NTIA Report TR-04-410

Gain Characterization of the RF

Measurement Path

J. Wayde Allen

U.S. DEPARTMENT OF COMMERCE

Donald L. Evans, Secretary

Michael D. Gallagher, Acting Assistant Secretary

for Communications and Information

February 2004
 CONTENTS

Page

ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1

1. INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2. WHAT KIND OF GAIN? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

3. MEASUREMENT UNCERTAINTIES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

4. MEASUREMENT METHODS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
4.1 Measuring Ga Using a Signal Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
4.1.1 Single-Point Calibration Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7
4.1.2 Two-Point Calibration Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4.2 Measuring Ga with a Noise Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.2.1 Uncertainties when Measuring Ga with a Noise Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.3 Using the Network Analyzer to Determine Ga . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.3.1 Uncertainties in Measuring Ga with a Network Analyzer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

5. SUMMARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

6. REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

iii
 GAIN CHARACTERIZATION OF THE RF MEASUREMENT PATH

∗
J. Wayde Allen

In radio frequency (RF) measurements the gain (or loss) of the signal path connecting the mea-
surement equipment to the measurement reference plane must be accounted for. This tutorial
paper discusses the various definitions of gain, and how to determine the gain using either a
calibrated signal generator, noise source, or network analyzer.

Key words: available power; cable loss; calibration; delivered power; gain; mismatch

1. INTRODUCTION

At the Institute for Telecommunication Sciences (ITS) systems such as the Radio Spectrum Measurement
System (RSMS) are often called upon to perform a wide range of radio frequency (RF) measurements. For
this reason, the RSMS is not a single system but rather a toolbox of equipment that can be configured
in a variety of ways. While this gives the RSMS a great deal of flexibility, it complicates the job of the
engineer who must provide an analysis of the inherent measurement uncertainties for a given test. The
uncertainties associated with each test configuration need to be dealt with on a case-by-case basis. The goal
here is to provide a collection of basic uncertainty analyses to be used as a starting point for determining
the overall test uncertainty. In a sense, this means creating a toolbox of uncertainty analysis methods that
can be adapted as needed. The measurement configurations and analyses that follow deal specifically with
the problem of accounting for the gain or loss in the RF path from where the signal is sampled to where it is
ultimately detected by the test instrumentation (power meter, spectrum analyzer, etc.). This arrangement
is shown schematically in Figure 1.

Since the test instrumentation can only respond to the signal delivered to it, the value indicated by the
test equipment must be corrected to remove the effects of the intervening two-port. This involves finding a
two-port gain in the form
Pmeas
Pi = (1)
G
where:

Pi is the power incident at the test plane,

Pmeas is power indicated by the test equipment, and

G is the gain of the intervening two-port.

Unfortunately, the term gain (G) can be defined in a number of ways, and often no information is provided
to indicate which definition is being used. This paper discusses the distinctions among different definitions
of gain, “available power gain” in particular, and describes a number of measurement methods.

∗ The author is with the Institute for Telecommunication Sciences, National Telecommunications and Information Adminis-

tration, U.S. Department of Commerce, Boulder, CO 80305

 Test Plane

Intervening 2-port
(Cable, Amplifiers, etc.)

Test Equipment
(Power Meter, Spectrum Analyzer, etc.)

Figure 1. Typical measurement system configuration.

2. WHAT KIND OF GAIN?

When one talks about the gain of a two-port device it is important to specify what one really means. It is
not as simple as forming the ratio of the output power versus the input power. As can be seen in Figure 2,
there are a number of powers that can be chosen to form this ratio. Here the powers P1a and P1d refer to the
powers available and delivered to port one of the two-port, whereas powers P2a and P2d describe the powers
available and delivered from port two of the same device. This means that one could offer the following
definitions for the gain of the two-port:
P2a
G1 = , (2)
P1a
P2a
G2 = , (3)
P1d
P2d
G3 = , (4)
P1a
P2d
G4 = . (5)
P1d

Of these gain definitions, equation 2 is known as the “available gain” (Ga ) of the two-port, equation 4 is
either the “signal gain” (Gs ) or “transducer gain” (Gt ), and equation 5 is commonly known as simply “power
gain” (G) [1]. To the best of my knowledge equation 3 is not named.

