You are on page 1of 7

Residual error models for the SOLT and SOLR

VNA calibration algorithms


J. Stenarson and K. Yhland
SP Technical Research Institute of Sweden
Box 857, 501 15 BonIs, Sweden
email: jorgen.stenarson@sp.se

Abstract
Uncertainty calculation of vector network analyzers (VNAs) using the SOLT or SOLR calibration algorithms is often performed
using residual directivity, match and tracking. In the literature the uncertainty equations are often stated without a derivation from
a proper model equation. In this paper we derive the model equations for both the SOLT and SOLR calibration, the two cases
do not result in the same model equation. The results are also compared to the commonly used expressions for uncertainty in
the EA guidelines for VNA evaluation. For one-port measurements our results confirm the expressions in the EA guide but for
two-ports there are significant differences. The symbolically derived model equations are verified using numerical simulations.

I. INTRODUCTION
Many calibration laboratories use the vector network analyzer (VNA) in their calibration of microwave impedance, attenuation
and S-parameters. Yet the treatment of the uncertainties in such calibrations is still unsatisfying.
There are basically two methods available for the uncertainty calculation of VNA calibrations. Either propagation of all
uncertainty contributions from their origin [1-3] or a black box treatment of the VNA [4].
The first method requires an extensive knowledge of the VNA and the calibration algorithm used as well as uncertainties
associated with the calibration standards. For the SOLT algorithm [5] an analysis based on the uncertainties in the calibration
standards is presented in [3]. However, since knowledge of calibration standard uncertainties is often unavailable, the second
method is used by many calibration laboratories. The second method is described in [4] (the EA guide) where the calibrated
VNA is evaluated as a black box using a set of air line reference standards. The evaluation results in a set of residual errors
associated with that specific combination of VNA, calibration standards and calibration algorithm.
However the EA guide essentially presents the final uncertainty equation without basing it on a model equation relating
the true value to the measured value as recommended in the guide to uncertainty in measurement (GUM) [6]. This has the
unfortunate consequence that it is difficult to determine for which calibration algorithms it is correct. To further complicate
matters the EA guide uses linear scale for reflection measurements and dB for transmission, with phase and complex valued
quantities only alluded to. The EA guide is vague on how the residual errors enter the measurements. Two situations can be
thought of, the residual errors are properties of the short open and load standards used in the SOLT and SOLR calibration
algorithms, or they are properties of the calibrated VNA. We have found this vagueness and lack of rigor in the presentation
of the EA guide to be unsatisfactory. We believe this method could benefit from a more complete study.
In this paper we present the model equations for the SOLT and the SOLR [7] VNA calibration algorithms. Furthermore we
present the series expanded versions that can be used to propagate the residual errors to an uncertainty in the device under
test (OUT). Our analysis shows that the uncertainty equations for reflection coefficient in the EA guide are identical to those
we derive for the SOLR algorithm using a 7-term model, but differ for SOLT. For transmission the differences are pronounced
for both SOLR and SOLT.
The analysis is based on the assumption that the residual errors are only caused by imperfectly known calibration standards
and not by poor directivity, match and tracking in the VNA hardware. Numerical simulations indicate that hardware performance
does not invalidate the expressions for SOLR, but further investigations are needed for SOLT.

II. VNA MODELS AND RESIDUAL ERROR MODEL


To allow for the propagation of the residual errors through the VNA calibration and error correction we need a model of
the VNA and the residual error terms.

