the measurement. The TPA re
moves the weighting effect of the receive line.
The measurement of intermodulation distortion is achieved
by notching out the 703-Hz tone prior to loopback. Therefore,
while of distortion products may be formed as a result of
transmission over the receive line, there are no ¡ntermodulation
products. Once again, the previously measured frequency re
sponse of the receive line is used by the TPA to compute
distortion.
Fig. 1. 5468 A Transponder works with 5453 A Transmission
Parameter Analyzer to characterize transmit and receive lines
of data circuits.
Three test signals are provided by the transponder to charac
terize the receive line. A pure 1015.625-Hz holding tone is
used to measure frequency shift, phase jitter, noise with tone,
and 1-kHz loss. Intermodulation distortion requires at least two
tones, and a third tone is added at V* power to better simulate an
actual data signal. Attenuation and envelope delay distortion
are measured using a broadband signal containing 16 tone
pairs. The transponder also provides a 600-ohm termination on
the receive line for no-tone noise measurements.
The 1 01 5.625-Hz tone must have low incidental phase modu
lation (<0.1°), stable amplitude (<0.05 dB drift), and an accu
rate frequency (±0.025 Hz). The frequency accuracy re
quirement implied that a crystal was necessary, while the low
phase modulation requirement ruled out a phase-lock loop. The
approach taken in the transponder is to use a transversal filter
to convert a stable digital clock into a sine wave with the de
sired Sec The circuit is illustrated in Fig. 2. Sec
ond-order distortion of les£ than 70 dB is typical of such a filter.
Three transversal filters are used to generate the intermodulation distortion test signal. The three frequencies are 703,
1172, and 1218 Hz.
The third test signal contains 16 pairs of sidebands spaced
±78 Hz about suppressed carriers spaced 250 Hz apart. Amp
litudes and relative phase differences must be stable and uni
form from one transponder to another. The 78-Hz modulation
signal is generated using the transversal filter approach. The
carriers are generated using a 63-clock-period (4 ms total
Subtracts 78 Hz from
Multi-Tone Signal
Output
Fig. 3. Tesf signa/ for envelope delay distortion and attenuaton measurements is generated by multiplying a 78-Hz sig
nal by a pseudorandom binary sequence (PRBS).
16
© Copr. 1949-1998 Hewlett-Packard Co.
Acknowledgments
Hewlett-Packard Instruments for
Checking Voice-Grade Telephone Lines
The authors wish to thank Peter Roth of HewlettPackard and David Favin of the Bell Telephone
Laboratories who jointly originated the idea of apply
ing digital techniques to the testing of data circuits
and who have contributed consistently throughout
the development. Ron Potter made significant contri
butions with regard to several of the more difficult
measurements. Special thanks are also due to Earle
Ellis for the application program, Al Low for some ex
cellent product design, and Dennis Kwan for the in
troduction to production. Pete Appel contributed to
helping complete this 5468A design as well as help
ing with the system programming. Finally, thanks
are due to Dave Snyder for his help in getting the first
prototype running and to Melba Lindgren for her sup
port of the programming effort.E
The August 1 974 issue of the Hewlett-Packard Journal contained a chart comparing
the capabilities of six Hewlett-Packard instruments designed to measure vahous
parameters of voice-grade telephone channels used tor data transmission All of
these Journal. have now been described in the Hewlett-Packard Journal. The
instruments, and the issues in which they appear are:
CCITT Standards
3552A Transmission Test Set (May 1975)
3770A Amplitude Delay Distortion Analyzer (November 1974)
3581C described Voltmeter (Related to 3580A Spectrum Analyzer, described in
September 1973)
North American Standards
3551A Transmission Test Set (May 1975)
4940A Transmission Impairment Measunng Set (August 1974)
5453A Transmission Parameter Analyzer with 5468A Transponder (May 1975)
3581C Selective Voltmeter (see above)
References
1. P.R. Roth, "Digital Fourier Analysis", Hewlett-Packard
Journal, June 1970.
2. E.O. Brigham, "The Fast Fourier Transform", PrenticeHall Inc., 1974.
3. "Transmission Systems for Communications", Bell
Telephone Laboratories, Inc., Fourth Edition, 1970.
4. E.G. Smith, "Glossary of Communications)", Tele
phony Publishing Corp., 1971.
