Вы находитесь на странице: 1из 7

Correction for Line Length Errors and

Center-Conductor-Gap Variation in the Coaxial


Multiline Through-Reect-Line Calibration
Arkadiusz Lewandowski, Member, IEEE, and Wojciech Wiatr, Member, IEEE
AbstractWe present an extension of the coaxial
multiline through-reect-line calibration method which
allows to correct for line length errors and centerconductor-gap variation. Our approach is based on
the multi-frequency formulation of the vector-networkanalyzer calibration problem which accounts for the
physical relationships between calibration standard Sparameters at dierent frequencies. We illustrate our
approach with experimental results for the coaxial
multiline through-reect-line calibration in the 7 mm
standard. We show that our calibration method yields
an improved measurement accuracy as compared with
the classical multiline through-reect-line calibration
method.
Index TermsVNA, coaxial multiline TRL calibration, multi-frequency VNA calibration, line length errors, center-conductor-gap variation

I. Introduction

HE multiline through-reect-line (TRL) method [1]


is commonly considered as the most accurate technique for the vector-network-analyzer (VNA) calibration.
The superior accuracy of this technique stems from two
facts: the use of redundant calibration standards which
reduces the uncertainty due to the VNA instrumentation
errors (i.e. connector nonrepeatability, cable exure, testset drift, and noise), and the foundation of the calibration
standard denitions solely on the dimensional measurements (line lengths) and some qualitative requirements
(uniformity of the lines, identical cross-sections of the lines
and identical reection coecients of the reects).
As the uncertainty due the VNA instrumentation errors can be reduced by increasing the number of calibration standards, the ultimate accuracy limitation in
the multiline TRL calibration results from errors in the
calibration standard denitions. These errors are caused
by the inaccuracy of the calibration standard parameters
(i.e. errors in line length determination) and the violation
of the requirements put on the calibration standard definitions (i.e. dierences in the characteristic impedance
and propagation constant of the lines, dierences in the
reection coecients of the reects, and nonrepeatability
of the connector interface).

Arkadiusz Lewandowski and Wojciech Wiatr are with the Institute of Electronic Systems, Warsaw University of Technology,
Nowowiejska 15/19, 00-665 Warsaw, Poland (phone: +48 22 234
7877; e-mail: A.Lewandowski@ieee.org, W.Wiatr@ise.pw.edu.pl).

978-1-4244-5714-4/10/$26.00 2009 IEEE

The errors in line lengths stem from the inevitable


dimensional measurement uncertainty and from the dierence between the measured length and the actual length
of the line when used in the measurement environment.
In the case of coaxial air-lines this dierence may be
attributed to the eects such as the thermal expansion
and contraction, or the mechanical compression of the
line outer conductor due to the torque applied to the
connectors [2].
The nonrepeatability of the coaxial connector interface
has two sources. In the case of coaxial devices with a xed
center conductor (direct through, reect) it is primarily
attributed to the displacement of the outer and inner
conductors, bending of the connector socket ngers and
variation of the contact resistance [3, 4]. We commonly
refer to the random errors caused by those phenomena as
the connector repeatability errors and classify them as the
VNA instrumentation errors.
In the case of coaxial air-lines which have an unsupported center conductor, the random variation of the
center conductor gap at the connector mating planes
becomes an additional source of the nonrepeatability of the
connector interface. This random variation is commonly
considered as the most important error source in the coaxial multiline TRL calibration, especially for millimeterwave coaxial connectors, such as the 1.85 mm or 1.0 mm
connector.
Eorts to reduce the uncertainty due to the errors in
calibration standard denitions focus primarily on the
development of precise dimensional characterization techniques for the transmission lines and the minimization
of the possible center-conductor-gap variation through a
tight control of the test-port pin-depth. Much research has
also been done on the modeling of the electrical properties
of the center-conductor gap [58].
In this work we present an extension of the multiline
TRL calibration method [1] which allows to correct for
errors in the determination of line lengths and the random
variation of the center conductor gap. Our method can be
thought of as a generalization of the self-calibration approach [9], employed in the classical TRL method [10, 11]
or in the unknown thru method [12]. The principle
of the self-calibration approach is to use the physical
relationships between calibration standard S-parameters
at one frequency as an additional piece of information

