Stray capacitance of a two-layer air-cored inductor

M.J. Hole and L.C. Appel

Abstract: A new analytic method for predicting the stray capacitance of a double layer coil with
conducting shield and air core is presented. The method involves construction of a two-layer turn-
to-turn circuit model, which includes the self-inductance and resistance of each turn, the mutual
inductance representing the flux linkage between all turns, and the capacitance between adjacent
turns and the capacitance between the outer turns and conducting shield. At high, low and zero
frequency the model is reduced to an effective capacitance, inductance, and resistance respectively:
for frequencies below the first self-resonant frequency the behaviour of the circuit is approximated
by an RLC circuit. A new result is a recursive calculation of the stray capacitance of the high
frequency circuit model. The method and stray capacitance value are compared to alternative
models and stray capacitance predictions, and the limitations of the different approaches discussed.
Comparisons with experimental measurements show good agreement at the fundamental self-
resonance and confirm that the effect of a conducting shield is likely to be small for a closely-
wound coil.

1 Introduction In this paper, a method for modelling a double-wound

coil with an air-core and conducting shield is presented. An
At high frequencies, the impedance of an inductor ceases to array of these coils (OMAHA) has been installed into the
be ideal and is dominated by parasitic capacitances. The Spherical Tokamak, MAST [6] for measuring magnetic
cumulative effect of these capacitances is commonly fluctuations up to 5 MHz. Indeed, magnetic fluctuation
described as a ‘stray’ capacitance, Cs . Of particular interest activity, believed to be compressional Alfven eigenmodes at
is the frequency at which the capacitive and inductive 3.5 MHz has already been detected using the OMAHA
impedances match, giving rise to a self-resonance, and array [7]. The function of the OMAHA coils favours a
providing an upper-frequency bound to operation as an double-wound winding configuration in order to maximise
inductor. magnetic pick-up away from the coil, with a large turns-
The magnitude of the stray capacitance is strongly density to maximise the coil inductance per unit length, and
dependent on the winding geometry and the proximity of whose first self-resonant frequency, f0 , is greater than
any conducting surfaces. Calculation of it can be carried out 5 MHz. The principal objective of this article is to determine
directly, by solving numerically the electrostatic field simple expressions for the stray capacitance. When
equations [1], or by distributed capacitance modelling, in combined with well-known expressions for the coil
which the inductor is represented by a network of inductance, these expressions provide an estimate of f0 ,
capacitance elements representing capacitances between and thus are a useful design tool. This work also
discrete conducting elements [2, 3]. Although the former complements a recent study by Appel and Hole [8], which
method is likely to be more accurate, particularly in the case provided a synopsis of coil in-situ remote calibration
of more complex geometries, the latter approach is techniques of Heeter et al. [9].
particularly useful for scoping studies. The model developed here builds principally upon the
In the past, measurements of the stray capacitance have work of Grandi et al. [2], in which a high frequency
been inferred from low frequency measurements of the equivalent circuit was used to generate expressions for the
inductance, Lcoil , coupled with measurements of the first stray capacitance of an n turn single layer air-cored
self-resonant frequency, f0 [2–4]. In instances where the solenoid. In their work, a distributed capacitance network
inductance is also frequency dependent (e.g., inductors with was used to determine an expression for the stray
a ferrite core), the addition of an external capacitor has been capacitance, valid at frequencies significantly above f0 ,
used to extract the inductance local to the self-resonant where the circuit model of the inductor retains only the
frequency [5]. effects of capacitive circuit components.
In this work, the single layer model is extended to provide
predictions for the stray capacitance of a two-layer solenoid,
thereby allowing comparison to the multiple layer dis-
r IEE, 2005
tributed capacitance model of Massarini and Kazimierczuk
IEE Proceedings online no. 20045217
[3]. Differences between the treatments are described in
doi:10.1049/ip-cds:20045217
Section 3. As in Massarini and Kazimierczuk, generation of
Paper first received 19th December 2004 and in final revised form 22nd July
2005
stray capacitance expressions for three or more layers is a
M.J. Hole is with the Research School of Physical Sciences and Engineering,
straight-forward generalisation of the two-layer result. The
Australian National University 0200, A.C.T. Australia advantage of distributed capacitance networks (e.g. Grandi
L.C. Appel is with EURATOM/UKAEA Fusion Association, Culham Science et al., Massarini and Kazimierczuk, this work), as compared
Centre, Abingdon, Oxon, OX14 3DB, UK to full numerical solutions of the electrostatic field equations
E-mail: [email protected] (e.g. Yu and Holmes [1]), lies in the model simplicity, which

IEE Proc.-Circuits Devices Syst., Vol. 152, No. 6, December 2005 565
affords a simple and fast computation of the the stray l
capacitance. Measurements of the circuit model parameters
have been obtained by a least squares fitting of the axial cut

impedance profile over the frequency range 0of o1:5f0 .

