A Simplified Noise Theory and Its Application to

the Design of LowFNoise Amplifiers*

Summary-Any noisy amplifier can be represented by an equiv-

portantsimplification,becausetheeffects of circuit
alent noiseless amplifier plus two noise generators either at the in-
changes, feedback, operating conditions, and other vari-
put or the output of the amplifier. The choice of two particular noise
ables upon e, and in are easily assessed, while their er-
generators (the equivalent short-circuit noise voltage and the open-
fects upon F o and X0 can be obscure.
circuit noise current) to characterize a noisy amplifier has a number
Forexample,thebiasresistors
of advantages over the concept of noise figure. The noise generators of a transistoram-
can easily be measured separately from the source noise, and the
plifier are in parallel with the input, and can obviously
optimum source impedance and the noise figure at any source im-
have no effect upon e,. However, they will increase the
pedance can then be calculated. Since the amplifier noise is measured
separately from the source noise, low noisefigurescanbe value of i,, and this will raise Foand lower Ro.Resistors
easily
measured.Theoptimumsourceimpedanceequalsthequotient in parallel with the input must be large with respect to
of
the two noise generators, and the noise 0 (not necessarils- with respect to the input impedance)
figure depends upon their
X
product. Neither feedback nor input impedance is a consideration in
so that their effect on Fo will be negligible. Resistors in
determining noise figure and optimum source impedance. series with the input must be small with respect to X.
Several transistor noise diagramsshow how the two noise genera-
for the same reason.
tors are affectedby emitter current, collector voltage, and frequency.
Noise diagrams can be used to select the most suitable amplifyingT o specify the noise performance of the amplifier ex-
actly, values are needed fory (the correlation coefficient
devices and optimum operating conditionsfor various applications.
between the e, and i, generators) and for X o , the reac-

T H E design of low-noiseamplifierscanbemade
much simpler than has been previously realized
by the use of equivalent short-circuit and open-
circuit noise generators as the measure
tivepart of theoptimumsourceimpedance.
termine X , , a curve of e, vs source reactance may be
plotted; X0 then equals the source reactance that gives
of the noisiness minimum e,. To determine y , noise figure need be meas-
T o de-

of an amplifier,ratherthanthenoisefigure as such. uredonlyonce, at theoptimumsourceimpedance;

This approach has two advantages: the magnitudes of then y is the only remaining unknown in the equation
the two equivalent noise inputs can be measuredeasi1-J-. for Fo. However, X . isusuallynegligible a t low fre-
a n d rigorousformulasfornoisefigureandoptinluln quencies, and y is bounded (0 _<y5 1) and usually lies
source resistance are most concisel-\. expressed in terms near 1: so thattheupperlimiton Fo (calculated as-
of the two noise-generator parameters.
Basic noise theory1 states that a nois>- amplifier can be
represented by an equivalent noiseless amplifier plus a
constant-currentnoisegeneratorinparallelwiththe
input, and a constant-voltage noise generator in series
with the input (Fig. 1). The magnitudes of these two
generators can be determined independently as follows 3
with the input terminals shorted, e, is responsiblefor qln
the entire noise output of the amplifier. To determine
the value of e,, the short-circuit noise output is com- Fig. 1-A noisy amplifier can be represented by an equivalent noise-
pared with the output produced bya small lulown input less amplifierplusaconstant-voltage noise generator in series
with the input and a constant-current generator in parallel n.ith
voltage large enough to mask the noise. To determine the input.
in, the noise output of the amplifier with the input ter-
minals open-circuited is compared with that produced
by a small known current at the input.
The equations in Fig. 2 show that the mininlum noise
figure of theamplifierdependsprimarilyuponthe
product of e, and in,while the optimum source resistance
depends upon the quotient of e, and in. This is an im-

Received by the PGrl, LIarch 16, 1961. ‘This material was pre-
sented at the NortheastElectronics Research and Engrg. hIeeting on
November 16, 1960, and appeared in the 3 E R E J I Record. Fig. 2--;llinimum noise figure F o andoptimumsourceimpedance
General Radio Co./( IVest Concord, hIass. Z0=R0+jXo can be written concisely in terms of the equivalent
H. A. Haus, et al., Representation of noise in linear twoports,” noise generators. The correlation coefficient betv-een the two gen-
P R O C . IRE, vol. 48,pp. 69-74; Januar!., 1960. erators y lies betn-een and 1.