A. VNA models for SOLT and SOLR


The SOLT calibration algorithm is designed for a three-sampler VNA architecture and thus the error model contains one
signal flow graph for the forward and one for the backward excitation, see Fig. 1. Since we use an ideal thru standard, the
estimation of ELF and E TF will only be affected by the SOL standards connected to port 1. Likewise, E LR and E TR will only

0-7803-9763-0/07/$20.00 ©2007 IEEE

Authorized licensed use limited to: MINCYT. Downloaded on April 25,2022 at 18:48:15 UTC from IEEE Xplore. Restrictions apply.
Fig. 1. Signal flow graph for the ten term VNA model used with the SOLT algorithm. E-terrns represent the VNA and S-terms represent the measured
device.

be affected by the SOL standards connected to port 2. In essence we get different system impedance definitions for the two
excitations, instead of for each port which is more natural.
The SOLR calibration algorithm is designed for a four-sampler VNA architecture and is based on an error-box formulation,
i.e, cascaded two-ports, and has only one signal flow graph , see Fig . 2. This method also assumes that the raw data used has
been switch corrected. See [8] for a description of switch correction. The unique property of this algorithm is the assumption
that the thru standard is reciprocal. Thus, it is possible to use non-mateable interfaces like male-male of the same connector
type, coaxial to waveguide, or different types of coaxial connectors.

ElO

E oo Ell

EOl

Fig. 2. Signal flow graph for the seven term VNA model used with the SOLR algorithm. E-terms represent the VNA and S-terms represent the measured
device.

The solution of the calibration problem, general properties of the two models and the relations between the two models is
found in [7-9].

B. Residual error model


Corrected measurements taken using a calibrated VNA show residual errors. The residual errors are mainly dependent on
errors in the knowledge of the calibration standards together with calibration algorithm. Therefore we will derive the model
equation under the assumption of perfect VNA hardware, we also do numerical simulations to verify the validity of the
equations for non-ideal hardware.
We have chosen to associate the residual directivity, tracking, and match (D, T and M) with the one-port calibration standards
short , open and load. A two-port containing D, T and M is cascaded in front of each standard. A signal flow graph of such
a connection is shown in Fig. 3, where f s , fa and f L are the true reflection coefficients of the standards. The reflection
coefficient at the calibration plane is the nominal value used for that standard when doing a calibration. It turns out that this
approach leads to the same uncertainty equation for one-port measurements as (2) in [4].

III. DERIVATION OF MODEL EQUATIONS


According to the method of determining uncertainties in GUM we need an expression that relates the true value of the
measurand to all relevant quantities. We derive the model equations as a function of the estimated S-parameters and the
residual errors . The series expansion of the model equation can then be used to obtain the uncertainty equation. Usually this
is done only to the first order.

Authorized licensed use limited to: MINCYT. Downloaded on April 25,2022 at 18:48:15 UTC from IEEE Xplore. Restrictions apply.
Refer en ce plan es 1
I
I
I I

I
I I

~ St andards ¢? D M
r s,r a ,r L
I I
I I
I I
I I

Calibration True 1+T

Fig. 3. Uncertainty model of one port standard s, and signal flow graph of standard plus residual errors

2 3 4 5

DUTraw
TrueDUT measured data
Data, Si;

Corrected Sij True Sij


as function of as function of
TrueVNA
error model, Eij true Sij and DTM corrected Sij and DTM
Standards raw
measured data sij=g(Sij,DTM) Sij=fl:Sij,DTM)

True
Standards

Residual errors
D, T,and M

Fig. 4. Flow chart for the propagation of the residual errors through the calibration algorithm and the error correction

