Get TRDoc 1
Get TRDoc 1
Get TRDoc 1
16. SECURITY CLASSIFICATION OF: 17.LIMITATION 18.NUMBER OF 19a. NAME OF RESPONSIBLE PERSON
OF ABSTRACT PAGES Richard A. Marr
a. REPORT b. ABSTRACT c. THIS PAGE 19b. TELEPHONE NUMBER (include area code)
Unclassified Unclassified Unclassified SAR 45 N/A
Standard Form 298 (Rev. 8-98)
Prescribed by ANSI Std. Z39.18
i
ii
Methods for Locating Stray-Signal Sources
in Anechoic Chambers§
‡ Anteon Coporation
§
This work was supported in part by the Air Force Office of Scientific Research.
Abstract
sources in anechoic chambers are investigated and applied in combination to actual measurement
data. Both methods use single-frequency near-field data collected on a planar surface and process
it to reconstruct field values (images) elsewhere. The first method, based on the fact that the
probe output satisfies the Helmholtz equation, uses plane waves to backpropagate the scan-plane
data and is well suited for FFT-based rapid reconstruction of images on planar surfaces parallel
to the scan plane. The second method uses the simple spherical-wave focusing technique and is
flexible in that it can be used to generate images on either planar or non-planar surfaces from data
collected on either planar or non-planar surfaces. When data and image points are both located on
a regular grid, the method can be implemented using the FFT-based fast convolution technique.
Both methods include a spatial filter for isolating selected plane wave spectrum components. An
asymptotic evaluation of the kernel of the plane-wave formulation reveals that the two methods
produce nearly identical images for the measurement parameters typically encountered in anechoic
chambers. The two methods are used in combination to successfully locate the strong multiple-
bounce stray signals that degrade the quiet-zone of a near-field bistatic RCS facility. Subsequent
scan data confirms that the suppression of these stray signals indeed substantially improves the
quality of the quiet-zone. The spherical-focusing method is then used to evaluate the effectiveness
of the various configurations of absorbers that are placed along the edges of a reflector to control
edge-diffracted fields. It is shown that the reduction of the edge-diffracted fields further improves
the quiet-zone .
2
1 Introduction
Radio anechoic chambers are used throughout the world to perform controlled measurements
of fields radiated by antennas and scatterers. Absorbers in the chamber reduce unwanted
reflections to levels that are much lower than could be achieved in uncontrolled environments.
Depending on the size of the chamber, the measurements can be performed in both the near
and far fields of the antenna or scatterer. When the far-field patterns of a large antenna
or the bistatic radar-cross sections (RCS) of a large target need to be determined, the near
fields are collected with an electromagnetic probe on a curve or surface in the chamber and
processed using appropriate theories to determine the far-field patterns and the bistatic RCS
values.
Despite the presence of absorbers, there will always be unwanted reflections (stray sig-
nals) in an anechoic chamber that may degrade the quality of the measurements. In antenna
measurements, reflections off walls and support structure can corrupt the probe output, and
thus result in errors in the computed antenna parameters. In scattering measurements, the
stray signals can manifest themselves in several ways. For example, in some RCS measure-
ments the scatterer is illuminated by a compact-range reflector, which is designed to produce
an approximate plane wave in the quiet-zone region. However, stray signals may corrupt the
quiet-zone field and thus result in an incident field that may substantially deviate from the
plane wave. As a consequence, the resulting far field of the scatterer may not represent a
good approximation to the RCS. Furthermore, even when the incident field is a good approx-
imation to a plane wave, the interaction between the scattered field and a support structure
may also contaminate the probe output, thereby reducing the accuracy of the RCS.
To achieve high accuracy of anechoic-chamber results, it is therefore important to have
methods available that can locate stray-signal sources. Planar measurements have been used
over the past 20 years to evaluate and improve the quiet zone of compact-range reflector
3
systems. Repjar and Kremer [1] used (without processing) amplitude and phase plots of
the quiet-zone field to identify and correct imperfections on the reflector. Filtered near-field
imaging methods were explored by Moghaddar and Walton [2] in a controlled experiment
with a metal tape on the reflector to obtain high-resolution images of the tape, which acts as
a strong stray-signal source. Lee and Burnside [3] showed that the field diffracted at the edges
of the reflector can cause significant errors in RCS measurements, and proposed a method
involving a flat plate to evaluate the compact range.1 In an overview article [6], Gupta
discussed various methods in both frequency and time-domains for mapping stray signal
sources in far-field antenna ranges. Chang et al. [7] employed a time-domain measurement
system to achieve a high-quality quiet zone despite the presence of strong stray signals.
Interesting results have been obtained recently by Wittmann and Francis [8] who measured
the quiet-zone field on a spherical surface and then obtained images from spherical expansion
coefficients.
The work discussed in this paper was initiated to evaluate and improve the quality of
the quiet zone of a cylindrical near-field bistatic RCS measurement facility [9], [10], [11]. In
this measurement facility, we use cylindrical near-field measurement techniques (originally
developed for antenna measurements [12], [13], [14]) to determine the bistatic RCS of a
scatterer located in the quiet zone of a compact-range reflector. A measurement of the quiet-
zone field over a plane revealed the presence of strong stray signals that significantly degrade
the quiet zone. The strongest of these stray signals was caused (as we shall demonstrate later)
by complicated multiple-bounce reflections. Moreover, we shall evaluate the effectiveness of
absorbers placed along the edge of a reflector in reducing edge diffractions. We shall employ
both the plane-wave backpropagation and near-field spherical-wave focusing reconstruction
1
To reduce the effect of stray signals in compact-range antenna near-field measurements, Burnside and
Gupta [4] proposed a method that involves measuring the field of the antenna under test in two different
positions. Mitchell [5] extended this method to three or more positions of the antenna under test and
broad-band signals.
4
techniques to process the 2D data collected on a planar surface in the quiet zone of the
reflector.
