Complementary frequency selective surfaces

D.S.Lockyer, J.C.Vardaxoglou and R.A.Simpkin

Abstract: A concept in frequency selective surface (FSS) technology that stems from Babinet's
principle, whereby a hybrid of two closely coupled FSS, a layer of conducting elements and a layer of
aperture elements are etched either side of a dielectric substrate, acronymed as complementary FSS
(CFSS) is introduced. There are two narrow passbands separated by a distinct null. The CFSS creates
electrically large elements from physically small ones to such an extent that a conventional iV2
resonator in free space at the lower passband resonant frequency would be over three times longer
than the dipole length employed in the CFSS. This passband is highly stable for normal and oblique
TE and TM incidences, with less than 2% frequency shift. A full wave modal analysis is developed
and two integral equations (IEs), an electric field IE and a magnetic field IE, are derived. These IEs
are coupled and this renders them suitable for the solution of the CFSS problem. The interlayer
dielectric region is critical to the location of the passbands as well as the in-band loss. Results are
presented for dipole and ring CFSS and compared with measured data up to 40GHz.

Introduction

Single layer frequency selective surfaces QFSS) formed from

a periodic array of simple elements backed by a supporting
dielectric layer are easy to construct, but limited in their
performance in terms of susceptibllity to changes of polarisation state and angle of incidence with frequency. Close
coupled FSS [I] and recodigurable FSS [2, 31 offer good
angular stability, but their performance is impaired when
altering the polarisation state. Aperture coupled microstrip
patch arrays, formed from three element layers, have been
studied with a view to obtaining an FSS like passband with
a good angular stability [4].
This paper introduces a concept and discusses results
arising from a hybrid of two closely coupled FSS, whereby
a layer of conducting elements and a layer of aperture elements are etched either side of a dielectric substrate, acronymed as complementary FSS (CFSS). The elements need
not necessarily be the same on either side of the substrate.
The CFSS takes advantage of the interaction between the
layers to produce strong fields in the separation region.
There are two narrow passbands separated by a distinct
null and the lower passband resonance appears at a frequency that is much lower than that of a single layer array
(aperture or conductor). As a result, t h s prhary CFSS
resonance is far from the grating region and, owing to the
complementary nature of the arrays, good angular and
polarisation stability can be achieved. A modal analysis
suited to closely coupled structures and based an Floquet
modes is presented. Results obtained from the electromag0IEE, 2000
IEE Proceedugs o h e no 2oooO799
DO2 10.1049/1pmap2CNO799
Paper frst r m i v d 14th March and m revised fonn 17th July 2000
D S Lockyer dnd J C Vardaxoglou are wth the Department of Electromc and
Electrical Engu7eenn& Loughborough Umversty, Loughborough, Leicestershre LEI1 3TU, UK
R A S m p h IS with Chelton Radomes Ltd, PO Box 55, Gunnels Wood
Road, Stevenage, Hertfordshe, SGl 2DB, UK
IEE Proc - M u o w Antennas Propag

Vol. 147, No 6, Deeernber 20110

netic model are compared with those measured for different CFSSs of linear dipoles and single rings. Comparisons
we made for both TE and TM incidences at 45",as well as
for normal incidence.
aperture array (element area Ae)

\
condiuctor array (element area Ac)

dielectric

Fig. 1 schrmatic representation of CFSS structure

Unit cell area A

Concept description

The structure of the CFSS is depicted in Fig. 1. It shows

the principle that the CFSS is made up of two separate,
complementary arrays. In isolation, when illuminated by an
incident plane wave, both arrays become resonant due to
the currents and fields i n d u d , and if the two sets of array
dimensions are identical, as in the eases studied here, the
resonant frequencies are the same. When they are in very
close proximity, the transmission responses produced by
the two layers interact with one another across the
frequency range, as typically shown in Fig. 2. Here, a
50 1