zWV

_BSV

Figure 1.
calibration

V$Sj$t

zWV

_BS

Model of the coaxial transmission line used in the

in the VNA calibration. In our approach we generalize


this principle and exploit also the physical relationships
between calibration standard S-parameters at dierent
frequencies. To this end, we model these relationships in
terms of the unknown line length corrections and the
center-conductor-gap widths. We then solve the VNA
calibration problem jointly at all measurement frequencies
which allows us to determine the unknown line length
corrections and the center-conductor-gap widths along
with the VNA calibration coecients. A similar approach
has already been applied in the sliding-load calibration
[13] and in the oset-short calibration with unknown oset
lengths [14].
In order to assure the uniqueness of the solution to the
VNA calibration problem (i.e. reference plane position,
propagation constant) we further impose some additional
constraints on the parameters identied in the VNA calibration. These constraints are derived from some intuitive
assumptions as to the statistical properties of the line
length errors and center-conductor-gap variation.
II. Coaxial transmission-line model
The model of a coaxial transmission line we use in
the calibration is shown in Fig. 1. Following the classical
multiline TRL method, we assume the all of the lines have
the same but otherwise unknown characteristic impedance
and propagation constant . We further assume that the
actual lengths of the lines ln , for n = 1, . . . , N , are
unknown and specify each one of them as the sum of
the known nominal length l0n and the unknown length
correction ln . These corrections are then determined
during the VNA calibration.
We model the electrical properties of the center conductor gap with a frequency-independent series inductance.
We rely here on the models of [5, 6] and the results of
electromagnetic simulations of [7]. We do not account for
the resonant behavior for small gaps (see [8]), as we think
that this behavior can be avoided by a skilled operator.
We split either gap into the part that belongs to the
calibration standard and to the VNA. We account for the
part that belongs to the calibration standard with the
series inductances LAn and LBn attached to both line

ends. These inductances are unknown and are determined


during the VNA calibration.
The part of the center conductor gap that belongs to
the VNA is lumped into the VNA calibration coecients.
Hence, the calibration reference plane coincides with the
outer-conductor joint plane.
To summarize, for a set of N transmission lines we have
two sets of unknown parameters: propagation constant values k at all measurement frequencies fk , for k = 1, . . . , K,
and the vector
T

= [l1 , . . . , lN , LA1 , . . . , LAN , LB1 , . . . , LBn ] , (1)


comprising the frequency-independent line-length corrections and center-conductor-gap inductances, where the
superscript T denotes the transpose.
III. Calibration algorithm
Our calibration algorithm is based on the formulation of
the VNA calibration as the maximum-likelihood systemidentication problem. In the following we rst review
the maximum-likelihood approach to the single-frequency
VNA calibration. We then introduce the multi-frequency
approach to the VNA calibration problem and show how
it can be applied to determine the line length corrections
and center-conductor-gap inductances comprised in the
vector (1).
A. Single-frequency approach
Let the vectors s and s denote the raw S-parameter measurement of a calibration standard and its S-parameter
denition at a xed frequency, respectively. We dene s
and s as real-valued vectors formed out of the real and
imaginary parts of the corresponding S-parameters. We
can write the relationship between these two vectors as
s = f (s, p) ,

(2)

where the vector p denotes the unknown VNA calibration


coecients and the function f characterizes the VNA
model used.
Let further the calibration standard denition s be
dependent on some unknown parameters, such as the
propagation constant of the line standard or the reection
coecient of the reect standard in the multiline TRL
calibration. That is,
s = g (c0 , c) ,

(3)

where c0 and c are the known and unknown parameters


of the calibration standard, respectively. Assuming that
the errors in raw measurements are negligible, we can now
describe the measurement of N calibration standards as
1

f (s1 , p) = g1 (c01 , c1 ) + s1
..
,
(4)
.

