Network Analyzer

Error Models
and
Calibration Methods
by
Doug Rytting

Slide 1

This paper is an overview of error models and calibration methods for

vector network analyzers.

1
 Presentation Outline
• Network Analyzer Block Diagram and Error Model
• System Error Model for Error-Correction
• One-Port Error Model and Calibration
• Two-Port Error Models and Calibration
• 12-Term Method
• 16-Term Method
• 8-Term Method
• Measuring S-parameters
• Accuracy of Error-Correction

Slide 2

A system error model will be derived from a generic network analyzer

block diagram. This error model will then be simplified to the standard
one-port and 12-term two-port models used the past 30 years.
Newer 16-term and 8-term models will then be introduced and the
modern calibration approaches described.
Measuring s-parameters without a Z0 termination will be covered. Also
the accuracy of error-corrected measurements will be outlined.

2
 Network Analyzer
Block Diagram
and
Error Model

Slide 3

First the block diagram for a network analyzer is described and the
hardware flow graph is defined.

3
Network Analyzer Block Diagram
RF
Source

a0 a3
IF IF

LO
Source
IF IF
b0 b3
Port - 1 Port - 2
a1 b2
Cable Cable
DUT
b1 a2

Slide 4

This is a generic block diagram of a 4 channel network analyzer. The

source can be switched to excite port-1 or port-2 of the device under test
(DUT). The switch also provides a Z0 termination for the output port in
each direction. Directional couplers are used to separate the incident,
reflected and transmitted waves in both the forward and reverse
direction. Mixers are used to down convert the RF signals to a fixed low
frequency IF. The LO source is tuned to the frequency of the RF + IF.
The s-parameters of the DUT can be defined as follows:
S11 = b1/a1, switch in forward direction
S21 = b2/a1, switch in forward direction
S12 = b1/a2, switch in reverse direction
S22 = b2/a2, switch in reverse direction

4
Block Diagram – Forward Direction

Port - 1 a1 b2 Port - 2
DUT
b1 a2
RF
Source
LO LO

A/D A/D A/D

IF IF

DSP DSP DSP

a0 b0 b3

Slide 5

This block diagram shows the measurement system switched to the

forward direction. Each of the IF signals are detected and digitized and
the real and imaginary terms are measured. From this data the
magnitude and phase can be calculated.
In most modern network analyzers the A/D digitizes directly at the IF
and the detection is done in the digital domain. The resultant digitized
versions of the DUT waves (a0, b0, and b3) are a scaled version of the
actual waves at the DUT (a1, b1, and b2).

5
 Error Model – Forward Direction

Cable & DUT Cable &

Port - 1 Connector Connector Port - 2
aS LS-1 a1 b2
LC LC
S21
MS M1 MC MC S11 S22 MC MC M2
L1-S S12
LC LC
LS-a0 LS-b0 L1-a0 b1 a2 L2-b3
L1-b0
NLa0 NLb0 NLb3
La0-b3

1+Aa0+NHa0 1+Ab0+NHb0 1+Ab3+NHb3

IF IF
a0 b0 b3

Branches With No Label = 1

Slide 6

From the block diagram a flowgraph can be developed showing all the
possible signal paths. These paths not only include the main desired
signals but the loss, match errors, and leakage errors, of the network
analyzer along with the cables, connectors, or probes that connect to to
DUT.
Also included in this model are the IF, A/D and detector non linearities
and the system noise.

6
 Error Model Definitions
a1 = Incident Signal at Port-1 S11 = Refl Coef of DUT at Port-1
b1 = Reflected Signal at Port-1 S21 = Forward Trans Coef of DUT
a2 = Incident Signal at Port-2 S12 = Reverse Trans Coef of DUT
b2 = Transmitted signal at Port-2 S22 = Refl Coef of DUT at Port-2

aS = Source Port M1 = Match at Port-1

a0 = Measured Incident Port M2 = Match at Port-2
b0 = Measured Reflected Port MS = Match of Source
b3 = Measured Transmitted Port MC = Match of Cables

LS-1 = Loss from Source to Port-1 NLa0 = Low Level Noise at a0

L1-S = Loss from Port-1 to Source NLb0 = Low Level Noise at b0
LS-a0 = Loss from Source to a0 NLb3 = Low Level Noise at b3
LS-b0 = Loss from Source to b0 (Directivity) NHa0 = High Level Noise at a0
L1-a0 = Loss from Port-1 to a0 (Directivity) NHb0 = High Level Noise at b0
L1-b0 = Loss from Port-1 to b0 NHb3 = High Level Noise at b3
L2-b3 = Loss from Port-2 to b3
La0-b3 = Loss from a0 to b3 (Leakage) Aa0 = Dynamic Accuracy at a0 (Linearity)
LC = Loss of Cables Ab0 = Dynamic Accuracy at b0 (Linearity)
Ab3 = Dynamic Accuracy at b3 (Linearity)

Slide 7

The above table gives the description of each of the branches and the
key nodes for the flow graph. This provides a very complete model for
the network analyzer. However it is possible to reduce the flow graph
without any loss in accuracy. This reduced flow graph is much easier to
analyze and will be discussed next.

7
 System Error Model
for
Error-Correction

Slide 8

The simplified system error model is described. This system model will
be used to develop the error correction procedure.