To our knowledge, these are the first published results of
both the predicted and measured stray capacitance of a
double layer solenoid. θ

2 Lumped-circuit model r
2R
Figure 1a is an illustration of the coil to be modelled. To z
describe the coil we introduce cylindrical co-ordinates
ðr; y; zÞ, with orientation as shown in Fig. 1a. The co-
ordinate origin has been placed at the terminal end of the
coil, and centred to the axis of the former. The non-
conducting former has radius R and length l, upon which
are layered two closely packed helical windings. In this shield
work the coil is taken to be sufficiently long such that the
fringing of the magnetic and electric fields near the end-
points of the coil may be ignored. In addition, the different a
winding pitch (the axial direction of each winding) between
inner and outer layers is ignored. Figure 1b is an axial
enlargement of four turns of Fig. 1a. As shown, the coil dielectric
wire has a metal diameter of 2a, and is coated with a thin
dielectric insulator of thickness ðd
2aÞ=2, where d is the
wire core
total diameter of the wire. Finally, Fig. 1c is a cross-section
of the coil, labelling the inner and outer layers.
In general, the electrical properties of a coil are governed
by both electrostatic and magnetostatic interactions, and a
lumped-circuit model must include both. Figure 2a is a
lumped circuit representation of the n turn coil, divided into
turn-to-turn elements. Such a discrete circuit model will be d 2a
valid providing the change in phase of a wave propagating
from any turn is small along its length. That is, shield
ð2pRÞ=l  1, with l the wavelength of excitation. The
b
circuit model includes capacitive couplings between the ith
and jth turns, the turn to shield coupling of the ith outer outer layer
turns, the self-inductance Lt;i and resistance Rt;i of each
turn, and the mutual-inductance M representing the flux- axial cut in shield
inner layer
linkage of each turn to all other n
1 turns. Only capacitive
couplings between adjacent turns have been retained: this
results from an extrapolation of the conclusions of Grandi
et al., who found that for a single layer solenoid the neglect
of capacitance between non-adjacent turns is reasonable
providing that jd=ð2aÞjo2. To simplify Fig. 2a, we next
neglect differences between the turn-to-turn capacitance Ctt
of the inner and outer layers (i.e. same z, different r) and
adjacent turns within the same layer (i.e. same r, different z).
We also neglect differences between the capacitance of
‘next’ nearest neighbours, which are henceforth labelled Ctt .
Finally, the variation in the turn-to-shield capacitance along former
the solenoid has been ignored. With these reductions, the shield
circuit model of Fig. 2a is reduced to Fig. 2b. In Section 3 c
we compute Ctt =Ctt ¼ 0:1 and jd=ð2aÞj ¼ 1:07 for MAST
coils, and so the neglect of capacitive coupling between Fig. 1 Illustration of a densely packed two layer inductor
more distant turns is not unreasonable. a Full geometry of the two layer windings and external shield
In principal, analysis of the circuit shown in Fig. 2 will b Magnification of four turns of the inductor, with a the wire radius
and d the diameter of the dielectric coated wire
yield the full circuit response including the first self-resonant
c Cross-section of the solenoid
frequency f0 , and higher resonances owing to reflected
waves coalescing with the same phase (i.e. transmission line
effects). However, assuming the primary purpose is to
establish f0 , a simpler approach is to model the coil inductors
characteristics in the two limits where the effect of the stray
m0 pn2r2
capacitance can be ignored (low frequency limit), and where Lcoil ¼ ð1Þ
the effect of the stray capacitance dominates (high l þ 0:9r
frequency limit). correct to within 1% providing that l=ð2rÞ40:4, where r
In the low frequency limit, Wheeler [10] developed a is the mean coil radius. The coil resistance computes as
simple approximate formula for circular cross-section Rcoil ¼ nRt ðoÞ, where o ¼ 2pf denotes a frequency
566 IEE Proc.-Circuits Devices Syst., Vol. 152, No. 6, December 2005
 R t,n L t,n M R t,n/2+1 L t,n/2+1 M M  ðn
1ÞLt . In Section 3 we show a posteriori that this
frequency is well above f0 . The coil characteristics over
Ct(n,n−1) both frequency limits is represented approximately by an
C t(1,n) Ct(n,2)
RLC circuit, comprising the series combination of a resistor
Ct(1,n−1) Ct(2,n−1) C t(n/2,n/2+1)
L t,1 R and coil inductance Lcoil , in parallel with the stray
R t,n/2 L t,n/2 M
R t,1
capacitance Cs .
M
As in previous treatments [2, 3], the effect of higher-order
Ct(1,2)
self-resonant phenomena on the value of stray capacitance,
C ts,1 C ts,2 C ts,n/2 when used (together with the low frequency inductance) to
predict the fundamental self-resonance, is neglected. Such
effects are outside the scope of this work, which is to
a
produce simple scoping design estimates for the stray
Rt Lt M capacitance. Qualitatively however, the presence of higher
Rt Lt M order resonances means that the measured fundamental
self-resonance f0 will be down-shifted from the predicted
Ctt
Ctt values computed in this work. To see this, we label
Ctt* Ctt* Ctt Ctt
Lt the successive resonant frequencies in the circuit f0 , f1 ,
Rt Lt M f2 , . . ., fN . In a frequency range local to each resonance, the
Rt M
circuit canp be described by an LC circuit, with
Ctt fi ¼ 1=ð2p ðLi Ci ÞÞ. In general, the highest resonance of
Cts Cts Cts Fig. 2b will be the self-resonance of the turn-to-turn
element. For this element LN  Lt þ M ¼ Lcoil =n, with
Lcoil  L0 . We combine L0 and CN to yield a prediction
b for the p fundamental self-resonant frequency, f0;pred ¼
Fig. 2 Lumped-circuit representation of the coil 1=ð2p ðL0 CN ÞÞ. As fN 4f0 and n  1, it follows that
a The labelled circuit components and couplings between the different CN oC0 and f0;pred 4f0 . That is, the predicted frequency
turns will be an upper estimate to the measured self-resonant
b The simplified circuit, with the assumptions made in this work frequency. More complete distributed circuit models,
motivated by communications [12] and semiconductor [13]
applications in the radio-frequency and microwave ranges,
are able to be fitted to the measured impedance profile, but
dependence arising from the skin-effect. At the low do not offer prescriptions for calculating the stray
frequency limit, Rt  RDC;l , where capacitance at the design stage. The development of such
rlt a predictive distributed model in realistic geometry would
RDC;l ¼ ð2Þ provide a more accurate design tool, able to resolve all the
pa2
coil resonances.
and where r is the conductor resistivity and lt the average
length of one winding. A general expression for Rt ðoÞ is
3 Calculation of stray-capacitance
found from Bartoli [11]


