Beruflich Dokumente
Kultur Dokumente
4, AUGUST1974 159
Absfract—The design of an integrated wide-band variable-gain mismatch in the quad transistors. On the other hand, the
amplifier with maximum dynamic range is approached by considering multiplier has a constant de output level which is most
three basic bipolar transistor configurations from which all others
suitable for cascading [5]. Gilbert’s quad also suffers
can be derived. The analysis of noise and of distortion shows the
from high distortion if the dynamic range is made large.
importance of transistor base resistance in all three circuits. On the
basis of these analyses, one configuration is shown to yield maximum In general, the output distortion can be reduced by
dynamic range, and this configuration is then used as the basis for using lower signal levels. However, the noise generated
the development of a new circuit called the improved automatic-gain in the amplifier then becomes important and limits the
control (age) amplifier. A unique biasing scheme allows a consider-
dynamic range. This noise is mainly due again to the base
able reduction in distortion and noise, together with a significant
resistances of the quad transistors and also to the col-
increase “in bandwidth compared with conventional circuits.
lector loads. A noise analysis for the three basic con-
figurations is given in Section II. It leads to the conclu-
I. INTRODUCTION sion that again the agc amplifier [Fig. 1 (a) ] is superior
v ARIABLE-gain
feedback
(age)
configurations
amplifiers
put signal is kept constant for all input
amplifiers are most often used in
as automatic-gain-control
where the amplitude of the out-
signal levels. The
to the
dynamic
sistors.
other
signal-to-noise
range
This
configurations,
upper
ratio,
that can
limit
there
and
exists
be achieved
value, then,
that
an
for
upper
with
depends
a given
limit
bipolar
only
output
to the
tran-
on the
ratio of maximum-to-minimum input signal amplitude transistor base resistance.
that can be handled is called the dynamic range. In order to realize effectively this optimum dynamic
Most wide-band variable-gain amplifiers can be de- range, the value of the total quad current ~EE has to be
rived from three high-performance amplifiers which all properly selected, Also, in order to maintain the optimum
consist of a quadruple of transistors driven by an input performance of the quad, the input pair [Fig. 1 (a)] has
pair. They are the age amplifier [1] shown in Fig. 1 (a), to provide a voltage-to-current conversion which is highly
the multiplier [2] shown in Fig. 1 (b), and Gilbert’s linear over a wide frequency range. This can be accomp-
variable-gain quad [3] shown in Fig. 1 (c). Considera- lished by a high pair current Ip. Thus, in order to achieve
tions are restricted to balanced circuit configurations be- optimum performance from the quad, a bias scheme must
cause of the resulting cancellation of even-order distor- be employed which allows differing values of 1P and ~EE.
tion [4]. Input and output ports are differential, and The agc amplifier which combines a low quad current ~EIS
input and output impedance are independent of amplifier with a high pair current 1P is the improved age amplifier.
gain. In order to reduce the noise contribution from fol- In Section IV the choice of the currents IBN and IP in the
lowing stages, the maximum gain is larger than unity and improved age amplifier is discussed.
is typically about 15 dB. This automatically excludes
II. ANALYSIS OF NOISE IN THE BASIC age AMPLIFIERS
varioloss circuits.
It has been previously [4] reported that the distortion The basic agc amplifiers all consist of a. quadruple of
in the basic amplifiers of Fig. 1 is mainly caused by transistors formed from emitter-coupled pairs. Therefore,
the base resistances of the quad transistors. For high- the dominant noise sources are examined first for an
dynamic range, the ag,e amplifier [Fig, 1 (a) 1 exhibits the emitter-driven pair (Fig. 2).