Authorized licensd use limted to: IE Xplore. Downlade on May 10,2 at 19:034 UTC from IE Xplore. Restricon aply.
I961 andXanderson Fulks: Design of Low-Noise Amplifiers 107

suming X o = O and 1) is quite accurate, and may bestintheregion of theintendedsourceimpedance,

actuallyexceedtheaccuracywithwhichthenoise and then to choose the optimum operating conditions
figure may be measured a t very low values of FO. for that source impedance. A convenient way of present-
Both e, and i, are independent of feedback, and may ing information about the noise generators is the noise
be taken outsideof the feedback loop with no change in diagram shown in Figs. 4-7. Both e, and i , are plotted
value. The proof is as follows: current feedback to the
inputobviously will not affect e, because i t will not lt30
change the voltage gainof the amplifier or the noise out-
Ikc
putwiththeinputshort-circuited.Sincethe i, gen-
erator is outside of a current feedback loop, it remains
unchanged.Ananalogous proof shows that e, and i,
are independent of voltage feedback as well. Since the
noisegeneratorsareindependent of feedback, so also
are Fo and Ro.Also the noise generators are approxi-
mately the same for a given device in any of the three
amplifier configurations. The generators appear to be a
property of the device and independent of the way i t Fig. 4-Noise-diagram of a typical low-noise transistor shoa‘‘1ng e,,
and i, as functions of emitter current. Minimum FOoccurs a t t h e
is used. emitter current which minimizes the product of e, and
At the optimum source resistance both noise genera-
torscontributeequallytothenoiseoutput of the ht30
amplifier.Atothersourceresistancesthenoisefigure
is given by

Fig. 5-Noise diagram of a transistor for which Fo minimizes at a

high source resistance. Conventional checkof noise figure a t
ohm source resistance would not reveal this transistor’s low-noise
potential.

Fig. 3-Comparison of a low-noise transistor with a low-noise vacuum

tube. The transistor hasa lower noise figure for source resistances
beloa 10,000 ohms, while the vacuum tube is superior for source
resistances above 10,000 ohms.

This dependence of noise figure on source resistance is

illustrated in Fig. 3. For source resistance much larger
or much smaller than the optimum, the noise figure de-
pendsupononlyone of thegenerators,andchanges Fig. 6-Koise diagram of a transistor with collector voltage as the
independent parameter. Both e, and in seem to be relatively inde-
linearly with source resistance. For such source resist- pendent of collector voltage up to several volts.
ances,as well as for all reactivesources, it is more
meaningfultorateanamplifierinterms of its noise
generators(inpv/cycle1’2 a.nd p p a / P ) , sincethese
numbersareindependent of source resistanceand
source temperature. The SNR is the ratio of the signal
totheappropriate noisegenerator w-hen thesource
impedance is much larger or smaller than e n / i n .
Both the e, and thei, generators vary widely between
vacuumtubesandtransistors,amongthedifferent
types in each category, and somewhat with the operat-
ing conditions of a particular device. Since nothing can
be done in the circuit to affect the generators, i t is im- Fig. 7-Soise diagram showing frequency spectrum of the e, and i,
generators. The two generators do not necessarily have the same
portant to choose the device whose noise performance is frequency spectrum.