1
f (sN , p) = gN (c0N , cN ) + sN

where sn , for n = 1, . . . , N denotes the error in the measurement of the n-th standard resulting from the connector
nonrepeatability, cable exure, and test-set drift.
Our objective now is to nd the vector of unknown
parameters

p
c1

(5)
= . ,
.
.
cN
based on the known parameters c0n of the calibration
standards and the raw measurements sn . To this end, we
rst need to make some assumptions as to the statistical
properties of the measurement errors sn . We dene the
joint measurement error as

s1

s = ... ,
(6)
sN
and assume that it is normally distributed with E (s) =
0, where E() is the expectation value operator. We further
write its covariance matrix as


(7)
s = E ssT = 2 V,
where the matrix V is known and 2 is the unknown residual variance (square of the residual standard deviation), to
be determined in the estimation procedure. With the use
of these denitions we can write the likelihood function


(8)
L , 2 = p (r ()) ,
where p (x) is the probability density function (PDF) of
the measurement error s, and

f 1 (s1 , p) g1 (c01 , c1 )

..
r () =
(9)
,
.
f

(sN , p) gN (c0N , cN )

are the residuals. We now obtain the unknown parameters


and 2 by maximizing the likelihood function (8). This
leads to the classical nonlinear least squares problem with
the solution [15]
T

=arg
min r () V1 r () ,

and

(10)

1 T 1
r() V r(),
(11)

where is the number of degrees of freedom, that is the


number of elements in the vector r () less the number of
parameters in the vector .
The estimate
of the residual standard deviation is
a convenient gure-of-merit quantifying the quality of
the calibration. It is a single number showing how well
the model of the measurement procedure we use (i.e.
the calibration standard denitions and the VNA model)
can reproduce the actual measurements of the calibration
standards.

2 =

B. Multi-frequency approach
In the multi-frequency approach to the VNA calibration
we rst divide the unknown parameters into two groups.
The rst group consists of the frequency dependent parameters of the calibration standards and the VNA calibration
coecients. For the set of frequncies fk , for k = 1, . . . , K,
we denote this group with the vectors 1 , . . . , K . The
second group contains the frequency independent parameters of the calibration standards and is represented by
the vector . For the multiline TRL calibration we consider here, this vector is dened by (1). With the use of
these denitions we write the likelihood function at the
frequency fk as


Lk k , , k2 = p (rk (k , )) ,
(12)
where the residuals rk (k , ) are dened similarly to (9).
Now, since the frequency-independent parameters in the
vector aect simultaneously the likelihood functions at
all frequencies, we can no longer solve the VNA calibration
problem independently at each frequency. Therefore we
write the multi-frequency likelihood function, that is
K




2
L 1 , . . . , K , , 12 , . . . , K
=
Lk k , , k2 (13)
k=1

We then estimate the parameters 1 , . . . , K , the correc2


,
tions and the unknown residual variances 12 , . . . , K
by maximizing this function. As the direct maximization is
infeasible due to the large number of unknown parameters,
we carefully exploit the structure of the optimization
problem so as to reduce its dimensionality. Details are
given elsewhere [16].
C. Identiability of the corrections
The line length corrections and center-conductor-gap
inductances captured in the vector cannot be uniquely
determined along with other parameters captured in the
vectors 1 , . . . , K . That is, there may exist more than one
solution to the problem of maximizing (13), yielding to
the same maximal value the likelihood function (13). For
example, for a given set of line length corrections ln , we
can add the same value l to all of the length corrections
to obtain another solution with the same value of the
likelihood function (13). Indeed, this additional length will
simply be compensated by the change of the propagation
constant which is captured as one of the elements in
the vectors 1 , . . . , K . A similar reasoning could also be
conducted for the center-conductor-gap widths.
In order to obtain a unique solution to the multifrequency VNA calibration problem we consider here, we
need therefore to impose some additional constraints on
the parameters captured in the vector . We can easily
derive such constraints from the statistical properties of
the corrections in the vector . Indeed, we expect the
mean value of line length errors to be zero. Also, it is
reasonable to assume that the mean center-conductor gaps