The paper is organized as follows. In Section 2.1 we present formulas for backpropagating
scan-plane data using a plane-wave spectrum. The backpropagated data (called the image)
reveals the locations of sources between the scan plane and the reflector. The formulation
is well-suited for FFT implementation and is based on the fact that the probe output satis-
fies the Helmholtz equation. In Section 2.2 we present a spherical-focusing formulation for
computing images behind a scan plane (this type of formulation is also known as near-field
focusing). The spherical focusing method can be implemented with FFT-based convolution
algorithms when both data and image points are on planar surfaces. Both formulations
include a spatial filter to isolate selected plane-wave spectrum components.
Section 2.3 compares the two formulations using both an asymptotic evaluation of the
kernel of the plane-wave formulation and numerical examples. In Section 2.4 we use the
physical interpretation of the plane-wave spectrum to design and interpret a filter function
that isolates the strong on-axis field of the reflector. With this component isolated, the
smaller stray signals are much easier to identify.
The theory developed in Section 2 is used in Section 3 to determine the origin of strong
stray signals in the cylindrical near-field bistatic RCS measurement facility. The location of
the stray-signal source is obtained by combining the fast plane-wave formulation with the
flexible spherical-focusing method. A summary is presented in Section 4.
Single-frequency data was collected in the anechoic chamber, so the theory presented in
the paper is formulated in the frequency-domain, and thus the range information contained
in the images is determined solely by the spatial extent of the planar data regions.
5
2 Plane-wave and spherical-wave methods for comput-
In this section we present and compare two formulations for computing backpropagated
time-harmonic electromagnetic fields, which we shall call images. The sources lie in the half
space x < xs < 0 and the resulting electromagnetic field is measured in a scan plane given
by x = 0, −y0 ≤ y ≤ y0 , 0 ≤ z ≤ z0 , as shown in Figure 1. The first formulation uses plane
waves as basis functions and is well suited for FFT implementation. The second formula-
tion uses spherical waves and can be implemented with FFT-based convolution algorithms.
Throughout, e−iωt time dependence is suppressed.
The electric field of a source that is located in the half space x < xs < 0 can be expanded in
terms of a spectrum of plane waves as follows [15], [16]
+∞ +∞
1 Z Z
E(x, y, z) = T(ky , kz )ei(ky y+kz z+γx) dky dkz , x ≥ xs (1)
2π
−∞ −∞
where the spectrum T(ky , kz ) can be expressed in terms of the field in any plane x = x1 ≥ xs
as
+∞ +∞
1 Z Z
T(ky , kz ) = E(x1 , y, z)e−i(ky y+kz z+γx1 ) dy dz, x1 ≥ x s . (2)
2π
−∞ −∞
Here γ is defined as
q
γ= k 2 − ky2 − kz2 , Re(γ) ≥ 0, Im(γ) ≥ 0 (3)
6
and k is the propagation constant. In principle, these formulas determine the electric field
throughout the source-free half space x ≥ xs from the value of the field in any scan plane
x = x1 ≥ xs . The (ky , kz ) integration extends over both propagating modes (ky2 + kz2 < k 2 )
and evanescent modes (ky2 + kz2 > k 2 ). The evanescent modes decay exponentially away from
a source and their amplitudes are too small to be accurately detectable in a scan plane that
is located more than a few wavelengths away. Therefore, the spectrum can be computed
accurately only for propagating modes.
The spectrum is given in (2) in terms of the electric field in the scan plane. In real
measurements, however, the probe output is known in the scan plane rather than the electric
field itself. We, therefore, present an expression for the probe output in terms of the spectrum
and a dyadic probe receiving characteristic that can be determined from the probe pattern.
As in [16, sec. 6.2], the probe moves without rotation and we introduce a vector probe
output
bp = byp ŷ + bzp ẑ (4)
where byp and bzp are the scalar probe outputs when some reference line fixed to the probe
is parallel to the y and z axis, respectively. (If the probe is an open-ended rectangular
waveguide, the reference line could be chosen parallel to the longest cross-sectional side of
the waveguide.) The dyadic receiving characteristic R̄(ky , kz ) is defined such that the vector
0 0 0
output is bp = R̄(ky , kz ) · T(ky , kz )ei(ky y +kz z +γx ) when the plane wave ei(ky y+kz z+γx) /(2π)
illuminates the probe and a reference point for the probe is at (x0 , y 0 , z 0 ). The dyadic receiv-
ing characteristic can be expressed in terms of the transmitting characteristic and far-field
pattern for the probe if the probe is reciprocal [16, sec.6.2]. If the probe is not reciprocal, the
dyadic receiving characteristic can be expressed in terms of the transmitting characteristic
and pattern of the adjoint probe.
We can use the definition of the dyadic receiving characteristic along with the plane-wave
7
expansion (1) to show that the vector probe output is
Z Z
bp (x, y, z) = R̄(ky , kz ) · T(ky , kz ) ei(ky y+kz z+γx) dky dkz (5)
ky2 +kz2 <k2
where we have limited the region of integration to propagating modes. This expression is
valid when multiple interactions between the probe and the source are negligible (that is,
the field scattered by the probe, rescattered by the surroundings, and returned to the probe
changes the output of the probe only negligibly). It was proven rigorously in [16, sec.6.2.1]
that this probe output satisfies the Helmholtz equation in free-space regions2 . Hence, the
probe output can be expanded in terms of plane waves and backpropagated just like the
electric field in (1) and (2).
We shall employ both a window function w(y, z) and a filter function f (y, z) (see Section
2.4) to the scan-plane data before backpropagating and thus introduce the pre-processed
scan-plane data
+∞
Z +∞
Z
q(y, z) = w(y 0 , z 0 ) f (y − y 0 , z − z 0 ) bp (0, y 0 , z 0 )dy 0 dz 0 (6)
−∞ −∞
+∞ +∞
1 Z Z
Q(ky , kz ) = q(y, z) e−i(ky y+kz z) dy dz. (7)
2π
−∞ −∞
1 Z Z
I(x, y, z) = Q(ky , kz )ei(ky y+kz z+γx) dky dkz , x < 0. (8)
2π
ky2 +kz2 <k2
2
This can be proven in the following way. First show that the operator (∇2 + k 2 ) can be moved inside
the integral signs in (5) in free-space regions. Then use the fact that each plane wave satisfies the Helmholtz
equation (∇2 + k 2 )ei(ky y+kz z+γx) = 0.