conducting rectangular patch is coupled with its free space

Babinet's complement, the aperture patch. Babinet's principle, relating to optics, states: 'when the field behind a screen
with an opening is added to the field of a complementary
structure, the sum is equal to the field when there is no
screen' [5]. The presence of the dielectric in our case means
that the separate elements are not direct Babinet complements, but the principle is preserved. It will be appreciated
that, for nonrotationally symmetric elements, such as the
linear dipole, the conductors and apertures must be rotated
by 90" with respect to each other, in order that they will
both be resonant. This does not apply to elements like rings
or square loops but, by shifting one array in relation to the
other, maximum interaction can be produced, ensuring the
currents and fields reach their peak values where they intersect. However, the CFSS principle is not limited by choice
of element [6]. There are two narrow passbands (at f r and
f,> separated by a distinct null at hull.
This null is mainly
produced by the arrays being partially reflective at that
frequency. This is not directly deduced from the Figure
since the interlayer coupling accounts for the position of
the null as well as its bandwidth. We have plotted the individual array responses for reference only, and it should be
realised that, due to the close proximity of the layers, the
CFSS response does not emerge by a simple addition of
two propagating forward and reverse waves (i.e. a standing
wave). This is true over most of the frequency range. At f r
a n d h the arrays interact very strongly and hgh evanescent
fields are generated in the dielectric region whch lead to
the CFSS becoming resonant. By changing the separation
distance, the passbands can get closer together or further
apart. The behaviour of these passbands is discussed in
detail in the examples given in Section 4.
fi

f,I

interface, this gives rise to two coupled IEs, whch are then
reduced to a matrix form [7]. These IEs are formed in
terms of spectral components of the current on the conductors at z = 0 (.?) and aperture fields within the apertures at
z = d (E). In t h s case a magnetic field integral equation
(MFIE) is formed for the fields withn the apertures and an
electric field integral equation (EFIE) for the currents on
the conductors. These are solved using the method of
moments (MOM) [SI for the unknown fields and currents.

I/.
. . '.

. ..

. ..

. ..

0 0

3 1 1

Tint
/

7 1 1

T+

Z=O

z=d

Fig.3 CFSS cross-section showkg qualitatively the modal coejfxients

In the following derivation we assume that the structure

is an infinite periodic array excited by a plane wave, incident at an arbitrary angle. This enables us to utilise Floquet's theorem and we assume the same numbers of
Floquet modes on each layer, represented by the subscript
indices p and q. We have used the same notation for the
expansion and meaning of symbols as in [7]. A section of
the CFSS is shown in Fig. 3, whch illustrates the conventions adopted for the various transmitted and reflected
fields in the analysis. The set of tangential field expansions
at the various
in terms of vector Floquet modes, Ypq(i)&pq,
regions are given by:
For z 5 0:

E-(r, z ) = Einc

R,pqej'Zq2@pq(~)&mpq(1)
mpq

-50-

-60

'""11

frequency, GHz

Fig.2 Comparison of CFSS t r m i s s w n respome with that of i n d v u

single layers

Simulated results

CFSS

_ _ _ conductors, single layer array

_ _ _ _ apertures, single layer array

H1( E , z ) =
3

Modal analysis

The frequency domain electromagnetic analysis of a CFSS

uses a set of Floquet expansions for the electric and magnetic fields. Through the boundary conditions at each interface a set of IEs are derived for the current (conducting
elements) and aperture field (perforated screen). When the
electromagnetic boundary conditions are applied at each
502

IEE Proc.-Microw. Antennas Propag.. Vol. 147, No. 6, December 2000

&+(E,

2)

(6)

VmpqT+e-'y:qZ'4'pq(r)2 x

and

mpq

(17)

where the incident plane wave has the form

2
' 0

~ i n c
-

~ i n ec - ~ ~ o o ~ ~ o o ~ m o o
mOO
(7)
m=l

The subscript m signifies TM (eqn. 1) or TE (eqn. 2) incidence. qmpqand ypq are the modal admittances and propagation constants, respectively.
To derive the two coupled IEs, expressions for the electric and magnetic, transmitted and reflected fields within
the dielectric region (Tmpq
and RTpq)must first be obtained.
These are found by the application of appropriate boundary conditions and the matching of the fields at z = 0 and z
= d. The boundary conditions are:
(i) At z = 0, the tangential electric field is continuous and
the tangential magnetic field is discontinuous. Further, the
total electric field is zero on the conductors.
At z = 0,
Einc(r,0 ) = E 1 (E, 0 )
(8)
using the orthogonality of the Floquet modes one arrives
at:
inc

Tmoo + R L p q = T m p q + R m p q

(9)