In order to get a reasonable amount of equations in this paper we only give the computational flow in the list below and
in Fig. 4. The computations were performed using the Mathematica software. Note that all computations are symbolic. The
numbers in the list correspond to the column numbering in Fig. 4, the alphabetic enumeration indices are implicit from top to
bottom.
The true values of the OUT and VNA error model are given in capital letters and the error corrected S-parameters are given
in lower-case, the estimated VNA error model is also given in lower-case.
l) Assumptions
a) The true S-parameters of the OUT are 8 11 , 8 12 , 8 2 1 and 8 22
b) The true values of the VNA error terms E ij in Figs . land 2 are unity for the transmission terms and zero for the
reflection terms, i.e, perfect hardware.
c) The true S-parameters of the calibration standards are r s.i . r 0 ,1 and I'L.I at port land r S,2 , I' 0 ,2 and I'L.2 at port 2.
For SOLT we assume unity for the transmission of the thru connection and zero for its reflection. For SOLR we
assume a reciprocal thru.
d) The residual error terms are Db T 1 and M 1 for the standards at port land D 2 , T 2 and M 2 for the standards at
port 2.
2) Calculations
a) Derive symbolic expressions for raw measurement data of the OUT in terms of E ij and 8 ij .
b) Derive symbolic expressions for raw measurement data of the standards in terms of E ij and I' x-
c) Derive symbolic expressions for the calkit values.
3) Solve the SOLT or SOLR calibration equations to obtain an estimated calset e ij of Fig. 1 or 2.
4) Use the estimated e-parameters to perform error correction of the raw measurement data for the OUT. This gives the
estimated S-parameters of the OUT as a function of 0, T, M and its true S-parameters.
5) Solve the previous equation to get the true S-parameters, 8 ij , of the OUT as a function of 0, T, M and its estimated
S-parameters, Sij'

A. SOLT model equation


Using the procedure outlined above, the model equation that shows the true S-parameters as a function of the estimated
S-parameters and residual errors for a SOLT calibrated VNA becomes

Authorized licensed use limited to: MINCYT. Downloaded on April 25,2022 at 18:48:15 UTC from IEEE Xplore. Restrictions apply.
8 11 =
Sll,T - D 1(1 + ~ST) + S22,T D r (1)
~T,a

8 12
- S12,T(1 + T2 ) (2)
-
~T,b
8 = 821,T(1 + T 1 )
21 (3)
~T,a
S22,T - D 2(1 + ~ST) + Sll,TD~
S22 = (4)
~T,b
where index T indicates data obtained after SOLT calibration and error correction and

~T,a = (1 - D1S22,T)(~a + M 1 s 11,T) + D1M1S12,TS21,T (5)


~T,b = (1 - D2S11,T)(~b + M 2 s 22,T) + D2M2S12,TS21,T (6)
~ST = Sll,T S22,T - S12,T S21,T (7)
~a = 1+T 1 - D 1M1 (8)
~b = 1+T2 - D 2M2 (9)

The uncertainty equation is based on series expanded versions of the model equations (1)-(4), here we provide the expressions
to second order for 8 11 , and 8 12
8 11 + (S12,T S21,T - 1)D1 - SI1,T M1 - Sll,TT1
=Sll,T

+ D 1T1 + 2S 11,TD1M1 + Sr1,T M l - 2S11,TS12,TS21,TD1M1


+ S12,T S21,T S22,TDI + 2SI1,T M1 T1 - S12,T S21,TD 1T1 + Sll,TTr (10)
8 12 =S12,T(1 + Sll,TD2 - S22,TM2 + SI1,TD~ + (1 - Sll,T S22,T - S12,T S21,T)D 2M2 +
+ S22,TM2T2 + S~2,TM?) (11)

8 21 and 8 22 can be obtained by index permutation.


Depending on the specific values of the DUT and the residual error terms, higher order terms in (10)-(11) can be omitted. To
obtain an uncertainty calculation, uncertainties have to be assigned to the residual error terms. Using the approach prescribed
by GUM [6], for complex valued quantities see [10, 11], we need to compute the sensitivity coefficients to each uncertainty.
These sensitivities can be found by inspection of (10)-(11).