AðoÞ þ iAðoÞ In the impedance domain, the capacitative coupling of any
Rt ðoÞ ¼ RDC;l <
tanhðAðoÞ þ iAðoÞÞ four turns (except those at the end) can be described by the
2ðNl2
1Þ
 circuit element shown in Fig. 3, where Zd ¼ 1=ðjoCts Þ,
þ ðAðoÞ þ iAðoÞÞ tanhðAðoÞ þ iAðoÞÞ Zg ¼ 1=ðjoCtt Þ, Za ¼ Zb ¼ 1=ðjoCtt Þ and the current
3 loops Ia;i , Ib;i , Ig;i , and Id;i are as drawn. For the final four
ð3Þ turns, the capacitive coupling is described by the same
For a compactly wound coil with d the spacing between circuit, except that Zb ¼ 0 on the right hand side. Here,
coils, AðoÞ can be written j is complex notation, with j2 ¼
1. The issue of model
p3=4 ð2aÞ3=2 completeness can be shown by current node analysis:
AðoÞ ¼ ð4Þ only three current loops contribute to the output coil
4 dðoÞd 1=2
with the skin depth dðoÞ given by
 1=2 Zα
2pr
dðoÞ ¼ ð5Þ
mrw m0 po Iα,i

where mrw is the conductor relative permeability. I β,i −1 I α,i−1 Zβ Zγ Zγ Zβ I α,i +1 I β,i+1
At the high-frequency limit, the inductance limbs (i.e. Rt ,
Lt , M paths) in Fig. 2 can be treated as an open circuit, I β,i
enabling a simpler calculation of the stray-capacitance Cs .
An estimate of the valid frequency range for this reduction
Zα
is jRt þ joðLt þ MÞj  j1=ðjoCtt Þj, with o ¼ 2pf the
angular frequency. That is, f  ft with I δ,i Zδ I δ,i+1
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi I δ,i −1 Zδ
ft ¼ 1=ð2p ðLt þ MÞCtt Þ ð6Þ
pffiffiffi p ffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
¼ n=ð2p Lcoil Ctt Þ ð7Þ
where Lcoil ¼ nðLt þ MÞ, and the mutual inductance Fig. 3 One element of the lumped-circuit representation showing
linking each turn with the other n
1 turns is constituent current paths

IEE Proc.-Circuits Devices Syst., Vol. 152, No. 6, December 2005 567
impedance: Id;i , Ib;i , and Ia;i . For all but the last circuit, 2.0
Kirchhoff’s voltage law may be written as the matrix
1.8
equation

Cs /C tt
ZIi
LIi
1 þ RIiþ1 ¼ 0 ð8Þ 1.6

where Ii is the column vector (Ia;i , Ib;i , Id;i ), and 1.4

0 1

Za 0 2Zd þ Za
0 0.2 0.4 0.6 0.8 1.0
Z ¼ @ 0 2ðZb þ Zg Þ 0 A ð9Þ ∗
C tt /C tt
2ðZa þ Zb Þ 0
Za
a
0 1
0 0 Zd 1.8
L ¼ @ Zb
Zb 0 A ð10Þ
Zb
Zb 0

Cs /C tt
0 1 1.6
0 0
Zd
R ¼ @ Zb Zb 0 A ð11Þ

Zb
Zb 0 1.4
10 −2 100 102 104
External connections to the coil are taken to be left of the Cts /C tt
circuit at i ¼ 1. At the last element on the right hand side
b
Ipþ1 ¼ 0, with Ib;p ¼ Id;p ¼ 0. Current conservation
implies Ip ¼ Cp
1 Ip
1 with Fig. 4 Plot of the normalised stray capacitance Cs/Ctt against C*tt/
0 1 Ctt and Cts/Ctt

1
1 0
a Normalised diagonal turn-to-turn capacitance Ctt*/Ctt with no shield
Cp
1 ¼ @ 0 0 0A ð12Þ Cts ¼ 0
0 0 0 b Normalised turn-to-shield capacitance Cts =Ctt with Ctt =Ctt ¼ 0:1
The different lines are: n ¼ 4 (dots), n ¼ 6 (dashes), n ¼ 8 (dash-
For the second and third last elements,
dot), and n ¼ 10 (solid)
Ip
1 ¼ ðZ þ RZ
1 LÞ
1 LIp
2 ð13Þ

Ip
2 ¼ ðZ þ RðZ þ RZ
1 LÞ
1 LÞ
1 LIp
3 ð14Þ

and so forth. This may be written more compactly as turn-to-turn capacitance that couples the shield and higher
order circuits. Figure 4b shows that the maximal effect of a
Ii ¼ Ci
1 Ii
1 ð15Þ shield is to increase the stray capacitance by around 20%.
with the Cs generated by the recursion relation However, this requires Cts =Ctt  10, which can only be
achieved by a very closely fitting shield; for a shield placed
Ci
1 ¼ ðZ þ RCi Þ
1 L; iop ð16Þ more than distance d from the coil, Cts =Ctt t1, and the
Solving recursively for I1 we calculate the output impedance increase in stray capacitance will be o7%.