8
The convolution theorem shows that the effect of the filter function in (6) can be achieved
by multiplying the spectrum by the Fourier transform
+∞ +∞
1 Z Z
F (ky , kz ) = f (y, z) e−i(ky y+kz z) dy dz (9)
2π
−∞ −∞
+∞
Z +∞
Z
Q(ky , kz ) = F (ky , kz ) w(y, z) bp (0, y, z) e−i(ky y+kz z) dy dz. (10)
−∞ −∞
The near-field data is available only in the finite scan plane, so the y − z integrations in
(6) and (10) extend over the region −y0 ≤ y ≤ y0 , 0 ≤ z ≤ z0 . Note that the filter function
can cause the integrations in (7) to extend beyond the scan plane.
All the plane-wave expansions can be discretized and computed efficiently with a 2D FFT
as follows. First introduce the sampled probe output in the scan plane by the expression
where the sample spacing ∆y and ∆z should be chosen small enough to capture all non-
negligible plane-wave components. To capture all propagating plane waves, ∆y and ∆z
should be less than λ/2. We define wemn to be the window function w(y, z) evaluated at the
same points as bp , and compute the windowed and filtered scan-plane data
N X
M
fd ([m − m0 ]∆y, [n − n0 ]∆z) wem0 n0 b
X
q
e mn = ∆y∆z e 0 0
mn (12)
n0 =1 m0 =1
where q is evaluated at the same points in the scan plane as bp . In (12) fd is the discrete
representation of the filter, which is different from the continuous representation f if the filter
contains delta functions. For example, if f (y, z) = δ(y − ∆y)δ(z), the discrete representation
9
is fd ([m − m0 ]∆y, [n − n0 ]∆z) = δm,m0 +1 δn,n0 , where δmn is the Kronecker delta: δmn = 0 for
n 6= m and δmm = 1.
The spectrum can now be computed from
N X M
∆y ∆z ei(s−1)∆ky y0 X
Q
f
st = e−i(s−1)(m−1)∆ky ∆y e−i(t−1)(n−1)∆kz ∆z q
e mn (13)
2π n=1 m=1
where ∆ky and ∆kz are the step lengths for ky and kz , and s and t are integers. The spectrum
f in (13) is evaluated at (k , k ) = (∆k [s − 1], ∆k [t − 1]) for s = −M/2, ..., 0, ..., M/2 − 1
Qst y z y z
and t = −N/2, ..., 0, ..., N/2 − 1, with M and N even. To fully exploit the FFT, the step
lengths should be chosen so that ∆ky ∆y = 2π/M and ∆kz ∆z = 2π/N . Zero padding can
be used to artificially increase M and N and thereby reduce ∆ky and ∆kz .
The image is finally obtained from the expression
M/2−1 N/2−1
∆ky ∆kz
ei(s−1)∆ky y ei(t−1)∆kz z eixγst Ust Q
X X
I(x, y, z) = f
st (14)
2π s=−M/2 t=−N/2
where
q
γst = k 2 − (s − 1)2 ∆ky2 − (t − 1)2 ∆kz2 (15)
is the propagation constant in the x direction, and Ust = 1 for (s − 1)2 ∆ky2 + (t − 1)2 ∆kz2 < k 2
and zero otherwise. The factor Ust is inserted to ensure that evanescent waves are not
included, even though the summation typically covers the rectangular area −k < ky < k,
−k < kz < k.
Assume that the number of image points is expressed as the product Nx Ny Nz , where
Ny > M and Nz > N . With zero padding the number of operations required by the 2D
FFT to compute the summations in (13) for Ny values of s and Nz values of t is on the
order of Ny Nz log2 (Ny Nz ). The number of operations required by the 2D FFT to compute
the formula (14) for fixed x and Ny values of y and Nz values of z is also on the order of
10
Ny Nz log2 (Ny Nz ). Hence, it takes on the order of Nx Ny Nz log2 (Ny Nz ) operations to compute
the image at Nx Ny Nz points in space.3
The formulas (2), (7), and (10) for the spectrum assume that the electric field or the
probe output is known on the planar surface. More general formulas express the spectrum
in terms of Huygens’ sources an arbitrary surface[16, Sec. 3.2.2]. However, these general
formulas cannot exploit the FFT and involve both tangential electric and magnetic fields.
Moreover, if the data surface is planar but the image surface is non-planar, the FFT cannot
be fully exploited even though the formulas (2), (7), and (10) hold. Therefore, we recommend
that the plane-wave method be used only when both data and image surfaces are planar.
If either the data or image surface is non-planar, we recommend that the spherical-focusing
method of the following section be used.
The image obtained with spherical focusing (also known as near-field imaging) is given by
+∞
Z +∞
Z
I(x, y, z) = q(y 0 , z 0 ) R e−ikR dy 0 dz 0 , x < 0 (16)
−∞ −∞
q
where R = x2 + (y − y 0 )2 + (z − z 0 )2 and q is the windowed and filtered scan-plane data
given by (6). We can discretize (16) to get
M X
N
e mn Rmn e−ikRmn
X
I(x, y, z) = ∆y∆z q (17)
m=1 n=1
where
q
Rmn = x2 + (y − [m − 1]∆y + y0 )2 + (z − [n − 1]∆z)2 (18)
3
Without the FFT, the number of operations would be on the order of Nx Ny Nz M N .
11
and q
e mn is given by (12).
The formula (17) determines the image at an arbitrary point in space from the data
in the scan plane. However, the spherical focusing formula allows the data to be given on
any surface, as long as Rnm denotes the actual distance from the data point to the image
point. We shall use this property of the spherical focusing formula in Section 3 to compute
a second-order image from a first-order image on a reflector surface (the first-order image on
the reflector surface is obtained from scan plane data).