P1a P1d P2a P2d

2-Port

Figure 2. Generalized 2-port showing available and delivered powers.

2
 Reflections from load (Γl) Reflections from generator (Γg)

bg
bg Γl
bg Γl Γg
bg Γl2 Γg
bg Γl2 Γg2
bg Γl3 Γg2
(etc.)

Figure 3. Diagram of multiple reflections at the measurement plane.

Things are further complicated by several other possible gain definitions:

Associated Gain (Gass ) The available gain of a device when the source reflection coefficient is the optimum
reflection coefficient (Γopt ) [1, pg. 29].
Insertion Gain (Gi ) The gain that is measured by inserting the DUT between a generator and a load.
The numerator of the ratio is the power delivered to the load while the DUT is inserted, Pd . The
denominator, or reference power Pr , is the power delivered to the load while the source is directly
connected [1, pg. 31].

Clearly it is important to specify the gain definition one is using.

Choosing the appropriate gain definition requires an understanding of the relationship between available
and delivered power. This is a direct result of what can be called the “microwave circuit theory analogue of
Thevenin’s or Norton’s Theorems” [2, pp 18–22] [3, pg 51]

b = bg + aΓg (6)

where:

b is the wave amplitude of the signal emerging from a source,

bg is the wave amplitude of the generator wave created by the source,

Γg is the reflection coefficient seen by energy incident on the source, and

a is the wave amplitude of energy incident on the source coming from the circuit the source is connected to.

By considering the multiple reflections of a transient signal across a reference plane as shown in Figure 3, we
can obtain a mathematical description of the mismatch term M . In this case, summing up all of the wave
amplitude terms directed to the right we get

b = bg + bg Γg Γl + bg (Γg Γl )2 + . . . (7)

3
which can be factored to give us
b = bg [1 + Γg Γl + (Γg Γl )2 + . . .]. (8)
Recognizing that the factor consisting of an infinite sum of reflection coefficients is a geometric series that
converges to (1 − Γg Γl )−1 as long as |Γg Γl | < 1 allows us to write this as
bg
b= . (9)
1 − Γg Γl

Note that by definition the power delivered to a passive termination is

Pd = |b|2 − |a|2 , (10)

and that
a = bΓl . (11)
This means equation 10 can be rewritten in terms of power as

Pd = |b|2 [1 − |Γl |2 ]. (12)

Substituting equation 9 into equation 12 results in the general expression:

|bg |2 (1 − |Γg |2 )(1 − |Γl |2 )
Pd = = Pa M (13)
(1 − |Γg |2 ) |1 − Γg Γl |2
where the available power Pa is given by
|bg |2
Pa = , (14)
1 − |Γg |2
and the mismatch M is
(1 − |Γg |2 )(1 − |Γl |2 )
M= . (15)
|1 − Γg Γl |2

With this in mind, let’s assume that we want to measure the power of some arbitrary signal source. Working
in terms of available power, the maximum power that this source can supply to a conjugate matched load is
Psource . Hence the maximum power available to the detector (power meter, spectrum analyzer, etc.) is:

Pout = Ga Psource (16)

where Ga is the available gain of the intervening two-port. This is diagrammed in Figure 4.

Pout
Source 2-Port Power
(Psource) Detector
M1 M2
Figure 4. Transfer of power to the detector using “available gain”.

It is still necessary to determine the power that will be “indicated” by the detector circuit due to the
impedance mismatch M2 and any calibration factors for the detector. However at this location in the mea-
surement system the engineer usually has control over system design and can design the two-port connecting
the signal source to the detector so that either the mismatch term M2 is known or is constant.

