1971 , Volume , Issue Jan-1971

1971 , Volume , Issue Jan-1971
© Copr. 1949-1998 Hewlett-Packard Co.
A New High-Speed Multifunction DVM
Plug-ins provide true rms ac capability as well as dc
and ohms. Reading speed is 1000 per second of ohms
and dc.
By Craig Walter, H. Mac Juneau and Lee Thompson
THERE is A NEED today for 'horizontal expansion' of the
capability to measure dc and ac voltage, and resistance —
more accuracy in general applications, with good repeat
ability. In addition, there is a need to reduce the difficulty
of eliminating errors under conditions such as making a
floating dc measurement in the presence of both common
mode and normal mode error signals, or avoiding large
errors when measuring distorted sinusoids or waveforms
without zero axis symmetry.
For bench users, the need is not to make an already
difficult measurement with greater precision, but there
is a need to make measurements of adequate resolution
with more ease and reliability. The measurement prob
lems of the bench user have often been ignored in favor
of 'greater' or 'more' rather than 'better'. Recently some
instruments have been compromised in favor of the 'sys
tem' aura — instrument optimization for system use at
the expense of, rather than for, the bench user.
The system user can generally find a unique solution
to his unique problem. Having found it, he can operate
his system properly. He has the time and generally the
capital to find unique solutions to his individual prob
The bench user has a difficult problem each time he
uses the instrument. A lash-up that provides optimum
results for a measurement one day cannot be expected
to yield the same results if the measurement problem
changes. The bench user's measurement problems vary
from day to day, and he seldom has time or money to
invent solutions for each problem.
Nevertheless, the problems associated with instrument
use in both applications are not totally unique to either.
A great degree of commonality exists. Providing solu
tions for the common measurement errors that exist in
bench applications can result in an instrument useful
for systems use at little or small expense to the bench user.
Indeed, the primary distinction between the two appli
cation areas often is in the speed (and end use) of the
measurement, not in the measurement itself. The instru
ment, in bench applications, is interfaced to a human;
in system applications, to a machine. If the same meas
urement can be made with greater speed (and if, in addi
tion, the data collected is easily transferred to a machine
as well as a human), the instrument can also satisfy many
system requirements. It should be possible, then, to pro
vide an instrument that could properly be called a hybrid
— optimized for bench and widely useful in systems.
Cover: Both half-module
and rack versions of the
Hewlett-Packard Model
3480A/B are shown. Its
reading speed is 1000 per
second for ohms and dc to
1000 volts. True rms ac is an
option. This article dis
cusses dc processing and
preconditioning as related to the Model 3480 A/ B.
The ac conversion technique will be covered in
detail in a future issue of the Hewlett-Packard
In this issue:
A New High-Speed Multifunction
DVM, by Craig Walter, H. Mac
Juneau and Lee Thompson
© Copr. 1949-1998 Hewlett-Packard Co.
page 2
Fig. multi-function DVM new Hewlett-Packard Model 3480A/B is a high speed multi-function DVM
capable of making 1000 readings per second up to WOO volts dc, and ohms down to 100
ohms full scale. It comes in both halt module and rack versions. Also shown here are
the Models 3481 A Butter Amplifier (with one 10 V dc range), 3482A DC Range Unit and
the 3484A Multifunction Unit.
Bench Features
Many features of the new HP Model 3480A/B, Fig. 1,
which may seem mundane by themselves, combine to
make the instrument easy to use. The display is easily
readable, function and range information and the instru
ment's functions are readily apparent. Autoranging is
complete through all ranges and functions. The first read
ing after an autorange cycle will be correct. The sample
rate is fully controllable, from a sample initiated by a
front panel pushbutton to >25 readings per second;
higher sample rates — to 1000 per second — can be initi
ated by external commands. Selectable filtering is pro
vided for normal mode signals so that the user has a
variety of choices between instrument settling time and
interference rejection. The instrument is fully protected
from damage from overvoltage, on any range, in any
function. Overload recovery of the amplifiers is fast
enough that a correct reading upon removal of the over
load is guaranteed, even though the reading is started at
the same time the overvoltage is removed.
Zeroing requirements are held to a minimum. The
instrument has accuracy commensurate with its resolu
tion — at moderate or extreme speeds. The ac converter
uses a thermopile to provide true rms conversion to elimi
nate common ac measurement errors. And of great im
portance, the instrument has minimum effect on the
device or circuit under test — all injected currents have
been eliminated or reduced to an insignificant level.
Measurement Speed
Terminology usually undergoes transformation when
applications are changed; hence, for system use, the
DVM is more properly called an A/D (analog to digital)
converter. There can be significant differences.
© Copr. 1949-1998 Hewlett-Packard Co.
For bench applications, the sequence of readings per
second should be quick compared with human reaction
time. A good performance criterion is the time required
for a single measurement, limited by the ability of the
analog circuits to respond to sudden input changes and
settle to the final value.
Most systems are without similar restrictions — data
can be absorbed at much faster rates. Thus, high reading
speed is a primary concern, provided, of course, that
response and settling times within the instrument are
commensurate with its reading rate.
Often speed is a major distinction. Because of the
design compromises generally required to obtain high
system speeds the system-designed analog to digital con
verter (A/D) is ordinarily less versatile than its bench
counterpart. It is task dedicated. A DVM, although slow
er, has varied functional capability and more versatile
signal preconditioning. Does the versatility necessary
for bench use preclude the measurement speed deemed
necessary for systems applications?
In many system applications, the Model 3480A repre
sents a solution to this apparent paradox. For systems
use it can be considered a comparatively slow A/D (its
maximum sampling speed is 1000/s) with 15 bit resolu
tion and much greater signal-conditioning capability than
the typical A/D. As such, it may be the best candidate
for certain system situations.
The digitizing technique used — successive approxima
tion — provides moderate speed at low cost. It is inher
ently simple and quite reliable. Reed relays in the A/D
converter are replaced with semiconductor switches. The
instrument accommodates plug-ins to provide signal pre
conditioning (giving great measurement versatility), and
the main frame contains the necessary power supplies and
the A/D. The performance of either section is comple
mented, not compromised, by the other.
To avoid placing an undesirable burden on the bench
user — added costs without benefit — the interfacing cir
cuitry required to communicate with other instrumenta
tion is not included within the basic instrument. These
interfacing circuits are, instead, available as options.
