1950 Solms: Logarithmic Amplifier Design 91

TABLE 11
DIFFE-RENTIAL-AmPLIFIER EQUIVALENT INPUT DRIFT, USING
HA-7500 FUSED UNITS
NUMBER 8(dc) co
A Vill MV 250C to 1250C 25"C 1250C '250C Rbi
Transistor Pair Vi,, At 25'C mv 27 35 3m,4A 2.2K
Measurements Calculations 10
2 18 23 5m.LA
1 2 9 2 3
3 18 +24 +7 +8 lKfl
4 5 - 8 -3.5 -2.7 5
5 6 - 5 -0.5 -i.7 Rbl=
7 19 -21 -6 -7 b2=0
8 9 +10 +2.5 +3.3 o Rb2'
10 16 - 8 -2.5 -2.7 0
270D

11 12 + 5 +1.5 +t.7 >

13 14 + 3 +1 +1
15 17 +18 +5 +6 -5 --j IKII

TABLE III -10

DIFFERENTIAL-AMPLIFIER EQUIVALENT INPUT DRIFT, USING 2.2K
GROWN UNITS
-15
1C1=IC2=0.5,nA
20 21 +77 +20 +26 0 MEASURED
22 23 + 4 +17 + 1.3 CALCULATED
24 25 - 7 - 3 - 2.3
26 27 -41 -25 -14 0 25 50 75 100 125 150
28 29 -27 -13 - 9
DEGREES C
30 31 +13 - 9 + 4.3
32 33 +26 + 9 + 8.7 Fig. 12-AVi.,, vs T (HA7523).

ogarit mic i8vmp i er esign

S. J. SOLMSt
INTRODUCTION special nonlinear design. The required nonlinearity is in
r- HE logarithmic amplifier has a wide applicability the
form of a transition from normal stage gain to un'ty
in electronic instrumentation. The logarithm is of gain at a certain signal amplitude. The amplifier ob-
special interest where it is necessary to compress tained by cascading
such stages is actually linear f..,'-om
wide dynamic range into a narrow range since it is the zero signal up
a
to the level at which the first transition
optimum function for representation of data with equal
takes place and is commonly called a "lin-log" amplifier
for this reason.
weight irrespective of the decade in which it occurs. For This paper presents a rigorous development of this
this reason, it has also found use in pulse receivers to
provide instantaneous automatic gain control. The log- method of approximating the logarithmic fuiiction. It
arithmic aniplifier is useful, too, in analog computer discusses the basic design of a lin-log amplifier and in-
circuitry as it provides a simple means for performing
cludes such problems as bandwidth and temperature
A i-rnnQ;-zf-nr 1;n_1no, nmn1;1;,:-r r],--z;0rn IlQincy,

Authorized licensd use limted to: Imperial Coleg Lond. Downlade on June 07,21 at 19:354 UTC from IE Xplore. Restricon aply.
92 IRE TRANSACTIONS ON INSTRUMENTATION December
the transition occurs at a breakpoint, Xb, Yb, determined y (OUTPUT)
by specifying the intercept a. The two line segments are
described by
y =mx for Xb _ > xX 0 (1)
y= sx + a for x. Xb. (2) Yb
The breakpoint of a stage is therefore located at y
Xb = a/(m - s), Yb = ma/(m - s). (3) /__
Suppose n identical stages are connected in cascade Kb X (INPUT)
and suppose the kth stage of the cascade is operating at Fig. 1-Amplitude characteristic of individual stage.
its breakpoint; its input is therefore Xb. If mr> 1, then the
stages preceding the kth stage will be operating in the PO
m-gain region. The input to the first stage corresponding
to the breakpoint of the kth stage (X1,k) is therefore ob- c