If, on the other hand, we choose to work in terms of the power delivered at the input to this same two-port,
we need to know something about the mismatch M1 between the source and the two-port. This would be
described by
P1d = M1 Psource , (17)

4
where:

P1d is the power delivered across the measurement plane,

M1 is the mismatch coefficient at the measurement plane, and

Psource is the power available power at the measurement plane.

The power delivered to the detector could then be found using the gain expression given in equation 3. The
power transfer equation in terms of delivered power would then be

Pout = M1 G2 Psource . (18)

This requires measuring values for both M1 and G2 .

3. MEASUREMENT UNCERTAINTIES

Before going any further, it is worth noting that no measurement is complete unless there is some specification
describing how close the measured value comes to reporting the truth. For example, what if the gain of a
two-port was measured to be 50 dB one time and 46 dB the next? Given this information it appears that
there is a 4 dB difference between these measurements. However, if this is a measurement of the same
two-port under the same conditions do these measurements have a discrepancy of 4 dB? Perhaps, but what
if the measurement system had an uncertainty of ±2 dB? In this case, it is entirely possible that the system
may be measuring correctly and still show a 4 dB difference between values. What if the signal path had
roughly 1 dB of gain or loss? Does a measurement of 1 dB ±2 dB mean anything at all? We need to know
the measurement uncertainty in order to know the effective resolution of our measurements, and to ensure
that our data and measurement process can actually support the conclusions we draw from them.

Unfortunately experiment design and uncertainty analysis is too broad a topic to cover here. There are,
however, a number of sources for more detailed information. See [4] and [5]; the analyses that follow are
based on these two references. The basic idea is to establish some estimate of the measurement uncertainty
inherent in the measurement system (systematic uncertainty) and to combine this with any uncertainties
caused by statistical or random fluctuations. However, the use of the term “systematic uncertainty” has
recently been dropped in favor of the following definitions [4]:

Type A Uncertainties arrived at by statistical methods,

Type B Uncertainties which are evaluated by other means.

This paper uses the modern Type A and Type B uncertainty definitions.

Finally, it is worth pointing out the difference between uncertainties expressed in linear (fractional or percent)
and logarithmic (decibel) terms. It is common to express the uncertainty in terms of a plus and minus value
(±∆x). This works well when using fractional or percent change. However, since the decibel scale is
logarithmic the plus value does not imply the same magnitude of change as the minus value. Indeed, a
statement of ±3 dB indicates a plus 100% and minus 50% change in value. This means that uncertainties
expressed in dB need to be converted to linear (fractional or percent) uncertainty before being combined
using traditional error propagation equations.

To convert between fractional uncertainty and dB uncertainty consider the dB representation of a given
value x with respect to a reference value y
x
dB1 = 10 log( ), (19)
y

5
and its fluctuation
x + ∆x
dB2 = 10 log( ). (20)
y
From this the change in dB can be found as
x + ∆x x x + ∆x
dB2 − dB1 = ∆dB = 10 log( ) − 10 log( ) = 10 log( ) (21)
y y x
or
∆x ∆dB
= 10( 10 ) − 1. (22)
x
Since the magnitude of the change depends on whether one is using plus or minus in the dB scale, the
convention taken through the rest of this paper is to assume the worst case change.

4. MEASUREMENT METHODS

4.1 Measuring Ga Using a Signal Generator

One common method for measuring gain is to use a calibrated signal generator to supply a source of known
available power at the input to the network. With this known input one then observes the power indicated
by the detector and forms the ratio. The basic test setup is described in Figure 5.

Signal P1a P1d P2a P2d

2-Port Power
Generator Detector

Figure 5. Test setup for using a signal generator to measure gain.