8
System Model – Forward Direction
e30
DUT
Port - 1 Port - 2
a1 b2
a0 b3
e10 S21 e32
e00 e11 S11 S22 e22
e01 S12
b0
b1 a2 Branches With No Label = 1

LS−b 0
Directivity e 00 ≅ Leakage e30 ≅ La 0−b 3
LS− a 0

Reflection LS−1 L1−b 0 LS−1 L1− a 0 LS−b 0 Transmission LS1 L2−b 3

e10e01 ≅ − e10e32 ≅
Tracking LS− a 0 (LS− a 0 )2 Tracking LS− a 0
LS−1 L1− a 0
Port-1 Match e11 ≅ M2 − Port-2 Match e22 ≅ M2
LS− a 0

There are also Errors caused by the Converter, IF, Cables and Connectors
There are also Six Terms in the Reverse Direction

Slide 9

The resultant system error model is the forward portion of the well
known 12-term error mode. Each of the branches have an accurate
relationship to the original hardware oriented flow graph presented
earlier. The 6 forward terms described above show a simplified set of
equations relating the two flow graphs.
The directivity error is caused primarily by the coupler leakage or
‘coupler directivity.’ This error is also increased by cable and connector
match errors between the measurement coupler and the DUT. The
reflection and transmission tracking is caused by reflectometer and
mixer tracking as well as cable length imbalance between the measured
ports. The match error is the mathematical ratioed port match error that
is not necessarily the ‘raw’ port match. The leakage error is through the
LO path of the mixers. It is not the leakage of the switch and this model
assumes the switch leakage is negligible.

9
 Improvements with Correction

ERRORS REMOVED ERRORS REMAINING

Port Match Noise and Residuals

Directivity Receiver Linearity

Tracking Drift after Error-Correction

Main Leakage Paths Stability after Error-Correction

Repeatability of Connectors, etc

Lower Lever Leakage Paths

Errors of Calibration Standards

Slide 10

A linear calibration procedure is applied to remove as many of the errors as possible. The
loss, directivity, match, and main leakage errors can be greatly reduced depending on the
accuracy of the calibration standards used.
However, the noise and linearity errors can not be reduced using a simple linear
calibration procedure. If fact the noise and linearity errors increase a small amount.
Residuals also remain that are caused by A/D quantization errors, clock leakage, etc.
Once the network analyzer is calibrated the drift, stability, and repeatability errors will
degrade the system performance. This usually means that the system will need to be
recalibrated at some interval depending on the system usage, environment and required
accuracy.
There are some lower level leakage paths in the RF hardware that are not modeled in
many of the error correction schemes.

10
Improvements with Correction

Slide 11

To see the improvements offered by error correction lets compare the

measurement results before and after error correction when measuring a
beaded airline. The two test cases will be with a response only
calibration, which will not remove the port match and directivity errors,
and a full two port calibration which will remove all the errors.
There is a twenty dB improvement in the reflection measurements when
the directivity errors are reduced.
The transmission tracking errors are greatly improved with error
correction. The error was mainly caused by port match. It should be
pointed out that the uncorrected response errors are only a tenth dB
which means the uncorrected return loss at both test ports must be at
least 20 dB. After correction they have been reduced 20 db more.

11
 One Port Error Model
and
Calibration

Slide 12

The one-port model will be first developed. This will then be used to
further develop the two-port model.

12
One Port: 3-Term Error Model
a0 b0
a0 Error a1
DUT
b0 Adapter b1
Perfect
Reflectometer 3 Error Terms

DUT
Port - 1
a1
a0 e00 = Directivity
1
e00 e11 Γ e11 = Port Match

e10e01 (e10e01) = Tracking

b0
b1

Slide 13

The most simple case is the one port 3-term error model. From this
model the same approach will be used for the two port cases.
In the one-port calibration procedure the model simplifies to just the
terms describing the directivity, port match , and tracking errors at port-
1.
The errors can be lumped into a fictitious error adapter that modifies the
actual DUT reflection coefficient which is then measured by a ‘perfect’
reflectometer.

13
 One Port: 3-Term Error Model
Measured Actual

b0 e00 - ∆eΓ ΓM - e00

ΓM = = Γ=
a0 1 - e11Γ ΓMe11 - ∆e

∆e = e00e11 - (e10e01)

For ratio measurements there are 3 error terms

The equation can be written in the linear form

e00 + ΓΓMe11 - Γ∆e = ΓM

With 3 different known Γ, measure the resultant 3 ΓM

This yields 3 equations to solve for e00, e11, and ∆e

e00 + Γ1ΓM1e11 - Γ1∆e = ΓM1

e00 + Γ2ΓM2e11 - Γ2∆e = ΓM2

e00 + Γ3ΓM3e11 - Γ3∆e= ΓM3

Any 3 independent measurements can be used

Slide 14

Solving the one-port flow graph yields a bilinear relationship between the actual
and measured reflection coefficient. The actual reflection coefficient is
‘mapped’ or modified by the three error terms to the measured result. This
equation can be inverted to solve for the actual reflection coefficient knowing
the measured result and the three error terms.
The three error terms can be determined by measuring three known standards
(such as an open, short and load) that yield three simultaneous equations. These
three equations can then be solved for the three error terms.
Another popular calibration standard is called e-cal. With this technique the
reflection coefficient is selected electronically from a set of pre measured states.
These were measured using a network analyzer calibrated with primary
standards. This is strictly a transfer standard but as long as the e-cal standard is
stable it is very accurate. This type of standard allows fast calibrations, ease of
use and less operator error.

14
 Two-Port Error Models
and
Calibration
12-Term Method
16-Term Method
8-Term Method

Slide 15

The classic 12-term model will be developed first. Then the more recent
16-term and 8-term models will be described.

15
 12-Term Error Model

b0

a0 a1

Forward b0 b1

Perfect Error
DUT
Reflectometer Adapter

a0,a3 Reverse a3 a2

b3 b2

b3

Slide 16

Many of the older and lower cost network analyzers use three couplers
instead of four for two port measurements. This puts the switch between
the incident and reflected couplers. If the switch characteristics change
as the switch is flipped from forward to reverse, the error adapter’s error
terms will change. Even some four coupler network analyzers use this
method as well, choosing not to use the fourth coupler for all
measurements.
This is the original technique uses in the 1960’s for the first automatic
network analyzers. This TOSL (through, open, short, load) 12-term
model has been used for many years and is still widely used today.