Having established an expression for Rmpqin terms of

and E , an expression for Tmpqcan now be developed using
eqns. 14 and 15. Thus:

Having derived the expressions for Rmpqand Tmpq,the

boundary conditions can now be applied at z = 0 and z = d
to obtain the integral equations related to each layer. First,
at z = 0, we force the total electric field on the conducting
regions of the unit cell (Ac)to be zero, in order to derive
the EFIE.
At z = 0:
E(E,O)= 0

T E A,

(19)

whence
(Tmpq

+ R m p q )q p q ( r ) E m p q = 0

(20)

mpq

Using eqns. 15 and 18 the EFIE at z = 0 is:

For the magnetic field:

H-(r,O)- E 1 (r,O)= .2 x J(r,O)

(10)

and again using the same orthogonality one obtains:

0

Tinc

Vmoo moo - V m p q R Z p q = 7 L p q T - w - V L p q R m p q
I

Jmpq

+A

1
~ v ~ O O T=
~ T~ mOp q ( V m p q

(21)

'pq (dEmpq

where A is the unit cell area and jmpq

signifies the Floquet
spectra of the current density. This can be written in terms
of an inner product, Jmpq= ( J ( ~ ) . KYp,(r))A,.
~~~,
Eliminating Rip, from eqns. 9 and 11 one arrives at:

The magnetic field integral equation (MFIE) at z = d is

obtained by applying the condition that the magnetic field
is continuous across the apertures (Ae):

H1(r,0) = H+ (50 )

+ V 0m p q )

T E A,

(22)

which leads to:

(ii) At z = d, both the tangential electric and magnetic
fields are continuous across the aperture regions. These
conditions can now be used to initially establish expressions
for the fields within the dielectric region. Thus:
El (E, d ) = E+ ( E , d )
Tmpne-jTiqd

+ R m p q e j 7 i q d = T+

mpq

e-jy:qd

(13)
=

E,pg

A
(14)
where E is the Floquet spectra of the electric field defined
in the aperture of the unit cey only.
Using Tmpqe-jhd+ Rmpqdkd= Epp#A and eqn. 12 the
following expression for Rmpqis obtained:

where

IEE Proc-Microw. Antennas Propag., Vol. 147, No. 6, December 2000

The above coupled IEs (eqns. 21 and 24) are Fredholm

type of the first kind with separable kernels and can be
conveniently combined into a single matrix equation. This
follows the same procedure as that described in [7], Ch. 4,
for multilayer arrays. In the case here it is the coeficient of
in the EFIE and J in the MFIE which account for the
coupling between the conductors and apertures.
SO3

Expanding the currents and aperture fields, and takmg

the inner product with their corresponding test basis functions, a resultant matrix equation is produced that takes the
following form:

cover the spectra of these bases. The maximum coupling

between the dipoles is achieved when their centres are coincident because the maxima of the dominant (cosinusoidal)
basis functions is at this point and it is desirable for optimum performance to have current and field maxima as
proximate as possible.

The submatrices Zll,Z2, are similar to those of individual

layers and Z12,Z2, represent the coupling between the
layers. The expressions for these are given in the Appendix,
Section 8. The order of the square matrix is IT + NT,
where IT and NT are the total current and field bases used
in the first and second layer, respectively.
4

Results from dipole and ring arrays

The key aspect of the CFSS is the coupling of the evanescent fields within the dielectric region. Ths is achieved by
having the elements closely spaced so that the fields due to
each strongly interact. Maximisation of the coupling
between the respective conducting and aperture elements is
achieved by the rotation of the elements by 90" with respect
to one another (in the case of dipoles for instance) or a relative shift of half a period in both x and y directions for the
ring elements. This does allow the current and field
maxima to strongly couple with one another. These cases
are shown in Figs. 4 and 8. It is important to appreciate
the significance of the rotation of the dipole aperture elements so that they are perpendicular to the incident electric
field. This is to satisfy the condition that both elements be
resonant in order to establish the interaction between the
induced currents and fields.

frequency, GHz

Fig.5 fi.anrisswn response of dipole CFSS: nomml incidence

-0-

measured
predicted (loss tangent 0.OOOl)
predicted (loss tangent 0.03)

-__

++

aperture (back layer)

(front layer)

UY

orientation of incident
electric field

T)
c

._

-10

c -15

.-U)

._
E

-20

Fig. 4

Geometry of d p l e CFSS