Authorized licensd use limted to: IE Xplore. Downlade on May 10,2 at 19:034 UTC from IE Xplore. Restricon aply.
on a logarithmicscaleagainstsomeindependentpa- variablessuch as current,voltage,temperatureand
rameter,suchasemittercurrentorcollectorvoltage. frequency, and thus facilitate the choice of the proper
Because of the log scale, minimum F , is indicated by amplifyingdeviceforeachapplication.Thepresent
the minimum sum of the e,L and in curves, while X0 is method of rating by noise figure is at best cumbersome,
proportional to the difference between the two curves. and can be incomplete when the source resistance at
survel- of different devices can be made quickly be- which the noise figure is measured is not both specified
causeaminimum of information is required on each and equal to Ro. The general use of noisegenerators
device tested. could considerably simplify the representation and ap-
blanufacturers could perform a service by publishing plication of amplifying devices where noise performance
noise diagrams of their devices for several independent is an important factor.

Average vs RMS Meters for Measuring Noise*

Summary-Itappearsthatthe controversy is stillaliveover and all processes--mechanical,chemical,atomic, elec-

whether average reading or root-mean-square reading meters should trical,oranyother,includingcommunication,imply
be specified as standard for the measurement of noise, This being
the case, it is worthwhileto consider the entire subject from the
and require the interchangeof energy betvieen one body
standpoint of basic fundamentals, to determinewhatarethe sig- and another. All measurements, then, have essentially
nificant quantities involved, and then proceed to investigate which one purpose: to determine the total amount of trans-
type of meter yields the most significant results. ferred energ)-, and/or therate a t which it is transferred.
The following is the result of such an investigation. The entire
discussion rests on 1) an axiom, that energy transfer is the funda- POXER
mental interaction within the universe, and a premise, that for the
Since to life in general and to human life in particu-
type of measurements under discussion (audio), all significant proc-
lar, time is of the essence, it is usually the rate of en-
esses are linear. Given these two starting points, the conclusion is
ergy transfer m-hich assumes predominance. That factis
reached that the meaningful quantities are found by rms measure-
ments. recognizedimplicitly by virtue of assigning a special
It is shownfurther, by concreteexample,thatmeasurements
name to the time rate of energy flow, namely power. I t
made with average reading meters can depart widelyfrom those
must be remembered, however, that the idea
made with an rms meter. This being the case, it is necessary that of power is
arrived at from the fact of energy transfer, and there-
measurement standards specify the use of rms meters. Those who
elect to use average meters, then, bear the responsibility fore, power is a derivative concept.
of deter-
mining the accuracy of their results in terms of the fundamentally
T o repeat, power is the time rate of energy transfer.
important quantities.
In the specific case of electrical phenomena, power is
often supposed to be the starting point. It isn’t, but

T H E FIRST point which must be investigated in

an\-discussion of measurementtechniquesis,
what are the significant factors? Once these are
known, i t is thenpossibletodetermine how bestto
provided its originsarekeptclearlyinmind,
beconvenientto so assume.Thereasonfortheas-
sumption is that “energy pile-up,” i n the form of heat
i t can

specifystandards of measurementn-hich will \field particularly, is one of the fundamental problems. Fur-
ther, since energy must be transferred to be useful, and
significant results, and the degree of approximation in-
since the rateof transfer determines how much work will
volved in alternative methods.
be done in a given time, the conceptof power becomes a
THE STARTINGPOINT potent tool for judging the utility of a given process.
\\‘heneveroneconsidersmeasurements of a n y k i n d OFT POWER
THE~ I E A S U R E M E N
whatever, the starting point must be the interchange or -111 energ)-measurementsproceedononecommon
.transfer of energ>-. The universe runs b\- energy transfer, basis; transformation. Whenever energy content orflow
is to be determined, some portionof i t is converted into
ReceiT-ed by the 1’GA4,January 3, 1961 revised manuscript re- another form, the representation of whichcanbeob-
ceived, April 27, 1961.
t RCA Victor Record Div., Indianapolis, Incl. servedbyhumanbeings.(Energy, as such, is notdi-

Authorized licensd use limted to: IE Xplore. Downlade on May 10,2 at 19:034 UTC from IE Xplore. Restricon aply.