However, is it sufficient to supply only a single known power, read the detector response and form the power
ratio (single-point calibration), or is it better to use two (or more) known levels from which the system gain
can be determined (two-point calibration)? Two possible gain curves are shown in Figure 6. Since the use
of a single known input and measured output power only identifies a single point on the curve it is clear that
this can only work for systems that have no zero point offset. In many real world measurement systems the
presence of detector offset, DC bias effects, or appreciable system noise can violate this condition. In these
cases, one must resort to the two-point method so that the zero point offset can be taken into account.

The following describes some of the key considerations between the single-point and two-point methods:

• Single-Point Method:
1. This is an absolute value measurement.
2. Assumes strict system linearity.
3. No zero point offset allowed.
4. Simple test.
• Two-Point Method:
1. Difference rather than absolute value measurement.
2. Identifies zero point offset (can be used to linearize the system).
3. More complicated since two points must be measured.

This is a good example of how the same hardware can be combined in different ways to produce very different
measurement methods.

6
 Pout

Pout = G1 * Pin + 0

Pout_2
Pout = G2 * Pin + b

Pout_1

Pin
Pin_1 Pin_2

Figure 6. Two possible gain curves.

4.1.1 Single-Point Calibration Uncertainty

In this case we know the available power supplied at the input to the two-port (Pa1 ), and the detector circuit
indicates the power delivered to it from the two-port (P2d ). Since the available gain is the ratio of the power
available at the output of the two-port to the power available at the input of the device, we make use of the
relationship
P2d = P2a M2 (23)
where:

P2d is the power delivered to the power detector,

P2a is the power available at the input to the detector, and

M2 is the mismatch between the two-port and the detector.

Solving for P2a and substituting this into equation 2 gives

P2a P2d
Ga = = . (24)
P1a M2 ∗ P1a
As long as we can obtain estimates for the variability of each of the terms in equation 24 (∆P2d , ∆M2 , and
∆P1a ), we can determine the Type B uncertainty by finding the following fractional uncertainties:

∂Ga 1 ∆P1a
uP1a = |( )( )∆P1a | = , (25)
∂P1a Ga P1a
∂Ga 1 ∆P2d
uP2d = |( )( )∆P2d | = , (26)
∂P2d Ga P2d
∂Ga 1 ∆M2
uM2 = |( )( )∆M2 | = . (27)
∂M2 Ga M2

7
Assuming that these uncertainties are independent we can combine them in quadrature to get a total Type B
uncertainty q
uB = u2P1a + u2P2d + u2M2 . (28)

In order to complete this analysis the Type A uncertainty is commonly found as the standard deviation
divided by the mean of a series of repeated gain measurements. Once this value (uA ) is known the expanded
measurement uncertainty can be found as
q
U = k u2A + u2B (29)
where k is defined by the National Institute of Standards and Technology (NIST) as the “coverage factor” [4].
The expanded uncertainty defines an uncertainty interval about which the measurement result is confidently
believed to lie, and the value of the coverage factor is chosen based on the desired level of confidence.
Typically, k is in the range of 2 to 3 with k = 2 defining an interval with approximately a 95% level of
confidence.

4.1.2 Two-Point Calibration Uncertainty

Supplying two known powers (Pin 1 , Pin 2 ) at the input to the two-port, and measuring the output responses
(Pa out 1 , Pa out 2 ) allows us to compute the gain as
Pa out 2 − Pa out 1
Ga = . (30)
(Pin 2 − Pin 1 )