lowest distortion and is thus the best choice. The dynamic
A. A’oise Analysis of an Emitter-Driven Pair
range of the multiplier [Fig. 1 (b) ] is limited due to
In a transistor, the following types of noise are con-
sidered [7] : the thermal noise power generated in the
Manuscript received October 26, 1973; revised February 18,
1974. This research was supported by the U. S. Army, Research base resistance, and the shot noise powers caused by the
Office, Durham, N. C., under Grant DA-ARO-D31-124-72-G52. de current flow in the base and the collector. For the
W. M. C. Sansen was with the Department of Electrical Engi-
emitter-driven pair (Fig. 2), six noise sources have to be
neering and Computer Sciences and the Electronics Research Lab-
oratory, University of California, Berkeley, Calif. W720. tie is considered (two per transistor), and also the thermal
now with the Laboratorium Fysica en Elektronica van de Half- noise associated with resistances Rs, R,g, and RL. The
geleidem, Katholieke Universiteit, Leuven, Belgium. noise contributions at the output due to the most im-
R. G. Meyer is with the Department of Electrical Engineering
and Computer Sciences and the Electronics Research Laboratory, portant noise sources are plotted versus the relative
University of California, Berkeley, Calif. 94720. voltage gain x in Fig. 3. These curves are the results of
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on November 2, 2008 at 07:20 from IEEE Xplore. Restrictions apply.
IEEE JOURNAL OF SOLID-STATE CIRCUITS, AUGUST 1974
v.. v~~
. .
.
Q2
+ R~
I
J IEE
f?
-VEE
(R,) (b)
Vccl
< QI Q1>
? !
-vE~
(c)
Fig. 1. (a) age amplifier [11. (b) Multiplier [21. (c) Gilbert’s
variable-gain quadruple [31.
Vcc
-&
‘L
t.
1“
Q3 1)4
q
VB+
. /1~ ~
+
-r!!
vi
‘V~E
RE
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on November 2, 2008 at 07:20 from IEEE Xplore. Restrictions apply.
5ANSEN AND MEYER: VARIABLE-GAIN.4MPLIFIER 161
straightforward circuit analysis by hand, verified by com- If the pair is used as an agc circuit for constant out-
puter noise analysis with CANCER [6]. put signal, then the curve of output sig~~al-to-noise ratio
The relative voltage gain z is determined by the dc &/No is precisely the inverse of the noise characteristic
control voltage V~ and is given by of Fig. 3. The output signal-to-noise ratio is a more
useful measure of noise than the noise figure because
(1) the noise figure depends on the circuit gain, which is
variable. At maximum gain, the amplitude of the input
At full gain (Z = 1 or 0 dB attenuation), the pair behaves signal reaches its lowest value, but at half the maximum
as a cascode amplifier with gain RL/RE. The total noise gain (where the amplitude of the input signal is twice
output is low and mainly due to transistor Q1 and re- this lowest value), the value of SO/NO is minimum. For
sistance RE. However, at half gain (x = 1/2 or 6 dB at- higher input signals, the attenuation increases, and so
tenuation), the currents in transistors Q~ and QA are does S./NO [1] until SO/NO levels out because of the noise
equal, and the noise due to the base resistances of Qs and due to RT,. Thus, for a specific minimum requirement in
Q1 causes a large output noise peak with a value given by Se/NO over the whole dynamic range, only the amount
of noise at half the maximum gain has to be examined.
v,. = ~kT(r,>m ~ V,.. (2) Under all other gain conditions, &JiVo is higher and
automatically satisfies the requirement.
where VT = kT/q != 26 mV at 302 K and At represents For the multiplier used as a variable-gain circuit [Fig.