Ia;0
Ib;0
Ia;1
Ib;1
 In the two extremes that a shield is absent and the shield
Zout ¼ Zb ð17Þ is adjacent to the inductor, the stray capacitance reduces to
Ia;0
Ib;0 Cs ¼ 1:45Ctt and Cs ¼ 1:72Ctt , respectively, where n  10
and extract the output stray capacitance, Cout ¼ 1=ðjoZout Þ. and Ctt =Ctt ¼ 0:1. These results compare well to earlier
Expressions for the output stray capacitance have been calculations by Massarini and Kazimierczuk, who found
checked in a number of physically relevant special cases. that for a two layer winding with n  10, the stray
The stray capacitance of a single layer solenoid without capacitance reduces to Cs ¼ 1:62Ctt for no shield, and Cs
shielding can be obtained by decoupling the inner and outer ¼ 1:83Ctt for a conducting core. Differences between the
layers. This involves setting Zg ! 1, Zd ! 1, and Zb ! expressions occur because in this model the windings lie in
1 of all but the last circuit, yielding Cs ¼ Ctt =ðn
1Þ, cylindrical shells centred at radius r ¼ R and r ¼ R þ d,
which is consistent with the results of Grandi et al. for a as opposed to the completely stacked layers of Massarini
single layer unshielded solenoid. If the turn-to-turn and Kazimierczuk, which p lie in cylindrical shells of radius
capacitance within layers dominates over the turn-to-turn r ¼ R and r ¼ R þ 2d= 3. Both treatments have error
capacitance between layers (i.e. Za =Zb ! 0, then associated with windings of opposite winding pitch. Turns
Cs ! ðn=2ÞCtt þ ðn
2ÞCtt . In the reverse ordering, where of opposite pitch do not lie compactly stacked, and so the
the turn-to-turn capacitance between layers dominates the spacing between layers will be underestimated in computa-
turn-to-turn capacitance within layers (i.e. Zb =Za ! 0), tions of Ctt in Massarini and Kazimierczuk. Whilst the
Cs ! Ctt . spacing between layers is correct in this treatment, the fact
Figure 4a plots the normalised stray capacitance against that the winding pitch is opposite between layers is
the normalised diagonal turn-to-turn capacitance for neglected. Thus, the stray capacitance, parameterised by
n ¼ 4, 6, 8 and 10, with the shield at infinity ðCts ! 0Þ. the turn-to-turn capacitance of touching wires, will be an
The swift convergence with increasing n results from the overestimate, as turns in different layers will in general be
series and diagonal turn-to-turn capacitance, which succes- non-touching. In turn, the predicted self-resonant frequency
sively reduce the contribution from higher turns. Figure 4b will be an underestimate.
plots the normalised stray capacitance against the normal- Finally, in order to compute the stray capacitance,
ised turn-to-shield capacitance for n ¼ 4, 6, 8 and 10, and expressions for Ctt and Ctt must be determined. Formal
Ctt =Ctt ¼ 0:1 (MAST case). As in Fig. 4a, the convergence solutions require calculation of the electrostatic field within
with increasing n results from the series and diagonal turn- the windings. As mentioned, this is outside the scope of this
to-turn capacitance. In the limit of large Cts the series turn- work, which is designed to provide simple scoping formulas.
to-turn capacitance is short circuited, and it is the diagonal Instead, we have used expressions for the capacitance in