Robert H. Perdriau
Bob Perdriau is product market
ing engineer for digital signal
analyzers at HP's Santa Clara
^•*L I Division. Before assuming that
f | I post in 1973 he had served as a
'Mi} I design engineer and as an appli
cations engineer for three HP
divisions. Born in Bosiun. Mas
sachusetts, he graduated from the
University of Massachusetts in
1963 with a BSEE degree. Except
for three years in the U.S. Army,
he's been with HP ever since. Bob
is married, has two children, and
lives in Los Altos, California, where
for the last two years he's been busy building a major addition
to his for With that project about finished, he's looking for
ward to having more time for his other interests, which include
fishing and hunting, woodworking, and bicycling.
HP Model 5453A Transmission Parameter Analyzer
Contact the factory or your local Hewlett-Packard sales office for specifications.
PRICES IN U.S.A.:
5453A Transmission Parameter Analyzer, $62.800.
5468A Transponder. $2500.
MANUFACTURING DIVISION: SANTA CLARA DIVISION
5301 Stevens Creek Boulevard
Santa Clara. California 95050 U.S.A.
Stephan G. Cline
Now a laser interferometer ap
plications engineer, Steve Cline
until recently was involved in
Fourier analyzer hardware and
software design. With HP since
1968, he wrote much of the softC% • J ware for the 5450A Fourier AnaII JT l| lyzer. helped design the 5470A
9 M >*^ Fast Fourier Processor, and served
•w • L. ^H as project leader for the 5471 A
ff 41 AvjB FFT Arithmetic Unit and the 5468A
Transponder. Born in Camp
McCoy, Wisconsin, he received
his BSEE degree from Michigan
State University in 1967 and his
MSEE degree from Stanford University in 1968. Especially
interested in meeting people from different cultures, Steve
enjoys travelling and serves as treasurer of American Field
Service, a foreign-student exchange program. His main rec
reational activity is golf, but he enjoys biking, too. He and his
wife live in Los Gatos, California.
Roger F. Rauskolb
Roger Rauskolb, a native of
Maehrisch-Ostrava, Czecho
slovakia, received his Dipl. Ing.
degree from the Technische
Hochschule of Darmstadt, Ger
many, in 1961. His career at HP
has been a varied one that began
in 1962 and includes service as a
microwave project engineer,
spectrometer project manager,
digital signal analysis (and 5453A)
group leader, and now, member of
the HP Laboratories technical
-»»^. staff. In 1965 he received his
MSEE degree from Stanford
University. A resident of Palo Alto, California. Roger is married
and has two daughters. He's an audiophile, a photographer, a
swimmer, and a skier. His interest in building a better world
goes back many years and currently expresses itself in his
membership in Project Survival, a group concerned with energy
problems and education for long-term survival on earth.
17
© Copr. 1949-1998 Hewlett-Packard Co.
A Precision Spectrum Analyzer for the
10-Hz-to-1 3-MHz Range
Adaptable to automatic systems or bench use, a new spec
trum analyzer has measurement resolution of 0.01 dB, passbands as narrow as 3 Hz, and a dynamic range of 70 dB.
by Jerry W. Daniels and Robert L. Atchley
THE SPECTRUM ANALYZER and the network
analyzer are assuming greater and greater im
portance as means of evaluating the performance of
electronic circuits and devices. The network ana
lyzer gives complete information about the ampli
tude and phase performance of linear networks
while the spectrum analyzer evaluates the amplitude
performance of both linear and nonlinear networks.
The spectrum analyzer is a single-channel instru
ment that selects and measures the amplitudes of the
individual frequencies that make up a complex sig
nal. It is thus able to detect and measure the distor
tion and intermodulation products of nonlinear net
works.
The network analyzer is a dual-channel instru
ment that compares the amplitudes and phases of
two signals, usually the input and output of a net
work or device. It is normally not suitable for mea
surements involving nonlinear networks because it
is designed on the assumption that only one fre
quency at a time will be at its input. The method of
heterodyning signals within the network analyzer
could cause spurious responses if the input signal
were distorted or otherwise contained more than one
frequency.
modulation responses are below the measurement
range of the instrument.