B. SOLR model equation


Using the procedure outlined above the model equation that shows the true S-parameters as a function of the estimated
S-parameters and residual errors for a SOLR calibrated VNA becomes

8 11 =
(Sll,R - D1)(~b + M 2 s 22,R) - M 2 s 12,R S21,R
(12)
~R,a
8
12
= 812,R JI+'Tlv'1+"T2 (13)
~R,a
821 = 821,R JI+'Tlv'1+"T2 (14)
~R,a

8 22 =
(S22,R - D2)(~a +M 1 s 11,R) - M 1 s 12,R S21,R
(15)
~R,a
where index R indicates data obtained after SOLR calibration and error correction and
(16)

Authorized licensed use limited to: MINCYT. Downloaded on April 25,2022 at 18:48:15 UTC from IEEE Xplore. Restrictions apply.
The series expansion of (12)-(15) becomes

8 11 =S11 ,R - D 1 - SIl ,R Ml - S11 ,RT1 - S I 2,R S 21,R M 2 +

+ D 1T1 + 2S 11,RD1M1 + S11 ,R M; + 2S11 ,R S12 ,RS21 ,RMIM2 + S12 ,R S21 ,R S22 ,RM?+
+ 2 sI1 ,R M1 T1 + S12 ,RS21 ,RM2T1 + S11 ,RT; + S12 ,RS21 ,RM2T2 (17)
T1 + T 2
- S11 ,RM1 - S22 ,RM2 + D 1M1 + D 2M2 + S11 ,R MI + 8 M +
2 2 2 2
8 12 =S12 ,R 1 -
( 2 22 2

M M 8 11 M 1 (3T1 + T 2) 8 22 M 2(T 1 + 3T2)


+ ( S11,R S22 ,R + S12 ,R S21 ,R ) 1 2 + 2 + 2 +
3Tr T 1T 2 3T? )
+8+-4-+8 (18)

821 and 822 can be obtained by index permutation. The uncertainty equations can be obtained using the same procedure as
outlined for SOLT.

C. Comparison between the SOLT and SOLR model equations


By comparing the model equations for the SOLT model (10)-(11) and the SOLR model (17)-( 18), we see the difference in
the uncertainty between SOLT and SOLR. For 8 11 of a SOLT measurement we see that the residual errors only depend on
the standards associated with port 1. This is caused by the separated VNA model where only the standards on port 1 are used
for the forward calibration and vice versa.
For transmission measurements the main difference is that the tracking term is not present to first order for SOLT but for
SOLR. This is because for the SOLT calibration the thru measurement is used to set the response for transmission measurements,
but in SOLR the short and open are used to set the response of both transmission and reflection measurements. In Fig. 5 we
show the deviation from an ideal thru in transmission of a SOLR calibration when using TRL characterized short and open
standards vs. using the data provided with the calkit by the manufacturer, in both cases the same TRL characterized load was
used. For a SOLT calibration we would get unity transmission in both cases.

- TRL characterized stds


- Manufacturer standard definitions
0.004

...... 0.003
I
tr5~
0.002

0.00 1

0.000 ~~.~~~~~~~~~:!.1
a 2 4 6 8 10 12 14 16 18
Frequency [GHz]

Fig. 5. Deviation from ideality for transmission measurements of a thru connection for SOLR calibration using TRL characterized standards and manufacturer
standard definitions.

D. Comparison to the EA guide


When considering only first order residual errors the SOLT and SOLR model equations both agree with the EA guide for
8 11 and 8 22 measurements on one-port DUTs or on two-ports with low transmission
However, for a two-port with high transmission the match of the other VNA port becomes important. Then we can see that,
to the first order, the expression for SOLR (17) is the same as (5) in [4]. This is not the case for SOLT (10).
In the EA guide only one second order term is included for reflection coefficient, it matches 2S11 ,RS12,RS21 ,RMIM2 for
SOLR whereas it does not agree for SOLT. In the EA guide reciprocity of the DUT is assumed since all expression use either
Sr2 or S~l instead of S12S21 as it should be.
For transmission, a mismatch term (m x ) is used in the EA guide instead of uncertainties from the residual match terms .
This mismatch term has been copied from the use in attenuation systems where a ratio is formed between the transmission
terms for the thru measurement and the DUT measurement. This mismatch term is then used to obtain a worst case error by
taking the modulus of each term