16
 12-Term Error Model
a0 b0
a0 a1
b0 b1
Forward
Forward Perfect DUT
Error
Model Reflectometer [S]
Adapter
a3 a2
b3 b2
b3 6 Error Terms

b'0
a'0 a'1
b'0 b'1
Reverse
Reverse Perfect DUT
Error
Model Reflectometer [S]
Adapter
a'3 a'2
b'3 b'2
a'3 b'3 6 Error Terms

Slide 17

The two-port case can be modeled in the same manner as the one-port.
A fictitious error adapter is placed between the two-port DUT and the
‘perfect reflectometer’ measurement ports. This error adapter contains
the 6 error terms for the forward direction. A similar 6 term model is
used in the reverse direction.

17
 12-Term Error Model
FORWARD MODEL

e30
DUT
Port - 1 Port - 2
a1 b2
a0 b3
1 S21 e10e32
e00 e11 S11 S22 e22
e10e01 S12
b0
b1 a2

e00 = Directivity
b0 S11 - e22 ∆S
e11 = Port-1 Match S11M = = e00 + (e10e01)
a0 1 - e11S11 - e22S22 + e11e22 ∆S
(e10e01) = Reflection Tracking
b3 S21
(e10e32) = Transmission Tracking S21M = = e30 + (e10e32)
a0 1 - e11S11 - e22S22 + e11e22 ∆S
e22 = Port-2 Match
e30 = Leakage
∆S = S11S22 - S21S12

Slide 18

Solving the forward flow graph yields measurements S11M and S21M.
These two equations contain all four actual S-parameters of the DUT
and the six forward error terms.

18
 12-Term Error Model
REVERSE MODEL
DUT
Port - 1 a'1 b'2 Port - 2
b'3
S21 e'23e'32
e'11 S11 S22 e'22 e'33
e'23e'01 S12 1
b'0 a'3
b'1 a'2

e'03

e'33 = Directivity
b'3 S22 - e'11 ∆S
e'11 = Port-1 Match S22M = = e'33 + (e'23e'32)
a'3 1 - e'11S11 - e'22S22 + e'11e'22 ∆S
(e'23e'32) = Reflection Tracking
b'0 S12
(e'23e'01) = Transmission Tracking S12M = = e'03 + (e'23e'01)
a'3 1 - e'11S11 - e'22S22 + e'11e'22 ∆S
e'22 = Port-2 Match
e'03 = Leakage
∆S = S11S22 - S21S12

Slide 19

Solving the reverse flow graph yields measurements S22M and S12M.
These two equations contain all four actual S-parameters of the DUT
and the six reverse error terms.
The forward and reverse equations combine to give four equations
containing the four actual S-parameters of the DUT and 12 error terms.
If the 12 error terms are known these four equations can be solved for
the actual S-parameters of the DUT.

19
 12-Term Error Model
⎛ S11M − e00 ⎞ ⎡ ⎛ S22M − e'33 ⎞ ⎤ ⎛ S − e30 ⎞ ⎛ S12M − e'03 ⎞
⎜ ⎟ ⎢1 + ⎜ ⎟ e'22 ⎥ − e22 ⎜ 21M ⎟⎜ ⎟
⎝ e10 e01 ⎠ ⎣ ⎝ e'23 e'32 ⎠ ⎦ ⎝ e10 e32 ⎠ ⎝ e'23 e'01 ⎠
S11 =
D
⎛ S21M − e30 ⎞ ⎡ ⎛ S22M − e'33 ⎞ ⎤
⎜ ⎟ ⎢1 + ⎜ ⎟ (e'22 − e22 )⎥
⎝ e10 e32 ⎠ ⎣ ⎝ e'23 e'32 ⎠ ⎦
S21 =
D
⎛ S22M − e'33 ⎞ ⎡ ⎛ S11M − e00 ⎞ ⎤ ⎛ S − e30 ⎞ ⎛ S12M − e'03 ⎞
⎜ ⎟ ⎢1 + ⎜ ⎟ e11 ⎥ − e'11 ⎜ 21M ⎟⎜ ⎟
⎝ e'23 e'32 ⎠ ⎣ ⎝ e10 e01 ⎠ ⎦ ⎝ e10 e32 ⎠ ⎝ e'23 e'01 ⎠
S22 =
D
⎛ S12M − e'03 ⎞ ⎡ ⎛ S11M − e00 ⎞ ⎤
⎜ ⎟ ⎢1 + ⎜ ⎟ (e11 − e'11 )⎥
⎝ e'23 e'01 ⎠ ⎣ ⎝ e10 e01 ⎠ ⎦
S12 =
D

⎡ ⎛ S − e00 ⎞ ⎤ ⎡ ⎛ S22M − e'33 ⎞ ⎤ ⎛ S − e30 ⎞ ⎛ S12M − e'03 ⎞

D = ⎢1 + ⎜ 11M ⎟ e11 ⎥ ⎢1 + ⎜ ⎟ e'22 ⎥ − ⎜ 21M ⎟⎜ ⎟ e22e'11
⎣ ⎝ e10 e01 ⎠ ⎦ ⎣ ⎝ e'23 e'32 ⎠ ⎦ ⎝ e10 e32 ⎠ ⎝ e'23 e'01 ⎠

Slide 20

This is the result of solving the four simultaneous measured s-parameter

equations. Note that each actual s-parameter calculated requires
measuring all four S-parameters as well as knowing the 12 error terms.

20
 12-Term Error Model
Calibration
STEP 1: Calibrate Port-1 using One-Port procedure

Solve for e11, e00, & (e10e01), Calculate (e10e01) from ∆e

STEP 2: Connect Z0 terminations to Ports 1 & 2

Measure S21M gives e30 directly

STEP 3: Connect Ports 1 & 2 together

S11M - e00
e22 =
S11Me11 - ∆e

e10e32 = (S21M - e30)(1 - e11e22)

Use the same process for the reverse model

Slide 21

The 12 error terms will now be determined. First solve for the 6 terms in
the forward direction. Then the same procedure can be used to solve for
the 6 reverse terms.
Step one calibrates port-1 of the network analyzer using the same
procedure used in the one-port case. This determines the directivity,
match, and reflection tracking at port-1 (e00, e11, and ∆e). From ∆e the
reflection tracking (e10e01) can be calculated.
Step two measures the leakage or crosstalk error (e30) from port-1 to
port-2 directly by placing loads on each of the ports.
Step three consists of connecting port-1 and port-2 together. Then
measure the port-2 match (e22) directly with the calibrated port-1
reflectometer. Then with the ports connected, measure the transmitted
signal and calculate the transmission tracking (e10e32).