tained by k-1 applications of (1). 'j

X,k = a/(m -s)rl. (4) YK xn :- msp
The output of the kth stage at its breakpoint is Yb. If >_|
m> 1, then all stages following the kth stage will be
operating in the s-gain region. The output of the nth x( TO FIRST STAGE)
stage corresponding to the breakpoint of the kth stage Fig. 2-Over-all amplitude characteristic of cascade.
(yn,k) is therefore obtained by n-k applications of (2).
APPROXIMATION ERROR
yn,k= a [MSn-k/ (m Eqs. (9) and (10) can also be evaluated for k =0 and
- s)
+ sn-k-l + sn-k-2 + * + sol. (5) k=n+l. The significance of the points Po and Pn+1 is
most easily shown by a numerical example. Fig. 3 is a
Eqs. (4) and (5) give the coordinates of the breakpoints
of the over-all amplitude characteristic shown in Fig. 2. graph of an over-all amplitude characteristic for n =3,
is drawn' of
it segments with
As k takes on integral values from1 hruh
m=5,
abscissa. a=1,The s=1, andline
straight Fig.a 2logarithmic
are curved
amplifier input given by (4) takes on values which form ii i.3adfr napoiaint h oaihi
ouputfunction represented napoiaint
i.3adfr h oaihi
a geometric
a gemetrc progression.
prgresion.If he corresponding
If the crresondig output by the straight line passing through
values given by (5) can be made to form an arithmetic
progression, then all. the breakpoints will lie on a single the points Po through P The approximation error is
' . .....................cyclic
logarithmic curve. To check this possibility, equate the byond thee Point
between P0 and PPn+' bbut increases without limit
and reay therefore be
increments in Ynk obtained by letting k take on, suc- taken tenaas thesuppernand lowerdlimit
the upper and lower pointshofethe
limit points
og
of the log-
cessively, the vvalues
cesivly,th aue p p, ' p+1,
p p+2.
1 p This
2 Tis aybe
may be rite
written aihi prxmto o nfr ylcerr
~~~~~~~~~~arithmicapproximation for uniform cyclic error.
f
Yn,p Yn,p+l Yn,p+l
= = -
Yn,p+2. (6) The approximation error is the same for each segment
and is dependent only on the stage gain m. This state-
Evaluating this by means of (5) leads to the intermedi- ment is proved in the Appendix along with a derivation
ate step of the error magnitude. The magnitude of the peak
deviation of the over-all amplitude clharacteristic to
m(snP + 5n-p-2 - 2sn-P-')/(r - ) either side of the "median line" of Fig. 3 may be re-
+ Sn-P-I Sn-P-2 = 0. (7) ferred to the input and expressed in db. Fig. 4 is a graph
of this error as a function of the stage gain in db.
If s$w4-0, division by Sn-' iS possible, with n and p being
dropped out from the development. After further sim- DYNAMIC RANGE
plification, The dynamic range may be defined as the ratio of the
s(rn- 1) = m -1. (8) largest to the smallest input signal for which the ap-
If rn H1, then s = 1. It is thus demonstrated that all the
proximation error is not exceded, i.e.,
Fro (9) thi ratiois
±1. x1,0/x1,
breakpoints fall on a single logarithmic curve if s = 1. Frm() hsatoI
From (4) and (5), the coordinates of the breakpoints of X = m8l (11)
the ver-llapliude harateriticfor = 1areIn decibel form, i.e., log X, M= 20 log m, the fol- A-=20
X1,k = a/(m- 1)mk-l (9) lowing convenient design equation results:
yfl,k=as[r/(r-1) +n-kif (10) A =(n+ 1)M. (12)

Authorized licensd use limted to: Imperial Coleg Lond. Downlade on June 07,21 at 19:354 UTC from IE Xplore. Restricon aply.
1959 Solms: Logarithmic Amplifier Design 93
l l l BREAKPOINT LINE /X/ over-all gain. From (10) and with the use of the same
MEDIAN LINE Yes assumptions, it is easily shown that for in-log amplifier
design the optimum number for maximization of small
TANGENT LINE signal bandwidth is
Yn2 |-X -- */Bno = In X-1 + [(ln X - 1)2 -I1/2 (14)
provided X > e2. For X < e2, the optimum number is one.