Since (16) is a convolution integral, the image can be computed efficiently with the FFT
when data and observation points are both at a regular grid on a planar surface. To see how
this can be done, assume that the observation points are
where x1 , y1 , and z1 determine the observation plane. For simplicity, the increments for
the observation points ∆y and ∆z are the same as the increments for the data points. The
integers p and q are positive and determine the individual observation points. With the
image at the observation points in (19) denoted by eIpq , and
q
R
e =
pq x21 + (p∆y + y0 − y1 − M ∆y)2 + (q∆z − z1 − N ∆z)2 (20)
M X
N
−ikRM +p−m,N +q−n
X
I
e
pq = ∆y∆z q
e mn R
e
M +p−m,N +q−n e , (21)
m=1 n=1
which can be computed with the FFT-based Matlab function conv2 as follows. Let a be the
data matrix of size M × N that contains the measured data in the scan plane. Let b be the
12
kernel matrix of size 2M × 2N that contains R e−ikR . Then the 2D convolution
N X
X M
c(p, q) = a(m, n) b(M + p − m, N + q − n) (22)
n=1 m=1
is determined from
where d = conv2(a, b). The data matrix a can be zero-padded to compute the image in a
plane of arbitrary size.
The number of operations required to compute the image with the spherical focusing
formula (17) at Nx Ny Nz arbitrarily-located points in space is on the order of Nx Ny Nz M N .
If both data and observation points are at regular grid points on parallel planar surfaces, the
FFT-based convolution algorithm requires on the order of Nx Ny Nz log2 (Ny Nz ) to compute
(21) at Nx Ny Nz observation points. Hence, for planar data and observation surfaces, the
spherical focusing method is as fast as the plane-wave backpropagation method.
The formulas (7)-(8) for plane-wave backpropagation are very different from the formula (16)
for spherical focusing. Indeed, the values for the images of the two methods are different.
Nevertheless, we shall now show that the two methods employ kernels that are asymptotically
similar and therefore produce almost identical images in many cases.
To compare the two methods we cast the plane-wave formulas into the same form as
the spherical-focusing formulas. Inserting the spectrum (7) into the formula (8) we get the
13
following expression for the image
+∞
Z +∞
Z
I(x, y, z) = q(y 0 , z 0 ) K(x, y, z, y 0 , z 0 ) dy 0 dz 0 , x < 0 (24)
−∞ −∞
1 Z Z
0 0
K(x, y, z, y 0 , z 0 ) = ei(ky [y−y ]+kz [z−z ]+γx) dky dkz
4π 2
ky2 +kz2 <k2
k
1 Z √
= u eiux J0 k 2 − u2 ρx du. (25)
2π
0
q
Here ρx = (y − y 0 )2 + (z − z 0 )2 and J0 is the Bessel function of the first kind of order zero.
The formula (25) shows that the kernel is a function of just two spatial variables: x and the
axial distance ρx . The corresponding kernel for spherical focusing is simply R e−ikR , where
q
R= x2 + ρ2x .
The images we compute are many wavelength away from the points where the data is
collected, so the distance R is much greater than a wavelength (the wavelength is 3 cm in
Section 3). Therefore, an asymptotic approximation for R → ∞ of the kernel is appropriate.
First assume that both |x| and ρx approach infinity. Then the large-argument approxi-
mation for the Bessel function
s
2
J0 (Z) ∼ cos(Z − π/4) (26)
πZ
√
s
1 Z
2
K(x, y, z, y 0 , z 0 ) ∼ u eiux √ cos k 2 − u2 ρ − π/4 du.
x (27)
2π π k 2 − u 2 ρx
By expressing cosine as a sum of two exponentials, and noting that for x < 0 only the
14
exponential with minus i produces a stationary point, gives
s
0 0 eiπ/4 Z 2 √
−iR(u|x|/R+ k2 −u2 ρx /R)
K(x, y, z, y , z ) ∼ u √ e du. (28)
4π π k 2 − u 2 ρx
ik|x| e−ikR
K(x, y, z, y 0 , z 0 ) ∼ . (29)
2πR2
The formula (29) was derived under the assumption that both |x| and ρx are large. By
employing a small-argument approximation for the Bessel function, it is easily shown also to
be a valid expression when ρx is small. Hence, the kernel for plane-wave backpropagation is
asymptotically given by (29) as R → ∞ except when |x|/R is close to zero.
Figure 2 shows the relative error in % of the asymptotic formula (29) when the wavelength
is 3 cm, and the spatial variables cover the region 0 < ρx < 3 m, −3 m < x < −0.3 m. The
error is less than 1% except when |x|/R is small, corresponding to the stationary point
u0 = k|x|/R being close to zero. For the image locations that will be considered in Section
3, the errors of the asymptotic formula (29) are negligible.
For our purposes we can therefore compare the kernels for plane-wave backpropagation
and spherical focusing by comparing (29) with R e−ikR . The two kernels have the same phase
behavior, as would be expected since they are based on the same wave equation. However,
the amplitudes of the two kernels are different: k|x|/(2πR2 ) for plane-wave backpropagation,
and R for spherical focusing. First we consider images of a point source. If the distance
between the point source and the scan plane is greater than the dimensions of the scan plane,
15
none of the amplitudes vary much during the image calculation. Hence, the two methods
produce similar images.
To verify this observation, we consider a numerical example involving 10 GHz synthetic
data in the scan plane, x = 0, −1.25m < y < 1.25 m, 0 < z < 2.5 m, sampled at ∆y = ∆z =
1.5 cm. The first data set is generated by a point source located 3 m directly behind the scan
plane at (x, y, z) = (−3 m, 0.5 m, 0.5 m). Figure 3 shows the normalized images obtained for
this point source from the plane-wave backpropagation formula (left) and from the spherical
focusing formula (right). Despite the differences between the two sets of formulas, the two
images are indeed indistinguishable. The second data set is generated by a point source
located off to the side of the scan plane at (x, y, z) = (−3 m, 3 m, 0.5 m). Figure 4 shows
the two corresponding images, which are also indistinguishable.
If the source has a considerable spatial extent, the amplitude factors can vary between
different regions of the image, and thus boost or attenuate parts of the image. To illustrate
this phenomenon, consider an extended source that consists of two equal-amplitude point
sources: Point source A is located at (x, y, z) = (−3 m, 0.5 m, 0.5 m) and point source B at
(x, y, z) = (−3 m, 3 m, 0.5 m). The resulting images are like the normalized images of Figures
3 and 4, except for the amplitudes. Since the spherical focusing method compensates for
1/R decay, the images of the two point sources will have almost equal amplitude. The factor
k|x|/(2πR2 ) of the plane-wave backpropagation kernel causes the amplitude of the image of
point source A to be about 2.3 times greater than amplitude of the image of point source B.