This is simply a more general form of equation 24. One could even approach the uncertainty analysis for
this case as the combination of two single point measurements as was done in the prior section; however, of
particular interest is that this two-point method gives us the option of analysing the uncertainties as a ratio
of differences rather than a ratio of absolute values. This is important since power differences can often be
measured more accurately than absolute levels. In fact, one could actually determine the gain without having
to know the absolute values of the input powers Pin 1 and Pin 2 , providing the difference between the levels
is known. However, let’s assume for this uncertainty analysis that we are given the absolute power values
for Pin 1 and Pin 2 . Let’s also assume that we are measuring the output power difference (Pout 2 − Pout 1 )
which can be written in terms of available powers (Pa out 1 , Pa out 2 ) as:
− Pout 1
Pout 2
Pdif f = Pa out 2 − Pa out 1 = (31)
M2
Assuming that the mismatch term M2 is constant, we only need to worry about the uncertainty in difference
itself (udif f ). Proceeding as for the single point case we have:
∂Ga 1 ∆Pdif f
udif f = |( )( )∆Pdif f | = , (32)
∂Pdif f Ga Pdif f
∂Ga 1 ∆P1a
uP 1a = |( )( )∆P1a | = , (33)
∂P1a Ga P2a − P1a
∂Ga 1 ∆P2a
uP 2a = |( )( )∆P2a | = . (34)
∂P2a Ga P2a − P1a
From this an estimate of the Type B uncertainty can be found as
q
uB = u2dif f + u2P 1a + u2P 2a . (35)
Again the Type A uncertainty (uA ) is commonly found as the standard deviation divided by the mean of a
series of repeated gain measurements, and the expanded measurement uncertainty computed as
q
U = k u2A + u2B (36)
where k = 2 is the most common coverage factor [4].

8
 Noise P1a P1d P2a P2d
2-Port Power
Source Detector

Figure 7. Test setup using a noise source.

4.2 Measuring Ga with a Noise Source

For low noise systems with enough sensitivity, the use of a calibrated noise source rather than a signal
generator is an option. This can be particularly useful when performing broadband measurements where a
single, narrowband frequency calibration using a signal generator may not be practical or desired. In this
case, the calibrated signal is created by some device or object emitting energy according to the Nyquist
Theorem [2, pg. 203] as if it were a blackbody radiator. This could be a simulated blackbody radiator such
as a temperature controlled resistor [6][7], a gas discharge tube, a celestial radio source such as the supernova
remnant Cassiopeia A [8], or an avalanche diode. The hardware configuration for this measurement is shown
in Figure 7. This is very similar to the setup used for the single frequency case discussed in Section 4.1;
however, in this case the available power from the noise source is represented by the expression
Pnoise = kT B (37)
where:

Pnoise is the noise power,

k is Boltzman’s constant (k = 1.38 × 10−23 [Joules/Kelvin]),

T is the “effective” blackbody temperature of the noise source in Kelvins, and

B is the measurement bandwidth in Hertz.

Again it is very important to recognize that the noise power described by Equation 37 is a true “available”
rather than delivered power [1, pg. 4][2, pg. 203].

An added complication for analysis of the circuit in Figure 7 is that noise sources internal to the two-port
network need to be included. Writing this in terms of available gain and power we have
P2a = Ga P1a + Na (38)
where:

P2a is the available noise power at port two,

Ga is the available gain of the two-port,

P1a is the available power at port 1, and

Na is the noise added by the two-port network.

By plotting this function in Figure 8, a few features of the model can be illustrated. First of all, if the
two-port could be connected to a blackbody noise source at absolute zero we would expect the internal noise
of the two-port to result in an available noise power of Na at port two of the device.

Finally, note that the measurement of Ga using a hot and cold noise source is really nothing more than using
two defined points in order to compute the slope of the equation for the line.

9
 P2a

P2a = Ga P1a + Na

P2a_hot

P2a_cold

Na

P1a
-Pe P1a_cold P1a_hot

Figure 8. Graphical description of the noise model.

So for two noise sources of known values P1ahot and P1acold we have

P2ahot = P1ahot Ga + Na (39)

and
P2acold = P1acold Ga + Na (40)
which can be solved for Ga to obtain
P2ahot − P2acold
Ga = . (41)
P1ahot − P1acold