21
 16-Term Error Model

a0 b0

a0 a1

b0 b1

Perfect Error DUT

Reflectometer Adapter [S]
Imperfect
Switch a3 a2

b3 b2

a3 b3 16 Error Terms

To remove the effects of an imperfect switch, use the procedure described later.

Slide 22

The complete system can be modeled using an error adapter between

the DUT and the actual measurements. This error adapter describes all
the possible stationary and linear errors that are caused by the system.
Since the adapter is a 4-port it contains 16 error terms. If these error
terms can be determined then the DUT S-parameters can be calculated
from the measured S-parameters.
When making ratio measurements one of the terms can be normalized to
1 leaving 15 error terms.
When using 4 couplers placed after the switch the errors of the switch
can be removed and will be discussed later.

22
 16-Term Error Model
a0 a1
a0
e10

e00 e20 e13 e11 DUT

b0 e01 b1 b1
b0 a1 e00, e33 Directivity
S11 e11, e22 Port Match
e10, e01, e32, e23 Tracking
e30 e03 e21 e12 S12 S21 e30, e03 Primary Leakage

S22 All others are lower level

a3 b2 leakage paths
a3 e23 a2 a2

e33 e31 e02 e22

e32
b3
b3 b2

One of the 16 error terms can be normalized to yield 15 error terms

Slide 23

There are 16 terms of which 8 are critical. These 8 include the

directivity, port match and tracking errors.
The leakage terms add 8 additional error terms to the model. Not only is
the traditional crosstalk term included, but switch leakage, signals
reflecting from the DUT and leaking to the transmission port, common
mode inductance, and other leakage paths are included. In a coaxial or
waveguide system, assuming the switch has high isolation, these errors
are small. But in a fixtured or wafer probe environment these errors can
be much larger.
In a wafer probe environment it is important that the error terms do not
change as the probes are moved around the circuit. If the error terms
change, the 16-term model then changes and the accuracy will reduce.
Again the error terms can be normalized so that for ratio measurements
there are only 15 error terms.

23
 16-Term Error Model
⎡b0 ⎤ ⎡ a1 ⎤
⎢b ⎥ T2 ⎤ ⎢a2 ⎥
⎢ 3⎥ = ⎡ 1
T
⎢ ⎥
Error Model ⎢
⎢a0 ⎥ ⎣T3 T4 ⎥⎦ ⎢ b1 ⎥
⎢ ⎥ ⎢ ⎥
⎣a3 ⎦ ⎣b2 ⎦

Measured S-Parameters SM = (T1S + T2)(T3S + T4)-1

Actual S-Parameters S = (T1 - SMT3)-1(SMT4 - T2)

Linear-in-T Form T1S + T2 - SMT3S - SMT4 = 0

With 15 or more independent observations the linear matrix

equation can be solved. TRL as well as TOSL calibration
methods are possible.

Slide 24

One method of solving this system is to use T-parameters (cascade

parameters) that relate the measured waves to the DUT waves. This
system is solved by partitioning the T matrix into 4 2x2 matrices and
then solving the matrix equations which yields the bi-linear form above.
This allows us to form a linear set of equations that can be solved for
the error terms.
By using enough known standards, a set of linear equations is created
that can be solved for the individual error terms. At least 5 two port
standards are required to solve the system of equations. However, most
of these can be degenerate two port standards that involve combinations
of opens, shorts, loads and transmission lines.

24
 8-Term Error Model

a0 b0

a0 a1
X
b0 Error b1
Adapter
Perfect
DUT
Reflectometer
Imperfect Y
Switch a3 Error a2
Adapter
b3 b2

a3 b3 8 Error Terms

To remove the effects of an imperfect switch, use the procedure described later.

Slide 25

The 8-term model can be derived from the 16-term model. First assume
that the leakage terms are all zero. Or that the two primary leakage
terms can be determined in a separate calibration step. Then assume that
the switch is perfect and does not change the port match of the network
analyzer as it is switched from forward to reverse. This assumption is
valid if there are 4 measurement channels that are all on the DUT side
of the switch. Then it is possible to mathematically ratio out the switch.
This mathematical approach will be discussed later.

25
 8-Term Error Model
a0 a1
a0
e10

e00 X Error Adapter e11 DUT

e01 b1
b0 a1
b0 b1 S11

S12 S21

a3 a2 S22
a3 b2
e23 a2

e33 Y Error Adapter e22

e32
b3
b3 b2

One of the 8 error terms can be normalized to yield 7 error terms

Slide 26

The flow graph consists of an error adapter at the input and output of the
DUT. For ratio measurements of S-parameters, the number of error
terms is reduced to 7 since the error terms can be normalized.

26
 8-Term Error Model

⎡b0 ⎤ ⎡ a1 ⎤
⎢b ⎥ T2 ⎤ ⎢a2 ⎥
⎢ 3⎥ = ⎡ T1 ⎢ ⎥
⎢a 0 ⎥ ⎢T T4 ⎥⎦ ⎢ b1 ⎥
⎣ 3
⎢ ⎥ ⎢ ⎥
⎣a 3 ⎦ ⎣b2 ⎦

⎡− ∆ 0 ⎤ ⎡e 0 ⎤
T1 = ⎢ X T2 = ⎢ 00
⎣ 0 − k∆Y ⎥⎦ ⎣0 ke33 ⎥⎦

⎡− e 0 ⎤ ⎡1 0 ⎤
T3 = ⎢ 11 T4 = ⎢
⎣ 0 − ke22 ⎥⎦ ⎥
⎣0 k ⎦

e10
k= , ∆ X = e00e11 − e10e01 , ∆Y = e22 e33 − e32e23
e23

Slide 27

This is one of many possible mathematical formulation for the 8-term

error model. Consider the error adapter as just one adapter between the
perfect measurement system and the DUT. Then model this error
adapter using the cascade T-parameters. This T-parameter matrix (T)
can be partitioned, as described in the 16-term method, into the four
diagonal sub matrixes T1, T2, T3, and T4. The 7 error terms are now
defined as ∆X, k∆Y, e00, ke33, e11, ke22, and k. Which are the tracking,
directivity, and match errors.