568 IEE Proc.-Circuits Devices Syst., Vol. 152, No. 6, December 2005
various limiting cases. For Ctt , the conductors are assumed radius ratio l=r ¼ 1:3, which is within the range of
to be immersed in a medium of permittivity ed [14] applicability for (1) (however since the formula is intended
plt ed for a single-layer winding, application to a double-layer
Ctt ¼   ð18Þ solenoid may result in an error which is more than 1%).
d
cosh
1 From (1), (2), (18), and (19), the derived parameters
2a are Lcoil ¼ 17:2 mH, RDC  nRDC;l ¼ 67:5 mO, and
Here, lt is the length of wire in one turn. For media in 20.2 pF Ctt  28:0 pF with Ctt ¼ 2:1 pF. Using Fig. 4
which ed  eair , the permittivity of air, (18) will significantly to obtain the ratio Cs =Ctt for n  10, one obtains
overestimate Ctt , as the electric field lines between the 29:3 pF  Cs  40:6 pF, with mean C  s ¼ 35:0 pF (overbar
windings will be different to those in a homogeneous denotes mean). The large uncertainty in the predicted stray
medium. Even when ed  eair however, (18) will slightly capacitance (716%) corresponds to the uncertainty in the
overestimate Ctt , as all electric field lines do not start and permittivity of the insulator.
end on adjacent conductors. That is, the presence of other The impedance profile of the coil was measured with a
conductors is not taken into account. Consequently, (18) ZVRE Rhodes and Schwarz 3 GHz network analyser [16].
should be taken as an upper bound to Ctt , and therefore a Measurements of the impedance, Zmeas , were obtained from
lower bound to f0 . For Ctt , we have used the equivalent the S11 reflection coefficient [17]. Figure 5 shows the
capacitance of the series combination of the capacitance of absolute value of the impedance, jZj, over a 200 kHz–
the insulated coating and the capacitance related to the air 15 MHz sweep range. Whilst the fundamental resonance f0
gap between the turns [2]. This yields is defined by IfZg ¼ 0, for practical reasons, the
plt eair fundamental resonance was measured to be the frequency
Ctt ¼  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ð19Þ at which <fZmeas g first peaked. The validity of this
ln F þ F 2
ð1 þ t=aÞ2eair =ed assumption can be examined by noting the predicted
impedance [11, 18] can be written
where t is the thickness of the dielectric coating and
pffiffiffi Rcoil þ joLcoil ð1
o2 Lcoil Cs
Cs R2coil =Lcoil Þ
d= 2a Zpred ¼
F ¼ ð20Þ ð1
o2 Lcoil Cs Þ2 þ o2 Cs2 R2coil
ð1 þ t=aÞ1
eair =ed ð22Þ
Taken together, (18) and (19), and Fig. 4 provide an p
approximate description for determining the stray capaci- If Cs R2coil =Lcoil  1 then at o ¼ 2pf0 ¼ 1= ðLcoil Cs Þ,
tance of the circuit in the absence and presence of shielding. one finds IfZpred g ¼ 0 and @<fZpred g=@o ¼ 0. Hence,
Finally, for completeness, we note that the self-resonant providing that the measured impedance profile can be
frequency of the RLC circuit is closely fitted by (22) and Cs R2coil =Lcoil  1 (proven a
. pffiffiffiffiffiffiffiffiffiffiffiffiffi . pffiffiffiffiffiffiffiffiffiffiffiffiffiffi posteriori), peaks of <fZmeas g lie at the fundamental
f0 ¼ 1 2p Lcoil Cs  1 2p Lcoil Ctt resonance f0 . Applying this technique to 10 measurements
pffiffiffi yields f0 ¼ 5:32 0:02 MHz, where the error of
¼ ft = n ð21Þ 0.02 MHz is inferred from the variance of the scatter of
and so f0  ft providing n  1. In turn, reduction of the ten f0 measurements. Higher order self-resonant frequen-
circuit model of Fig. 2a to Fig. 2b with open circuit cies, discussed elsewhere (e.g. Rhea [12]), can also be
inductance limbs, for the purposes of calculating Cs at identified, at 30 MHz and 74 MHz.
frequencies above f0 was reasonable.
The expressions used in this work for Ctt and Ctt differ to 10 2
earlier treatments in several respects. In Massarini and
Kazimierczuk, the electric field path in an infinite compact
lattice was assumed, and expressions for the capacitance per
unit surface area between connecting field lines derived. The
turn-to-turn capacitance was found by integrating over all
possible paths. In Grandi, where a single layer solenoid was 10 0
| Z |, kΩ

modelled, (19) was used for Ctt , and (18) with ed ! eair used
to describe Ctt for windings with large air spacing between
turns. For the double layer, oppositely wound coil
considered here, an infinite fixed lattice structure cannot
be assumed: turn-to-turn capacitance calculation requires a
full 3-D Maxwell electrostatic simulation (e.g. [1]). Given 10− 2
the application of this model is to provide simple scoping
formulas, we have instead followed the simplified approach
of Grandi. 0 20 40 60 80 100
frequency, MHz
4 Experimental validation
Fig. 5 Frequency dependence of the absolute value of the coil
To test the stray capacitance model, the characteristics of impedance, 7Z7, as measured by a ZVRE Rhodes and Schwarz
a candidate double-layer coil to be used in MAST were network analyser
measured. The specifications of the coil were n ¼ 30,
l ¼ 14 mm, a ¼ 0:38 mm, d ¼ 0:81 mm, with the coil
wire coated in a t ¼ 25 mm layer of polyamide/imide Figures 6a and b show an enlargement of the magnitude
insulator with relative permittivity 3:9  ed =e0  5:4 [15]. of the imaginary part of Z, jIfZgj, about f0 . The frequency
The outer diameter of the wound coil was 23 mm, giving a response below f0 is dominated by the coil inductance, and
mean turns radius, r, of r ¼ 10:7 mm, and coil length to immediately above it by the coil capacitance. To obtain

IEE Proc.-Circuits Devices Syst., Vol. 152, No. 6, December 2005 569
 10 2 structures in the test board, which cannot be adequately
represented by the cylindrical wire model of (3)–(5). The
|ℑ{ Z}|, kΩ predicted coil inductance is accurate to within 1% of
10 0 the fitted value, consistent with the estimates of accuracy of
(1) as proposed by Wheeler [10]. The fitted capacitance
lies within the range of the predicted values, with
10 − 2 mean C  s;pred ¼ 35:0 pF, 8% below the fitted value. A
0 5 10 15
more detailed comparison between the predicted and mea-
frequency, MHz
sured stray capacitance would require the permittivity
a
of the insulator to be measured. Finally, the predicted
inductance, Lcoil;pred ¼ 17:2 mH, and stray capacitance
10 2 Cs;pred ¼ 35:0 5:7 pF combine to yield the prediction
|ℑ{ Z}|, kΩ