Constraints of such a form can easily be accounted for


in the maximization of (13) with the use of Lagrangian
multipliers or direct elimination [17].
Now, it is important to note that the constraints (14) are
met by the parameters in the vector in the probabilistic
sense, that is, we have E (C) = 0. Consequently, the
actual values of the parameters meet the constraint (14)
only approximately. However,
the error resulting from this
approximation is roughly by N smaller than the error
due to the use of erroneous nominal lengths of the lines,
and decreases as the number number of lines N increases.
IV. Experiments
We illustrate our approach with experimental results
for the coaxial multiline through-reect-line calibration
in the 7 mm standard. We used six in-house-made airlines, a direct thru connection, and an 1.27 mm long oset
short as the reect standard. We measured the calibration
standards in the frequency range 0.1 18 GHz with the
HP 8722D vector network analyzer.
In Fig. 2 we compare the residual standard deviation obtained from the classical multiline TRL [1] (solid blue) and
from our method, with only line length corrections (solid
green), and with both line length and center conductor
gap corrections (solid red). We see that performing the line
length corrections leads at all frequencies to the residual
standard deviation smaller by 25 50 % as compared with
the classical multiline TRL. Correction for the centerconductor gap variations provides a further improvement
in the residual standard deviation by attening out the
ripples which yields the values that are consistently at least
two times smaller then for the classical multiline TRL.
The sharp peaks in the residual standard deviation around
9.5 GHz and 11.5 GHz are due to the sampler-bounce eect
in the VNA receivers which signicantly deteriorates the
accuracy of raw VNA measurements.
In order to determine how the reduction of the residual
standard deviation improves the accuracy of the calibration, we performed the residual analysis (see [15]) and
determined the standard deviation of all of the estimated
parameters. To illustrate the results of this analysis, we
compare in Fig. 3 the eective source match and the
eective tracking for port one of the VNA, as obtained
from the classical multiline TRL calibration (solid blue
line) and from our method when determining only the
line length corrections (solid green), and when determining
both the line length corrections and the center-conductorgap inductances (solid red). We observe that correcting for
the line length errors leads to much smaller values of both
parameters as compared with the classical multiline TRL
method: the eective source match improves by 4 9 dB,

((x

iR$IZsjSRstIsISI%$s$Bt

on both line ends are equal. These two conditions lead to


a set of linear equality constraints which can be written
as
C = 0.
(14)

Zj$]VN
tbS8YBISwBtjkSjtYD
tbS8YBISwjtYSmSs}D

((;
((n
((1
((d
(

E
M d( d1 d;
3JZt{kS>p<9H

dE

dM

Figure 2. Residual standard deviation as obtained from the classical


multiline TRL (solid blue) [1], and from the new method with only
line length corrections (solid green), and with both line length and
center-conductor-gap corrections (solid red).
Table I
Nominal and corrected air-line lengths along with the
standard uncertainty from the residual analysis.
Length
Nominal
Corrected
[mm]
[mm]
6.09
6.103
13.95
13.961
18.80
18.817
49.07
49.059
55.26
55.254
74.98
74.957

Line
L1
L2
L3
L4
L5
L6

Standard
uncertainty
[m]
0.7
0.7
0.6
0.7
0.7
0.7

Table II
Inductances of the center conductor gap along with the
standard uncertainty from the residual analysis.

Line

LA

LB

L1
L2
L3
L4
L5
L6

[pH]
1.72
0.37
3.07
1.62
4.60
2.15

[pH]
2.23
0.08
3.21
1.03
5.73
0.51

Standard
uncertainty
[pH]
0.13
0.13
0.13
0.13
0.13
0.13

and the eective tracking by over 50 %, compared with


the classical multiline TRL method. We then see that
correcting for the center-conductor-gap variation leads to
a further reduction of both parameters by as much as 4 dB
in the case of the eective source match, and up to 60 %
in the case of the eective tracking. This improvement,
however, is not as signicant as the improvement resulting
from the correction for line length errors.
In Tab. I we compare the nominal lengths of the lines
with the corrected lengths obtained from our calibration
approach. The nominal lengths were measured with a
digital caliper with the uncertainty 30 m. The corrected
lengths dier from the nominal lengths by up to 23 m.
In Tab. I we also show the standard deviations of the

((1(

Zj$]VN
tbS8YBISwjtYD
tbS8YBISwjtYSmSs}D

Tx(

ePP{$%Ss{,$tS>I H

ePP{$%SRBZ{S8s{YS>I H

T;(

TE(

T(

TM(

((dE
((d1
(((M
(((;
(

E
M d( d1 d;
3JZt{kS>p<9H

dE

Zj$]VN
tbS8YBISwjtYD
tbS8YBISwjtYSmSs}D

dM

(a)

E
M d( d1 d;
3JZt{kS>p<9H

dE

dM

(b)

Figure 3. Eective source match (a) and eective tracking (b), as obtained from the classical multiline TRL (solid blue) [1], and from the
new method with only line length corrections (solid green), and with both line length and center-conductor-gap corrections (solid red).