27
 8-Term Error Model

Measured S-Parameters SM = (T1S + T2)(T3S + T4)-1

Actual S-Parameters S = (T1 - SMT3)-1(SMT4 - T2)

Linear-in-T Form T1S + T2 - SMT3S - SMT4 = 0

Expanding Yields:

e00 + S11S11Me11 - S11∆X +0 + S21S12M(ke22) +0 +0 = S11M

0 + S12S11Me11 - S12∆X +0 + S22S12M(ke22) +0 - S12Mk =0

0 + S11S21Me11 +0 +0 + S21S22M(ke22) - S21(k∆Y) +0 = S21M

0 + S12S21Me11 +0 + (ke33) + S22S22M(ke22) - S22(k∆Y) - S22Mk =0

Slide 28

Using this approach the measured S-parameters formulation is a

‘bilinear matrix equation.’ The equation can be ‘inverted’ to solve for
the actual S-parameters. And most important the relationship can be put
in linear form. Expanding this matrix equation for the two-port case
yields 4 equations with 4 measured S-parameters, 4 actual S-parameters,
and 7 error terms. Note that these 4 equations are linear with regards to
the 7 error terms.
This approach is particularly attractive for multi-port measurement
systems. The matrix formulation does not change as additional ports are
added.

28
 8-Term Error Model
Using the cascade parameters in matrix form yields

MEASURED ACTUAL

TM = TXT TY T = TX-1 TM TY-1

1 ⎡ − ∆S S11 ⎤ 1 ⎡ − ∆M S11M ⎤
T= ⎢ TM =
S 21 ⎣− S 22 1 ⎥⎦ ⎢
S 21M ⎣− S 22M 1 ⎦⎥

∆ S = S11S 22 − S12 S 21 ∆ M = S11MS 22M − S12MS 21M

1 ⎡− ∆ X e 00 ⎤ 1 ⎡− ∆ Y e 22 ⎤
TX = TY = ⎢− e
e10 ⎢− e
⎣ 11 1 ⎦
⎥ e 32 ⎣ 33 1 ⎥⎦

∆ X = e 00 e11 − e10 e 01 ∆ Y = e 22 e 33 − e 32 e 23

1 ⎡ − ∆ X e 00 ⎤ ⎡ − ∆ Y e 22 ⎤ 1
TM = =
(e10 e 32 ) ⎢⎣− e11 1 ⎥⎦ ⎢⎣− e 33 ⎥
T ATB
1 ⎦ (e10 e 32 )

Slide 29

Another approach is to cascade of the input error box (X), the DUT, and
the output error box (Y). The measured result of this cascade is most
easily calculated by using the cascade matrix definition (T-parameters).
This formulation was used by Engen and Hoer in their classic TRL
development for the six-port network analyzer. And is the same
approach used in the HP 8510 network analyzer.
From the last equation in the slide above, the 7 error terms are easily
identified. There are 3 at port-1 (∆X, e00, and e11) and 3 at port-2 (∆Y, e22,
and e33) and one transmission term (e10e32).
The calibration approach require enough calibration standards to allow
at least 7 independent observations of the measurement system.

29
 8-Term Calibration Examples
Seven or more independent known conditions must be measured
A known impedance (Z0) and a port-1 to port-2 connection are required
Thru (T) or Line (L) with Unknown equal Reflect (R) Line (L) with known
TRL & LRL known S-parameters on port-1 and port-2 S11 and S22
[4 conditions] [1 condition] [2 conditions]
Thru (T) or Line (L) with Unknown equal Reflect (R) Known Match (M)
TRM & LRM known S-parameters on port-1 and port-2 on port-1 and port-2
[4 conditions] [1 condition] [2 conditions]
Thru (T) or Line (L) with 3 known Reflects (XYZ)
TXYZ & LXYZ known S-parameters on port-1 or port-2
[4 conditions] [3 conditions]
Traditional Thru (T) with 3 known Reflects (OSL) 3 known Reflect (OSL)
TOSL known S-parameters on port-1 on port-2
(Overdetermined) [4 conditions] [3 conditions] [3 condition]
Line (L) with known 2 unknown equal Reflects Known match (M)
LRRM S-parameters (RR) on port-1 and port-2 on port-1
[4 conditions] [2 conditions] [1 condition]
Unknown Line (U) with 3 known Reflects (XYZ) 3 known Reflects (XYZ)
UXYZ S12 = S21 on port-1 on port-2
[1 condition] [3 conditions] [3 conditions]

Slide 30

There is a number of calibration techniques that have been developed

based on the 8-term error model. Seven or more independent conditions
must be measured. There must be a known impedance termination or a
known transmission line. And port-1 and port-2 must be connected for
one of the measurements.
The list of calibration approaches can be much longer than the ones
shown above. And there continues to be new and novel ways to solve
for the seven error terms and calibrate the system.
The 8-term error model approach has yielded more accurate calibration
methods as well as simplified the calibration process. TRL and LRL
provide the best accuracy. The other methods simplify the calibration
steps compared to the older TOSL 12 term model. In one case (UXYZ
above) the thru standard does not need to be known as long as it is
reciprocal.

30
 Measuring
S-parameters

Slide 31

A method for measuring s-parameters without terminating the ports

with a Z0 load will now be described.