Hence, for a large constant-amplitude source that extends out to about 45◦ from the x
axis, the image values near the x axis will be about twice as large as the image values near
the edge for the plane-wave formulation. If desired, one can compensate for this variation in
the amplitude across an image with a filter. However, this weak variation of image values is
often negligible compared to the effect of the filters that are applied to the scan-plane data.
For the reflector surface and the other objects of interest in Section 3, the images cre-
16
ated with the two methods are nearly indistinguishable. We shall therefore use plane-wave
backpropagation whenever the data and image surfaces are parallel planes, and spherical
focusing when either the data or image surface is nonplanar.
The formulas in the previous sections contain filters that can enhance or reduce the effect
of selected components of the scan-plane data. We shall use this capability to eliminate
a strong plane-wave component from a compact-range reflector that propagates in the x
direction. This component can obscure the weaker features caused by stray signals. It is the
weaker features that we want to identify and eliminate to achieve a better quiet zone.
Since the main component of the plane wave is at (ky , kz ) = (0, 0), we see from (10)
that the filter should be designed to have a null in this direction. Hence, we should demand
F (0, 0) = 0. There are many filters with this property. We chose a filter that has a simple
representation in the spatial domain:
f (y, z) = δ(y − ∆y)δ(z) + iδ(y)δ(z − ∆z) − δ(y + ∆y)δ(z) − iδ(y)δ(z + ∆z) (30)
where ∆y and ∆z are the spatial sampling rates in the scan plane. The filter function (30)
is a 2D generalization of a 1D filter function used in [2]. The filtered scan-plane data at a
given point is simply a linear combination of the scan-plane data at the four closest points.
The Fourier transform of this filter function is
sin kz ∆z i sin ky ∆y
F (ky , kz ) = − (31)
π π
which clearly satisfies F (0, 0) = 0. Figure 5 shows the amplitude of the spectrum of this filter
function for ∆y = ∆z = λ/2. The filter is evaluated as a function of the spherical coordinate
17
angles θx and φx , where cos θx = −x̂ · k/|k|, and φx is the corresponding azimuthal angle in
the y − z plane. Note that the amplitude of the filter function is almost independent of the
azimuthal angle φx , especially for small values of θx .
stray-signal sources
Data was collected in the scan plane located at the center of the quiet zone of the compact-
range reflector shown in Figures 6, 7 and 20. This reflector is an offset paraboloidal segment
with a total of eighteen serrations in its original configuration. The four serrations along
the bottom edge have been removed as shown in Figure 20. The vertex of the paraboloid
lies about 2 cm above the center of the bottom edge. Consequently, the feed at the focal
point is exposed to the collimated field from this region. The upper eight edges of the
side serrations are constant-radius curves centered on the vertex, while the lower edges are
straight cuts pointing toward the vertex. The curved serration edges and the bottom edge
have been covered with absorber in strategic locations, so that the curvature can here only
be seen on the lowest serrations. The purpose of the absorber will be described shortly. The
reflector is an older design, approximately 4.4 m wide and high including serrations, with
an advertised quiet zone of 1.2 m both in width and height. A dual-polarized, orthomode
feed, with a 3 dB-beamwidth of approximately 50 at λ = 3cm located at the focal point of
the reflector, the reflector, which in turn creates a plane wave propagating in the positive
x direction. The total field in the quiet zone is measured by mechanically scanning an
identical, dual-polarized orthomode probe over the scan plane, just above the origin of the
coordinate system in Figure 6. Data is taken on the planar surface in a square-wave fashion.
To save additional time, the field is sampled while the probe is in continuous linear up or
18
down motion in the z direction. At the end of each sweep the vertical scanner mount steps
in the y direction to the next horizontal position. Data points are spaced 1 cm apart in both
y and z directons. The fully polarimetric instrumentation includes an HP8530 Microwave
Receiver. All transmitter/receiver functions, probe motion and data acquisition are under
computer control for long-term unattended operation. Some preliminary near-field bistatic
RCS measurements have been described in [9], [10], and [11].
The planarity of the field in the quiet zone, given an ideal paraboloidal reflector surface,
is typically degraded by stray radiation, the most likely source of which is serration-edge
diffraction. To be effective, serration edges have to be more than a few wavelengths long.
Rays from the feed, striking an edge at an angle β relative to the edge’s local tangent, lead
to a conical bundle of diffracted rays. The cone of half angle β has the tangent as its axis
and the incident ray continuing on as one of its diffracted rays. Care must be taken to orient
the serration edges in such a way as to minimize the number of diffracted rays entering the
quiet zone. It therefore becomes evident that the curved serration edges of the reflector
under study are less than optimal. A simulation by Lee and Burnside [19] evaluates the
effects of straight-edge serrations on diffracted signal levels in the quiet zone of a square
reflector. After selecting the length and number of serrations per reflector side, each of their
orientations can be adjusted such that the maximum cross section in the quiet zone remains
free of diffracted rays. As a first-order approximation, the edges must point toward the
center of the reflector. This is most closely achieved with many long and narrow serrations.
However, to limit their number in order to hold down construction cost as well as tip and
root diffraction, a compromise must be made.
When measuring field quality in the quiet zone, it is important to use a field probe with
sufficient beamwidth in order not to discriminate against edge-diffracted radiation. This
would make the quiet zone look more uniform or larger than it really is. The scan plane of
size 2.5 m height and 2.34 m width is located 6 m from the vertex of the reflector (Figure
19
6). While the probe is at y = 1.17 m, it receives stray radiation from the tip of the opposite
serration (y = - 2.2 m) at an angle of φ = 29.1o and vice versa. Similar values can be derived
in the vertical direction. This necessitates the use of the wide-beamwidth orthomode probe.
A photo of the pyramidal-horn probe originally used to specify the quiet-zone parameters of
this reflector suggests a 3 dB-beamwidth of 13.6 x 10.7.