f0 ¼ 6:5 0:5 MHz. If the capacitance p of the test board and

10 1 cable is added in parallel, then 1=2p ðLcoil ðCs þCtest ÞÞ ¼
5:5 0:3 MHz. As with Cs , the measured value for f0 ¼
5:32 0:02 MHz lies within the predicted range. The
10 0 predicted impedance profile, correcting for the test board,
5.0 5.2 5.4 5.6 5.8 6.0
frequency, MHz is also shown in Figs. 6a and b.
b
It is useful to compare these predictions to those of
Massarini and Kazimierczuk [3]. Using their expression for
Fig. 6 Measured, fitted and predicted magnitude of the imaginary a two layer stacked coil structure in Massarini and
part of Z, jIfZgj, about the fundamental resonant frequency, f0 Kazimierczuk [3], Cs ¼ 1:62Ctt , together with (15) for Ctt
Solid points are the measured data, the dotted and dashed lines are the in that article, yields 17.7 pF r Cs r 21.5 pF, with mean
fitted and predicted profiles, respectively. At frequencies lower than 19.6 pF. The mean represents a
46% error from the
f0 ¼ 5:32 0:02 MHz, the coil is inductive (i.e. IfZg40), and
measured value, thereby casting doubt on the accuracy of
immediately above it, the coil is capacitative (i.e. IfZgo0)
a Vertical line denotes the upper frequency limit used in fitting to an
the calculations of Massarini and Kazimierczuk [3] in
RLC parallel circuit predicting the stray capacitance on double wound induc-
b Enlargement of Fig. 6a about the self-resonant frequency tors. It should be noted that the experimental measurements
of Massarini and Kazimierczuk [3], which quoted an
error of
17.2% for the predicted stray capacitance
fitted values for the RLC circuit model parameters (i.e. compared to the measured value, were based only on a
frequency-independent measured values of RDC , Lcoil , and single layer inductor. The difference between the results
Cs ), a least squares fit to the residue of Massarini and Kazimierczuk [3] and our working stems
from a difference in Cs =Ctt (1.62 against 1.45) and C  tt ,
rfit ðf Þ ¼ jðZpred ðf Þ
Zmeas ðf ÞÞ=ðZpred ðf Þ
 