The emphasis during development was to capitalize on
the digitizing speed made available by the successive
approximation technique without compromising bench
performance or increasing the instrument's basic cost; to
solve those problems common to all traditional measure
ments, not those related to instrument use in a single
application — bench or system.
Measurement Errors
Errors associated with instrument use commonly fall
within three groups. One group is caused by the measure
ment circuit's interaction with its surroundings. The most
common sources are normal mode and common mode
generators which, directly or by magnetic coupling, in
duce unwanted currents in the measuring loop; the
sources can be magnetic fields from other instrumentation
or voltages generated because of the flow of relatively
large currents in the ground connections among instru
ments. A second error group is caused by interactions
between the measuring instrument and the circuit or de
vice under test. Common mode or normal mode sources
exist within virtually all instrumentation, and may
force 'injected currents' into the circuits being measured;
this may occur between 'high' and 'low' or between the
instrument's chassis and its other input terminals. These
currents can create errors by flowing through unbalanced
impedances in the input circuit or, more importantly, may
actually upset or change the characteristics of the circuit
under test. Third among error sources is those caused
directly by the instrument. The most obvious are errors
in amplification, attenuation, or conversion that somehow
modify the information being sought so that the data
presented as absolute may actually be in error.
Unfortunately, reducing the errors of one group does
not guarantee reduction of the others. In fact, the opposite
is often true. These sources of error, although classified
separately, must be treated simultaneously to minimize
the entire error matrix.
CM and NM Sources— Outside the DVM
The most general measurement situation is shown in
Fig. 2. A floating (above earth or chassis ground) meas
urement is to be made across a resistance bridge. Al
though the instrument ground and the source ground are
on the same line, a voltage generator (the common mode
source) will exist between them. The difference in the
ground voltage is primarily due to induced currents and
ground currents that flow between the two physically
isolated grounding points. This current creates a voltage
difference (the ground line or plane will always have
some impedance) whose magnitude depends on the hook
up and the environment into which it is placed.
The common mode generators (ac and dc) will not,
by themselves, cause measurement errors if the instru
ment impedance from chassis to each of its input termi
nals is infinite. Unfortunately, this is not generally the
© Copr. 1949-1998 Hewlett-Packard Co.
R, ¡s Parallel Combination
To High of R! and R3
Rb is Parallel Combination
of R i and R4
DC Common Mode is the
Voltage Across R*
b = R2//R4
To Low
Guard Should Go Here, but
this Point Doesn't Exist
in Actual Circuit
Fig. 2. A general measurement situation where a floating
measurement is made across a resistance bridge. Its
Thévenin equivalent is shown.
Possible Common Mode Currents
3480A DVM
normal mode errors — the common-mode signal has been
converted to one in series with the primary measurement
loop, that is, to a normal mode error signal. (The treat
ment of common mode errors, once the conversion to
normal mode occurs is then identical to that for normal
mode errors.) These signals create undesirable errors —
either dc offsets or a time varying voltage which may
cause the DVM's display to 'rack.'
Because the instrument's 'high' terminal is generally a
point or line rather than a plane, its impedance to chassis
is generally quite high and common mode errors in this
input lead can generally be ignored. If, however, the im
pedance from high to low (> 1010n, <50 pF for each of
the 3480's plug-ins) is not extremely high and a guard is
either not available or used, significant errors can result
— a 60 Hz common mode voltage of 10 V, an imbalance
R of 1 kn, and an input capacity of 1000 pF combine to
generate an error of 3.7 mV. Normal mode filtering can,
of course, reduce the ac errors but not without a corre
sponding increase in measurement time (the response
time of the filter must be added to the instrument's basic
digitizing time).
The instrument impedance from low to chassis is
usually much lower because of the instrument's physical
construction; 'low' will generally be a plane — capacities
of several thousand picofarads are not unusual. Guarding
must then be used to reduce these errors. Connecting the
guard, Fig. 4, will effectively bootstrap that portion of the
impedance between low and chassis that is terminated on
guard. It can be of little help, obviously, for that imped
ance not interrupted by guard.
nection Shunts Common Mode Current Away
From Source Resistances
Mode Source
Fig. 3. Possible common mode currents from external
sources. Assume Z6 « Z, and Z,.
case. Because impedance is finite between the instru
ment's high and low terminals and the point to which the
common-mode generator is referenced (the instrument's
chassis), Fig. 3, current will flow through each of the
imbalance resistors. The resulting voltage drop across
these imbalance R's will enhance or oppose Eln, creating
© Copr. 1949-1998 Hewlett-Packard Co.
Common Instrument Ground (Chassis)
Mode Source Added ¡, Add¡tjonal
strumentation Without
Isolation is Used.
Fig. 4. Proper guard connections will shunt most com
mon mode current away from source resistances.
The degradation in Z6 that is so undesirable can easily
occur when the instrument is interfaced to other equip
ment. If, for example, the data generated by the DVM is
to be used to provide hard copy, a printer or other record
ing device will be electrically tied to the DVM's data
output lines. If, too, the instrument is to be remotely
programmed or externally commanded, the program
source must be electrically connected to the DVM. As
the DVM's data programming lines are referenced to the
instrument's low terminal, a floating measurement will
be impossible unless this added instrumentation can also
be floated. Floating this instrumentation, unfortunately,
will reduce ZCl and cause deterioration of the system's
CMR. Injected currents from low to chassis will also be
significantly increased — unless the output or controlling
circuitry is completely isolated from its chassis. If the
output and control lines can be so referenced, and if iso
lation can be provided within the DVM, these system
errors can be eliminated. The Model 3480A/B and its
plug-ins have digital output and programming options
for necessary electrical isolation without degrading other
performance criteria (measurement speed, susceptibility
to electrical interference). The isolated programming op
tion also provides program storage.
The physical architecture of the Model 3480A/B has
eliminated the necessity for costly 'box-within-a-box-construction' (all internal circuitry surrounded by guard) —
yet CMR for the Model 3480A/B and any of its plug-ins
is >80 dB at 60 Hz for a 1 kn imbalance. Physical spac
ing between the internal circuitry ('low') and chassis is as
large as practicable. Where large spacings are impractical,
individual shields are employed. 'Box-within-a-box' con
struction is necessary only when much higher levels of
CMR are required. While reducing errors caused by ex
ternal CM voltages, this type of construction may accen
tuate measurement errors caused by injected currents.