The new spectrum analyzer has accuracy and pre
cision normally not associated with spectrum ana
lyzers. It has an amplitude resolution of 0.01 dB, an
accuracy of ±0.05 to ±1.15 dB, depending on the sig
nal level and frequency, a dynamic range of 70 dB
and a measurement range of 150 dB. The analyzer's
passband is selectable in steps from 10 kHz down to 3
Hz, the stability of both the analyzer and the synthe
sizer used as the local oscillator being such that the
3-Hz bandwidth is practical even at 13 MHz. The fre
quency of a signal component can be determined
well within 1 Hz. Typical measurement results are
shown in Fig. 2.
Some of the measurements for which this high-pre
cision instrument is especially useful are harmonic
and intermodulation distortion in amplifiers, powerline sidebands and harmonic levels in oscillators, RF
and LO feedthrough in mixers, and frequency transla
tion errors in communications repeaters. It can also
serve as a frequency-response test set for high-preci
sion measurements, such as amplitude errors in the
up-converted channels of multiplex communica
tions systems.
A New Spectrum Analyzer
Remote Control
Some two years ago, a network analyzer for measur
ing the magnitude and phase characteristics of sig
nals in linear networks over a frequency range of 50
Hz to 13 MHz was introduced (Model 3570A).1 The
accuracy and ready adaptability of this instrument to
systems use have now been incorporated in a new
spectrum analyzer, Model 3571A, for measurements
of complex signals over this same frequency range
(and down to 10 Hz). This new instrument (Fig. 1)
performs waveform signal analysis with full assur
ance that all internally-generated image and inter-
Besides being operable from the front panel, the
new spectrum analyzer is also programmable
through the HP Interface Bus.2 Every front-panel
switch position (except for the ON/OFF switch) is as
signed an ASCII code so it can be selected by a sys
tem controller. The analyzer can thus be incorpo
rated into calculator-controlled automatic measure
ment systems (Fig. 3) that can manipulate the data so
it can be presented in more meaningful form. For ex
ample, it can function as a high-precision distortion
analyzer by providing a mathematically exact sum•6
© Copr. 1949-1998 Hewlett-Packard Co.
Fig. 1. Model 3571A Tracking
Spectrum Analyzer (lower unit)
works over a 10-Hz-to- 13-MHz
range using one of the HP syn
thesizers (upper unit) as a local
oscillator. The combination of
analyzer and synthesizer is known
as the Model 3044A Spectrum
Analyzer.
spectrum analyzer but it differs in the characteristics
of its selective filters, which have a rounded re
sponse curve that minimizes ringing during a fre
quency sweep, rather than the wave analyzer's
squared-off response curve.
Because the front-panel of the new Model 3571A
Tracking Spectrum Analyzer resembles neither the
traditional spectrum analyzer nor a wave analyzer, a
look at the controls can be informative. First of all,
there is no tuning control on the instrument itself. It
was designed to work with the offset frequency sig
nal from either the HP Model 3320A/B Frequency
Synthesizer or the Model 3330A/B Automatic Fre
quency Synthesizer. Tuning the synthesizer tunes
the analyzer, the frequency of the synthesizer corre
sponding to the center frequency of the analyzer's
mation of the individually-measured distortion pro
ducts. An automatic system not only speeds measure
ments, removes the tedium from repetitive measure
ments, and facilitates a high degree of data manipula
tion, but it also provides a means for enhancing mea
surement accuracy by using calibration routines to
store the results of reference measurements and then
using these to correct actual measurements.
Information Display
To obtain the high resolution that the accuracy of
this instrument makes possible, it has a digital read
out rather than the CRT display commonly asso
ciated with spectrum analyzers (however it has an
analog output for a CRT display). Superficially it re
sembles the traditional wave analyzer more than a
dBV
20
3BV
-40
-40--
-60- -
60--
-•O- -
5th order
7th order
80--
l  °  ° J / J t M U . ^ .  · + M ^
Ã-
1
1
U * '
100- •
' * v * ^ - J j ^
1 — r — I
1
1
H
97 98 99 100 101 102 103 104 105 106kHz
H — I — h
- 1 kHz 11 MHz - 1 kHz
19
© Copr. 1949-1998 Hewlett-Packard Co.