After installation and adjustment of the reflector system with its full complement of ser-
rations, preliminary quiet-zone scans were conducted. The planar scanner pictured in Figure
15 can be reconfigured as a cylindrical near-field scanner, which is its primary application.
For initial measurements, only the vertical scanner itself is absorber covered, but much of
the additional hardware is not. The latter is partially shielded against direct radiation from
the reflector by the transmitter/feed housing. Figure 8 displays the raw scan-plane data
recorded with the probe set to measure the horizontal electric field component. The plot
is linear in amplitude and shows a general taper from the center toward the edges. This is
due to the beam pattern of the feed and the variable range attenuation between feed and
reflector. The circular red and yellow area of 1.2 m diameter is roughly the region where
the target under test can be placed. A multitude of ripple patterns overlay the basic beam
pattern. The predominant one indicated with arrow in the plot is a ”smiley” pattern which
indicates a degraded quiet zone. We shall proceed to identify the cause of these and other
ripples.
Suppressing the planar field with the filter in (30) leads to the residual scan-plane data in
Figure 9. The plane-wave component is now significantly reduced while two slanting regions
of higher amplitude marked with red and white ellipses dominate. To further investigate the
origin of these residual field components plus any others not as clearly visible in Figure 9,
plane-wave backpropagation yields the four different images in Figure 10 in planes parallel
to the scan plane at x = -2 m, x = -2.75 m, x = -4 m, and x = -5.75 m (recall that the scan
plane is at x = 0). The image at x = -2 m shows that the two features marked in Figure
20
10 have merged together over the 2 meter distance. Scattered field from the top edge of the
transmitter/feed enclosure and spillover radiation from the feed are also visible. The top
edge of the enclosure is at x = -2.14 m and the feed is at x = -2.34 m. The feed is visible to
the probe in the upper part of its travel. The overall outline of the reflector is discernable at
a somewhat reduced size. At x = -2.75 m all the previous features still exist but are displaced
from their previous locations. In particular, the two regions associated with the ellipses have
separated again. Actually, the paths of the two sets of stray signals cross approximately at
the position of the feed. At x = -4 m and at x = -5.75 m these features move even further
apart. The plane at x = -5.75 m is located immediately in front of the reflector. Here the
reflector image has grown to nearly full size and clearly shows diffracted signals originating
from the curved edges of all side serrations prior to absorber treatment. Other stray sources
are also apparent. Backpropagated radiation from the feed had increased in size at x = -4 m
with a noticeably lower position and largely disappeared at x = -5.75 m below the plotted
area.
Next we compute an image on the reflector. The reflector surface is a paraboloid given
in the coordinate system of Figure 6 by the equation
Figure 11 shows the amplitude of the image on the reflector surface obtained with the
spherical-focusing formula, (17). The curved edges of the serrations are clearly visible, as
are the two vertical features (enclosed by ellipses) seen in Figure 9. The feature on the right
bounded by the white ellipse in Figure 11 is related to the feature on the left bounded by
the white ellipse in the filtered scan-plane data. A complementary statement holds for the
features bounded by the red ellipses.
In part, the image in Figure 11 has a simple physical interpretation: non-uniform currents
21
on the reflector surface cause the field in the scan plane to deviate from a perfect plane
wave. As expected from high-frequency scattering theory, the edges of the reflector create
strong non-uniform currents. High-frequency scattering theory cannot, however, explain the
currents associated with the features enclosed inside the ellipses, since these currents are
located in regions where the reflector surface is perfectly smooth. Nor can these features be
explained from the incident field of the feed, which has almost spherical symmetry. Hence,
these currents must be caused by additional sources that illuminate the reflector.
We have plotted the phase of the image on the reflector surface in Figure 12. The phase
has a very regular pattern in the regions bounded by the ellipses, where the non-uniform
current has a large amplitude. Hence, we can expect the non-uniform currents to be caused
by a simple source distribution.
To determine the origin of these non-uniform currents, we employ spherical focusing on
only that part of the reflector-surface image in Figures 11 and 12, which lies inside the white
ellipse. The amplitude of the resulting image is shown in Figure 13 in the plane y = 1 m,
along with projections of the scan plane and the reflector. The image focuses at a point
about 0.3 m below the bottom edge of the scan plane. Figure 14 displays a slice of the image
in the orthogonal plane z = -0.3m and shows that the peak is at the end of the scan plane.
Hence, the source of the non-uniform current inside the white ellipse of Figures 11 and 12 is
located about 0.3m below the lower corner of the scan plane.
The probe support structure in Figure 15 consists of dual horizontal rails spaced 30.48
cm apart, on which the column for the scanning probe moves. As shown, the probe sits
at the corner of the scan plane (y, z) = (-1.25 m, 0 m). The central part of the rails is
shielded against direct radiation from the reflector by the absorber-covered transmitter/feed
enclosure. Only the two ends of the rails are illuminated by the planar wave. Figure 16
schematically shows the ray paths that create one half of the observed pattern in Figure 9.
The exposed end of the rail is illuminated by the planar field and reflects energy back to the
22
reflector. This energy reflects off the reflector thus creating the signal component within the
white ellipse. Similarly, the signal in the red ellipse is generated by the opposite end of the
rail.
Completing the scanner assembly by installing the absorber skirt thereby shielding the
rails (Figure 17), results in the disappearance of the smiley feature in the scan-plane data of
Figure 18. To further verify that we have eliminated reflections from the rails, we acquired
new scan-plane data and recomputed the image on the reflector surface from the filtered scan-
plane data. The image in Figure 19 confirms the absence of the high amplitude currents
in the regions bounded by the ellipses in Figure 11. Hence, the absorber skirt has indeed
significantly reduced the reflections from the rails.
Next we address the problem of stray radiation from the periphery of the reflector. In
Figure 19 the curved edges of the higher side serrations are strong contributors. The side
serrations are an integral part of the reflector surface, while the four top and four bottom
serrations are bolted on. The edges of the bolted-on serrations do not show in Figure 19,
however, the transitions between reflector surface and these serrations do. There are slight
alignment discontinuities and gaps between the two surfaces as seen along the top. The
strong bottom center response is due to a slightly upward-tilted edge of this region of the
reflector. The strong bottom side responses are caused by rays grazing the sides of pyramids
along the upper edge of the absorber covering the reflector support structure. These are
then redirected by the reflector into the quiet zone.