Where guarding is required within the instrument (the
transformer, power supply heat sink, and plug-in covers
are the principal guard shields) the necessary care has
been taken to eliminate time-varying voltages in their
vicinity. The result is an extraordinarily favorable set of
tradeoffs. The complete measurement problem has been
taken into account. Errors from all sources, not just a few,
have been reduced together in appropriate amounts.
Normal Mode Filtering
Normal mode filtering, again, can be used to reduce
these errors, but measurement speed is reduced. The most
obvious solution to this tradeoff is to use a digitizing
technique that provides filtering — i.e., integration. Hence,
the recent popularity of dual-slope DVM's. This com
promise does not solve the entire measurement problem
unless the injected currents are also minimized. Those
currents may do nothing to the instrument because of its
inherent rejection, but can and do create subtle and seem
ingly mysterious changes in the circuit under test. Inte
gration is also only effective for noise whose frequencies
are related to (and multiples of) the converter's integration
period. Although filtering by integration can, theoreti
cally, give superior results at these discrete frequencies,
Fig. 5, little help is afforded the user whose noise is not
exactly synchronized to the DVM's integration period.
Moreover, an ideal converter using integration is limited
to a maximum sample rate of 60 Hz (and a correspond
ing minimum aperture time of 16.6 ms) if the CM fre
quency is exactly 60 Hz.
l/10sGate(100ms) I Essentially 'infinite1 46 dB
Typical Integrating
DVM (100 ms Period)
Fig. 5. Normal mode rejection at discrete frequencies
characteristic of a typical integrating DVM.
Normal mode rejection can also be achieved by passive
or active filtering or by a combination of filtering and
frequency conversion. A chopper stabilized amplifier is a
good example. Filtering may occur before, within or after
this input amplifier. All have relative advantages and dis
advantages — none is an ideal solution to the general
measurement problem. The degree of filtering required
for different applications will, of course, be different as
will the measurement times desired. There will always be
a speed/rejection tradeoff, stated or implied. Rather than
restrict the user to a fixed compromise — the amount of
filtering and the delays that are a necessary consequence
— filtering in the plug-ins for the Model 3480A/B is
selectable (see Specifications). The user can choose the
© Copr. 1949-1998 Hewlett-Packard Co.
compromise that best suits his needs.
Strong magnetic fields near the DVM can contribute
both common mode and normal mode errors if care is not
exercised in the design of the instrument. Here, filtering
may be of no direct benefit, for the injected currents may
be induced in the instrument's filter or in the circuitry
following the filter or the integrator (if the DVM uses
integration). A five gauss 60 Hz field (typically found
near the primary power section of most instrumentation)
can induce a peak-to-peak error as large as 30 ¿uV if the
circuitry within the field encloses an area of one square
inch. Loops can be generated by the improper layout or
design of either or both of the DVM's high and low leads.
Shielding will be of little help if the field is large enough
to cause saturation.
Reducing these errors within the Model 3480A/B and
its plug-ins is accomplished in several ways. All the input
wiring — high and low — form tightly twisted pairs to re
duce the area that may enclose the field. Where twisted
pairs are impractical, compensating loops have been
added. Wirewound resistors, notorious for their ability to
sense magnetic fields, even when 'non-inductively' wound,
have been replaced in sensitive areas by precision metal
film resistors.
CM and NM Sources Inside the DVM
Not all common mode currents (and the normal mode
errors they create) come from the circuit being measured.
Some common mode error sources are generated within
the measuring instrument. These are caused by currents
induced into the ground or guard shields by voltages
referenced to chassis, or into chassis from sources refer
enced to low. These internal common mode sources are
generally constant current sources and force 'injected
currents' into the circuits being measured, Fig. 6. Direct
measurement errors are the normal result, but by upset
ting the circuit under test or by changing its characteristics
during the measurement period, indirect errors often oc
cur — these errors, because they occur only when the
device under test is connected to the DVM, are often
impossible to isolate and identify.
The most common source of these injected currents is
the instrument's power transformer. The transformer, to
reduce its capacity from low (secondary) to chassis (pri
mary) is guarded. Capacity from chassis to low is inter
rupted by a guard shield between the two windings, Fig.
7. If the guarding is complete (C, quite small) the injected
current flowing through low will be correspondingly small
(if E, is 100 V at 60 Hz and d = 10 pF, the result cur
rent — 400 nA — will develop 400 pV across an Rb of 1 k,
Fig. 6).
Injected Currents
(From Transformer Primary)
Fig. 6. Internally generated common mode sources ref
erenced to chassis.
The injected current flowing through the guard termi
nal — caused by C2 — is of no consequence as it shunts the
measurement circuit. If, however, the guard is connected
to low, all the injected current will flow through the meas
urement circuit. This error can be hundreds or thousands
of times larger than the one previously calculated (C2 ^
1000 pF). This error source can be reduced by placing
an additional shield between the existing guard shield and
the transformer primary winding.
Other sources of similar magnitudes exist between the
transformer's secondary winding and its guard shield,
Figs. 7 and 8. These inject current from low to guard.
This injected current will only go through Rb if the guard
is properly connected; if tied to low the current is shunted
around the measurement circuit. So connecting the guard
to maximize CMR will also maximize the effect of this
particular injected current. Connecting it to minimize the
injected current will correspondingly reduce the CMR.
The need for this tradeoff can be eliminated if still
another shield — between guard and low — is added to the
transformer, Fig. 9.
The transformer construction in the Model 3480A/B
incorporates all three shields to reduce all of these injected
currents. Both the primary and secondary windings are
completely enclosed in 'box' shields tied to their respec
tive grounds. A guard shield between these two boxed
windings is used as a guard to maximize CMR. In this
manner injected currents are limited to a few nanoamps.
Shielding requirements for the transformer must also
extend to all primary and secondary wiring — in fact to all
time varying voltages within the DVM. The primary wir-
© Copr. 1949-1998 Hewlett-Packard Co.
Fig. 7. A typical guarded transformer and capacitances
Fig. 8. Internally generated common mode sources ref
erenced to 'low.'
ing is physically isolated and shielded from circuitry
referenced to low. The secondary wiring and all the recti
fication and regulation circuitry used for the DVM's in
ternal power supplies are also shielded and isolated from
chassis (the instrument's frame).