(Ctt ¼ 24:1 pF against Ctt ¼ 12:1 pF) respectively. Physi-
þ Zmeas ðf ÞÞj ð23Þ cally, the value of C  tt in Hole and Appel is higher because
was performed, with Zpred given by (22). The frequency the whole structure is assumed to be completely immersed
range has been restricted to 0of o1:5f0 , so as to avoid in a dielectric.
phenomena associated with the second self-resonant Table 1 shows calculated values for Ctt , Ctt and Cs for
frequency. In each fit, the components RDC , Lcoil and Cs different models, and over the range of permittivity of the
R 1:5f polyamide/imide (PAI) insulator, 3:9  er  5:4. For
were varied to minimise the integral Lfit ¼ ð 0 0 rfit ðf Þdf Þ= completeness, we also show predictions for Hole and
ð1:5f0 Þ using a simplex search algorithm [19] implemented Appel, and Massarini and Kazimierczuk [3] in the limit that
in Matlab. Initial values were taken to be the predicted the entire structure is immersed in a dielectric (with
values, whilst the exit condition was DIfit o10
8 with DIfit permittivity eair ¼ er e0 ), where e0 is the permittivity of free
the change in Ifit between iterations. For this fractional space. In these cases, the predictions of the two models
exit condition, Ifit o1:8 10
2 for all experiments. Least overlap more closely: differences are due to details of the
squares fitting over 10 experiments yields: RDC ¼ electric field path, which have not been calculated precisely
115 3 mO, Lcoil ¼ 17:29 0:01 mH, Cs ¼ 51:75 0:25 pF,
where the variance in RDC , Lcoil and Cs has been taken from
Table 1: Predictions for Ctt , Ctt and Cs across different
the scatter of the fits. The good agreement between the models and permittivity
fitted (dotted line) to the measured (solid points) impedance
profiles in Figs. 6a and b provides confidence in the circuit eair er Ctt Ctt ; y Cs
model. For completeness, we note that at f 0 , Rcoil ¼ 4:7 O,
yielding Cs R2coil =Lcoil  7 10
5p  1, and so the funda-
Hole and Appel e0 3.9 20.2 pF 2.1 pF, – 29.3 pF
mental resonance f0 ¼ 1=ð2p ðLcoil Cs ÞÞ corresponds to Hole and Appel e0 5.4 28.0 pF 2.1 pF, – 40.0 pF
@<fZpred g=@o ¼ 0 in (22). Hole and Appel 3:9e0 3.9 20.2 pF 7.7 pF, – 33.5 pF
To compare the fitted capacitance to the predicted Hole and Appel 5:4e0 5.4 28.0 pF 10.7 pF, – 46.2 pF
capacitance, the value of RDC and Cs must be reduced by Massarini and e0 3.9 10.9 pF –, 0.18 rad 17.7 pF
RDC;test ¼ 31:1 mO and Ctest ¼ 13:78 pF, being the d.c. Kazimierczuk
resistance and capacitance of the connecting test board and
Massarini and e0 5.4 13.3 pF –, 0.15 rad 21.5 pF
cable measured (by a similar fitting technique) in isolation. Kazimierczuk
This yields RDC;fit ¼ 84 3 mO, Lcoil;fit ¼ 17:29
Massarini and 3:9e0 3.9 20.7 pF –, 0.36 rad 27.9 pF
0:01 mH, Cs;fit ¼ 37:97 0:25 pF, values which compare Kazimierczuk
well to the predicted values RDC;pred ¼ 67:5 mO, Lcoil;pred ¼
Massarini and 5:4e0 5.4 23.8 pF –, 0.36 rad 38.6 pF
17:2 mH, Cs;pred ¼ 35:0 5:7 pF. The discrepancy between
Kazimierczuk
RDC values may be the result of additional conducting
570 IEE Proc.-Circuits Devices Syst., Vol. 152, No. 6, December 2005
in either structure. The similarity in predictions of Cs The practical use of these estimates is that, when
between Massarini and Kazimierczuk with eair ¼ er e0 and combined with existing expressions for the coil inductance,
Hole and Appel with eair ¼ e0 occurs because in both cases Lcoil , an estimate of the fundamental self-resonant frequency
Ctt (which in Hole and Appel is the dominant capacitance) of the coil, f0 , can be obtained. Because our model: (1)
is calculated assuming immersion of the structure in the assumes that touching turns are immersed in a dielectric of
dielectric. permittivity ed 4e0 , and (2) does not account for the
Finally, the effect of a shield was studied by measuring opposite winding pitch of the layers, the predicted values of
the variation in the frequency of the fundamental resonance. Cs and f0 are likely to be upper and lower limits,
In these experiments, two thin copper sleeves of axial length respectively. Nonetheless, the expressions do provide a
8 mm and 11 mm, and with an axial cut, were wrapped useful scoping tool to determine the maximum frequency
around the terminal end of the coil, and molded to minimise operation as an inductor.
the air gap (see Fig. 1 schematic). The axial cut ensured that Experimental results were presented for a coil designed to
currents in the shield did not affect the axial magnetic field, measure high frequency oscillatory magnetic fields in a
and thereby the coil inductance. The effects of conducting fusion plasma. The measured value of the stray capacitance
shields on the coil inductance have been examined elsewhere was determined by a least squares fit to an RLC circuit
[20]. In the shield experiments described here, the measured model over the frequency range 0  f  1:5f0 . Our results
fundamental resonance changed to f0 ¼ 5:37 0:02 MHz show that the measured stray capacitance is within the
and f0 ¼ 5:36 0:04 MHz for the wide and narrow range of the predicted stray capacitance: uncertainties in
shields, respectively. the permittivity of the insulator give a range of 716%, with
Correcting for the capacitance of the test board and the mean value of the stray capacitance C  s ¼ 35:0 pF,
cable, corresponding fits to RDC , Lcoil and Cs yielded which is 8% lower than the measured value. A more
RDC ¼ 75:8 1:9 mO, Lcoil ¼ 17:28 0:02 mH, Cs ¼ 37:00 detailed comparison between the predicted and measured
0:36 pF, and RDC ¼ 76:2 0:1 mO, Lcoil ¼ 17:17 stray capacitance would require the permittivity of the
0:02 mH, Cs ¼ 37:69 0:65 pF, for the wide and narrow insulator to be measured. When combined with an
shield, respectively. In both cases, the experimentally expression for the coil inductance, the predicted self-
derived values of Cs are equal to the unshielded stray resonant frequency f0 is within the range of the measured
capacitance, to within error bounds; whilst the difference in value, with mean 4% larger than the measured value. Our
self-resonant frequencies between the shielded and un- results compare well to predictions of other models. For
shielded cases is attributable to the change in coil example, the analysis of Massarini and Kazimierczuk [3]
inductance. A possible reason why the stray capacitance predicts a stray capacitance and self-resonant frequency
has not changed is that each turn is almost-touching the 46% below and 25% above the measured values,
shield, which is locally flat. Hence, we would expect respectively. In both treatments, the higher mean of the
Cts =Ctt  1, and so Cs =Ctt  1:55 (see Fig. 6b), yielding predicted self-resonant frequency is also consistent with a
31:3  Cs  43:4 pF, with mean C  s ¼ 37:4 pF, which is frequency upshift owing to the presence of higher-order
2.4 pF above from the unshielded prediction of 35.0 pF. resonances discussed in Section 2.
Whilst the difference exceeds the uncertainty in the Finally, the effect of a shield on the value of Cs is
measurements, the uncertainties in Cts =Ctt preclude a predicted to be insignificant for the usual case of closely-
definitive comparison of theory to measurements. A series spaced windings when Cts =Ctt o0:1. In the case of widely
of more dedicated experiments, possibly involving the use of spaced windings where Cts =Ctt 410, Cs is predicted to
a conducting paint would be required to fully explore the increase by not more than 20%. Experimental measure-
upper limit of Cs =Ctt ¼ 1:7, shown in Fig. 4b. ments using the test coil confirm that a close fitting shield
(with axial cut) has a negligible effect on the stray
5 Conclusions capacitance.
New results in this work are:
A simple distributed circuit model has been developed to
compute the stray capacitance, Cs , of a double layer air- 1. The construction of a simple model for the stray
cored solenoid. The use of recursive circuit modelling capacitance of a two-layer, tightly packed inductor.
techniques provide a prescription for generating algebraic The model builds upon existing expressions for the turn-
expressions for the stray capacitance of an arbitrary number to-turn capacitance, but differs to earlier multi-layer
of coil turns. Providing that the number of coil turns is models because the coils are not assumed to be
larger than 10, our analysis shows that the stray capacitance compactly packed.
is proportional to the turn-to-turn capacitance of two wires 2. Predictions for the unshielded stray capacitance and self-
immersed in a dielectric of permittivity ed : for no shield, resonant frequency, which overlap the measured range.
Cs ¼ 1:45Ctt , with a shield, Cs  1:7Ctt . Qualitatively, To our knowledge, these are the first published results of
these results consolidate earlier findings [2, 3] that the stray both the predicted and measured stray capacitance of a
capacitance is only weakly dependant on the number of double layer solenoid. More precise predictions would
turns, and converges quickly with increasing number of require measurement of the permittivity of the insulator.
turns, n. In a single-layer solenoid [2] it is the capacitive 3. Predictions, validated by experiment, that the effect of a
coupling with the shield that dominates the stray capaci- close-fitting conducting shield on the value of Cs and f0
tance, and is responsible for the swift convergence with is weak.
increasing n. In a two layer solenoid swift convergence with
n is achieved without the shield, and is due to the capacitive 6 Acknowledgments
coupling between the first few turns in the inner and outer
layer. We hypothesise that this result will extend to This work was funded jointly by the United Kingdom
solenoids with an even number of layers, where the Engineering and Physical Sciences Research Council and by
terminals are at the same end of the coil. In solenoids with EURATOM. The authors would like to thank R. Martin
an odd number of layers, the capacitance will in general be for providing a test coil, and the referees for useful feedback
dominated by the electrical path through the shield. and advice.