Authorized licensed use limited to: MINCYT. Downloaded on April 25,2022 at 18:48:15 UTC from IEEE Xplore. Restrictions apply.
1 + IM1 s 11 + IM2 s 22 + IM1M2S12S211 + IM1M2S11S221
mEA =- ---------------------
1 1
(19)
1-IM1M21

The use of the mismatch term is a consequence of trying to obtain expressions for the uncertainty not by a formalized
approached based on how the measurement procedure is done but rather on trying to create the equations in an ad hoc way.
Looking at (11) and (18) we can see that the expressions share similarity to the EA version but there are differences. In
particular the directivity is important for SOLT due to the thru calibration that transports the directivity to the opposite ports
residual match, there are also terms associated with residual tracking which are ignored. The denominator in (19) is not present
at all in either SOLT or SOLR.

E. Verification of model equations


The model equations have been verified numerically. Again we have used the computational flow chart in Fig. 4 but this
time only the first four steps. The numerical values for the error terms have been taken from actual calibrations of a network
analyzer with Type-N connectors or randomly with the distributions given below.
1) Numerical assumptions
a) The true S-parameters of the DDT 8 11 , and 8 22 had a magnitude of 0.04 , 8 12 = 8 21 and a magnitude of 1 and
they all had random phase.
b) The true values of E RF, E TF, E TR, ERR, E 0 1 , E 10 , E 23 , and E 32 were either nominally one with a random phase
plus an added complex perturbation with a complex standard deviation of 0.05 or directly taken from the results
of a VNA calibration.
The true values of E DF, E SF, ELF, E LR, E SR, E DR, E oo, Ell, E 22 , and E 33 either had zero mean and a complex
standard deviation of 0.05 or were directly taken from the results of a VNA calibration.
c) As true values for fS,l, fO,l, fL,l, f S,2, f O,2 and f L,2 we used the definitions from the calkit used to obtain the
calibration terms. For SOLT we assume unity for the transmission of the thru connection and zero for its the
reflection. For SOLR we assume a reciprocal thru.
d) The residual error terms D 1 , T 1 , M 1, D 2 , T 2 and M 2 had a zero mean and a complex standard deviation of 0.01.
2) Numerical calculations
a) Simulate raw measurement data for the DUT.
b) Simulate raw measurement data for the standards.
c) Calculate the calkit values.
3) Calculate an estimated calset eij of Fig. 1 or 2.
4) Use the estimated e-parameters to perform error correction of the raw measurement data for the DUT. This gives the
estimated DUT S-parameters Sij.
The estimated DUT S-parameters Sij and the values of the residual error terms D 1 , T 1 , M 1, D 2 , T 2 and M 2 are entered
in (1)-(4) for SOLT and in (12)-(15) for SOLR. The results should be the true S-parameters of the DUT and they are thus
compared to the DUT S-parameters entered in step la).
For SOLR the comparison was successful to within the numerical precision of the computer for both the random error terms
and the error terms from a real VNA. For SOLT it was successful to within 10- 7 for 8 11 and 8 22 in both cases and 10- 2 for
8 12 and 8 21 when using simulated error terms and 10- 3 when using error terms from a real VNA calibration. When setting
the VNA hardware terms for ELF and E LR to zero and repeating the process we get perfect agreement within the numerical
precision of the computer for both sets of error terms. The non-zero ELF and E L R is not a problem for 8 11 and 8 22 where
the error is neglible for practical uncertainties, but for 8 12 and 8 2 1 it will cause non-reciprocity. The numerical simulations
indicate that the non-recprocity, 8 12 - 8 21 , is approximately of the same size as the errors caused by ELF and E LR and could
thus be used to obtain an estimate of the uncertainty contribution necessary to cover the error.
This problem is also present when using the uncertainty equations of the EA guide, since it does not take the raw performance
into account either.