Timing, gating, and external display circuitry must
also be shielded (and guarded, if necessary) from the
instrument chassis. Logic circuitry that generates internal
timing and gating is located on a single printed-circuit
card in the middle of the instrument — boards on either
side shield it from chassis. The sample rate generator that
initiates each sample is coupled to the plug-in by a steady
state voltage, not one that is time varying. Time varying
voltages on the 'mother board' are shielded and guarded
from the chassis by another board below. This board,
physically attached to the mother board, also provides the
mechanical stiffening necessary to insert and extract the
other plug-in cards from the mother board.
Gas discharge display tubes are also isolated. Because
of the differences in the glow area of the various segments
(glow tube cathodes), each has a different sustaining
voltage. Because of the differences in the spatial arrange
ment of the segments, every unlit segment assumes a
unique voltage — a voltage that will be dependent on the
lighted segment. These voltage variations can be in excess
of 40 V. When the display is changed, these voltages also
change, and create large voltage transients that can gen
erate large injected currents (the capacity between the
Nixie segments and the instrument's chassis is relatively
large because of the glow tube's large surface area). Isola
tion is achieved by depositing a metallic coating (tied to
low) on the inside of the plastic window that is the front
of the mainframe. This conformal coating, although it
does provide significant attenuation of the broadband
noise generated by these display tubes, is not sufficient to
reduce the injected currents to the nanoamp levels de
sired. Therefore, buffering between the decoder drivers
and the D/A logic has been added. The display is changed
only once — at the completion of each reading. From a
visual standpoint, buffering is not required because of the
relatively small digitizing time of the A/D — the human
eye could not detect the change in the voltage displayed
during digitizing.
Shielding has also been added to the board on which
the D/A converter and the comparator are located.
Shielding is required here because the physical spacing
between the components on the board and top cover is
not sufficient to reduce the injected currents from these
sources to acceptable levels. These components are delib
erately near the top of the instrument to provide easy
accessibility to calibration potentiometers. The heat sink
on the regulator card must of necessity, be tied to guard
(its capacity to the instrument top cover cannot otherwise
be tolerated). Here again, all time varying circuitry refer
enced to low has been physically removed from the vicin
ity of this shield to reduce injected currents. Time varying
lines between the mainframe and plug-in are either
shielded (and guarded, as necessary) or have voltage
levels reduced commensurate with the injection desired.
This method of guarding and shielding to minimize
injected currents without the necessity of sacrificing CMR
in actual use has also been used in the plug-ins, the iso-
© Copr. 1949-1998 Hewlett-Packard Co.
Fig. 9. Manner in which the transformer in the Model
3480A is shielded and guarded.
lated digital output option (for the mainframe), and the
isolated programming option (for the plug-ins) . The total
injected current from all sources has been reduced to a
few nanoamps — a level sufficient to eliminate any error
when moderate unbalances are used, regardless of the
guard connection.
Current injection from the instrument high terminal to
low will also cause an obvious error. Leakage or injected
currents are dc rather than time varying (currents used
to bias the input amplifier or currents from improperly
shielded power supply voltages). The total input error
involves not only the instrument's input resistance but
also this dc leakage current. An instrument with 1012f2
input resistance may have a leakage current of 1 nanoamp. If a source resistance of 1 Mn is used, little loading
error will result, but the offset caused by the leakage
current will be 1 millivolt. The plug-ins for the Model
3480 A/B have an initial offset current of < 10 picoamps;
its change with temperature (perhaps of more impor
tance) is less than 1 picoamp/°C.
Input impedance (and offset current) of the instrument
is also constant with time or sample rate. 'Kickback' cur
rents that might adversely affect measurements from rela
tively high source impedances have been eliminated.
DC Signal Preconditioning
Conditioning the signal voltage to the nominal value
required by the A/D (10 V) within the accuracies desired
(±0.005%) can be achieved easily. It is more difficult
though, if characteristics other than accurate amplifica
tion or attenuation are also required. Some of the more
obvious requirements are: moderate bandwidth (20 kHz
@ A = 40 dB); wide dynamic range (0 V to ±15 V at
the instrument's input); extremely high input resistance
( > 1010n) ; very low offset voltage and current «1 ¡J.V,
1 pA at the amplifier input); and low sensitivity to power
supplies, source and load impedances, temperature and
All but the requirement for bandwidth are associated
with the design of any low level dc amplifier. The actual
operating characteristics desired are fast recovery from
overload ( >50 /is) and a slew rate and settling time fast
enough to make useful the A/D digitizing time. These
requirements imply a bandwidth in excess of 20 kHz. To
satisfy the bandwidth requirement only is simple. But to
satisfy both the requirement for dc preconditioning and
for bandwidth requires greater sophistication.
Chopper amplifiers (which up-convert the signal to
some low carrier frequency, amplify, then down-convert)
will not normally satisfy the bandwidth requirement. Such
amplifiers are normally used only to amplify dc and lowfrequency voltages near dc. To amplify the higher fre
quencies, an ac-coupled amplifier can be paralleled, but
this is, of course, more complicated and more costly.
Up-conversion to a much higher frequency carrier (mega
hertz), as in a parametric amplifier, accommodates the
frequency range requirement, but adequate amplifier ac
curacy and dc stability (variation of the offset voltage
and current at the amplifier input with time and tempera
ture) are difficult to achieve.
A direct-coupled amplifier of unusual design was the
final choice for the instrument, satisfying performance
requirements at reasonable cost.
A matched pair of field-effect transistors is used for the
differential input stage, Fig. 10. The FET offers both the
high input resistance and the small leakage current re
quired. Bipolar devices, though not necessarily limited
by their lower input resistance (although lower than the
FET, it is boosted by the amplifier's loop gain) have con
siderably more leakage current. The FETs are operated
in a balanced common drain configuration to achieve
minimum sensitivities of offset voltage and gain to varia
tions in power supply and device parameters. High CMR,
>80 dB, is not readily obtainable in a common source
configuration because of the inherent mismatch of the
two discrete devices.
To reduce parameter variations caused by temperature
fluctuations, the FET environment is temperature con
trolled at a temperature higher than the maximum ex-
© Copr. 1949-1998 Hewlett-Packard Co.