Quite often the calibration of the noise source is expressed as an “equivalent” blackbody temperature. This
is not usually the actual physical temperature of the device, but rather the temperature a true blackbody
radiator would need to have the same power output as the noise source as described by the Nyquist relation.
For devices such as noise diodes that can easily be turned on and off it is also common to use the diode
“on” temperature to represent P1ahot and the “off” temperature to represent P1acold . If this is done, the
“off” temperature usually assumes that of the connected two-port and detector. This has the advantage of
establishing the condition where there is no net power flow from the noise source to the connecting two-port
as long as these are both at the same temperature. In this case, the unbiased noise source is simply treated
as a passive resistor (blackbody) and accepted practice is to assume that the blackbody temperature of this
device is then approximately 290 Kelvin. This temperature is commonly represented using the notation To .
Rewriting equation 41 in terms of the noise source temperature T and To results in the following expression.
P2ahot − P2acold
Ga = . (42)
kB(T − To )

Noting that the detector in Figure 7 can only respond to delivered power (P2d ), and that there may be a
detector calibration factor to consider, the “indicated” power can be described by the expression

Pindicated = ηP2d = ηM2 P2a (43)

10
where:

Pindicated is the power indicated by the detector,

η is the calibration factor for the detector, and

M2 is the mismatch between the detector and the two-port.

Letting Phot and Pcold represent the powers indicated by the detector with hot and cold noise sources
connected to the circuit, allows us to rewrite equation 42 as
Phot Pcold
ηM2hot − ηM2cold
Ga = . (44)
kB(T − To )

Here it is important to consider the design and operation of the measurement system. For example, the
powers Phot and Pcold could be measured independently, and assuming that values could be determined
for the detector calibration factor η and the mismatches M2hot and M2cold , using equation 44 to determine
the available gain is straightforward. However, the need to determine the quantities M2hot , M2cold , and
η combined with the implication that the system can make accurate power measurements relative to true
power is somewhat problematic. In most cases, the preferred approach would be to design the system to
measure the power difference (Phot − Pcold ). It is important however, to note that

Phot − Pcold = η(M2hot P2ahot − M2cold P2acold ), (45)

and this is proportional to the numerator in equation 44 only if

M2hot = M2cold . (46)

4.2.1 Uncertainties when Measuring Ga with a Noise Source

To compute the uncertainties in the gain measurement described by equation 44 we need to find:

Phot Pcold
udif f the uncertainty in the difference ηM2hot − ηM2cold ,

ubandwidth the uncertainty in the measurement bandwidth B,

uT the uncertainty in the noise source temperature T , and

uTo the uncertainty in the system temperature To .

Since Boltzman’s constant k is a universally accepted physical quantity it can be ignored as an uncertainty
source for this analysis.

Most test equipment, such as a spectrum analyzer, is designed primarily to measure power differences rather
than absolute powers. So it will be assumed that the system is carefully designed to meet the mismatch
requirements of equation 46. Furthermore, the following analysis assumes that the measurement detector
is well behaved so that not only is M2hot = M2cold , these mismatch terms and the calibration factor η are
constant. This simplifies the uncertainty analysis since it means the only variability comes from the difference
term, Phot − Pcold . This results in
∂Ga 1 ∆(Phot − Pcold )
udif f = |( )( )∆(Phot − Pcold )| = . (47)
∂(Phot − Pcold ) Ga (Phot − Pcold )

11
Expressions for ubandwidth and uTo are

∂Ga 1 ∆B
ubandwdith = |( )( )∆B| = (48)
∂B Ga B
and
∂Ga 1 ∆To
uTo = |( )( )∆To | = . (49)
∂To Ga T − To
However the noise source term is a bit more complicated. Assuming the calibration is done using a noise
diode the uT term is the result of a number of separate uncertainty mechanisms that must be accounted for.
This results in the expansion of the single uT term into:

uTcal the noise source calibration uncertainty,

uTamb ambient temperature effects on the noise source output,

uTV the noise source power supply stability,

uTinterp uncertainty caused by interpolating between calibration points.