Fig. 2. Typical spectra recorded
by a 3045A calculator-controlled
spectrum analyzer system based
on the Model 3571A Tracking
Spectrum Analyzer. The recording
at far left was made during a twotone intermodulation test of an
amplifier using input frequencies
of 101 and 102 kHz. The oddorder intermodu/ation products
are clearly shown. The recording
at near left shows the output of a
double-balanced mixer fed by a
high-level 11 -MHz carrier through
the LO port and a mixture of
1-kHz and low-level 180-Hz tones
through the RF port. The up-con
verted 180-Hz signal is clearly
resolved about the 1 1-MHz carrier.
frequency components from stronger frequencies
close by. Because of the stability of the instrument,
two signals only 15 Hz apart but with an amplitude
difference of 55 dB can be resolved. Line-related side
bands more than 70 dB down can be resolved.
Measurement results may be displayed in a variety
of measurement units. With the DISPLAY REFERENCE
switch in the dBV position, the display reads dB with
respect to 1 volt no matter what the input impedance
may be. With the switch in the dBm position, the
instrument displays the measured power in the selec
ted input impedance (either 50ÃÃ, 75Q, or an external
600ÃÃ). With the switch in the RELATIVE position the
instrument displays a dB reading relative to a pre
viously established reference. For example, a reading
in either the dBV or dBm position can be stored as a
reference by pressing the ENTER OFFSET button, es
tablishing this signal level as the 0.00 dB level. Then
with the switch moved to the RELATIVE position, all
further readings are displayed as so many dB above or
below this reference. This is handy for reading the
level of harmonics with respect to the fundamental.
This arrangement also allows the user to calibrate
the instrument with respect to some other impe
dance level. With an external termination of the de
sired impedance attached, the user supplies a known
0-dBm signal and presses the ENTER OFFSET button.
Subsequent measurements made with the switch in
the RELATIVE position will then be direct reading in
terms of this impedance level.
The analog equivalent of the stored reference is
subtracted from the analog output voltage allowing
expanded-scale visualization of a portion of the spec
trum on a CRT display or X-Y recording.
Fig. 3. The Model 3571 A Tracking Spectrum Analyzer also
functions under programmable calculator control through the
HP Interface Bus, giving an automatic measurement and datareducing system that shortens test time and decreases the
possibility of measurement error. A packaged calculatorbased system that includes the analyzer is known as the
Model 3045/4 Automatic Spectrum Analyzer.
passband (it is for this reason that it is known as a
Tracking Spectrum Analyzer).
The controls that are on the analyzer have to do
with bandwidth, signal level, units of measurement
in the display, and input impedance. The input impe
dance is selectable to allow use of the analyzer in a
variety of measurement situations. The 50ÃÃ and 75Ã1
input impedances match a wide range of high-fre
quency devices and are especially useful for measure
ments in communications systems. The IMfl input
impedance allows the user to supply his own termin
ation for other impedance levels, and it also allows a
conventional oscilloscope probe to be used for highimpedance circuit probing.
Full-scale input amplitude ranges are from 3.16V
rms to 1 mV rms in eight 10-dB steps with a full
70-dB dynamic range on each step. An OVERLOAD
indicator flashes if the signal exceeds the input
range.
As mentioned earlier, the passband is selectable
from 3 Hz to 10 kHz in a 1-3-10 sequence. The wider
passbands permit relatively fast sweeps over a wide
band for a quick, overall look at a spectrum. The nar
row passbands make it possible to isolate low-level
Block Diagram Overview
As other spectrum/wave analyzers do, the Model
3571A heterodynes the input signal to an interme
diate frequency for narrowband filtering. A block dia
gram is shown in Fig. 4. The input signal is mixed
with the synthesizer offset frequency to derive a
20-MHz intermediate frequency (the synthesizer off
set frequency is precisely 20 MHz higher than the
synthesizer main output). The up-conversion to 20
MHz places the image frequencies in a range of 40 to
53 MHz, which are easily attenuated more than 80 dB
by a low-pass filter at the input.
The 20-MHz output of the mixer is filtered to re
move unwanted mixer products and then down-con
verted to 100 kHz for the filtering that establishes the
instrument's passband. An IF of 100 kHz was chosen
to permit the use of narrow-band crystal filters.
The filtered IF is then processed by an amplifier
whose output is logarithmically proportional to the
input. This amplifier, a hybrid 1C similar to that used
20
© Copr. 1949-1998 Hewlett-Packard Co.