Bootstrap Zener
(Voltage Shifter)
Fig. 10. Input stage ot the
Model 3482A and Model 3484A
dc preconditioning amplifier.
pected ambient. An integrated circuit is used as an 'oven'
to maintain the FET at constant temperature. The mono
lithic 1C has within it all the circuitry normally associated
with an oven and its control circuitry — heaters, tem
perature sensors, and amplification. The FET dice,
mounted atop the 1C, assume the temperature of the
larger chip. Although the temperature control does oper
ate open loop (the sensing devices are within the 1C, not
the FETs), the resultant thermal gain (AVGS without
temperature control/AVGs with temperature control) can
be quite high (AT > 100). A high thermal gain, however,
does not guarantee a reduced offset voltage temperature
coefficient. The thermal gains of the two devices must be
matched because of their large initial TC (—600 /¿V/°C),
i.e., if the two devices are ideally matched initially but
AT = 100 for one side and 1000 for the other, the net TC
will be 5.4 fiV/°C, not zero. Compensation must be used
to reduce the effects of open loop control.
The device is constructed with obvious symmetry
about the two FET dice, Fig. 1 1 . Thermal gradients
across the face of the 1C have been reduced by an
anodized aluminum heat sink (0.001" X 0.030") be
tween the 1C and the FET chips. The epoxy used for
mounting the FETs and the 1C is thermally conductive
although electrically resistive. The aluminum bonding
wires (1.5 mil diameter) used for connecting the FETs
are thermally bootstrapped. Such bootstrapping is re
quired to minimize the effect of the heat conduction
through the bonding wires — >60% of the heat lost. If
heat is lost unevenly, the gradients that result are severe
enough to seriously degrade both the absolute value of
thermal gain achieved and the resulting match in thermal
gain between devices.
The FETs are operated at 80°C, a temperature high
enough above ambient to allow regulation when the
instrument is operated at elevated temperatures. Com
pensation reduces the offset current initially to < 10 pA
and attains a composite TC of <1 pA/°C. To keep the
resulting offset current independent of input voltage level,
the compensation circuitry is bootstrapped.
Even though the temperature of the FET is controlled,
the offset voltage temperature coefficient of the FET,
combined with the rest of the amplifier, may still be
greater than desired. To reduce this to <±1 /¿V/°C the
entire amplifier is temperature compensated. The ampli
fier's temperature is varied and its TC calculated.
Resistive compensation is added, that is the amplifier's
TC is changed and the amplifier is rerun until the desired
TC is achieved. Compensation is achieved by varying the
VBE match of the first bipolar gain stage. Its TC match
will change by approximately 1 /xV/°C for each 300 /¿V
mismatch in its base-emitter voltage. (1) Resistors are
added in series with the emitters of these transistors to get
the required mismatch. As the stage is driven by a current
source, the voltage drops across the emitter resistors act
as small batteries to mismatch AVBE.
To take full advantage of the DVM's reading speed,
field effect transistors are used for dc range switching.
© Copr. 1949-1998 Hewlett-Packard Co.
FET Chips
AL Strip
Fig. compensation. Construction of the FETs on a chip with temperature compensation. Heaters
are on the same chip.
Input voltages from 10 to 1000 V are divided down to
10 V by reed relays. Ranging is then by FETs. The FET
'on' resistance, although several hundred ohms, is in
series with the amplifier's input impedance ( > 10°n) and
thus creates little error. The leakage current from the
inverting side of the amplifier (and from the 'off switches)
flows through the 'on' switch and, on the two lower
ranges, 10 kn, Fig. 12. Even though this leakage current
can be as large as 1 nA, it is relatively constant because
of the controlled environment of the FET used as the
amplifier input stage. The offset it creates can be removed
initially by zeroing. Any changes with temperature are
accounted for by the compensation technique previously
Fig. 1 2. DC amplifier gain switching.
© Copr. 1949-1998 Hewlett-Packard Co.
What is the
HP Model 3480 A?
The Model 3480A/B is a 4-digit digital voltmeter (with
50% overrange), an A/D converter with moderate speed
which, when combined with one of its plug-ins, can provide
multifunction measurement capability without many of the
limitations created by traditional measurement errors.
The 3480A/B mainframe (the 'A' is a half module; the
'B', a full rack module) uses plug-ins — Models 3481A, 3482A
or 3484A. The Model 3484A, with all options, has five dc and
true rms ac voltage ranges, and six ohms ranges. The Model
3482A has the same dc capability as the Model 3484A (it
cannot, however, be expanded to provide ac and ohms). The
Model 3481A has only a single dc voltage range. All plug-ins
fit either mainframe configuration.
Successive approximation is used for A/D conversion.
Because of the design of the analog processing portions of
the instrument (within the plug-in) and the means employed
for data or programming transfer, reading and recording
speeds up to 1000 per second are possible without per
formance degradation.
A true rms ac converter (an option within the 3484A) en
ables accurate voltage measurements to be made of wave
forms with frequency components from dc to 10 MHz. The
converter eliminates significant errors (when other conver
sion techniques are used) caused by small amounts of har
monic distortion present in most sinusoidal signals. Accurate
measurement of non-sinusoidal phenomena is also possible
— the full scale crest factor is 7:1. Because the converter
can be dc coupled, it also measures the rms value of a com
bined ac and dc signal. A dual-thermopile makes the con
version and is 30 times more sensitive than a single thermo
couple. This sensitivity permits accurate measurements on
the 100 millivolt range.
DC input errors between the high and low input terminals
are virtually eliminated by the combination of a constant
input resistance of >10'° ohms and a leakage current of
<10 pA. A three position input filter can be used to reduce
or eliminate measurement errors caused by normal-mode
noise. Errors caused by common-mode noise can be re
duced by using the guard. Injected currents flowing from
the low and guard terminals to chassis have been signifi
cantly reduced to minimize the effect the instrument has
on the device or circuit under test.
System options include isolated or non-isolated BCD out
puts and isolated programming inputs. Everything (except
terminal selection) on the DVM is programmable. A two
range, three terminal, dc ratio option is also available. The
variety of possible configurations available and the innate
operating features allow the user to easily adapt the instru
ment to his specific needs while simultaneously reducing
measurement errors.
Fig. 13. FET range switch.
Ql and its associated components drive Q2 'on' or
'off', Fig. 13. When the range line is open (high), Ql
is off and the gate is biased to the negative supply through
R4 and CR1. Grounding the range line reverses biases
CR1 — Q2 turns on and is zero biased through R3.