We decided to remove the four bottom serrations and to mitigate the effects of stray radi-
ation sources by installing custom shaped absorber in critical locations. Pyramids cut from
18 inch absorber panels were quartered and mounted along the troublesome serration edges,
covering the edges themselves and creating smaller, more numerous metallic ”serrations”
pointing toward the reflector center. Figure 20 is a photograph of the final configuration.
Figure 21 in comparison with Figure 18 demonstrates the improvement of the quiet-zone
23
scan achieved with these modifications. For example, the concentric narrow ripple struc-
ture due to backward radiation and scattering from the feed has been greatly diminished.
Backpropagating the scan-plane data to the reflector surface in Figure 22 reveals that most
stray radiation sources there have been eliminated or substantially reduced. The top-edge
discontinuities were left untreated for the time being.
As stated earlier, the reflector is not of virtual-vertex design, as the paraboloid vertex
is on its surface. Therefore, as one can see from Figure 6, the feed and the feed enclosure
intercept the lower part of the collimated field and scatter in the direction of the quiet zone.
Furthermore, it makes it difficult to implement an effective fence against backlobe radiation,
directly from the feed to the quiet zone and indirectly scattered off the housing into the quiet
zone, as this proposed fence would itself be illuminated by the collimated field producing
additional scattered field. A virtual-vertex design would have allowed an extended vertical
fence behind the feed for better backlobe-radiation suppression. We implemented a thin
forward tilting wedge fence over the feed for some improvement. The image at x = -2 m in
Figure 10 identifies the feed as a source of stray radiation in the scan plane along with the
top edge of the enclosure. The remaining ripple structure in Figure 21, largely concentric
rings centered on an origin below the lower edge of the plot, suggests the same. The feed as a
point source would lead to a concentric interference pattern with the planar field in the quiet
zone, with the ring spacing narrowing in upward direction. The actual situation involves four
sources: (i) diffraction from a 1.22 m long top ridge of 4-inch absorber pyramids covering
the enclosure and illuminated by the planar field, (ii) diffraction by the same structure
illuminated by backward radiation from the feed, (iii) direct backward radiation from the
feed, and (iv) scattering of the plane field off the feed. Inspection of the ripple curvature
in Figure 21 suggests a source wider than a point source at the feed location. We did
not investigate this further, though the techniques presented can be employed to achieve
additional improvement in the quiet zone.
24
4 Summary
Acknowledgements
25
References
[1] A.G. Repjar and D.P. Kremer, “Accurate evaluation of a millimeter wave compact
range using planar near-field scanning,” IEEE Transactions on Antennas and Propa-
gation, vol. AP-30, pp. 419-425, May 1982.
[2] A. Moghadder and E.K. Walton, “Imaging of low level signals in a compact range,”
Proceedings of Antenna Measurement Techniques Association (AMTA) Annual Meet-
ing and Symposium, Philadelphia, PA, pp. 14.3-14.8, October 1990.
[3] T.H. Lee, W.D. Burnside, “Stray signal requirements for compact range reflectors based
on RCS measurement errors,” IEEE Transactions on Antennas and Propagation, vol.
AP-39, pp. 1193-1202, August 1991.
[4] W.D. Burnside and I.J. Gupta, “A method to reduce stray signal errors in antenna
pattern measurements,” IEEE Transactions on Antennas and Propagation, vol. AP-42,
pp. 399-405, March 1994.
[5] R.L. Mitchell, “On the reduction of stray signal errors in antenna pattern measure-
ments,” IEEE Transactions on Antennas and Propagation, vol. AP-43, pp. 629-630,
June 1995.
[6] I.J. Gupta, “Stray signal source location in far-field antenna/RCS ranges,” IEEE An-
tennas and Propagation Magazine, vol. 46, pp. 20-29, June 2004.
[7] D.C. Chang, C.H. Liao, and C.C. Wu, “Compact antenna test range without reflector
edge treatment and RF anechoic chamber,” IEEE Antennas and Propagation Maga-
zine, vol. 46, pp. 27-37, August 2004.
[8] R.C. Wittmann and M.H. Francis, ”Test-chamber imaging using spherical near-field
scanning,” Proceedings of Antenna Measurement Techniques Association (AMTA) An-
26
nual Meeting and Symposium, 2001 Antenna Measurement Techniques Association
Conference, Denver, CO, October 2001.
[9] R.A. Marr, T.B. Hansen, U.W.H. Lammers, R.V. McGahan, and T.J. Tanigawa, ”Far-
field bistatic RCS from near-field measurements,” Proceedings of Antenna Measure-
ment Techniques Association (AMTA) Annual Meeting and Symposium, 2003 Antenna
Measurement Techniques Association Conference, Irvine, CA, October 2003.
[10] T.B. Hansen, R.A. Marr, U.H.W. Lammers, T.J. Tanigawa, and R.V. McGahan,
“Bistatic RCS calculations from cylindrical near-field measurements, Part 1: Theory,”
Submitted to IEEE Transactions on Antennas and Propagation.
[11] R.A. Marr, U.H.W. Lammers, T.B. Hansen, T.J. Tanigawa, R.V. McGahan, “Bistatic
RCS calculations from cylindrical near-field measurements, Part 2: Experiments,”
Submitted to IEEE Transactions on Antennas and Propagation.
[12] W.M. Leach and D.T. Paris, “Probe compensated near-field measurements on a cylin-
der,” IEEE Transactions on Antennas and Propagation, vol. AP-21, pp. 435-445, July
1973.
[13] D.T. Paris, W.M. Leach, Jr., and E.B. Joy, “Basic Theory of probe-compensated near-
field measurements,” IEEE Transactions on Antennas and Propagation, vol. AP-26,
pp. 373-379, May 1978.
27
[16] T.B. Hansen and A.D. Yaghjian, Plane-Wave Theory of Time-Domain Fields, New
York: IEEE Press, 1999.
[17] N. Bleistein and R.A. Handelsman, Asymptotic Expansions of Integrals, New York:
Dover, 1986.