4.1 Dipole arrays

Of all the potential elements for use in the CFSS, the
simplest is the h e a r dipole which has already been
employed for the close coupled and reconfigurable FSS in
both conductor and aperture cases [l-31. Fig. 4 shows the
array geometry and element orientation with respect to a
vertical incident electric field. The material used in the
construction of the arrays consists of sheets approximately
x 300"
which are made up of a layer of dielec280"
tric (polyester) 7 0 p thick sandwiched between two layers
of 1/4 oz copper. The dielectric has a relative permittivity of
3. The unit cells of each array are identical, being on a
square lattice with periodicities D, and Dy equal to 4mm.
The element dimensions are also identical havmg w1 = W 2
= 0.3mm and L1 = L2 = 3.5mm. In the computer model
three entire domain basis functions were used for each
layer. For the conductors cos(ny/Ll), sin(2xy/Ll) and
cos(3qdLI) were used and for the apertures the modes
cos(m/L2), sin(2m/L2) and cos(3m/L2) were used. In each
layer lpl = 141 = 13 was found to be sufficient per unit cell to
504

-25

-30

10

30

20

40

50

60

frequency, GHz

Fig.6 Trammission response of dipole CFSS: TE:45"incidence

a-

--_

measured
predicted (loss tangent 0.ooOl)
predicted (loss tangent 0.03)

Figs. 5-7 show a comparison of measured and theoretical data for the dipole CFSS described above at normal
incidence and oblique TE:45" and TM:45" incidences. The
measuring range available to us is 1240GH2, and thus we
have chosen the array dimensions to include at least the
lower passband and the null. It can be seen that, overall,
the agreement with measured results is good. The passband
resonance is stable at around 13.2GHz with the null
between 23 and 24GHz. It is also a 3J6 resonance as
opposed to a 3J2 resonance of a single layer dipole FSS.
This means that more elements can be packed into a given
surface area for a given resonant frequency. This is attracIEE Proc.-Microw. Antennas Propag., Vol. 147, No. 6, December 2000

tive in FSS applications, such as frequency selective wave

guides (FSGs) [9], which have limited utilisation space for
the array. The stability of the lower passband is due to the
resonance being far removed from the influence of grating
lobes. According to our model the first grating mode
begins at 44GHz for both TE and TM incidences, even
though the way it manifests itself in the transmission coefficient response in each case is different due to the polansation change.

4.2 Ring arrays

In this Section a CFSS from single ring elements is investigated. Rings have been extensively studied as elements for
FSS, see, for example, [lo]. They exhibit good cross-polar
performance and allow close packing of the elements.
Owing to the ring symmetry the maximum coupling
between the currents and the fields can be achieved by a
relative shift between layers. Fig. 8 shows the arrangement
used for the ring CFSS. The elements are identical and are
arranged on a square lattice. Coupling is maximised by
introducing the shifts in two directions: DSX and DSY.
That ensures the peak values of the currents and fields
intersect. The dimensions used in this case are W =
0.15mm; Rin = 1.1";
Dx= DJ,
= 3".
The relative shifts
are of half a lattice period in each of the two perpendicular
directions such that DSX = DSY = 1.5mm. As previously,
the dielectric was 7 0 p thick with E,. = 3.

\ apertures

'

frequency, GHz

conductors

Fig.7 Trunsmkswn response of dipole CFSS: TM:45"incidence

-0-

__-

measured
predicted (loss tangent 0.0001)
predicted (loss tangent 0.03)

Fig. 7 shows the comparisons for TM:45" between measured and predicted results. High angular stability is demonstrated for both oblique TM and TE incidences, to about
1%, which was not observed for other noncomplementary
closely coupled structures. The resonant loss for TM:45"
incidence is about 2dB, which is the lowest of all the dipole
CFSS incidences and contrasts with the TE:45" case, where
the loss is about 4dB. This loss is primarily due to the dissipative power loss in the dielectric, which depends on the
field coupling, which is in turn dependent on the angle of
incidence and polarisation. In addition, the reflection loss
at the CFSS face due to mismatching between the structure
and free space was better than 15dB. However, the
coupling between the individual screens may alter the path