Paul Baird and Ken lessen contributed significantly to
the project's definition. George Latham was responsible
for the 3480A/B's electrical design, Dave Luttropp for
its mechanical design. Jim Arnold's suggestions greatly
increased the instrument's serviceability. John Hettrick
bore the primary responsibility for coordination of the
production transfer, Gregg Boxleitner the responsibility
for production and electrical design of the 3481 A. Karl
Waltz and Barry Taylor were jointly responsible for the
3482A, Mike Aken for the 3484A. Jerry Blanz was
responsible for the mechanical design of the plug-ins.
Approbation must also be given to Larry Lopp and Jerry
Harmon who contributed to the SFET's development and
to Larry Linn who made many significant contributions.
Too numerous to name but nonetheless indispensable are
the many other people involved in layout and manufac
turing. £
1 A. H. Hoffait and R. D. Thornton, "Limitations of Tran
sistor DC Amplifiers," Proceedings of the IEEE, February
© Copr. 1949-1998 Hewlett-Packard Co.
Electrical Isolation: Coupling from Low to Chassis
Transformer coupling is used to transfer the information
used for programming or digital output — from circuitry ref
erenced to low, to or from circuitry referenced to chassis.
This solution offers both speed and reliability — neither
achievable with reed relays. It has one inherent disadvan
tage, however. The successive approximation technique,
unlike integration, presents the data in a parallel, rather
than serial format. The information to be transferred is in
its final form. Data transfer, then, requires a separate trans
former for each bit — 32 for data, 15 for programming.
Integrating or digitizing techniques that use voltage to
frequency converters need provide transfer only for their
clock pulses (a single line). Decoding is then done in the
'out guard,' or chassis section of the instrument. Program
ming isolation is inherent in the reed relays typically used,
so additional transfer is not necessary.
The costs involved in implementing a multiple transformer
scheme at first appear prohibitive. But to convert first from
the parallel format to serial, transfer, and then reconvert is
also costly. Serial transfer also requires clocking to main
tain cogency. At the speeds deemed necessary for data
transfer, the injected currents caused were considered ex
cessive. The use of light isolators, although attractive, is
still somewhat expensive.
To reduce costs, the transformers used are simply pairs
of molded RF chokes. DC isolation is achieved by mounting
the chokes and their respective circuits on interfacing PC
boards — one referenced to low and one to chassis. Power
for the isolated or chassis side is provided by a separate
winding on the transformer and by a separate regulator —
both isolated from the instrument's ground (low).
The basic coupling circuit is shown below.
these common mode voltages may be switched at rapid
rates (when used with a scanner switching large high, low,
and guard voltages), the shielding and the guarding it im
plies are necessary.
A small capacity «1 pF) exists between each of the two
coils used for transfer. In the example shown, a large and
relatively fast common mode voltage (caused by switching)
may inject enough current into the base of the transistor
to cause the latch to change state. This injected current —
caused by the external circuitry, not by the DVM — can be
eliminated if a shield, tied to chassis, is used to interrupt
the capacity between coils. If information is to be trans
ferred in the opposite direction, the shield must be tied to
low. Unless these shields are properly terminated, current
injection will be enhanced and noise sensitivity reduced.
These shields, although providing the reduced sensitivi
ties desired, greatly decrease CMR. An additional shield,
tied to guard, is added to obtain a net reduction in capacity.
A compromise between CMR and noise sensitivity is re
quired, however, as common mode voltages may also exist
between guard and low or guard and chassis. As the largest
voltage change allowed is between low and chassis, only
shielding (no guarding) is incorporated on output lines; the
resulting decrease in CMR is tolerable (the instrument's sys
tem specification, regardless of option or plug-in, is 80 dB
at 60 Hz with a 1 k imbalance in either input lead). Guarding
of the input lines is tolerable because of the reduced voltage
allowed between low and guard.
Li and L; (100 /¿H and 220 /iH, respectively) provide a cur
rent step-up of approximately 2:1 — low to chassis. A high
current, low-voltage pulse, on transfer, is forced into the
transistor's base causing saturation and a resultant change
in state of the latch used to provide storage. Programming
transfer is accomplished in an identical manner — the
grounds are simply reversed. The coefficient of coupling,
M, is 0.3 to 0.5.
It is imperative that the programming and digital output
circuitry remain insensitive to externally changing voltage;
otherwise, false triggering or programming, or a change in
the output data could invalidate the measurement being
made. As the low and earth grounds can vary by as much
as 700 V (the maximum voltage that can be tolerated from
low to guard is 200 V; 500 V from guard to chassis), and as
A multilayer flex circuit is used to implement the shielding
and guarding between the coils. The shields are constructed
in a manner that eliminates any eddy currents that could be
induced in a solid sheet that would result in unacceptable
coupling losses.
Double shields at different potentials are offset to reduce
© Copr. 1949-1998 Hewlett-Packard Co.
HP Model 3480A/B
(With 3481A Buffer Amplifier)
HP Model 3480A/B
(With 3484A Multifunction Unit)
90 days (25°C, <95% R.H.):
±(0.01% of reading +0.01% of range)
0°C to 55°C: ±(0.001% of reading +0.0003% of range) per °C
Reading Period: 950 /is.
Reading Rate: Variable from 1 to 25 per s plus manual with front
panel controls; 0 to 1000 per s with external trigger.
RESPONSE TIME: 1 ms. Reads to 'within 1 count of final reading
when triggered coincident with step input voltage.
COMMON MODE REJECTION: >80 dB, dc to 60 Hz with 1 kS2 in
either lead.
±1000.0 mV.
±10.000 V.
±100.0 V.
±1000.0 V.
OVERRANGE: 50% on all ranges. ±1200 V max input.
RANGE SELECTION: Manual, automatic or remote.
AUTOMATIC RANGING: Upranges at 140% of range; downranges at
10% of range.
Reading Period: 950 /¿s.
Reading Rate (without range change): Variable from 1 to 25 per s
plus manual with front panel controls; 0 to 1000 per s with ex
ternal trigger.
Autorange Time:
Filter Out: 4 ms per range change.
Filter A: 200 ms per range change.
Filter B: 1 s per range change.
Response Time: (without range change):
Filter Out: 1 ms to within 1 count of final reading when triggered
coincident with step input voltage.
Filter A: 200 ms to within 1 count of final reading.
Filter B: 1 s to within 1 count of final reading.
100 mV, 1000 mV, 10 V ranges: >1010 S2.
100 V, 1000 V ranges: 10MS2±0.1%.
of the peak common-mode voltage to the resultant error in reading
with 1 kii unbalance in either lead.