31
 Measuring S-parameters
a0 b0

a0 a1

b0 b1
Forward

Perfect Error DUT

Reflectometer Adapter [S]
Reverse
a3 a2

b3 b2

a3 b3

Forward Reverse

b0 = S11Ma0 + S12Ma3 b'0 = S11Ma'0 + S12Ma'3

b3 = S21Ma0 + S22Ma3 b'3 = S21Ma'0 + S22Ma'3

Slide 32

Normally we measure S-parameters by terminating one port with a Z0

impedance and then measure the input reflection and transmission
coefficients then flip the switch and make the same set of measurements in
the reverse direction. However, with 4 couplers we can measure both the
forward and reverse terminating impedances of the switch. This allows us to
“ratio out” or remove the effects of the imperfect switch.
Note that the measurement channels are all on the DUT side of the switch.
This allows the measurements of the incident and reflected signals at both
outputs of the switch and the switch match errors can be calculated.
Mathematically the S-parameters of the system generate 4 equations. 2 in the
forward direction and 2 in the reverse direction. These 4 equations can then
be solved for the 4 measured S-parameters. This general way of measuring
S-parameters does not require the DUT to be terminated in a Z0 environment.

32
 Measuring S-parameters

By defining

a′0 a
Γ1 = and Γ2 = 3
b′0 b3

b0 b'0 b3 b'0 b0 b'0

− Γ2 − Γ1
a0 a'3 a0 a'3 a0 a'3
S11M = S12M =
d d

b3 b'3 b3 b'3 b3 b'0

− Γ2 − Γ1
a0 a'3 a0 a'3 a0 a'3
S 21M = S 22M =
d d

b3 b'0
d = 1− Γ1Γ2
a0 a'3

Slide 33

Solving the 4 previous equations yield the above results. Note that the
equations are written to allow ratio measurements by the network
analyzer. Typically the network analyzer is more accurate making
measurements this way. Noise and other common mode errors are
reduced.
Using this method for measuring the 4 S-parameters requires 6 ratio
measurements. The additional two measurements are required to
remove the effects of the switch. However, these two additional
measurements (Γ1 and Γ2) need only be made during the calibration step
since they do not change during measurement assuming the switch is
stable.

33
 Accuracy
of
Error-Correction

Slide 34

The accuracy of error corrected measurements will now be briefly

discussed.

34
 Accuracy of Error Correction

Slide 35

The error corrected measurement system can be nicely described using flow graphs. The
flow graph after error correction is very similar to the one before correction but with the
error terms reduced.
The device under test will be degraded by the following residual errors and hardware
imperfections. The residual microwave errors (δ, τ1,2 and µ1,2) are mainly determined by
the errors of the calibration standards and are respectively the residual directivity, residual
tracking and residual match. Subscript 1 applies to port-1 and subscript 2 applies to port-2.
Ct1,t2 describes the cable transmission coefficient change and Cr1,r2 describe the change in
the cable reflection coefficient. Rt1,t2 characterizes the connector transmission repeatability
error and Rr1,r2 characterizes the connector reflection repeatability error. The low level
noise (NL1,2) of the converter determines the sensitivity of the system, and a high level
noise of the LO and IF (Nh1,2) contribute to the trace noise on the measurement data.
The front end and IF hardware will drift with time and temperature as characterized by the
stability terms (S1,2). The nonlinearities of the system with measurement level are
described by the dynamic accuracy (A1,2).

35
 Accuracy of Error Correction

Slide 36

The flow graph can be solved and simplified to show the total system
uncertainty. These equations calculate the magnitude uncertainty (∆S11
and ∆S21) with each error term defined by its absolute magnitude. The
first part of the equation describes the systematic errors and these errors
typically add up in a worst case manner. The random, drift and stability
errors are typically characterized in an RSS fashion in the second part.
These equations do not include second order effects.

36
Accuracy of Error Correction

Slide 37

The reflection residual microwave errors are mainly determined by the

quality of the calibration standards. If the assumed values of the
standards and their measured value are known, then the residual errors
can be calculated. These approximate residual errors only apply to
TOSL and TRL type calibrations.

37
 Accuracy of Error Correction
APC-7 (7 mm Coax) at 18 GHz

Residual OSL OSL

TRL TRM
Errors Fixed Load Sliding Load

Directivity
-40 dB -52 dB -60 dB -40 dB
δ

Match
-35 dB -41 dB -60 dB -40 dB
µ

Reflection
Tracking ± .1 dB ± .05 dB ± .01 dB ± .01 dB
τ

Slide 38

This table gives the tradeoff in accuracy for various coax calibration methods. The example is for
APC-7 mm but can be scaled to other connector types. The relative differences stay about the same.
The OSL (Fixed Load) cal is the least expensive and easy to use. The tradeoff is that it has the
lowest accuracy, but this may be fine for many measurements. The directivity error is determined
by the load. The match and tracking errors are mainly determined by the the phase error of the
open.
The OSL (Sliding Load) is the traditional calibration that has been used for many years for accurate
measurements. It is fairly expensive and sometimes the sliding load is not as easy to use. OSL
calibrations are popular for one-port measurements.
With the OSL calibrations the short and open determine most of the match and tracking error and
this error does not change dramatically with improved directivity values.
The TRL method provides the best accuracy and particularly for the match and tracking terms. It is
not always easy to use and the precision lines are reasonably expensive. The TRM calibration does
not change the directivity error over the OSL (load-fixed) method but reduces the match and
tracking errors. The TRL and TRM calibration methods need a two port measurement system in
order to calibrate. The TRM calibration is the easiest to use.

38
Accuracy of Error Correction

Slide 39

The first graph shows the transmission accuracy after error correction
using APC-7 connectors. The main error at high levels is due to
compression. At low levels it is primarily due to noise and uncorrected
leakage. This graph assumes that S11 and S22 of the DUT are zero.
The APC-7 reflection magnitude accuracy using the sliding load
calibration is shown in the second graph. The main error at small
reflection coefficients is due to the sliding load residual match. At high
levels the sliding load match and the open circuit phase error are the
primary contributors. This graph assumes S12 of the DUT is zero.