[18] H. Weyl, “Ausbreitung elektromagnetischer Wellen uber einem ebenen Leiter,” Ann.
Physik, vol. 60, pp. 481-500, 1919.
[19] T. H. Lee and W. D. Burnside, ”Performance trade-off between serrated edge and
blended rolled edge compact range reflectors,” IEEE Transactions on Antennas and
Propagation, vol. AP-44, pp. 87-96, January 1996.
28
Figure Captions
1. The scan plane x = 0, −y0 ≤ y ≤ y0 , 0 ≤ z ≤ z0 in which the data is known. The sources are
located in the half space x < 0.
2. The relative error in % of the asymptotic expression for the kernel of the plane-wave backpropa-
gation method. The wavelength is 3 cm.
3. The image of a point source located at (x, y, z) = (−3 m, 0.5 m, 0.5 m).
13. The amplitude of the image in the x-z plane at y = 1 meters (side view).
14. The amplitude of the image in the x-y plane at z = -0.3 meters (top view).
15. A photo of the feed, feed enclosure, the rail that supports the probe, and a white ball placed at
the center of the image in Figures 9 and 10.
16. A schematic of the mechanism that generates the smiley pattern in the scan plane. One end of
the probe rail (marked by the white ball) scatters significant energy back to the reflector, which
re-reflects that energy towards the opposite side of the scan plane and thereby creates one side of
29
17. A photo of the feed and feed enclosure. The rail that supports the probe is now covered by
absorbers.
18. The amplitude of the raw scan-plane data. The rail that supports the probe is covered by absorbers.
19. The amplitude of the image at the reflector surface. The rail that supports the probe is now
covered with absorbers.
21. The amplitude of the raw scan-plane data. Absorbers are now placed along the edges.
22. The amplitude of the image on the reflector surface. Absorbers are now placed along the edges.
30
z
Scan
plane
Sources
Figure 1: The sources located in the half space x < xs < 0 behind the scan plane in
which the field is measured.
3
9
2.5 8
Relative error in %
7
2
ρx (meters)
1.5 5
4
1
3
0.5 2
1
0 0
-3 -2.5 -2 -1.5 -1 -0.5
x (meters)
Figure 2: The relative error in % of the asymptotic expression for the kernel of the
plane-wave backpropagation method. The wavelength 3 cm.
0.6 1
0.9
0.8
0.55
0.7
z (meters)
0.6
0.5 0.5
0.4
0.45 0.3
0.2
0.1
0.4 0
0.4 0.45 0.5 0.55 0.6 0.4 0.45 0.5 0.55 0.6
y (meters) y (meters)
0.6 1
0.9
0.8
0.55
0.7
z (meters)
0.6
0.5 0.5
0.4
0.45 0.3
0.2
0.1
0.4 0
2.9 2.95 3 3.05 3.1 2.9 2.95 3 3.05 3.1
y (meters) y (meters)
Figure 4: Image of a point source at (x,y,z)=( −3m, 3m, 0.5m ) obtained with
plane-wave backpropagation (left) and spherical focusing (right).
z θx= 30o
90
0.5
60 θx = 5o
120 θx = 1o
150 0.25 30
y
180 0
210 330
240 300
270
Reflector 2.34 m
Scan
plane
2.50 m
Feed
x
3.66 m
Axis of parabola
Feed
2.16 m 2.29 m
1.96 m enclosure
Floor
Figure 7: Three-dimensional views of the reflector, focal point, and scan plane.
Axis units are meters.
2.5 0.38
0.36
2
z (meters) 0.34
0.32
1.5
0.3
1 0.28
Smiley
pattern 0.26
0.5
0.24
0.22
0 -1 -0.5 0 0.5 1
y (meters)
2.5
0.07
2
0.06
z (meters)
0.05
1.5
0.04
1 0.03
0.02
0.5
0.01
0 -1 -0.5 0 0.5 1
y (meters)
4 4
3 3
z (meters)
z (meters)
2 2
1 1
0 0
-1 -1
-2 -2
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
y (meters) y (meters)
4 4
3 3
z (meters)
z (meters)
2 2
1 1
0 0
-1 -1
-2 -2
-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
y (meters) y (meters)
3
z (meters)
-3 -2 -1 0 1 2 3
y (meters)
5 3
4 2
1
3
Radians
z (meters)
0
2
-1
1
-2
0
-3
-3 -2 -1 0 1 2 3
y (meters)
Figure 12: The phase of the image on the reflector surface.
Scan
plane
z (meters)
Reflector
Floor
x (meters)
Figure 13: The amplitude of the image in the plane y = 1 meters (side view).
1.5
1
y (meters)
0.5
0
y
-0.5
Scan
-1
plane
-1.5
Reflector
-2
-6 -5.5 -5 -4.5 -4 -3.5 -3 -2.5 -2 -1.5 -1 -0.5 0 0.5 1
x
x (meters)
Figure 14: The amplitude of the image in the plane z = −0.3 meters (top view).
Image
center
Probe
Feed
Rail
Figure 15: A photo of the feed, the feed enclosure, the rail that supports the probe,
and a white ball placed at the center of the image.
Feed
Figure 16: A schematic of the mechanism that generates the smiley pattern in the scan
plane. One end of the probe rail (marked by the white ball) scatters significant energy
back to the reflector, which re-reflects that energy towards the opposite side of the
scan plane and thereby creates one side of the smiley pattern.
Probe
Feed
Absorber
covering
the rail
Figure 17: A photo of the feed, the feed enclosure. The rail that supports the probe
is now covered with absorbers.
2.5 0.38
0.36
2
0.34
z (meters)
1.5 0.32
0.3
1 0.28
0.26
0.5
0.24
0.22
0 -1 -0.5 0 0.5 1
y (meters)
3
z (meters)
-3 -2 -1 0 1 2 3
y (meters)
0.36
2
0.34
z (meters)
1.5 0.32
0.3
1 0.28
0.26
0.5
0.24
0.22
0 -1 -0.5 0 0.5 1
y (meters)
3
z (meters)
-3 -2 -1 0 1 2 3
y (meters)
Figure 22: The amplitude of the image on the reflector surface.