20-33 MHz
From Synthesizer
Input
Amplifier
20 MHz
100 kHz L°9
Amplifier
Analog
Output
10 Hz-13 MHz v
19.9 MHz
Overload
Detector
A-to-D K D-to-A
Converter I Converter
1 MHz
From
Synthesizer
Fig. Spectrum Analyzer. block diagram of the Model 357 T A Tracking Spectrum Analyzer.
in other HP instruments,3 converts the signal vol
tage level to the equivalent dB level by compressing
signals in proportion to their amplitude.
The detector is a peak-to-peak type. The detector
output, a dc voltage proportional to the log of the sig
nal amplitude, is smoothed in a low-pass filter and
then provided at a rear-panel connector as a Y-axis
output for use by a CRT display or by an X-Y plotter
(an X-axis output is available from the associated
synthesizer). A front-panel switch can slow the filter
response by a factor of 20 to smooth noisy signals.
The detector output is also converted to an equiva
lent digital number by an analog-to-digital converter
and sent to the digital processor. The number is proc
essed in accordance with the format established by
the settings of the front-panel switches. Offset, dB,
dBV, dBm 50ÃÃ, dBm 75ÃÃ, dBm 600Ã1, and input
range all affect the number that is finally displayed.
The difference between the displayed number and
the raw digitized number is converted to an analog
voltage and applied as an offset to the rear-panel ana
log output voltage. The analog output is thus consis
tent with the digital readout in terms of measure
ment units (0.1 V = 1 dB).
Analog Circuit Details
Now to examine some of the considerations in
volved in the design of this instrument. The dynamic
range of a spectrum analyzer is limited by noise at
the low end and intermodulation distortion at the
high end. The design of the input circuits is directed
towards maximizing the difference between these
two extremes.
Low noise is achieved by use of a J-FET buffer am
plifier which also gives high input impedance. The
use of the complementary-symmetry configuration
obtains high linearity and very good frequency re
sponse. By making the "straight-through" input
range 10 mV rms, the noise level allows the desired
70-dB range to be obtained with the widest band
width (10 kHz).
The output of the input amplifier is monitored by
the overload detector, a peak detector driven by an
amplifier that has greater than 60-MHz bandwidth to
enable response to out-of-band signals. As long as
the front-panel overload indicator is not illuminated,
the user is assured that the input signal is within
the linear range of the amplifiers, which means that
internally generated distortion and intermodulation
products are more than 80 dB below the full-scale in
put range.
Only Two Conversions
The frequency conversion to 20 MHz occurs in an
active double-balanced mixer. After bandpass filter
ing, the 20-MHz signal is presented to the second
mixer for conversion down to 100 kHz. The second
mixer's local oscillator frequency (19.9 MHz) is
phase-locked to a 1-MHz signal from the synthesizer
and is thus in precise relationship to the first local os
cillator frequency.
Since the 100-kHz IF is derived by mixing a 20MHz signal with 19.9 MHz, a 19.8-MHz signal in the
first IF channel would also be converted to 100 kHz.
Normally, this situation would be avoided by having
another IF conversion between the 20-MHz and
100-kHz IP's. Fortunately, we were able to avoid the
extra cost of a third IF channel, not to mention the ad-
© Copr. 1949-1998 Hewlett-Packard Co.
Attenuating the Classical Attenuator Problem
The classical attenuator problem is encountered anytime a
device under test is placed between a single-ended source
and a single-ended detector. It may manifest itself in several
ways:
• Apparent detector inaccuracies at low signal levels;
• Reduction of dynamic range at low frequencies;
• Spurious responses caused by common-mode signals.
The basics of the problem are outlined in the drawing. This
represents a signal source driving an attenuator that is moni
tored by a detector. Coaxial cables are used, and the test is
being conducted at frequencies in the audio range (at high fre
quencies, the coaxial cables behave more like baluns and the
problem is not so acute).
To simplify the discussion, the attenuator is set for infinite at
tenuation. It is easily seen that return currents through the cable
shield to the signal source can develop a voltage, ea, across
the finite resistance of th
tages across Rc2 and Z
If the detector input resistance, RD, is high, the voltage drop
across Rc2 is seen by the detector, so a residual signal is mea
sured even with infinite attenuation.
When (RL2 + RD) » Rc2,
(R s
(RC
From the figure it is easily seen that if Z were zero, which
means that the source and detector would be referenced to the
same ground, the full voltage of ea would be measured by the
detector. Consider a 50fi system using two 4-foot lengths of
RG/58U cable with an infinite attenuator between. What would
be the real attenuation?