DC: >80 dB.
AC (50-60 Hz):
Filter Out: >80 dB.
Filter A: >110 dB.
Filter B: >160 dB.
NORMAL MODE REJECTION (NMR): NMR is the ratio of the peak
normal mode signal to the resultant error in reading.
Filter Out: 0 dB.
Filter A: >30 dB at 50 Hz and above.
Filter B: >80 dB at 50 Hz and above.
Manual or Remote.
OHMS, Option 042
1 000.0 Q
10.000 kS2
100.00 kiJ
1000.0 kS2
10.000 MO
OVERRANGE: 50% on all ranges.
RANGE SELECTION: Manual, automatic or remote.
AUTOMATIC RANGING: Upranges at 140% of range; downranges at
10% of range.
90 days (25°C ±5°C, <95% R.H.):
1000 Ã! thru 1000 kS! ranges: ±(0.01% of reading +0.01% of range).
100 G range: ±(0.02% of reading ±0.05% of range).
10 Mfi range: ±(0.1% of reading +0.01% of range).
Reading Period: 950 /¿s.
Reading Rate (without range change): Variable from 1 to 25 per s
plus manual with front panel controls; 0 to 1000 per s with ex
ternal trigger.
HP Model 3480A/B
(With 3482A DC Range Unit)
±1000.0 mV.
±10.000 V
±100.0 V
±1000.0 V
OVERRANGE: 50% on all ranges. ±1200 V max input.
RANGE SELECTION: Manual, automatic or remote.
AUTOMATIC RANGING: Upranges at 140% of range; downranges at
10% of range.
Reading Period: 950 iis.
Reading Rate (without range change): Variable from 1 to 25 per s
plus manual with front panel controls; 0 to 1000 per s with ex
ternal trigger.
Autorange Time:
Filter Out: 4 ms per range change.
Filter A: 200 ms per range change.
Filter B: 1 s per r&nge change.
Response Time (without range change):
Filter Out: 1 ms. Reads to within 1 count of final reading when trig
gered coincident with step input voltage.
Filter A: 200 ms to within 1 count of final reading.
Filter B: 1 s to within 1 count of final reading.
100 mV, 1000 mV, 10 V ranges: >1010 fi.
100 V, 1000 V ranges: 10MÃ2 ±0.1%.
of the peak common-mode voltage to the resultant error in reading
with kl2 unbalance in either lead.
DC: >80 dB.
AC: (50-60 Hz):
Filter Out: >80 dB.
Filter A: >110 dB.
Filter B: >160 dB.
NORMAL MODE REJECTION (NMR): NMR is the ratio of the peak
normal mode signal to the resultant error in reading.
Filter Out: 0 dB.
Filter A: <30 dB at 60 Hz and above.
Filter B: <80 dB at 60 Hz and above.
Manual or Remote.
© Copr. 1949-1998 Hewlett-Packard Co.
Response Time (full scale step input):
100 !2 thru 100 kSi ranges (no filtering): 1 ms. Reads to within 1
count of final reading.
100 kS2 range (Filter A): 200 ms to within 1 count of final reading.
10 M'.J range (Filter A): 2 s to within 1 count of final reading.
Note: may to noise generated in the unknown resistance, filtering may
be required for quiet readings with inputs >100 kl.'. Response times
with below are proportional less than those shown for inputs below
full scale.
VOLTAGE ACROSS UNKNOWN: 1 V at full scale on all ranges.
FLAG (Print Command): Line remains 'High' during reading period.
Line changes to 'Low' to Indicate completion of reading period
and remains 'Low' until start of next reading period.
PROGRAM FLAG (3482A Option 021, 3484A Option 041 only): Line
remains 'Low' until Program is executed. Line then goes 'High'
upon execution, then 'Low' after programming is completed (~1 ms).
Non-isolated Remote Control is standard on the 3481A, 3482A and
ISOLATED REMOTE CONTROL (3482A Option 021, 3483A Option 041)
3482A Option 021, 3484A Option 041 will operate only with 3480A/B
mainframes equipped with Isolated Digital Output (Option 004). Iso
lated Remote Contr.ol for the 3481A is provided in the mainframe
Isolated Digital Output option.
True rms ac Voltage Option 043
1000.00 mV.
10.000 V.
100.00 V.
1000.0 V.
The Digital Output Options provide measurement data outputs in digital
form for printer and systems applications. In addition, input lines are
included to remotely control triggering of the 3480A/B.
Non-isolated Digital Output is available both as a factory-installed
option (3480A/B Option 003) and a field installable accessory
(HP 11147A).
POWER: 115 V or 230 V ±10%, 50 Hz to 400 Hz, 60 W max (including
plug-in, options, normal environmental conditions).
INPUT TERMINALS: High, Low and Guard terminals on both front
and rear panels of 3481A, 3482A and 3484A. Front/Rear selector
switch on front-panel of plug-in. High and Low Ratio Reference Input
terminals on 3480A/B rear panel. Low Ratio and Low Input termi
nals are electrically common.
3480A Basic Instrument: 11 Ibs, 12 oz (5,25 kg)
Including Options: 12 Ibs, 8 oz, (5,7 kg)
Shipping: 17 Ibs, (7,75 kg)
3480B Basic Instrument: 12 Ibs, 12 oz (5,71 kg)
Including Options: 13 Ibs, 8 oz, (6,15 kg)
Shipping: 18 Ibs, (8,1 kg)
3481A Net Weight: 2 Ibs, 11 oz (1,2 kg)
Shipping: 5 Ibs, (2,3 kg)
3482A Basic Instrument: 4 Ibs, (1,8 kg)
Including Options: 4 Ibs, 4 oz, (1,9 kg)
Shipping: 7 Ibs, (3,15 kg)
3484A Basic Instrument: 4 Ibs, 6 oz (1,97 kg)
Including All Options: 6 Ibs, 2 oz, (2,76 kg)
Shipping: 8 Ibs, (3,6 kg).
HP 11148A Plug-In Extender Cable for servicing all plug-ins.. $ 45
HP 11149A Remote Control Cable for all plug-Ins
$ 25
The following accessories add optional capabilities not included
with not basic instrument. Optional capabilities which are not
listed as accessories can be ordered only at the time of initial
purchase. The Isolated Remote Accessory, HP 11 151 A, can be
used only when the 3480A/B has the Isolated Digital Output
Option 004, which is not available as an accessory.