CIRCUIT DESIGN
-4-? LOG X,
LOG xi Nonlinear feedback in a transistor operational ampli-
Fig. 3-Over-all characteristic in semilogarithmic plot.
.fier in the common emitter configuration, as in Fig. 5, .
is one possibility for approximating the required two-
_._ - - - _- slope characteristic. This configuration has the advant-
a04_l_-
z
_
_
.
- -- I - -
-
-
-
age that with sufficient feedback the output impedance
is much smaller than the input impedance which is es-
- .- 1 { - - sentially R. This makes possible the design of individual
U a/stages without complications due to cascading and it
02 _ > / _also reduces the interaction of a stage on its neighbors
G~i 2 _ in the alignment process.
0

<~~~
~RI El DI
0 4 8 12 16 20 24 28 2
M (STAGE GAIN IN DB)
Fig. 4-Approximation error vs stage gain.
Rg
IEx Eyl
Sometimes the dynamic range of the output is of in- - _i_ _
terest. Defining this as Yno/yn, +1, from (10),
Ey
,um=+ n(m-1). (13)
BANDWIDTH E- SRIR2/(R1+R2)Rg
For signals below Pn, the lin-log amplifier is linear and = R2/REx
the frequency response can be described conventionally.
ea
For signals above P,n nonlinear distortion makes ordi- Fig. 5-Stage design using nonlinear feedback.
nary frequency response measurements invalid. The
high-frequency response above Pn could be measured For cascaded common emitter stages, there is an opti-
using an incremental sinusoidal signal mounted on a mum value for Rg in the interstage location to maximize
variable amplitude pedestal many cycles in width. How- the stage gain-bandwidth product.2 The optimum Rg for
ever, if the stage gain is controlled by feedback, the the configuration of Fig. 5 is
stage bandwidth will usually increase when a stage RgJ= (1.22rb11W,Cc) 12 (15)
enters the.unity gain region. The cascade bandwidth for
an incremental signal is, therefore, least for signals be- where rb' is the extrinsic base resistance, wa is the alpha-
low P,,. The specification of bandwidth using a signal cutoff radian frequency, and C, is the collector capaci-
below Pn is thus convenient and conservative. tance. For this value of Rg, the gain-bandwidth product
If the input dynamic range is specified, the choice of is
the number of stages determines stage gain and approxi- G= wa/1.22[1 + (rX,Caxl.22)112]2 (16)
mation error. It is well known that in amplifier design c (
there is an optimum numhber of stages for maximization Because of the effect onCn, a high collector voltage is
of bandwidth if over-all gain is specified. This assumes desirable to maximize bandwidth. A low collector volt-
stages of constant gain-bandwidth product and it as- age may be desirable in the input stages for best noise
sumes that the bandwidth of a cascade is inversely pro- figure. However, the collector voltage must be chosenl to
portional to the square root of the number of stages permit the maximum voltage swing required in a given
which is an approximation for stages with minimal over- stage. For the stage of order k, this is
shoot.1 The optimum number is 2 ln A, where A is the
Ek= Ea[k ± m/(m -1)]. (17)
1 G. E. Valley, Jr., et al., Tube Amplifiers," M.I.T. Rad.
Lab. Ser., McGraw-Hill Book Co., Inc., New York, N. Y., vol. 18,
"4Vacuum 2 G. Bruun, "Common &nitter transistor video amplifiers,"
PP. 77-80; 1948. PROC. IRE, vol. 44, PP. 1561-1571; November, 1956.

Authorized licensd use limted to: Imperial Coleg Lond. Downlade on June 07,21 at 19:354 UTC from IE Xplore. Restricon aply.
94 IRE TRANSACTIONS ON INSTRUMENTATION December
In a bipolar stage, the quiescent collector voltage must RECOVERY CHARACTERISTICS
be at least Ek. (Bipolar design is helpful for the mini- While the use of ac coupling in high-gain multistage
mization of recovery transients, as explained below-) amplifiers eliminates the drift problem associated with