These can be written using the increments of the variable T as:

∂Ga 1 ∆Tcal
uTcal = |( )( )∆Tcal | = , (50)
∂Tcal Ga T − To

∂Ga 1 ∆Tamb
uTamb = |( )( )∆Tamb | = , (51)
∂Tamb Ga T − To

∂Ga 1 ∆TV
uTV = |( )( )∆TV | = , (52)
∂TV Ga T − To

∂Ga 1 ∆Tinterp
uTinterp = |( )( )∆Tinterp | = . (53)
∂Tinterp Ga T − To

where:

∆Tcal is the variation in the noise source temperature due to the calibration uncertainty,

∆Tamb is the variation in the noise source temperature caused by ambient temperature fluctuation,

∆TV is the variation in the noise source temperature caused by power supply variations, and

∆Tinterp describes how much variation can exist in the noise source output value (T ) due to interpolating
between calibration points.

Assuming that these uncertainties are independent we can combine them in quadrature to obtain an estimate
of the total Type B uncertainty
q
uB = u2dif f + u2bandwidth + u2To + u2Tcal + u2Tamb + u2TV + u2Tinterp . (54)

12
The Type A uncertainty would commonly be found as the standard deviation of a series of repeated gain
measurements. Once values have been obtained for both the Type A (uA ) and Type B (uB ) uncertainties
the expanded uncertainty of the measurement can be computed as
q
U = k u2A + u2B (55)

where the coverage factor k = 2 representing a 95% confidence limit is commonly used [4].

4.3 Using the Network Analyzer to Determine Ga

Laboratories equipped with vector network analyzers have the option of computing the available gain of
a two-port from measurements of the network’s scattering parameters (S11 , S12 , S21 , S22 ) and reflection
coefficient of the signal source (Γs ). In this case, the power available from the signal source at the input to
the two-port is described by equation 14, and the power at the output of the intervening two-port can be
described by either [1, pg. 29]

|bg |2 |S21 |2 (1 − |Γ2 |2 )

P2a = (56)
|(1 − Γs S11 )(1 − Γ∗2 S22 ) − Γs Γ∗2 S12 S21 |2

or

|bg |2 |S21 |2
P2a = (57)
|1 − Γs S11 |2 (1 − |Γ2 |2 )

where

S12 S21 Γs
Γ2 = S22 + . (58)
1 − S11 Γs

Since the available gain is given by the ratio of the available output power to the input power as described
in equation 2, we can compute the gain using either [1, pg. 29]

(1 − |Γs |2 )(1 − |Γ2 |2 )

Ga = |S21 |2 (59)
|(1 − Γs S11 )(1 − Γ∗2 S22 ) − Γs Γ∗2 S12 S21 |2
or
1 − |Γs |2
Ga = |S21 |2 . (60)
|1 − Γs S11 |2 (1 − |Γ2 |2 )

Here it can be readily seen how the available power gain depends only on the source impedance and network
parameters and is independent of the load connected at the output of the two-port.

4.3.1 Uncertainties in Measuring Ga with a Network Analyzer

In this case we are faced with five variables (Γs , S11 , S12 , S21 , S22 ) that are complex numbers. The S-
parameters are also correlated rather than independent variables, which complicates matters even more. In
a case like this, it is usually simplest to write a computer program to solve the gain equation and vary
the input variables to empirically determine the gain uncertainty. The difficult part to this approach is
deciding how to vary the input variables. One option is to estimate the uncertainties with which the network
analyzer can determine the real and imaginary components of S-parameters and reflection coefficients. This
information could be gleaned either from the specifications given by the manufacturer of the network analyzer,

13
from a statistical analysis of measuring a number of devices with known network parameters using the test
equipment at hand, or from participation in a measurement assurance or comparison program of some kind.