HP 11147A Non-isolated Digital Output for 3480A/B
HP 11151A Isolated Remote Control for 3482A
3 4 8 4 A ( r e q u i r e s 3 4 8 0 A / B O p t i o n 0 0 4 )
$ 2 0 0
H P 1 1 1 5 2 A O h m s C o n v e r t e r f o r 3 4 8 4 A
$ 2 0 0
H P 1 1 1 5 3 A A C C o n v e r t e r f o r 3 4 8 4 A
$ 8 0 0
H P 3 4 8 0 A V 2 M o d u l e M a i n F r a m e
$ 8 0 0
H P 3 4 8 0 B F u l l R a c k W i d t h M a i n F r a m e
$ 9 0 0
Main Frame Options:
O p t i o n
0 0 2
R a t i o
$ 2 0 0
O p t i o n
0 0 3
D i g i t a l
O u t p u t
$ 2 0 0
O p t i o n 0 0 4 I s o l a t e d D i g i t a l O u t p u t
$ 3 7 5
HP 3481A Buffer Amplifier (includes Single Range DC Voltage
a n d N o n - i s o l a t e d R e m o t e C o n t r o l )
$ 3 5 0
HP3482A DC Range Unit (includes 5 Range DC Voltage and
N o n - I s o l a t e d R e m o t e C o n t r o l )
$ 7 0 0
Option 021 Isolated Remote Control (requires Main Frame
with Option 004, HP 11149A Remote Cable furnished) $200
HP 3484A Multifunction Unit (includes 5 Range DC Voltage and
N o n - I s o l a t e d R e m o t e C o n t r o l )
$ 9 0 0
Option 041 Isolated Remote Control (requires Main Frame
with Option 004. HP 11149A Remote Cable furnished) $200
O p t i o n
0 4 2
O h m s
C o n v e r t e r
$ 2 0 0
O p t i o n 0 4 3 T r u e R M S A C C o n v e r t e r
$ 8 0 0
Loveland, Colorado 80537
OVERRANGE: 50% on all ranges. 1500 V peak max input.
RANGE SELECTION: Manual, automatic or remote.
AUTOMATIC RANGING: Upranges at 140% of range; downranges at
10% of range.
Reading Period: 950 jus.
Reading Rate (without range change): Variable from 1 to 25 per s
plus manual with front panel controls; 0 to 1000 per s with ex
ternal trigger.
Response Time (full scale step input, without range change):
AC Coupled: 1 s to within 5 counts of final reading.
DC Coupled: 15 s to within 5 counts of final reading.
CREST FACTOR: 7:1 at full scale. 70:1 at 10% of full scale.
Mainframes, Plug-ins and Options
DC Ratio 3480A/B Option 002
DISPLAYED RATIO: Display in all functions is proportional to the ratio
of the input voltage to the external 10 V dc reference voltage ap
plied to rear-panel Ratio terminals.
ACCURACY (with respect to external reference voltage):
10 V or 100 V ±5% external reference: Same as basic instrument
accuracy specifications.
10 V, 100 V +5% to +35% or 10 V, 100 V -5% to -13%:
Add ±0.02% of reading to basic instrument accuracy specifications.
INPUT CHARACTERISTICS (ratio reference terminals):
INPUT VOLTAGE: +10 V or +100 V (referenced to Low side of
10 V Ratio Range: 100 k!2 ±1.5%.
100 V Ratio Range: 100 kG ±0.5%.
Remote controls are selected by application of a 'Low' state (logical
'0') to the remote lines through a rear-panel connector.
ENCODE (external trigger): Initiates a measurement period. Actu
ated by application of 'Low' state for >50 /is. Line must be in
'High' state >50 tis before applying 'Low' state. Minimum time
between ENCODE commands: 1 ms.
INHIBIT (Interface Hold): Disables front-panel Sample Rate control.
RATIO SELECT (Non-isolated Remote Control only): Selects Ratio
Measurement (if mainframe has the Ratio option).
FILTER SELECT (3482A, 3483A only): Selects Filter A or Filter B;
one line per filter.
RANGE SELECT (3482A, 3484A only): Selects measurement range;
one line per range.
FUNCTION SELECT (3484A only): Selects measurement function; one
line per function.
PROGRAM (3482A Option 021, 3484A Option 041 only):
Accepts program commands when 'Low' state is applied for >50 us.
Prevents changes in previously selected program when 'High' state
is applied for >50 /is. Does not affect operation of ENCODE line.
A minimum of 1 ms must be allowed between PROGRAM and
ENCODE commands.
© Copr. 1949-1998 Hewlett-Packard Co.
H. Mac Juneau
Mac received his BSEE from
• Swarthmore College in 1961
I and his MSEE and Ph.D. from
I the University of Minnesota in
1 1965 and 1967 respectively.
After graduation Mac came to
Hewlett-Packard and worked on
ac converters for DVM's, a job
— which has kept him occupied
for three years.
Outside working hours, Mac
spends his time woodcarving
and welding.
Lee Thompson
I Lee Thompson received his
BSEE degree from the University
^M^^_ aaï. oÃ- Texas at Austin in 1966.
f He joined HP's Loveland Division
P ^ J that same year as a product
designer. Lee did the product
design and some circuit design
Jy ^3^,^ on the rms converter in the
HP 3450A before becoming
involved with the true rms
J c o n v e r t e r i n t h e H P 3 4 8 4 A , w h e r e
he has worked primarily on the
amplifier design.
* Lee received his MSEE degree
from Colorado State University in 1968 as a participant in
the HP Honors Cooperative Program. He is a member of
Tau Beta Pi.
Craig Walter
Craig earned a BSME from
Stanford University in 1961,
then continued at graduate
school while working part time
for Hewlett-Packard. After
receiving his MSEE in 1963,
Craig joined the HP Loveland
Division. In 1968 he assumed
responsibility for the four-digit
DVM program.
Craig's interests include
sailing, bridge and wood
working. He is also an ardent
sports enthusiast.
Hewlett-Packard S. A. 1 21 7 Meynn - Geneva, Switzerland • Yokagawa-Hewlett-Packard Ltd., Shibuya-Ku, Tokyo 1 51 Japan
Editor: R. Assistant: Jordan Editorial Board: R. P. Dolan. H. L. Roberts, L. D. Shergalis Art Director: Arvid A. Danielson Assistant: Maridel Jordan
© 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