the noise power bandwidth. For increasing attenuation, 1 (b)], the maximum output noise is again due to the
the current in transistor Q4 decreases such that the con- base resistances of the quad. However, whereas for the
tribution at the output of the base resistance noise de- agc amplifier of Fig. 1 (a) the base resistance noise peaks
creases proportionally. at half the maximum gain and decreases at higher at-
If the pair current ID is very low, the base resistance tenuation levels, the base resistance noise for the multi-
noise peak given by (2) may become lower than the shot plier also reaches its peak at half the maximum gain,
noise of Q4 (which decreases only as ~~B), or lower than but then remains the same for higher attenuation because
the noise output of R~ and R~, which does not depend the currents in all four quad transistors are then nearly
on 1~. This is only true for very low values of In at equal in magnitude. This leads to the surprising result
which wide-band performance is hard to obtain. As a that S,/NO is minimum at high attenuation. The reverse
consequence, higher currents are usually chosen, and holds for the agc quad. It can thus be concluded that in
the maximum noise peak is thus entirely due to the base noise performance, the multiplier is inferior to the agc
resistances in the pair. It is given by (2) and occurs at quad since the noise peak due to base resistance, although
half the maximum gain. in amplitude the same for both, extends over all high
attenuation levels in the multiplier.
B. Comparison of the Noise Performance of the For Gilbert’s quad [Fig. 1 (c)], the :maximum output
Three Basic Quads noise voltage due to base resistance occurs at maximum
The noise output power of the agc quad is simply twice gain and equals 2@ times the noise peak for the agc
that of the emitter-driven pair. It is thus found from quad if base resistance and quad current are assumed to
Fig. 3 by shifting the curves up by 3 dB. Experimental be equal. Used in an agc loop, Gilbert,’s quad thus ex-
data show good agreement with the calculated curves. hibits its minimum SO/NO for its smallest input signal
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on November 2, 2008 at 07:20 from IEEE Xplore. Restrictions apply.
162 IEEE JOURNAL OF SOLID-STATE CIRCUITS, AUGUST 1974
unity. Maximum gain AV{~.x is achieved for all points on resultant dynamic range DR jumps up towards the po-
line AB. Lines parallel to Al? thus represent constant tential dynamic range DRP which is the maximum pos-
–A. lines (e.g., the AO,~.x –20 dB line). sible for the chosen value of VO. It is more practical from
Neglecting the effects of distortion and noise, the maxi- the viewpoint of the designer, however, to let the value
mum dynamic range DR,W for a specific output. signal of VO vary for a given minimum requirement n, and to
level VO equals the potential dynamic range DEP, which find out what maximum dynamic range can be achieved
is given by for each value of VO. The answer is given in Fig. 5 where
DR is plotted versus VOfor a given value of n and a con-
stant total quad current I~H.
DItP = 20 log,, ~Z dB (4)
o The maximum dynamic range DR1l is obtained at out-
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on November 2, 2008 at 07:20 from IEEE Xplore. Restrictions apply.
SANSEN AND MEYER: VARI.4BLE-GAIN AMPLIFIER 163
‘“_———_—l
g 40 –
-..._..-.r__-...
4x or 12dB
DRM
-1 60
z
.
%
.
s
,s
E ,
: 20
due 10 lc-
S
/:
due 10RL
OR limited by noise.,.
! ‘ore’ fi.%tJ
I I 1 I
o
‘0.01 0,I I 10 100
0.1 I 10 100
Output voltage v, [mV) Quad current IEE ( mAl
2VF
DR. = — (6)
nw’’lcT(r~,,, + r,,,,) Af (8)
and vO,,Lis then given by Since the circuit is balanced, ideally, no second-order
distortion is present.