Rci = Rc2 = 20 mil and Rs + RL1 = 100Ã1. IfZwereequal
to zero, then:
e^
(20 x 1Q-J) (20 x 1Cr3)
»-80 dB
100(40 x 10-3)
Increasing Z to 111 yields:
(20 x 1Q~3) (20 x 10~3)
=>-108dB
100( 1. 0004)
thus, a small increase in Z results in a significant reduction in
eD . A similar analysis shows that common-mode voltages
caused by ground loops are also reduced by increasing Z.
Increasing Z, however, would not allow the barrel of the frontpanel BNC connector, which is connected to signal ground, to
be tied directly to earth ground. The classical attenuator error
was reduced in the Model 3571 A without fully floating the input
connector by making Z a saturable-core inductor wound with
#17 wire. This has practically zero impedance at dc but a finite
impedance at frequencies where the classical attenuator prob
lem exists. On the other hand a large powerline signal through
Z, such as might occur with a grounding error, would saturate
the core, reducing the impedance of Z to less than one ohm.
This is why the input to the Model 3571 A is not described as
"floating", but as "isolated at low frequencies".
The potential reduction in measurement errors achieved by
this arrangement is shown by the graph below. This was made
by the Model 3571 A measuring the output of a 120-dB attenua
tor fed by a one-volt signal supplied through a 4-foot cable
(bandwidth: 3 Hz; range: 1 mV; smoothing: on).
Z) + Rcl (RC2 + Z)
and,
The object is to reduce the detector signal, eD, to zero, or at
least to insignificant proportions. This would occur if either
RC1 or Rc2 were zero, but this would be difficult to achieve.
Increasing (Rs + RL)) and/or Z would also reduce eD but Rs
and RL1 are fixed by the measuring system, which leaves Z as
the only variable available for manipulation.
dBV
Detector
Source
-70 -p
80
With Signal Ground
Connected to Chassis Ground
90
100
-110
With Z Between
Signal and
Chassis Grounds
120-
130
10 Hz 100 Hz 1k
ditional problems with intermodulation and noise
that another mixer would introduce, by use of a
20-MHz filter that attenuates 19.8 MHz more than 80
dB.
The 20-MHz filter consists of two cascaded twopole crystal filters, one of which is shown in Fig. 5.
Stagger-tuning the crystals gives a bandwidth of 30
kHz. However, at 19.8 MHz the currents through the
•
10 k
100 k 1 MHz
shunt capacitances of the two crystals are exactly
equal and of opposite phase, cancelling at point A
and giving a transmission zero at 19.8 MHz. Signals
at 19.8 MHz are attenuated more than 50 dB in each
stage, more than adequate to meet demands.
IF Stability
The 100-kHz IF is where all of the bandwidth selec22
© Copr. 1949-1998 Hewlett-Packard Co.
ly programmable with all control executed through
the PS register. During local control, the front-panel
switches are parallel loaded into the PS register but
under remote control the front-panel controls are
locked out and the contents of the PS register may
be changed only by data from the HP-IB.
Pushing the ENTER OFFSET switch during local con
trol sets the data flag. The controller checks to see if it
is in local control, and finding that it is, takes the pre
sent dBV reading, which has been stored in memory,
and places it into memory as the reference for a rela
tive dB display.
Data coming from the HP-IB is parallel loaded into
the 8-bit input (I) register. The controller uses a quali
fier to sense this condition and upon receiving this
information it shifts the data serially through an opti
cal isolator to the Q register for decoding (both the
HP-IB input and HP-IB output circuitry are isolated
from measurement ground). After the data has been
deciphered, it may be used to program an arbitrary
offset into memory, or to initiate some immediate
command. It could also be recognized as an unused
command and be ignored.
The controller also calculates an offset voltage to
be subtracted from the log amp output so the analog
output will correspond to the display. During the
time that the controller is not in a measurement rou
tine or data entry, the controller uses the digital-toanalog portions of the successive-approximation digi
tizer to construct an analog voltage. For a given dis
play reference and input range, the number is con
stant so a follow-and-hold circuit can retain the ana
log voltage while the controller is occupied with
the other routines. This analog signal is then added to
the normal (straight through) analog output to obtain
the required offset. Because only the offset is obtained
by the D-to-A converter, small pertubations in the
signal amplitude are transferred to the analog output
signal. The resolution then is that of the log amplifier,
rather than the digitizer.
tion and most of the gain occurs. The filters are all
5-pole synchronously-tuned types with a -3dB to
— 60dB shape factor of 10. The response curve is ap
proximately Gaussian. The three widest passbands
are derived from high-Q LC tanks while the rest are
crystal derived. A single set of five crystals is used
with loading resistors to broaden the bandwidth
when required.