Having selected R, on bandwidth considerations, Ea direct coupling, it introduces the problem of recovery
should, therefore, be made as small as possible in order from transients.3 The coupling time constants may be
to minimize standby power consumption. made large compared to the pulse width in order to
TEMPERATURE COMPENSATION minimize pulse droop. This reduces the amplitude of the
multiple undershoots following the pulse but prolongs
The diode bias B1 (Fig. 5) determines Ba according their duration. Alternatively, all the time constants may
to the relationship. be made long while one is. made comparable to the pulse
Ea-= E(m - 1)/m, (18) width. This results in a large undershoot but fast re-
covery towards the baseline and high attenuation of
but B3 is affected by the temperature coeficient of the subsequent overshoots.
diode or the residual coefficient remaining after imper- If the sae-shtac
If.t stage-by-stage unity-gain feedback paths in a
fect compensation. Therefore, El may be error by lin-log amplifier are provided for signals of one polarity
in
adA T, where ad is the residual temperature coefficient in only but are omitted for signals of the opposite polarity
volts/C and AT is the temperature change. The cor- the amplifier will have full gain (mn) for undershoots.
responding error in Ea is adA T(m - 1) /m volts. The effect These will, therefore, produce unnecessarily large arti-
on the breakpoints can be calculated from (9) and (10). facts in the output and may drive some stages into
From (22), the slope of the breakpoint line on semi- saturation or cutoff. Bipolar design provides the lin-log
logarithmic coordinates is Ea/log m. Provided m is in- characteristic for signals of either polarity and thus
dependently stabilized, the per cent change in slope is handles the undershoots in a controlled manner. For
therefore the same as the per cent change in Ea which is example, in a bipolar lin-log amplifier, if the input time
e = lOOadAT(m - 1)/mEa. (19) constant is made very short compared to the other time
constants in the cascade in order to speed recovery, the
Thus, given a permissible e, the minimum Ba can be decay of an undershoot towards the baseline will ap-
calculated. proximate a linear ramp, the result of logarithmation of
Another factor to be considered in choosingba is that the exponential decay. It should be noted that, while the
the feedback diode current excursion should be sufficient position of the short time constant does not matter in a
to provide the diode impedance change necessary to cascade of linear stages this is not true for the lin-log
attain the dual gain characteristic without too gradual amplifier because of nonlinearity.
a transition. It is desirable to have the diode impedance
change over a wide range and to have the impedance
limits set by resistors in series and parallel with the
diode. This design procedure separates the stage gain A lin-log amplifier with an 80-db dynamic range has
from dependence on the diode temperature-impedance been designed using transistor circuitry. For X =80 db,
relationship. It also makes feasible provision for align- about 16 stages are optimum for maximization of band-
ment of the over-all amplitude characteristic by stage- width. However, for n= 7, about 83 per cent of the band-
by-stage adjustment of Ea. width for 16 stages is obtained and this is done with an
This requirement determines a minimum value for Ea approximation error less than 1 per cent of the input
which can be derived from the diode equation dynamic range in db. Seven transistor operational am-
plifiers in common emitter configuration are used in
I = Io(exp qV/ckT - 1). (20) cascade. Rg was chosen for maximization of bandwidth