RJBE 1
“m =
v— (7) For specific values of output signal-to-noise ratio n
DR~ “72”
and distortion d, the maximum dynamic range depends
This value of DR.Tl is the highest that can be achieved on the quad current 1~~, as it is shown in Fig. 6. For in-
with this type of bipolar transistor, It only depends on termediate values of l~n, the maximum dynamic range is
the base resistances of the quad transistors, whereas the optimum, as given by (6). For lower va Iues of IE~, noise
optimum output voltage vO,,, also depends on the quad due to R,, reduces the maximum dynamic range, whereas
current 1~~ and load RI,. If, for example, all four base for higher values of Ifl~, the maximum dynamic range is
resistances equal 100 Q and A~ = 4.5 MHz, then n“ DR,W decreased by distortion. The optimum value of Imn is thus
equals 83 dB from (6). If the required output signal-to- obtained at the edge of the distortion-limited region,
noise ratio is 40 dB, then 43 dB is the maximum dynamic which is designated by 1~~1. at low frequencies. For ex-
range. This is obtained at VO = 10.7 mVr~, if Rz = 500 fi ample, 1~~1,, equals 12 mA for c1 = 0.01 (1 percent IMs)
and 1~~ = 4 mA from (7). and rI)/P = 1. Actually, any current in the flat region
If the chosen output voltage VOis greater than vO,,,,the yields the same value of DR1,, but the output signal
potential dynamic range decreases [see Fig. 5 or expres- level VOis proportional to l~,j [see expression (7)] and is
sion (4) ] and so does DRJ1. Noise does not limit the thus better chosen as high as possible.
dynamic range, and the output signal-to-noise ratio is At high frequencies, the distortion increases consider-
then always greater than the required value of n. On the ably [4] with frequency. For a specific value of maxi-
other hand, if the amplitude of the output signal VO is mum distortion d, the optimum value of 1~~ (Fig. 6) thus
reduced to values below v~,~,, the dynamic range is lim- decreases with increasing frequency and is now desig-
ited.. by base resistance noise such that the dynamic range nated by 1~~,,,. If, for example, f~ = 6 MHz, the maxi-
becomes independent of v, at a value 12 dB below DRl~. mum dynamic range decreases by 20 dB at 60 MHz,
This is caused by the equal slopes (20 dB/dec) of line unless 1~~ ]s lowered to l~~i~i = 1.2 mA. Also at high
AB and the curve of noise due to base resistance (Fig 4). frequencies, distortion due to feedthrough combines with
Indeed, when VO decreases, point B moves down and to the distortion due to base resistance and becomes the
the left along line AB. The noise curve moves only to the dominant limitation of dynamic range. This is treated in
left such that CD remains the same. If the amplitude of more detail in the next section.
output signal voltage is reduced further, the dynamic
B. High-Frequency I,imitations
range is limited by noise due to Rz or 1.4, whichever is
higher. In this region, both the values of DRJ1l and VO At high frequencies, the maximum dynamic range is
have become very low and thus are of little interest. usually not reached because of signal feedthrough via
Distortion also limits the dynamic range which can be the junction capacitances of Q, and Qfi [Fig. 1 (a) 1. A
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on November 2, 2008 at 07:20 from IEEE Xplore. Restrictions apply.
104 IEEE JOURNAL OF SOLID-STATE CIRCUITS, AUGUST 1974
40 I I I ,
10dB
20dB
.30dB
40dB
/
I I I
-20
20 - 10dB
z
E
.:
‘o - 30 dB
1 I I
-20,
10
. 100 loco
Frequency ( MHz)
first-order estimate of this feedthrough is obtained in the following circuit considerations. At high gain, the output
following way. Under high attenuation conditions, tran- transistors behave as current-driven common base stages.
sistor Q3,6 is on, and thus represents at its emitter an The cutoff frequency is thus about J!”. However, under
impedance level which is never higher than ~E3,!3. Tran- maximum attenuation conditions, the output transistors
sistor Q4,5 is thus driven from a voltage source with QA and Q, exhibit a capacitive input impedance consist-
source impedance rB3,6 at. high frequencies and acts as a ing mainly of their junction capacitances CjE and Cjo.