39
Appendix

Slide 40

40
 Example: TRL

Slide 41

The best know calibration method using the 8-term model is TRL. We
will now review this calibration method. The math nomenclature is
slightly different in this review.
The first step involves separating the system into a perfect reflectometer
followed by a 4-port error adapter. This error adapter represents all the
errors in the system that can be corrected. It can be split into two 2-port
error adapters, X (at port-1) and Y (at port-2), after removing the
leakage (crosstalk) terms as a first step in the calibration. Since X and Y
are 2-ports it would appear there are 8 unknowns to find, however since
all measurements are made as ratios of the b's and a's, there are actually
only 7 error terms to calculate. This means that only 7 characteristics of
the calibration standards are required to be known. If a thru (4 known
characteristics) is used as one of the standards, only 3 additional
characteristics of the standards are needed.

41
 Example: TRL

Slide 42

It is convenient to use T-parameters because it allows one to represent

the overall measurement, M, of the DUT, A, as corrupted by the error
adapters as a simple product of the matrixes,
M = XAY.
In a similar manner, each measurement of three 2-port standards, C1, C2,
and C3 can be represented as M1, M2, and M3.
M1 = XC1Y
M2 = XC2Y
M3 = XC3Y

42
 Example: TRL

Slide 43

While there are 7 unknowns, measuring three 2-port standards yields a

set of 12 equations. Due to this redundancy, it is not necessary to know
all the parameters of all the standards. X and Y can be solved for
directly plus 5 characteristics of the calibration standards.

43
 Example: TRL

Slide 44

All 4 parameters of C1 must be known but only 2 parameters for C2 and

one for C3 if S11 = S22. The simplest of all known standards is a through
line, so let C1 be a thru, C2 a Z0 matched load or line with S12 = S21 and
C3 a high reflect with S11=S22. If needed, impedance renormalization can
be used to shift to a different impedance base. The other parameters of
C2 and C3 can be solved from the calibration process.
For this calibration method there are several combinations of standards
that fit the requirements. However, there are also choices that generate
ill-conditioned solutions or singularities. In choosing appropriate
standards, one standard needs to be Z0 based, one needs to present a
high mismatch reflection, and one needs to connect port-1 to port-2. In
addition, all three standards need to be sufficiently different to create
three independent measurements.

44
 Example: TRL

Slide 45

There are several possible strategies in choosing standards. For the first standard (C1), the
use of a zero length thru is an obvious selection. But a non-zero length thru is also
acceptable if its characteristics are known or the desired reference plane is in the center of
the non-zero length thru. This standard will determine 4 of the error terms.
The second standard (C2) needs to provide a Z0 reference. In this solution, only the match of
this standard needs to be of concern. Its S21 and S12 can be any value and do not need to be
known. In fact, they will be found during the calibration process. This opens up the choices
to a wide range of 2-port components, such as a transmission line, pair of matched loads, or
an attenuator. This standard will determine 2 of the error terms.
For the final standard (C3) only one piece of information is needed. This could be an
unknown reflection value for the same reflection connected to each port (S11 = S22). Since
the other standards have been well matched, this standard should have a higher reflection.
This standard determines the last error term.
The table shows a partial list of possible calibration configurations with appropriate three
letter acronyms.

45
 Example: TRL

Slide 46

The unknown device characteristics can be easily calculated by knowing

the parameters of the X error adapter, the known standard C1, and the
measured data of the test device and measured data for C1.

46
Example: Unknown T, Known A & B
1 ⎡− ∆ X e00 ⎤ ⎡ − ∆ Y e22 ⎤ 1
TM = T = ATB
(e10 e32 ) ⎢⎣−e11 1 ⎥⎦ ⎢⎣−e33 ⎥
1 ⎦ (e10 e32 )

(e10 e32 )TM = ATB

det[(e10 e32 )TM ] = det[ ATB]

(e10 e32 )2 detTM = det[ AB], since detT = 1, because S21 = S12

Therefore

detAdetB
(e10 e32 ) = ±
detTM

Slide 47

This example is for the unknown thru calibration method (UXYZ). This
is most easily developed using the cascaded t-parameter formulation.
With 3 known standards at port-1 and port-2, 6 conditions are provided.
The thru standard with S21 = S12 provides the 7th required condition.
The key to solving this approach is that the determinant of T is unity for
the passive thru calibration connection (S21 = S12).

47
Example: Unknown B, Known A & T
1 ⎡− ∆ X e00 ⎤ ⎡ − ∆ Y e22 ⎤ 1
TM = T = ATB
(e10 e32 ) ⎢⎣ −e11 1 ⎥⎦ ⎢⎣−e33 1 ⎥⎦ (e10 e32 )

1
B = T −1A −1TM = D, and D is completely known
(e10 e32 )

⎡ ∆Y e22 ⎤
⎢− ( e e ) (e10 e32 ) ⎥⎥ = ⎡D11 D12 ⎤
⎢ 10 32
⎢D ⎥
⎢− e33 1 ⎥ ⎣ 21 D 22 ⎦
⎢⎣ (e10 e32 ) (e10 e32 ) ⎥⎦

Therefore

1 D12 D 21 D11
(e10 e32 ) = e22 = e33 = − ∆Y = −
D 22 D 22 D 22 D 22

Slide 48

This example calibration approach (TXYZ) requires 3 known standards

on port-1 but none on port-2 as long as the thru connection is known.
The known thru connection provides 4 observations. The 3 known
reflection standards connected to port-1 complete the required 7 known
conditions.