Frequency drift in narrowband filters can cause
problems. When using the 3-Hz bandwidth, a drift of
1 Hz, although only 0.001% at 100 kHz, would cause
significant measurement errors. This problem was
minimized by incorporating a 100-kHz oscillator in
the 19.9-MHz phase-locked local oscillator circuit
(Fig. 4). The 100-kHz crystal of this oscillator is the
same type used for the IF filters; in fact, all six crys
tals are supplied as a matched set. Therefore, any
drift in the IF center frequency is matched by a com
pensating drift in the local oscillator.
Digital System Details
The digital machine in the 3571A is a 16-bit binary
serial processor using 8K of ROM. It has four major
functions: the measurement routine, the data entry,
the data output, and the calmlation of the analog
output.
The measurement routine is the data gathering
process. Here the controller commands the analog-todigital converter to digitize the output of the log am
plifier. The output of the digitizer is then manipu
lated according to the program in the program storage
(PS) register. The controller does this by interrogating
the "PS" register and checking to see what amplitude
range is in use. It then subtracts or adds a number to
correct the reading to dBV (IV = OdBV).
The controller now must check to see if the answer
is to be displayed in dBV, dBm, or dB relative to a
stored number. If it is to be displayed in dBV, it con
tinues with the binary-to-BCD conversion for dis
play. If the displayed answer is to be in dB relative to
some reference, this reference, which was stored
in memory by the data routine, is subtracted from the
dBV answer. If the program calls for a dBm display,
the controller again corrects the dBV answer by the
appropriate factor for the selected terminating impe
dance (50Ã1, 75Ã1, or 600Q).
In any case the binary answer must be converted to
BCD for display purposes. This is done by hardware.
The binary data is shifted into a binary-to-decimal
converter with the most significant bits first. At the
end of the shifting a BCD answer is stored in the regis
ter and is latched into the displays. It may also be
output to the HP interface bus (HP-IB).
The data entry routine is primarily used to bring in
data from the HP-IB for remote control of the 3571 A.
The 3571 A, except for the power switch, is complete
Fig. 5. 20-MHz IF filter precedes the second mixer. The
adjustable capacitors balance the crystal distributed capa
citance to achieve a zero at 19.8 MHz
23
© Copr. 1949-1998 Hewlett-Packard Co.
Acknowledgments
References
Group Leader Paul Thomas was responsible for the
basic block diagram and he contributed to the design
of the log amplifier and the method of phase-locking
to an external synthesizer. Howard Hilton was re
sponsible for the input amplifier, overload detector,
image filter, and 100-kHz reference oscillator. The
100-kHzIF was designed by Tom Rodine. Product de
sign was by Jim Saar. Virgil Leenerts was responsible
for the front-panel design and provided the lowfrequency isolation scheme (see box, page 22). £
1 . G.E. Nelson, P.L. Thomas, and R.L. Atchley, "Faster GainPhase Measurements with New Automatic 50 Hz-to-13
MHz Network Analyzers," Hewlett-Packard Journal,
October 1972.
2. D.W. Ricci and P.S. Stone, "Putting Together Instru
mentation Systems at Minimum Cost," Hewlett-Packard
Journal, January 1975.
3. R. Jeremiasin, "Logarithmic Amplifier Accepts 100-dB
Signal Range," Hewlett-Packard Journal, March 1974.