IV. DISCUSSION
Non-ideal behaviour of the thru connection has not been considered in this paper. There can be discontinuities in the thru
connection which will not be the same when measuring a DUT. Especially in the case of SOLT calibration these will have
an impact on the calibration. In the case of SOLR calibration such discontinuities will not be a problem as long as they are
reciprocal.
The model equations were derived assuming perfect hardware, numerical simulations have been performed to verify the
accuracy of the equations for non-ideal hardware. These simulations show that the propagation of the residual error terms
through the SOLT calibration is sensitive to nonideal values of ELF and E L R, this needs further study. The discrepancy is of
the same order as the non-reciprocity in the measurements (assuming reciprocal DUT), thus this non-reciprocity can be used

Authorized licensed use limited to: MINCYT. Downloaded on April 25,2022 at 18:48:15 UTC from IEEE Xplore. Restrictions apply.
as an uncertainty contribution. The perfect numerical agreement for SOLR should not be taken as a statement that SOLR
calibrations are completely independent of VNA hardware, but it indicates more robustness than the SOLT algorithm has. The
simulation of the measurement process used in the verification is not complete enough to uncover all problems associated with
imperfect hardware.

V. CONCLUSION
We have presented model equations for use in determining uncertainty in VNA measurements when doing SOLT and SOLR
calibration. The model equations were derived assuming perfect hardware and with the residual errors associated with the
calibration standards. The residual error terms associated with one-port calibration standards propagate differently through the
SOLT and SOLR calibration algorithms. For one-port measurements the results confirm the expressions in the EA guide but
for two-ports there are significant differences. Thus the uncertainty equations in the EA guide need to be modified depending
on which calibration algorithm is being used.
More work is needed to establish more complete and valid models when VNA hardware is imperfect, this is increasingly
important at higher freqeuncies where hardware performance decreases. We believe it is necessary to do so considering the
calibration algorithm in use and not just assume all calibrated VNAs are the same regardless of which calibration method was
used.

ACKNOWLEDGMENT
The authors thank the Swedish metrology council for financial support.

REFERENCES
[1] B. D. Hall, "Calculating measurement uncertainty for complex-valued quantities," Measurement Science and Technology, vol. 14, no. 3, p. 368, 2003.
[2] N. Ridler and M. Salter, "A generalised approach to the propagation of uncertainty in complex S-parameter measurements," in 64th ARFTG Microwave
Measurements Conference, Fall 2004, pp. 1-14, 2004.
[3] U. Stumper, "Influence of TMSO calibration standards uncertainties on VNA S-parameter measurements," Instrumentation and Measurement, IEEE
Transactions on, vol. 52, no. 2, pp. 311-315, 2003.
[4] "Guidelines onthe evaluation of vector network analysers (VNA)," Tech. Rep. EA-10/12 (rev.OO), European co-operation for Accreditation, 2000.
[5] Agilent, "Applying error correction to network analyzer measurement," Tech. Rep. AN 1287-3, 2002.
[6] ISO, "Guide to the expression of uncertainty in measurement," tech. rep., International organization for standardization, 1993.
[7] A. Ferrero and U. Pisani, "Two-port network analyzer calibration using an unknown "thru"," Microwave and Guided Wave Letters, pp. 505-507, 1992.
[8] R. B. Marks, "Formulations of the basic vector network analyzer error model including switch terms," in 50th ARFTG CONFERENCE DIGEST, vol. 50,
IEEE, 1997.
[9] J. Stenarson and K. Yhland, "Automatic root selection for the unknown thru algorithm," in ARFTG, (San Francisco), pp. 150-155, ARFTG, 2006.
[10] N. M. Ridler and M. J. Salter, "An approach to the treatment of uncertainty in complex S-parameter measurements," Metrologia, vol. 39, pp. 295-302,
2002.
[11] K. Yhland and J. Stenarson, "A simplified treatment of uncertainties in complex quantities," in Conference on Precision Electromagnetic Measurements,
vol. 1, (London), pp. 652-653, IEEE, 2004.

Authorized licensed use limited to: MINCYT. Downloaded on April 25,2022 at 18:48:15 UTC from IEEE Xplore. Restrictions apply.

You might also like