(1x

d((E

Zj$]VN
tbS8YBISwjtYD
tbS8YBISwjtYSmSs}D

Zj$]VN
tbS8YBISwjtYD
tbS8YBISwjtYSmSs}D

(1(

d((;
(dx
(d(
d((1
((x
d
(

(
1

E
M d( d1 d;
3JZt{kS>p<9H

dE

dM

(a)

E
M d( d1 d;
3JZt{kS>p<9H

dE

dM

(b)

Figure 4. Propagation constant for the coaxial multiline TRL calibration in the 7 mm standard: (a) imaginary part normalized to the
free-space propagation constant 0 , (b) real part; as obtained from the classical multiline TRL (solid blue) [1], and from the new method
with only line length corrections (solid green), and with both line length and center-conductor-gap corrections (solid red).

length corrections determined in the residual analysis.


These standard deviations quantify the precision with
which we can determine the line length corrections1 , and
for all of the lines are less then 0.7 m, which indicates that
we can very precisely determine the line length corrections.
In Tab. II we put together the center-conductor-gap
inductances obtained from our method along with the
standard uncertainties determined in the residual analysis.
In most cases the precision with which we determine the
inductances is better than 10 %. For the 7 mm coaxial connector standard, the inductance of the gap is 0.136 pH/m
[5, 6]. Thus, the maximum gap inductance we obtain
corresponds to the center conductor gap of around 40 m.
In Fig. 4 we compare the real and imaginary part of
1 In order to determine the accuracy of these corrections, we would
have to account for the error of the constraints (14). This will not be
discussed here.

the propagation constant determined with the use of the


classical multiline TRL calibration (solid blue line), and
with the use of our method when determining only the
line length corrections (solid green), and when determining
both the line length corrections and the center-conductorgap inductances (solid red). We see that correcting for
the errors in the line lengths leads to a slightly dierent
estimate of both the real and imaginary part of the propagation constant, as compared with the classical multiline
TRL calibration. We further note that the correction
for the center-conductor-gap variation aects only the
imaginary part of the propagation constant. This can be
explained with the fact that the electrical character of
center conductor gap is primarily of the reactive nature
[6]. We also observe that with the use of our approach we
can better detect the sampler-bounce eect, manifesting
itself as the sharp peaks in the real part of the propagation

d((

Tn(

x(

T;(

(
Tx(
Tx(
TE(
T(
TM(

Zj$]VN
tbS8YBISwjtYD
tbS8YBISwjtYSmSs}D

Td((

Zj$]VN
tbS8YBISwjtYD
tbS8YBISwjtYSmSs}D

Tdx(
T1((

E
M d( d1 d;
3JZt{kS>p<9H

dE

dM

(a)

E
M d( d1 d;
3JZt{kS>p<9H

dE

dM

(b)

Figure 5. Corrected reection coecient of a matched termination: (a) magnitude, and (b) phase with the linear part removed, as obtained
from the classical multiline TRL (solid blue) [1], and from the new method with only line length corrections (solid green), and with both
line length and center-conductor-gap corrections (solid red).

constant. These peaks are not present in the result from


the classical multiline TRL calibration.
In Fig. 5 we compare magnitude and phase (with a
linear part removed) of the corrected reection coecient
of a matched termination, as determined with the classical multiline TRL calibration (solid blue line), and with
the use of our method when determining only the line
length corrections (solid green), and when determining
both the line length corrections and the center-conductorgap inductances (solid red). We see that the agreement
between all of the methods is very good. We expect the
uncertainty of our result to be smaller than in the classical
multiline TRL method due to smaller residual errors of our
approach.
V. Conclusions
We presented an extended coaxial multiline TRL calibration method which corrects for the errors in the
determination of line lengths and the variation of the
center conductor gap. Our method employs the multifrequency approach to the VNA calibration which, unlike the classical VNA calibration methods, allows to
account for the relationships between calibration standard
S-parameters at dierent frequencies. We showed that
including the information about those relationships in the
VNA calibration makes is possible to correct for the errors
in line lengths and for the random variation of the center
conductor gap in the coaxial air-lines. The experimental
results we presented for the coaxial multiline calibration
in the 7 mm standard show that our approach leads to
the signicant improvement of the calibration accuracy,
as compared with the classical multiline TRL method [1].
VI. Acknowledgment
This work has been supported by the grant N N505
360836 of the Polish Ministry of Science and Higher