length through the dielectric, which, in turn, will influence
the dissipation loss and partially determine the Q. When
the loss tangent is reduced to 0.0001, there is negligible loss,
and that demonstrates the CFSS is well matched. The
manufacturers of the polyester specified a loss tangent of
0.005, but at a frequency of a few kHz and, upon
commencing these investigations, we assumed that value.
However, to match the theory with measured results we
needed to use a loss tangent of 0.03. There may also be a
contribution from conducting losses, which we have not
included in our model. An investigation was performed
into the loss mechanism, and it was found to be related to
the Q of the CFSS. Keeping the same loss tangent but
increasing the Q yields higher passband loss. This was
borne out by comparison with a single layer array, where
the Q is much lower. This trend is not specific to CFSS,
and will be reported in detail elsewhere.
IEE Proc -Microw. Antennas Propug., Vol. 147, No. 6, December 2000

Fig.8 Geomepy ofring CFSS

frequency, GHZ

Fig.9

Trunsmisswnresponre of Ying CFSS: normal incidence

__-

measured
predicted (loss tangent 0.OOOl)
predicted (loss tangent 0.03)

3-

In modelling the rings two sets of bases have been used,

cosnla and sinn2a,where a represents the angular position
of a point on the circumference of the ring element from 0
to 2x. n, and n2 are integers representing the order of the
basis function. In this particular case M , = 0, 1, 2, 3 and n2
= 1, 2, 3 in both arrays and it was found that lpl = 141 = 13
Floquet mode indices were again sufficient. Figs. 9-11
show a comparison of measured and predicted results for
the array described, at normal incidence, TE:45" and
505

TM:45". The primary passband is centred at about

13.8GHz for normal incidence, but at TM:45" incidence
this moves to 14.2GHz. This represents a percentage shift
in frequency of just under 2%, and, while not as good as
the dipole CFSS, this is far superior to a conventional
planar ring FSS. In both the dipole and ring cases, the shift
observed is introduced linearly as the angle of incidence
increases. Further, the better than 2% stabihty holds for
both TE and TM cases which was not the case for the
noncomplementary close coupled surfaces. It is also worth
noting the slight change in the bandwidth as the incidence
conditions change. Overall, the agreement between theory
and measurement is good.

frequency, GHz

Fig. 10 Transmission response of ring CFSS: TE:45"incidence

A-

measured
predicted (loss tangent 0.0001)
predicted (loss tangent 0.03)

nance is much further away from their frequency of influence, which in this case is about 58GHz. At normal
incidence, the upper passband is out of the measurement
range, but it moves to 33GHz for 45" oblique incidence for
both TE and TM polarisation. This demonstrates the
destabilising effect of the grating lobes and highlights the
advantage of having a resonance as far away from this
region as possible.

4.3 Effect of separation distance between arrays

The separation distance between the two element layers d
has already been identified as a key issue in the design and
performance of the CFSS. The closer the layers are to one
another, the higher will be the coupling between them and
hence the stability will be improved. Fig. 12 shows, for the
dipole CFSS (same geometry as the results in Section 4.1),
how increasing the separation distance alters the frequency
response of the CFSS and consequently how the arrays
become decoupled from one another. The curve for a separation of 7 0 p is displayed as a reference for the others.
When d is increased to 200p-11, the two passbands begin to
move closer together, although they are still individually
well defined, with the main passband being at around
17.5GHz. However, as the separation is further increased,
the main passband essentially disappears and the stability
noted at close separations is lost. For d = lmm, the upper
passband has collapsed and the lower one is beginning to
split into two separate responses around 20GHz and
32GHz, and this is a clear indication of the arrays decoupling. When the separation reaches 2mm (and above) the
most prominent feature is the very narrowband null at
30GHz. The response seems to lock on near to t h s
frequency for separations up to lOmm, while the original