second-order high-pass filter consisting of CjB, rBj ~jc, They are driven, however, by a current source shunted
a,nd RL, all belonging to QA,E. In this way, the feed- by the inductive input impedance of the other pair of
through is described by transition frequency ff given by transistors Q~ and QG. Peaking can thus occur. This can
be avoided, however, if the quad is biased at the quad
1/3
current given by [9]
“ = & ( T~3,f,TBr;ciEciC )
(9)
I . 2VT
—— (11)
EG —
where Cj~, Cfc, and w all refer to transistors Q4 and Q5. r~
where A is the attenuation (e.g., A = 100 for –40 dB In order to maintain the high performance of the quad,
attenuation). Expressions (9) and (10) have been veri- the input pair has to provide a voltage-to-current con-
fied by measurements on the agc amplifier [Fig. 1 (a)] version which is highly linear over a wide frequency
and by use of the circuit analysis program TIME [8]. range. This can only be accomplished by a high pair
The agreement is satisfactory except at very high fre- current Iv or a high value of degeneration resistance RE
quencies (above 300 MHz in Fig. 7) where transistors [Fig. 1 (a)]. Since a high R~ decreases the gain and
Q, and Q. can no longer be treated independently and causes a high noise output, the first solution is indicated.
where collector-substrate capacitances become important. On the other hand, the optimum dynamic range and the
Fig. 7 shows the transfer characteristic of the agc amp- flat transfer characteristic can only be realized at fairly
lifier [Fig. 1 (a)] for two different values of the quad low quad currents. As a consequence, different currents
current IE,E. For a high quad current, peaking occurs in are needed in the pair and the quad. The agc amplifier,
the transfer characteristic, which can be explained by the which combines a low quad current IEE with a high pair
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on November 2, 2008 at 07:20 from IEEE Xplore. Restrictions apply.
SANSEN AND MEYER : VARIABLE-GAIN AMPLIFIER 165
vcc 5
7
Hi
3.7k
8k
QIO
1.4k
Qll 8
. .
2.2k
2:k
II
1 A I
-L
-VEE
Fig. 8. Improved agc amplifier (ZP = IQ + Im).
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on November 2, 2008 at 07:20 from IEEE Xplore. Restrictions apply.
166 IEEE JOURNAL OF SOLID-STATE CIRCUITS, AUGUST 1974
mask processing. Two types of n-p-n transistors are used transistor variable-gain amplifiers,” IEEE J. Solid-State Cir-
in the signal path. The first is a small-size transistor and
caits, vol. SC-$, pp. 275-282, Aug. 1973.
[51 J. Addis, “Three technologies on one chip make a broadband
is designed to keep signal feedthrough as low as possible amplifier,” Electronics, pp. 103–107, June 5, 1972.
(Q Q5, etc.]. The other transistor type has a double base [61 L. Nagel and R. Rohrer, “Computer analysis of nonlinear
circuits, excluding radiation (CANCER) ,“ IEEE J. Solid-
stripe for the purpose of low circuit noise and distortion
State Circuits, vol. SC-6, pp. 166-182, Aug. 1971.
(~1, Q2, Q3, QG) . [71 E. M. Cherry and D. E. Hooper, ‘(The design of wide-band
The biasing scheme of the improved age amplifier has transistor feedback amplifiers,” Proc. Znst. Elec. Eng., vol.
110, pp. 375-389, Feb. 1963.
to ensure that the total quad current I~D, which is the
[81 F. S. Jenkins and S.-P. Fan, *’TIME—A nonlinear DC and
sum of the emitter currents of Qs, QA, Qb, and QG, is time-domain circuit-simulation program,” IEEE J. Solid-
stabilized with respect to transistor parameters. This State Circuits, vol. SC%, pp. 182-188, Aug. 1971.
[91 W. M. C. Sansen and R. G. Meyer, “Characterization and
must be achieved even when IEE is small compared with
measurement of the base and emitter resistances of bipolar
the total pair current Ip which flows throug,h Qls. Other- transistors,” IEEE J. Solid-8tate Circuits, vol. S(2-7, pp. 492-
wise, any inaccuracy in the determination of IP would 498, Dec. 1972.
Authorized licensed use limited to: NATIONAL INSTITUTE OF TECHNOLOGY WARANGAL. Downloaded on November 2, 2008 at 07:20 from IEEE Xplore. Restrictions apply.