Education.
References
[1] R. Marks, A multiline method of network analyzer
calibration, IEEE Trans. Microw. Theory Tech.,
vol. 39, no. 7, pp. 12051215, July 1991.
[2] M. Horibe, M. Shida, and K. Komiyama, Sparameters of standard airlines whose connector is
tightened with specied torque, IEEE Trans. Instrum. Meas., vol. 56, no. 2, pp. 401405, April 2007.
[3] J. Juroshek, A study of measurements of connector
repeatability using highly reecting loads (short paper), IEEE Trans. Microw. Theory Tech., vol. 35,
no. 4, pp. 457460, 1987.
[4] J. Miall and K. Lees, Modeling the repeatability of
Type-N connectors using Microwave Studio, in Proc.
19th ANAMET Meeting, 2003.
[5] T. E. MacKenzie and A. E. Sanderson, Some fundamental design principles for the development of
precision coaxial standards and components, IEEE
Trans. Microw. Theory Tech., vol. 14, no. 1, pp. 29
39, Jan 1966.
[6] W. C. Daywitt, A simple technique for investigating
defects in coaxial connectors, IEEE Trans. Microw.
Theory Tech., vol. 35, no. 4, pp. 460464, 1987.
[7] B. B. Szendrenyi, Eects of pin depth in LPC 3.5
mm, 2.4 mm, and 1.0 mm connectors, in Proc. IEEE
MTT-S International Microwave Symposium Digest,
vol. 3, 1116 June 2000, pp. 18591862.
[8] J. P. Homann, P. Leuchtmann, and R. Vahldieck,
Pin gap investigations for the 1.85 mm coaxial connector, in Proc. European Microwave Conference, 9
12 Oct. 2007, pp. 388391.
[9] H.-J. Eul and B. Schiek, A generalized theory and
new calibration procedures for network analyzer self-

[10]

[11]

[12]

[13]

[14]

[15]
[16]

[17]

calibration, IEEE Trans. Microw. Theory Tech.,


vol. 39, no. 4, pp. 724731, April 1991.
B. Bianco, M. Parodi, S. Ridella, and F. Selvaggi,
Launcher and microstrip characterization, IEEE
Trans. Instrum. Meas., vol. 25, no. 3, pp. 320323,
February 6 1976.
G. F. Engen, Calibrating the six-port reectometer
by means of sliding terminations, IEEE Trans. Microw. Theory Tech., vol. 26, no. 12, pp. 951957, Dec
1978.
A. Ferrero and U. Pisani, Two-port network analyzer
calibration using an unknown thru, IEEE Microw.
Guided Wave Lett., vol. 2, no. 12, pp. 505507, Dec.
1992.
G. Vandersteen, Y. Rolain, J. Schoukens, and A. Verschueren, An improved sliding-load calibration procedure using a semiparametric circle-tting procedure, IEEE Trans. Microw. Theory Tech., vol. 45,
no. 7, pp. 10271033, July 1997.
W. Wiatr and A. Lewandowski, Multiple reect
technique for wideband one-port VNA calibration,
in Proc. International Conference on Microwaves,
Radar & Wireless Communications MIKON 2006, 22
24 May 2006, pp. 3740.
Y. Bard, Nonlinear parameter estimation. Academic
Press, New York, 1974.
A. Lewandowski, Multi-frequency approch to
vector-network-analyzer
scattering-parameter
measurements,
Ph.D.
dissertation,
Warsaw
University of Technology, 2009.
R. Fletcher, Practical methods of optimization. Wiley, 2008.

Вам также может понравиться