48
 8-Term to 10-Term -Forward DUT
Port - 1 Port - 2
a1 b2
a0 b3
e10 S21 e32
e00 e11 S11 S22 e22 e33
e01 S12 e23
b0 a3
b1 a2
DUT
Port - 1 Port - 2
a1 b2
a0 b3
e10 S21 e32
Γ3
a3
e00 e11 S11 S22 e22 e33
Γ3 =
e01 S12 e23 b3
b0
b1 a2
DUT
e32 e 23 Γ3
e′22 = e 22 +
Port - 1 Port - 2
a1 b2
a0
S21
b3 1 − e33 Γ3
1 e10e'32
e00 e11 e'22 e32
e′32 =
S11 S22

b0
e10e01 S12 1 − e33 Γ3
b1 a2

Slide 49

The 12-term and 8-term models describe the same system. So there must
be a relationship between them. First lets reduce the 12 term model to 10
terms by removing the crosstalk terms which can easily be measured in a
separate step. Then the 8-term model can be modified as shown above.
First the 4th measurement channel (a3) is removed by defining the ‘switch
match’ as Γ3. Then the forward model error terms for port match and
transmission tracking (e´22 and e´32) can be calculated. This gives the
standard forward model if we form the two products e10e01 and e10e´32 and
normalize e10 to 1.

49
 8-Term to 10-Term -Reverse
DUT
Port - 1 Port - 2
a1 b2
a0 b3
e10 S21 e32
e00 e11 S11 S22 e22 e33
e01 S12 e23
b0 a3
b1 a2
DUT
Port - 1 Port - 2
a1 b2
b3
a0 e10 S21 e32
Γ0 =
b0 Γ0 e00 e11 S11 S22 e22 e33
e01 S12 e23
b0 a3
b1 a2
DUT
e e Γ Port - 1 Port - 2
e′11 = e11 + 10 01 0 a1 b2
1 − e00 Γ0 S21 e23e32
b3

e01 e'11 S11 S22 e22 e33

e′01 = e23e'01
1 − e00 Γ0 b0
S12 1
a3
b1 a2

Slide 50

The same procedure can be used for the reverse model.

There is a constraining relationship for the 8-term and 10-term models. This is
the same as saying that the 8-term model can reduce to 7 terms when making
ratio measurements. And the 10-term model can reduce to 9 terms.

8-term: (e10e01)(e23e32) = (e10e32)(e23e01)

10-term: [e10e01 + e00(e´11 - e11)][e23e32 + e33(e´22 - e22)] = (e10e´32)(e23e´01)

50
 Vector Network Analyzer References

Basic Error Correction Theory

• S. Rehnmark, “On the Calibration Process of Automatic Network Analyzer Systems,” IEEE Trans. on Microwave Theory
and Techniques, April 1974, pp 457-458.
• J. Fitzpatrick, “Error Models for Systems Measurement,” Microwave Journal, May 1978, pp 63-66.
TRL and Self Calibration Techniques
• G. F. Engen and C. A. Hoer, “Thru-Reflect-Line: An Improved Technique for Calibrating the Dual 6-Port Automatic
Network Analyzer,” IEEE Trans. on Microwave Theory and Techniques, MTT- 27-12, Dec. 1979, pp 983 – 987.
• H. J. Eul and B. Schiek, “A Generalized Theory and New Calibration Procedures for Network Analyzer Self-Calibration,”
IEEE Trans. on Microwave Theory & Techniques, vol. 39, April 1991, pp 724-731.
• R. A. Speciale, “A Generation of the TSD Network-Analyzer Calibration Procedure, Covering N-Port Scattering-Parameter
Measurements, Affected by Leakage Errors,” IEEE Trans. on Microwave Theory and Techniques, vol. MTT-25, December
1977, pp 1100-1115.
8-Term Error Models
• Kimmo J. Silvonen, “A General Approach to Network Analyzer Calibration,” IEEE Trans. on Microwave Theory &
Techniques, Vol 40, April 1992.
• Andrea Ferrero, Ferdinando Sanpietro and Umberto Pisani, “Accurate Coaxial Standard Verification by Multiport Vector
Network Analyzer,” 1994 IEEE MTT-S Digest, pp 1365-1367.
• Andrea Ferrero and Umberto Pisani, “Two-Port Network Analyzer Calibration Using an Unknown ‘Thru’,” IEEE
Microwave and Guided Wave Letters, Vol. 2, No. 12, December 1992, pp 505-506.
• Andrea Ferrero and Umberto Pisani, “QSOLT: A new Fast Calibration Algorithm for Two Port S Parameter Measurements,”
38th ARFTG Conference Digest, Winter 1991, pp 15-24.
• H. J. Eul and B. Scheik, “Reducing the Number of Calibration Standards for Network Analyzer Calibration,” IEEE Trans.
Instrumentation Measurement, vol 40, August 1991, pp 732-735.
16-Term Error Models
• Holger Heuermann and Burkhard Schiek, “Results of Network Analyzer Measurements with Leakage Errors Corrected with
the TMS-15-Term Procedure,” Proceedings of the IEEE MTT-S International Microwave Symposium, San Diego, 1994, pp
1361-1364.
• Hugo Van hamme and Marc Vanden Bossche, “Flexible Vector Network Analyzer Calibration With Accuracy Bounds Using
an 8-Term or a 16-Term Error Correction Model,” IEEE Transactions on Microwave Theory and Techniques, Vol. 42, No. 6,
June 1994, pp 976-987.
• A. Ferrero and F. Sanpietro, “A simplified Algorithm for Leaky Network Analyzer Calibration,” IEEE Microwave and
Guided Wave Letters, Vol. 5, No. 4, April 1995, pp 119-121.
Switch Terms
• Roger B. Marks, “Formulations of the Basic Vector Network Analyzer Error Model Including Switch Terms,” 50th ARFTG
Conference Digest, Fall 1997, pp 115-126.
Overview Papers – Available upon Request
• Douglas K. Rytting, “An Analysis of Vector Measurement Accuracy Enhancement Techniques,” Hewlett Packard RF &
Microwave Measurement Symposium and Exhibition. Includes detailed appendix. March 1982.
• Douglas K. Rytting, “Advances in Microwave Error Correction Techniques,” Hewlett Packard RF & Microwave Symposium,
June, 1987.
• Douglas K. Rytting, “Improved RF Hardware and Calibration Methods,” Hewlett Packard detailed internal training
document later released to the public.
• Douglas K. Rytting, “Network Analyzer Error Models and Calibration Methods,” 62nd ARFTG Conference Short Course
Notes, December 2-5, 2003, Boulder, CO.

51