As far as writing the computer program goes, consider that each complex variable consists of a real and
imaginary component (a + jb) for which there are two possible uncertainty offsets. Using our estimate ∆x
for the determination of each real and imaginary component results in the values:

a+ = a + ∆x, (61)

a− = a − ∆x, (62)
b+ = b + ∆x, (63)
−
b = b − ∆x. (64)
which means that for each complex variable in our gain equation there are four possible perturbations:

a+ + jb+ , (65)

a+ + jb− , (66)
a− + jb+ , (67)
− −
a + jb , (68)

and that for our gain equations 59 and 60, which depend on five complex variables, there are 45 = 1, 024
possible combinations that need to be computed in addition to the “measured” gain value. Finding the
maximum difference between these “fluctuations” and the measured value determines the Type B uncertainty
(uB ) in the measurement.

As usual, the Type A uncertainty (uA ) would commonly be found as the standard deviation of a series of
repeated gain measurements. The expanded uncertainty would then be computed as:
q
U = k u2A + u2B (69)

where the coverage factor k = 2 would commonly be used to indicate a 95% confidence interval [4].

5. SUMMARY

It is clear that when building a measurement system it is important to account for the gain (or loss) in the
RF measurement path, that there are a number of ways of expressing gain, and that the definition used
must be consistent with the manner in which the gain term will ultimately be used. Also, it is important
to note that how the measurement is performed has an effect on the results. A comparison of some of the
basic features of the gain calibration methods discussed can be found in Table 1.

It should also be clear that the uncertainties attainable by any given technique depend not only on the
equipment at hand, but also on how it is used. It should be noted that the uncertainty analyses presented
here are minimal. They are intended only as a baseline from which a more complete analysis for a specific
experiment can be derived.

14
 Table 1. Comparison of Measurement Methods

Measurement Method Advantages Disadvantages

Signal Generator: single-
point method • Simple test to perform, • Assumes strict system lin-
minimum time and equip- earity.
ment.
• No zero point offset al-
• Narrowband calibration. lowed.
• Suitable for low gain sys- • Requires absolute power
tems. measurements which bur-
dens the detector system.

Signal Generator: two-

point method • Is a difference measure- • Is a more complicated test.
ment. This puts less de-
mand on the detector cir-
cuit.
• Zero point offset is allowed.
• Narrowband calibration.
• Suitable for low gain sys-
tems.

Noise Source
• Wide bandwidth character- • Only suitable for high gain,
ization. low noise systems.
• Noise sources tend to be • Not suitable if a narrow-
small and easily integrated band calibration is needed.
into a test system.

Network Analyzer:
Gain computation from • Gain is relatively sim- • Requires the use of complex
S-parameters ple to compute from and expensive equipment.
S-parameters.
• Requires measurement of
• Results in a complete math- five complex variables.
ematical model of signal
propagation through the
two-port.

15
 6. REFERENCES

[1] “Fundamentals of RF and microwave noise figure measurements,” Agilent Application Note 57-1, July
1983.
[2] G. F. Engen, Microwave Circuit Theory and Foundations of Microwave Metrology, IEEE Electrical
Measurement Series 9, London, Peter Peregrinus Ltd., 1992.
[3] D. M. Kerns and R. W. Beatty, Basic Theory of Waveguide Junctions and Introductory Microwave
Network Analysis, International Series of Monographs in Electromagnetic Waves, London, Pergamon
Press, 1967.
[4] B. N. Taylor and C. E. Kuyatt, “Guidelines for evaluating and expressing the uncertainty of NIST
measurement results,” NIST Technical Note 1297, National Institute of Standards and Technology,
1994.
[5] J. R. Taylor, An Introduction to Error Analysis - The Study of Uncertainties in Physical Measurements,
Mill Valley, CA, University Science Books, 1982.
[6] W. C. Daywitt, “Design and error analysis for the WR10 thermal noise standard,” NBS Technical Note
1071, National Bureau of Standards, 1983.
[7] W. C. Daywitt, “A coaxial noise source for the 1 GHz to 12.4 GHz frequency range,” NBS Technical
Note 11074, National Bureau of Standards, 1984.

[8] J. D. Kraus, Radio Astronomy 2nd Edition. Powell, Ohio, Cygnus-Quasar Books, 1986.

16