In the region where Io is not a significant component of and inductive compensation was used in the m-gain
the diode current, the minimum change in voltage re- feedback path of each stage to provide an additional
quired to change the diode incremental resistance by the bandwidth improvement. Bipolar design is used in
factor mp is order to control the recovery characteristics. All cou-
pling time constants are of the same order of magnitude
\V = (ckT/q) ln rnp (21) and provide a small-signal low-frequency response
where ckT/q is a constant usually between 26 and 52 my which is down 3 db at 100 cps. The small-signal high-
for germanium or silicon diodes. A ratio of at least rn is frequency response is down 3 db at 2.5 mc.
required to produce the required change in stage gain. For X =80 db and n =7, the stage gain is 10 db, or
The factor p determines the sharpnless of the gain rn= 3.16. For the final stage (k =7), from (17), the col-
transition. For example, for m=3.16, p=2O and ckT/qr lector voltage swing is B7=8.46 Ba. For B7=1 volt,
= 26 my, then Av V= 110 my which can be taken as the
minimumn allowable value for Ba in this case. 3Valley, Op. cii., pp. 114-142.

Authorized licensd use limted to: Imperial Coleg Lond. Downlade on June 07,21 at 19:354 UTC from IE Xplore. Restricon aply.
1959 Solms: Logarithmic Amplifier Design 95
E, = 118.2 mv. This value of Ea is obtained by use of for positive inputs is very similar. The responses for
forward bias on the 1N252 silicon diodes used in the 1000-Asec and 1-,usec pulses vs amplitude are shown in
feedback path. The forward bias of about 0.33 volt is Fig. 7. The multiple baseline crossings during recovery
provided by DR435 germanium diodes which are always from a 1000-Asec pulse are very prominent since they
in the conducting state and which also provide tempera- appear on a logarithmic scale. The slow return to the
ture compensation for the feedback diodes. The forward baseline following the 1-,usec pulse, commencing about
bias on the silicon diodes is not sufficient to produce ap- 35 db down for zero db input level, could be caused by a
preciable conduction, and the feedback path is still com- rather small deviation in the amplifier frequency re-
parable to a reverse-biased ideal diode. The lOOK sponse characteristic. Such a deviation would ordinarily
potentiometers provide a positive or negative correction be negligible in a linear amplifier and serves to illustrate
for the bias in each stage for each polarity and allow for the fact that such effects become very noticeable on the
alignment of the amplifier to conform to the desired logarithmic scale.
over-all amplitude characteristic. This characteristic The dependence of the amplitude characteristic on
is shown for negative input polarity at three values of duty factor is shown in Fig. 8. This effect is due to in-
the ambient temperature in Fig. 6. The characteristic sufficient time for recovery before the arrival of a suc-

1.0-- B... ooF

0. --- _ _.
06
>00-
4

022
080 70 60 50 40 30 20 10 0
INPUT DB BELOW 0.173 V
Fig. 6-Amplitude characteristic of lin-log amplifier at different tem-
peratures. Pulse width 3 Asec, repetition rate 100 pps. Battery
also at specified temperature.

IJS PULSE 1000 WS PULSE

--
_ -
W-
--_ s
rn ___

SWEEP
-
-~

,IJS/DIV.
_

SWEEP O,cOOOp5/DIV.
-I
Fig. 7-ulsersponseof lin-og ampifier.Fig. 8-Effect of duty factor on amplitude response.

Authorized licensd use limted to: Imperial Coleg Lond. Downlade on June 07,21 at 19:354 UTC from IE Xplore. Restricon aply.
96 IRE TRANSACTIONS ON INSTRUMENTATION December
e * e * * e e e e e e * * e
_LDOKu0~IOODK IO
IOKMOK dOOK0/OOK KHIODK IOOK i4100Kl
IN252
.'4K
. .7K ¢4.7
INDICAT
K 3.9K
IN252 477
R 82.7KK
AND IN52 47K
K
IN252 4.7
4.7K 4.7K