passbands have completely vanished. This holds for
frequencies up to about 44GHz, beyond which the Floquet
modes couple, since they are above cutoff in the dielectric.
0 -

Or

I
-5 -

U
c--10

.0

8 -15 0
._
U1

.-

-20 -

-25

-300

frequency, GHz

Fig. 12 Effect of increasing separation between CFSS layers

frequency, GHz

Fig.11 Tranmisswn response of rizg CFSS: TM:45"incidence

-0-

__-

measured
predicted (loss tangent 0.0001)
predicted (loss tangent 0.03)

__

70p.m
200p.m
1"

-0-

2"

--e

Conclusions

Once again, the ring CFSS exhibits loss at resonance of

around 2-2.5dB. The same arguments apply here with
regard to the nature of these losses as with the dipole CFSS
above. The effect of the higher order grating lobes is less
apparent for the rings because the upper frequency reso506

The plane wave transmission response and angular stability

properties of a complementary FSS structure have been
assessed. A modal analysis was developed and implemented
using a MOM technique. It was found that the CFSS delivIEE Proc.-Microw. Antennus Propug.. Vol. 147, No. 6, December 2000

ered powerful performance in terms of angular stability

with all cases of oblique plane wave incidence, up to 45,
resulting in a frequency shift at resonance of less than 2%
when compared to the case at normal incidence. This is due
to the primary passband being situated suitably far enough
from the grating lobe region so as the destabilising effect
does not interfere, even at oblique angles of incidence. The
CFSS creates electrically large elements from physically
small ones to such an extent that a conventional ?J2 resonator in free space at the CFSS resonant frequency would
be over three times longer than the dipole length employed
in the CFSS. The dipole CFSS, for example, resonated at
around 13GHz using elements that are 3.5 in length. In
the absence of dielectric loading, these elements would
normally resonate at around 42GHz. The grating lobes
would strongly influence the resonance at this frequency,
but the CFSS resonance at 13GHz is virtually unaffected.
This makes the CFSS a very attractive proposition for use
in applications where filtering is required over a narrow
bandwidth, and the signals to be fdtered are not from the
same direction. It was observed that, due to the hghly
coupled fields within the separating dielectric region, the
loss in the passband was mainly a dissipative power loss in
the dielectric that could only be reduced by the use of a low
loss dielectric substrate. Nevertheless, CFSS structures are
made of simple elements and are easy to construct using
conventional PCB etching techniques.
6

9 ORTA, R., TASCONE, R., TRINCHERO, D., LOUKOS, G., and

VARDAXOGLOU, J.C.: Dispersion curves and modal fields of a
waveguide with FSS inserts, Electron. Lett., 1995,30, pp. 1072-1075
10 VARDAXOGLOU, J.C., and PARKER, E.A.: Plane wave illumination of concentric ring frequency selective surfaces, IEE Proc. H,
Microwaves, Antennas and Propagation, 1985, 132, (3), pp. 176180

Appendix

The expressions for the matrix equation in terms of the

spectra of the basis functions of the currents hl and field h2
are:
NT

IT

c.3~~~

C i h l i and E =

J =
i=l

n=l

Acknowledgment

Ths work was partially funded by British Aerospace.

References
VARDAXOGLOU, J.C., and LOCKYER, D.S.: Modified FSS
response from two sided and closely coupled arrays, Electron. Lett.,
1994,30, (22), pp. 1818-1819
LOCKYER, D.S., MOORE, C.M., SEAGER, R., SIMPKIN, R.,
and VARDAXOGLOU, J.C.: Coupled dipole arrays as reconfigurable frequency selective surfaces, Electron. Lett., 1994, 30, (16), pp.
1258-1259
LOCKYER, D.S., and VARDAXOGLOU, J.C.: Reconfigurable
FSS from two layers of slotted dipole arrays, Electron. Lett., 1996,32,
(6), pp. 512-513
POUS, R., and POZAR, D.M.: A frequency-selective surface using
aperture coupled microstrip patches, IEEE Trans., 1991, AP-39, (12),
pp. 17661769
BALANIS, CA.: Antenna theory: analysis and design (John Wiley,
New York, 1997,2nd edn.)
SIMPKIN, R., and VARDAXOGLOU, J.C.: A frequency selective
surface. UK Patent application 9412551.5, 1994
VARDAXOGLOU, J.C.: Frequency selective surfaces: analysis and
design (Research Studies Press, UK; Wiley, New York, 1997)
HARRINGTON, R.F.: Matrix methods for field problems, Proc.
IEEE, 1967,55, (2), pp. 136149

IEE hoc.-Microw. Antennus Propug., Vol. 147, No. 6, December 2000

507