IEE Proc.-Circuits Devices Syst., Vol. 152, No. 6, December 2005 571
7 References 9 Heeter, R.F., Fasoli, A.F., Ali-Arshad, S., and Moret, J.M.: ‘Fast
magnetic fluctuation diagnostics for Alfv!en eigenmode and magneto-
hydrodynamic studies at the Joint European Torus’, Rev. Sci.
1 Yu, Q., and Holmes, T.W.: ‘A study on stray capacitance modeling of Instrum., 2000, 71, (11), pp. 4092–4106
inductors by using the finite element method’, IEEE Trans. 10 Wheeler, H.A.: ‘Simple inductance formulas for radio coils’, Proc.
Electromagn. Compat., 2001, 143, (1), pp. 88–93 IRE, 1928, 16, pp. 1398–1400
2 Grandi, G., Kazimierczuk, M.K., Massarini, A., and Reggiani, U.: 11 Bartoli, M., Reatti, A., and Kazimierczuk, M.K.: ‘Modelling iron-
‘Stray capacitances of single-layer solenoid air-core inductors’, IEEE powder inductors at high frequencies’. Proc. IEEE Industry Applica-
Trans. Ind. Appl., 1999, 35, (5), pp. 1162–1168 tions Soc. Ann. Meeting, 1994, pp. 1225–1232
3 Massarini, A., and Kazimierczuk, M.K.: ‘Self-capacitance of induc- 12 Rhea, R.W.: ‘A multimode high-frequency inductor model’, Appl.
tors’, IEEE Trans. Power Electron., 1997, 12, (4), pp. 671–676 Microw. Wirel., 1997, 9, (6), pp. 70–80
4 Bartoli, M., Noferi, N., Reatti, A., and Kazimierczuk, M.K.: 13 Horng, T.-S., Wu, J.-M., Yang, L.-W., and Fang, S.-T.: ‘A novel
‘Modeling winding losses in high-frequency power inductors’, modified-T equivalent circuit for modeling LTCC embedded inductors
J. Circuits Syst. Compat., 1995, 5, (4), pp. 607–626 with a large bandwidth’, IEEE. Trans. Microw. Theory Tech., 2003,
5 Yu, Q., Holmes, T.W., and Naishadham, K.: ‘RF equivalent circuit 51, (12), pp. 2327–2333
modeling of ferrite-core inductors and characterization of core 14 Smythe, W.R.: ‘Static and dynamic electricity’ (McGraw-Hill, New
materials’, IEEE Trans. Electromagn. Compat., 2002, 44, (1), York, 1950)
pp. 258–262 15 GoodFellow Catalogue, http://www.goodfellow.com/csp/active/
6 Hole, M.J., Akers, R.J., Appel, L.C., Buttery, R.J., Brickley, C., STATIC/E/Polyamide-imide.HTML, 2004
Conway, N.J., Gryaznevich, M., Hender, T.C., Kwon, O.J., Valovic, 16 Rhodes and Schwarz website, http://www.rsd.de, 2005
M., Medvedev, S., Patel, A., Saarelma, S., Taylor, D., Wilson, H.R. 17 Sander, K.F., and Reed, G.A.L.: ‘Transmission and propagation of
and the MAST Team: ‘Ideal MHD stability of the mega-ampere electromagnetic waves’ (Cambridge Univeristy Press, 1978)
spherical tokamak’, Plasma Phys. Control Fusion, 2005, 47, (4), 18 Bartoli, M., Reatti, A., and Kazimierczuk, M.K.: ‘High-frequency
pp. 581–613 models of ferrite core inductors’, Proc. IEEE Int. Conf. on Industrial
7 Appel, L.C., Akers, R.J., Fullop, T., Martin, R., and Pinfold, T.: Electronics (IECON), 1994, pp. 1670–1675
‘Observations of CAE’s on MAST’. Proc. 31st EPS Conf. on 19 Lagarias, J.C., Reeds, J.A., Wright, M.H., and Wright, P.E.:
Controlled Fusion and Plasma Physics, 2004, P4.195 ‘Convergence properties of the Nelder-Mead simplex method in low
8 Appel, L.C., and Hole, M.J.: ‘Calibration of the high-frequency dimensions’, SIAM J. Optim., 1998, 9, (1), pp. 112–147
magnetic fluctuation diagnostic in plasma devices’, accepted, Rev. Sci. 20 Simpson, T.L.: ‘Effect of a conducting shield on an air-core solenoid’,
Instrum., 2005 IEEE Trans. Magn., 1999, 35, (1), pp. 508–515

572 IEE Proc.-Circuits Devices Syst., Vol. 152, No. 6, December 2005