IN252 IN252 IN252
12uh _ Fg i tastor,1bp h up 1 12uh clh
12
t 1< 4620 N252 IN252 N25 t DbOsvN252 N5 l l N5

ceedingpulseand,b e o i25fd 25 desin300 is

;0 S t 25u) (24)
300 240 dtuf i t
tN 4 25 SU9393 9245+UXN9 N 393 N?f 393 N393 N 393||||
o 51Ke IK the pproxim i e2rror2 firs n he10
tK m
0
e7K b5eiO; draw
to:2nK
2netKo V seg

19 DIODE BIAS ADJUSTMENT FOR POSITIVE INPUT POLARITY. IICOM.

small,orwhere3th duyfco+scntn,o
e DIODE BIAS hr h
ADJUSTMENT FOR NEGATIVE INPUT POLARTY. F ( - )lo 1 1
variatin withARROWutyINDICATES
facor doe no3mttrThs0ffc Yee = 30a
CLOCKWISE ROTATION AND tNCREASE OF INTERCEPT
o + +30026
3wol no4epeeti 0adrc transistor
Fig. 9-Lin-log ope bipolardesign. Llog43
amplifier schematic. Power supply, 12m RM12R
ou 2-Dlogm cells. m-
mercury

ceeding pulse and, because of it, the Fig. 9 design is Substituting (25) in (24), the maximum deviation in the
eqaIonouseful
primarily th in
lgrthmic cuv
appications psIn
where through
the duty factorthis ymen in th seioaIthi
direction th input plt andthiiscalen
is obtainled.Thi line,
Fig. labele
3.
Pk pons Fro
small, where
or (9
the an
duty (10)
factor tis
is is
constant, or where the "tngn
i lie
i in Fig
4 t s 3, is paale
g to th
b log breakpoint m)
variation with duty factor does not matter. This e[ect
would not be present in a direct coupled design.
(m log e logn 12
(27
1)og
APPENDIX Notice that p drops out and that the error is, there-
he aproimaion
To ealute rro, frst indthefore, the same for each segment. A single straight line
equation of the logarithmic curve passing through the ment in the
g oivenbythe diFf ance (22) aThisis lb
Dtangent lTne" Sn
semelogarEthmRc
F og.
thEs
plot, and line, labeled
3, ESparallel to the breakpoint
a line.T
yn = ~ log xi + ~log (M-JM
. (22) From (22), it is seen that the slope of the breakpoint -

log m m I log m
line in the semilogarlthmo cI plot S a/log m. Then
Let the endpoints of a straight line segment of the over- (y[Rlog m)/2aA is the peak deviation to either side of the
all amplitude characteristic be called P[ and P 4+]. "median line" of Fig. 3 referred to the log xi axis. Sub-
Setting k-p and k=p+l (9)in and (10), the coordi- stituting from a5d (26) multiplying by 20, the error in
Ye pintFi.
nates 3.oendpoinlts of this segment are obtained. Jnd
of the db C.0 Broekert
[1] referred "Sm,oaihi
to the input is obtained.
ie This is called a and
ApiirAayin
These coordinates determine the equation of the seg- is plotted in Fig. 4 vs the stage gain in db. (M = 20 log m)
ment joininglogm Ye= P,,+,:
Pp and m1-og[2]5. . Vn Vori,"irwv -0r (m -1) log eeevr, log m .. Ra. Lab
Yn mpx1 + a( p).
= - (23) e log m m-1] (27) =1-Lo .

The deviation of this line from the logarithmic curve is

given by the difference of (22) and (23). This is labeled REFEcRENCES
Ye in Fig. 3. [6] J.Design
c. Broekert, "Some Logarithmic Video Amplifier Analysis and
Techniques," Stanford Electronics Labs., Stanford, Calif.
a Tech. Rept. No. 152-1; 1957.
ye --log xlmpxl [2] S. N. Van Voorhis, 'Microwave Receivers," M.I.T. Rad. Lab.
log m Ser.,583-606;
McGraw-Hill Book Co., Inc., New York, N. Y., vol. 23,
1
b
rr rr n ~
~~~~~~~~~~~~~~~~pp.
~~[3] R. V.vol.
1948.
Alred and A. Reiss, "An anti-clutter radar receiver," J.
+a
Im log m
_log (m-1)Mn n +p .(24) IEE, 45, pt. 3, p. 459; 1948.
[4] J. Croney, "A simple logarithmic receiver," PROC. IRE, vol. 39,
pp. 807-813; July, 1951.
Let dy,ldxl 0, and solve for xi. ceiver,"PROC. IRE, vol. 42, pp. 1307g1314;
August 1954.hmc e
[6] H. Kihn and W. E. Barnette, "A linear-logarithmic amplifier for
XI = a(log e)/mP log m. (25) ultra-short pulses," RCA Rev., vol. 18, pp. 95-135; March, 1957.

Authorized licensd use limted to: Imperial Coleg Lond. Downlade on June 07,21 at 19:354 UTC from IE Xplore. Restricon aply.