ABRIDGED SPECIFICATIONS
FREQUENCY RESPONSE: - 0 25 OB from 10 Hi lo 13 MHi referei
HP Model 3571A Tracking Spectrum Analyzer
Frequency
RANGE- 10 Hi to 13MHz
SELECTIVITY: 3 Hz W 10 K
» Shape li
Amplitude
ME SUREMENT RANGE 150 dB
DV AMtC RANGE: 70 dB
SE SmviTY Nominally 126 dBV ai 1 NH; on all ranges Mai
45 d6V on 60 dBV range wi!h 3 Hz BW from 0 1 to 10 MHz
33 30 A, B
AB OLUTE ACCURACY: calibrated to 33308 3330A or external
LINEARITY
30 06
o ma»
input level at 250 kHz
RESPONSE TIME: 0 4 ms ( 10 *Hi BW) lo 1 25s (3 Hz BW). with SMOOTHING on.
becom«s some 20 • longer up to max ol 2 5 s
DISTORTION RESPONSES. 80 dB below lull scale
SPURIOUS RESPONSES: 70 dB I3330A.B) of 60 dB I3320A B) Below full scale
POWER LINE RELATED RESPONSES' 70 dB on - 10 dBV throogh 40 08V
•anges 60 dB on 50 dBV range and 50 dB on 60 dBV range
IF REJECTION: 80 dB
OUT-OF-BANO REJECTION: 70 dB
DISPLAY RANGE - 1 99 99 dB
READING RATE: 4 readings per second
ANALOG OUTPUT: 10 OBVDC ± 13 5V range
SYNTHESIZER INPUTS irear panel) 20-33 MHz tracking signal and 1-MHz
.
60 d
Input Characteristics
13MHz. 50 orims 30 dB return loss lo 13 MHz
ATTENUATOR ACCURACY: • 0 07 dB per slop total accumulation 0 15 dB
General
OPERATING TEMPERATURE • 20 C lo • 30 C can work trom 0°C lo -55
with degraded accuracy
POWER: 100 120 220 240V 10*. -S". 46-66 Hz. 230 VA max
DIMENSIONS: 425mmW - 133mmH • W3mmD(l6B5 • 522 • 21 27,ncne;
WEIGHT 16 7 Kg 137 IDS)
PRICES IN U.S.A.:
Model 3044A opt 100 (standard Model 3571A| $6250
Model 3044A opt 200 I3571A and 3320A Synthesize-) $6495
Model 3045A opt 100I3571A 33306 Automatic Synthesizer 9620A Calculat
with pertinent ROMs interface and cabkng) $22 400
MANUFACTURING DIVISION: LOVELANO INSTRUMENT DIVISION
PO Box 301
B15 Fourteenth Slreet S W
Loveland Colorado 80537
Robert L. Atchley
Bob Atchley joined HewlettPackard in 1968, going right to
work on the Model 3570A Network
Analyzer where he eventually
assumed responsibility for the
digital processor. He did the same
for the Model 3571 A Spectrum
Analyzer. Married, and with two
children, Bob once took his family
to Bangkok, Thailand, (1967) to
work on a study project for Colo
rado State University, where he
had earned BSEE and MSEE
degrees. Bob also dabbles in
photography.
Jerry W. Daniels
Jerry Daniels worked on aero
space projects while earning
BSEE and MSEE degrees at the
University of California at Berkeley
in a work-study program. He
joined Hewlett-Packard in Janu
ary 1969 where he worked on the
mixer and D-to-A converter for the
Models 3320A/B and 3330A/B
Frequency Synthesizers. He then
contributed to the Model 3570A
Network Analyzer before moving
on to the 357 1 A. Married, but with
no youngsters, he enjoys skiing
and backpacking.
Bulk Rate
U.S. Postage
Paid
Hewlett-Packard
Company
Hewlett-Packard Company, 1501 Page Mill
Road, Palo Alto, California 94304
Volume 26 • Nu
Technical Information from the Laboratories of
Hewlett-Packard Company
Hewlett-Packard S.A . CH-1217 Meyrin 2
Geneva. Switzerland
Yokogawa-Hewlett-Packard Ltd.. Shibuya-Ku
Tokyo 151 Japan
C A 8LACK8URN
Administrative Services. Typography • Anne S. LoPresti
European Production Manager • Michel Fogiia
20910.
dele'te
your nameVorri bur MM ^ ÃÃ J.c f,,c=ol. jl.tlj?,__ . is lable lit peels off)
/*"* I I A K \f~^ off) f^l A I "1 I 1 R F-Ç Ç • To change your address or delete oryour
ñámélrorrT'ou'rtTfJItfiQIÃ,
v_y I! /\ I N vj L V_X I f\ \— J I — ' 11 d. O O . Send changes to Hewlett-Packard Journal. 1501 Page Mill Road. Palo Alto California 94304 USA Allow 60 days
© Copr. 1949-1998 Hewlett-Packard Co.
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement