You are on page 1of 8

IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 54, NO.

11, NOVEMBER 2006 3541

A Microwave Tomography Approach for a


Differential Configuration in GPR Prospecting
Raffaele Persico and Francesco Soldovieri

Abstract—The capabilities of a microwave inverse scattering al- is known from the lot of papers published within the frame-
gorithm for ground penetrating radar (GPR) data processing are work of the diffraction tomography (DT). For example, in [8]
analyzed. The algorithm is based on the Born approximation (BA) a multibistatic configuration on a lossy half space has been con-
and exploits a differential measurement configuration. This con-
sidered, in [13] a multistatic (single view and multiview) con-
figuration exploits measurements achieved as the difference of the
electrical fields gathered in two symmetrical positions with respect figuration on a half space has been dealt with, in [14] a mul-
to the source. We will discuss how this choice allows to mitigate tistatic (multiview) configuration in free space has been coped
some bad effects of uncertainties on the knowledge of the back- with, in [15] a bistatic configuration in free space (with the pos-
ground scenario but, on the other hand, affects in a negative way sibility to rotate both the object under test and the position of the
the properties of the relevant scattering operator. receiving antenna) has been examined, in [16] and [17] multi-
Index Terms—Born approximation, differential configuration, static multiview configurations in inhomogeoneous media have
inverse scattering, microwave tomography. been considered. Finally (but of course the list could be much
longer), in [6] an explicit comparison among the capabilities of
three different measurement configurations (multiwiew/multi-
I. INTRODUCTION static, single view/multistatic and multibistatic) has been per-
formed in a two dimensional half space geometry by exploiting
F OR a number of applications, inverse scattering algorithms
[1] can be applied to ground penetrating radar (GPR) [2],
[3] data, gathered either in time or in frequency domain. Gener-
suitable spectral algebraic relationship between the data and the
object function to be retrieved.
With regard to the dielectric profiles retrievable in the frame-
ally, a GPR prospecting involves a large investigation domain,
work of a differential configuration, here we propose an anal-
and this makes appealing fast and efficient inversion algorithms.
ysis that follows the same guidelines of [5]–[9]. In particular,
In particular, in this paper, we make use of the Born approxima-
it will shown that a differential configuration allows to achieve
tion (BA) [4]–[9], that often allows to obtain an acceptable lo-
an amount of independent data smaller compared to that avail-
calization and sizing of the buried objects even in cases beyond
able within a more classical multibistatic configuration. This en-
the range of validity of the Born model itself. This is related to
tails a “reduction” of the functional space where the unknown
the fact that an inversion model based on BA qualitatively tends
contrast function can be looked for. Also, we will show that
to “preserve” the support of the unknown function, as it has been
the class of retrievable objects has a meaningful dependence
noted in 9 and 10. To this pros, let us also note that examples of
on the offset between the antennas, whereas this dependence
strong scattering objects retrieved thanks to a BA-based inver-
is somehow weaker in the case of a multibistatic configuration
sion model have been proposed several times [6], [8].
[6]. Therefore, the convenience to choose a differential configu-
In this paper we propose a linear inverse scattering BA-based
ration should be evaluated also by exploiting the a-priori infor-
algorithm able to deal with the differential measurement tech-
mation (if any) about the geometrical features of the targets.
nique in frequency domain, where the datum is made up of the
The paper is organized as follows. In Section II, the formula-
difference between the total field data collected at two symmet-
tion of the problem under BA is given for both multibistatic and
rical points with respect to the source. To our knowledge, a dif-
differential configurations. In Section III the question of the ex-
ferential configuration has been first introduced and analyzed
traction of the scattered field from the total field is investigated.
in [11], [12] where the authors discussed (in time domain) the
The reconstruction capabilities of a BA based inversion model
role of the distance between the antennas in dependence on the
in the case of a multibistatic and a differential configurations
background scenario and on the targets.
are theoretically compared in Section IV by means of DT spec-
As it will be shown, a differential configuration can coun-
tral relationships. Section V is devoted to confirm the theoret-
teract some negative effects of slow spatial variation of the back-
ical expectations of Section IV in realistic conditions thanks to
ground scenario, that arise uncertainty on the incident field.
a singular value decomposition (SVD) analysis of the scattering
However, in general, the measurement configuration has a
operator. In Section VI numerical examples are presented. Con-
meaningful effect on the class of retrievable profiles too, as it
clusions follow.

Manuscript received September 12, 2005; revised April 18, 2006.


The authors are with the Istituto per il Rilevamento Elettromagnetico
dell’Ambiente Consiglio Nazionale delle Ricerche (IREA-CNR), I-80124
II. FORMULATION OF THE INVERSE SCATTERING PROBLEM
Napoli, Italy (e-mail: persico.r@irea.cnr.it).
Color versions of Figs. 3–9 are available online at http://ieeexplore.ieee.org. Fig. 1 depicts the geometry of the problem. The background
Digital Object Identifier 10.1109/TAP.2006.882161 scenario is made up of two homogeneous half spaces. The upper
0018-926X/$20.00 © 2006 IEEE
3542 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 54, NO. 11, NOVEMBER 2006

where is the offset between the source and observation points,


is the source point and, of course, is the observation
point. Moreover.

(2)

is the contrast function, i.e., the relative difference between the


unknown dielectric profile and the background permittivity,
Fig. 1. Geometry of the problem. is the wavenumber in the soil. is the
incident (or unperturbed) field, i.e., the electric field in absence
of any buried scattering object. The expressions of the Green’s
function and of the incident field are given in Appendix A.
For the differential configuration, under BA the datum of the
problem is achieved as

(3)

Fig. 2. Pictorial of multibistatic (2A) and differential (2B) configurations.


The inverse scattering problem consists in retrieving the con-
trast function from the inversion of the integral relationship (1)
in the case of a multibistatic configuration, whereas it consists in
region is made up of vacuum. The lower region shows a rela- reconstructing the contrast function from the inversion of the in-
tive dielectric permittivity , possibly complex to account for tegral relationship (3) in the case of a differential configuration.
losses. The problem at hand concerns the determination of the
(possibly complex) dielectric profile in the probing rect- III. THE PROBLEM OF THE DATA
angular domain D, embedded in the lower region. The source of In electromagnetic inverse scattering the datum is the field
the incident field is a filamentary electric current . Source and scattered from an unknown permittivity profile illuminated by
observation are at the interface within an observation a known source. However, the receivers gather a signal related
domain . With regard to the position of the ob- to the total field present in the observation point. Therefore,
servation points, two possibilities will be considered and com- the problem of extracting the scattered field from the total field
pared. The first one is the multibistatic (or common offset) con- data arises. This is a problem particularly serious in GPR ap-
figuration, customarily exploited in GPR prospecting [3], where plications, because the unavoidable uncertainties on the back-
source and observation points are linked to each other by means ground scenario can make unreliable the subtraction of a numer-
of a fixed offset [see Fig. 2(a)]. The second configuration is de- ically calculated incident field from the measured total field. An-
picted in Fig. 2(b), and we will name it “a differential configura- other idea to achieve the scattered field data could be to subtract
tion.” In this configuration, for each source point, two opposite from the gathered data some other data measured elsewhere, and
(symmetrical) observation points are considered, and the datum specifically in an area where it is a-priori known that no object
is made up by the difference between the two signals. With re- is buried. However, apart from the difficulty to know a-priori
gard to both the considered configurations, the data will be mul- where an “empty” subsoil zone is, in many realistic cases the
tifrequency in a fixed frequency range . Time background scenario shows “slowly variable” (with respect to
dependence is assumed. the wavelength) spatial characteristics, and therefore the inci-
Under the linear model provided by BA [4]–[9], the scattered dent field in the current observation point is different from the
field within a multibistatic configuration is given by [6]: incident field measured elsewhere. Incidentally, there is no the-
oretical error with assuming a “mathematical” background that
does not vary with the observation point, but this amounts to in-
clude the inhomogeneity of the “physical” background scenario
in the contrast function to be retrieved, which can cause severe
problems, as described, e.g., in [18]. The conclusion is that only
(1) in very favorable cases the scattered field can be retrieved by
PERSICO AND SOLDOVIERI: A MICROWAVE TOMOGRAPHY APPROACH 3543

subtracting a numerically calculated, or off-line measured, inci- IV. PROPERTIES OF THE INVOLVED OPERATOR:
dent field from the available total field data. In GPR prospecting, DIFFRACTION TOMOGRAPHY
a common technique to extract the scattered field from the total In this section we analyze some relevant properties of the
field is that performed by gating the first part of the received two linear integral relationships (1) and (3), that correspond to
signal [19] in time domain. the multibistatic and differential configurations, respectively. In
However, this operation does not lead to a “perfect cancel- particular, we will focus on the differences between the class of
lation” of the incident field from the data, and in particular it retrievable profiles in the two cases, by means of DT based con-
entails that the shallower layers of the soil cannot be reliably siderations. This analysis requires the simplifying hypotheses
reconstructed. In particular, one substantially renounces to re- that the soil is lossless and that the buried objects are deep
trieve objects within a depth of the order of [3] enough. The theoretical expectations “gained” in this analysis
will be then confirmed in the next section by means of a numer-
(4) ical SVD based analysis performed in the more realistic situa-
tion of lossy soil.
At the moment, let us consider a multibistatic configuration,
where is the propagation velocity of the light in free space and let we substitute in (1) the expressions of the Green’s func-
and is the frequency band of the signal. This occurs because tion and of the incident field as given in Appendix A. After a
represents the spatial “extent” of the probing pulses. number of passages (reported in [6]), BA leads to a spectral al-
In frequency domain, to the best of our knowledge, there is gebraic relationship between the contrast and the scattered field.
still no popular technique for the extraction of the scattered field. This relationship is given by
Therefore, if one has at disposal data in frequency domain, at the
moment the only available choice seems a gating in time domain (6)
after Fourier back-transforming the data. In this paper, we pro-
pose an alternative possibility that does not require any trans- where
formation in time domain and, therefore, is not affected by the
assumptions and/or approximations involved in this procedure.
With reference to Fig. 2(b), the idea consists in assuming as data
not any longer the scattered field, but rather the difference be-
tween the field values in two symmetric points with respect to
(7)
the source. In particular, the whole system is symmetric with
respect to the source point, so that transmitting antenna sub-
stantially radiates the same field toward the left-handed and the being the imaginary part of the roots not positive. Moreover
right-handed directions. Then two identical receiving antennas
gather the signal at the opposite offsets and with respect (8)
to the source point. Finally, after propagating along two iden-
tical paths, the two signals are subtracted. If no buried object
and are the wavenumbers in the upper and
is present, clearly the differential signal is equal to zero. This
lower half space, respectively. In (6)–(8), is the conjugate vari-
means that, on condition that at most slow variations of the back-
able of the source position , and are the conjugate vari-
ground scenario are present, a differential configuration enables
ables of the horizontal abscissa and of the depth respectively,
us to get rid of the incident field directly from data in frequency
and the hats in (6) stand for Fourier
domain, providing a signal only related to the scattered field. In
transform.
formulas (with an obvious meaning of the symbols) we have
In particular this requires that is real, and in its turn this
limits the retrievable part of the spectrum of the contrast to a
spectral region of finite extent. The retrievable spectral area is
described in detail in [6] for an infinite observation domain,
(5)
whereas in [5], [7] and [20] the effects of the truncation of the
observation domain have been accounted for. Therefore, here
because . Let us note that the validity of (5) is in- we just depict quantitatively this spectral area in Fig. 3. In par-
dependent from the extent of the frequency band of the incident ticular, Fig. 3 refers to a case with an observation line two
signal. We find also interesting to stress the fact that, in the case meters long and a band of frequencies ranging from 200 up to
of a prospecting on a masonry or more in general on an inhomo- 600 MHz. The investigation domain D extends from the depth
geneous layered medium, a “simple” gating in time domain is of 0.1 m up to the depth of 1.8 m of depth and the soil exhibits
affected by the problem of the multiple “answers” associated to a relative permittivity equal to 9.
the several interfaces between the layers. This makes the inci- From Fig. 3, it can be understood that the class of retrievable
dent field not associated to a single reflection as in the case of a profiles within a multibistatic configuration is expected to un-
half space. As a consequence, in these cases time domain gating dergo a low-pass filtering with regard to the horizontal spatial
might erase the incident field unsatisfactorily, whereas the dif- variations and a band-pass filtering with regard to the vertical
ferential procedure of (5) still provides a rigorous cancellation. spatial variations.
3544 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 54, NO. 11, NOVEMBER 2006

the class of retrievable profiles, because drives the choice of


the spatial frequencies (further than the spectral axis ) that
will be most “penalized.”
From a practical point of view, this means that a proper choice
of should be somewhat related to a-priori information (if any)
about the horizontal size of the objects looked for. For example,
let us suppose an object whose horizontal extent is with
. In this case, when moving the differential measure-
ment set-up on the object [see Fig. 2(b)], after the left-hand edge
of the object there is a long path where the retrieved differen-
tial signal is very low, and this occurs up to the right-hand edge
of the object. The reconstruction, in this case, will be consti-
tuted by two separate “two-peaks” at the same depth. From a
more mathematical point of view, we can note that a horizontally
long object is expected to have a spectrum particularly concen-
trated about the low horizontal spatial frequencies (in particular
about spatial frequencies lower than ). How-
Fig. 3. Diffraction tomography retrievable spectral area in the case of the multi-
bistatic configuration.
ever, the more is little the more penalizes the
spectrum of the contrast about low values of (and specifically
values up to the order of ).
A further insight about the features of the retrievable objects
From a practical point of view, this means that the reconstruc-
along the horizontal direction can be achieved by introducing
tion of a buried homogeneous object “follows” its horizontal
some simplifications in (9). In particular, let us now neglect
variation, whereas it is expected to reproduce the steeper dis-
the variability of and let us reduce (9) to a one dimen-
continuities represented by the upper and lower bounds of the
sional case by eliminating the variability of the contrast along
object along the depth [5]–[8], [16]–[18].
the depth. Under these assumptions, (9) reduces to
This premix is important in order to give an insight about what
is expected when adopting a differential configuration. Specif-
ically, also in the case of a differential configuration the rea- (10)
soning starts from (6). In particular, let us note that in (6) the
offset appears only in the phase factor . There- Moreover, in order to account for the finiteness of the avail-
fore, in the case of a differential configuration the spectral rela- able spatial frequency band (related in particular to the trunca-
tionship (6) evolves straightforwardly in tion of the observation domain [20]), we suppose that (10) is
assessed only throughout a limited interval of horizontal spa-
tial frequency . At this point, let us consider again
the large object, characterized by an actual contrast
, being . As prompted, the differential
signal is meaningful only in the neighbors of the two bound
(9) points and . Due to this fact, and since
the current calculations are merely qualitative, at this point we
assimilate the signal gathered in this situation with the signal
Equation (9) means that, unlike the multibistatic configuration gathered in correspondence of a couple of Dirac’s functions
(where the offset influences the retrievable spectral frequency .
set only by means of a relatively unimportant phase factor), After this further passage we are able to retrieve the solution
in the framework of a differential configuration the offset of the inverse problem in closed form
“weights” the spatial frequencies of the retrievable spectral
area, and therefore it is expected to have a stronger effect on
the achievable reconstruction.
In general, the retrievable spectral area is decreased because
some spatial frequencies are penalized in correspondences of
the lower values of the weighting function . This
make us expect a decrease in the dimension of the class of
retrievable objects. In particular, the factor
[see (8)] affects the central part (about the
axis) of the spectral set of Fig. 3. This means that, unlike the
multibistatic case, within a differential configuration we expect
that the reconstruction tends to become band-pass (or possibly
multiband) along the horizontal direction too. In particular, the
(11)
value of the offset is expected to have a strong influence on
PERSICO AND SOLDOVIERI: A MICROWAVE TOMOGRAPHY APPROACH 3545

Fig. 4. Differential one dimensional qualitative reconstruction of a large object.

A graph of is given in Fig. 4, with


m, and m. From Fig. 4 we can observe that the Fig. 5. Behavior of the singular values for the four considered configurations.
Solid line:M1 ; dashed line: M1 ; dashed-dotted line D1 ; dotted line
location of the bounds of the objects correspond to null points D1 .
in the reconstructed image , whereas the maxima of the
two sincs are shifted of with respect to the bounds of
the object. The extent of the object is important with regard to
this point, because only if the object is quite larger than the
“tails” of the functions centered about each edge are neg-
ligible about the other edge. This reasoning also depends on the
available band, i.e., on the value of (which drives the width
of the lobes of the sincs). However, an increasing of the band
makes narrower the sincs but does not shift the position of their
maxima.

V. PROPERTIES OF THE INVOLVED OPERATOR:


SVD BASED ANALYSIS
In this section, we confirm the theoretical expectations of the
previous section by means of a numerical SVD of the involved
operators (1) and (3). In particular, we focus on a multifrequency
case wherein the frequency band 200–600 MHz is ranged with a
Fig. 6. Spectral contents for the four considered configurations M1 M1
; ;
step of 25 MHz. In the following examples the investigation do- D1 D1 .
;
main is a rectangle of 1.6 m 1.7 m, starting from the depth of
0.1 m. The relative permittivity of the soil is 9 and its electrical
conductivity is 0.01 S/m. The observation domain is 2 m long cases and , coherently with the fact that the differen-
(at quote 1 cm), and the measurement set-up is moved along it tial configuration entails a reduction of the achievable amount
with a spatial step of 0.05 m. These parameters are the same that of independent data.
underlie the spectral area of Fig. 3 but for the addition of losses In Fig. 6 the spectral content of the main singular functions is
in the soil. shown in all the four cases. The spectral content [5]–[7], [20] is
In order to perform the numerical SVD, the integral relation- the sum of the moduli of the Fourier Transforms of the singular
ships (1) and (3) have been discretized via a method of moments functions related to the singular values above a fixed threshold.
(MoM) by exploiting a functional basis made of 31 Fourier har- In formulas, the spectral content sp is expressed by
monics along the horizontal direction and 31 rectangular steps
along the depth, and point matching has been exploited. With
regard to the offset between source and observation points, two (12)
cases have been considered, one with m and
another with . With the given parameters, both
the multibistatic and the differential configurations have been where
taken into consideration. At last, therefore, we have four cases.
For simplicity of notation, we will label them as (multi- (13)
bistatic with offset ), (differential with offset )
and D . In Fig. 5 we show the behavior of the singular values
in the four cases. As it can be seen, the singular values curves and is the th singular function in the space of the unknowns
in the cases and are higher than the curves in the (again, the hats stands for Fourier transform). Here, N is chosen
3546 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 54, NO. 11, NOVEMBER 2006

as the number of singular values greater than dB with re-


spect to the maximum singular value (which means that here sp
is related to the reconstruction capabilities of a regularized in-
version with truncation of the singular values at dB). From
Fig. 5, we clearly see that the spectral contents in the cases
and are very similar to each other and appear quite “com-
pact,” whereas vertical null stripes appear in both the cases
and . In particular, in the case shows a unique ver-
tical null stripe, because actually has been calculated such
as the period of the function is approximately
equal to the “horizontal” extent of the spectral area of Fig. 3. In-
stead, the spectral content in the case shows narrower and
(twice) nearer vertical null stripes, coherently with the fact that
.

VI. NUMERICAL EXAMPLES


Fig. 7. Reconstructions under multibistatic and differential configuration
In this section we present numerical results to backup the con- without and with background error.
siderations of the previous sections. The reconstructions are ob-
tained from a truncated singular value decomposition (TSVD)
of the linear scattering operators with a regularization threshold actual scattered field plus the actual incident field ,
at dB [21]. The numerical SVD is constructed by means of at last the estimated scattered field is given by
MoM and point matching, as described in Section V.
The first example puts into evidence the advantages offered
by a differential configuration with regard to counteracting the
(14)
effects of inaccuracy in the knowledge of the background sce-
nario. The target is a square object sized 0.3 0.3 m and cen-
tered in an investigation domain D sized 2 2 m . The investi- being .
gation domain starts from the depth of 25 cm. The data have The reconstructions are shown in Fig. 7, and four cases
been computed according to the exact model of the electro- are considered: The first one is multibistatic with an exact
magnetic scattering (as reported in [22]) along an observation estimation of the background scenario (ME), the second one
domain at the interface, 2.16 meters long. The spatial step is differential with an exact estimation of the background
is 4 cm, the offset between the antennas is 8 cm (meant as scenario (DE), the third one is multibistatic with a wrong
and cm for the differential configuration). In order to sim- estimation of the background scenario (MW) and the fourth
ulate the data, the forward scattering problem has been solved one is differential with an wrong estimation of the background
by means of an iterative code based on conjugate gradient and scenario (DW).
made efficient by an extensive use of FFT algorithms. This is From Fig. 7, we can see that, if we have at disposal a reli-
possible due to the convolution structure of the Green’s func- able calculation of the scattered field (in particular an accurate
tion [23]. The frequency band is 200–600 MHz and the fre- knowledge of the background scenario) the two reconstructions
quency step is 25 MHz. The relative permittivity of the ME and DE agree with the theoretical expectations. In fact, the
soil has a real part equal to 9 and an imaginary part given by ME case is characterized by a low pass horizontal filtering so
, where is the current frequency, is that the upper side of the object is well reconstructed, whereas
the dielectric permittivity of the free space and is an electrical the reconstruction in the DE exhibits two peaks in correspon-
conductivity of 0.02 S/m. The square object shows a relative dence of the edges of the upper side of the object. However,
permittivity , where if we have an uncertainty about the incident field (in partic-
. ular if we do not have at disposal an accurate knowledge of
In particular, we have made use of a model with the correct the background scenario), then the differential configuration ap-
permittivity of the soil and a second model with a per- pears much more robust. In fact, the MW reconstruction is very
mittivity of the soil . This second model corresponds poor because the artifact in the deeper part of the observation do-
to an error in the knowledge of the dielectric characteristic of main gives a dominant contribution on the reconstructed image.
the host medium. In addition, in both cases we suppose not to This happens because the term of (14) is stronger than the
have a reliable measure of the entity of the losses, and so we useful term associated to the buried object. For the homol-
make use of lossless models with . With regard to the ogous differential reconstruction DW we observe an undesired
multibistatic configuration, in the cases with a wrong estimation shift of the reconstruction along the depth due to the incorrect
of the soil permittivity, the scattered field in frequency domain dielectric permittivity assumed in the model. However, the DW
is calculated by subtracting the erroneously estimated incident reconstruction is certainly clearer and cleaner than the MW re-
field (corresponding to ) from the correctly cal- construction, because the term does not affect the data
culated total field. Since the correct total field is given by the any longer. Let us note that this example can be representative
PERSICO AND SOLDOVIERI: A MICROWAVE TOMOGRAPHY APPROACH 3547

Fig. 9. Reconstruction of a horizontal long void with model data (upper panel)
Fig. 8. Reconstructions of a horizontally long object under the four considered
configurations M1 M1 D1 D1
; ; ; .
and exact data (lower panel).

data with high model error. The object is a void with the same
of a realistic situation where slow spatial variation of the back- location and size of that of Fig. 8. The spatial step of the mea-
ground scenario occurs, so that we have an erroneous estima- surements is 0.05 m and the offset is 0.1 m, while the other pa-
tion of the permittivity of the soil because we have measured it rameters are unchanged with respect to the previous examples
“elsewhere.” of Fig. 8. Fig. 9 depicts a comparison between the differential
In order to show an example that puts into evidence the reconstructions obtained with no model error (upper panel) and
considerations worked out in Sections IV and V about the with exact (non linear) data (lower panel). As it can be seen,
properties of the involved operators, the reconstruction of an despite of the high model error, the lower plot of Fig. 9 allows
object horizontally “long” with respect to the offset is proposed. to point out “the geometry” of the object in a similar way with
Since we are focused on the properties of the operator, in respect to the case of no model error.
this example model data (evaluated according to BA) will be
exploited. The reconstruction of an object sized 1 0.2 m VII. CONCLUSION
at the depth of 0.6 m is considered (see Fig. 8). The object
shows a relative permittivity equal to 9.1 and a conductivity In this paper we have discussed a microwave tomography ap-
equal to 0.01 S/m. The parameters of the test are the same proach exploiting a differential configuration for GPR data in
of those accounted for the SVD considered in Section V. In frequency domain. This differential configuration allows to mit-
particular, the object is a weak scatterer and this justifies the igate some bad effects of uncertainties on the reference scenario
use of model data. The same four cases , but, on the other hand, changes and somehow worsens the prop-
and of Figs. 5 and 6 have been considered, and the erties of the involved scattering operator. Also, we have shown
reconstructions are shown in Fig. 8. From Fig. 8, it can be that a-priori information about the size of the looked objects
seen that the reconstruction in the multibistatic cases can be particularly important in the framework of a differential
and are very similar to each other, and this confirms configuration. Future works will be devised to do experimental
the fact that the value of the offset plays a marginal role tests. In particular, situations of interest for a differential con-
in this case. The differential reconstructions, instead, exhibit figurations could be those with shallow object [24] and/or with
a marked band-pass filtering behavior along the horizontal objects embedded in a wall.
direction, so that they are characterized by two peaks about
the edges of the objects. The reconstruction is worse APPENDIX A
than the reconstruction due to the smaller amount of The Green’s function of the problem is given by [22]
independent data (see Fig. 5). Finally, it also can be noted that
the differential reconstructions presents two null points at the
lateral edges of the object, whereas the “maxima” in the case
are farther from the edge of the object with respect to the
case , according to the qualitative reasoning of Section IV.
Finally, we find interesting to present a further test-case in (A1)
order to show the reliability of the reconstruction approach and
the validity of the previous considerations also in the case of where and are defined in the main text [see (7)].
3548 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 54, NO. 11, NOVEMBER 2006

The incident (or unperturbed) field is given by [22] [17] T. J. Cui, W. C. Chew, X. X. Yin, and W. Hong, “Study of resolution
and super resolution in electromagnetic imaging for half-space prob-
lems,” IEEE Trans. Antennas Propag., vol. 52, no. 6, pp. 1398–1411,
Jun. 2004.
[18] R. Persico and F. Soldovieri, “Effects of uncertainty on background
permittivity in one dimensional linear inverse scattering,” J. Opt. Soc.
Amer. Part A, vol. 21, no. 12, pp. 2334–2343, Dec. 2004.
(A2) [19] G. F. Stickley, D. A. Noon, M. Cherniakov, and I. D. Longstaff, “Gated
stepped-frequency ground penetrating radar,” J. Appl. Geophys., vol.
43, pp. 259–269, 2000.
where is the impressed current and is the frequency. [20] R. Persico, F. Soldovieri, and G. Leone, “A microwave tomographic
approach for multibistatic configuration: The choice of the frequency
ACKNOWLEDGMENT step,” IEEE Trans. Instrum. Meas., Dec. 2006, to be published.
[21] M. Bertero and P. Boccacci, Introduction to Inverse Problems in
The authors thank Dr. L. Crocco and Dr. I. Catapano for pro- Imaging. Bristol and Philadelphia: Institute of Physics Publishing,
viding the numerical code for the forward problem. 1998.
[22] D. Lesselier and B. Duchene, “Wavefield inversion of objects in strat-
ified environments: From backpropagation schemes to full solutions,”
REFERENCES in Review of Radio Science 1993–1996, W. R. Stone, Ed. Oxford,
[1] D. Colton and R. Kress, Inverse Acoustic and Electromagnetic Scat- U.K.: Oxford Univ. Press, 1996.
tering Theory. Berlin, Germany: Springer Verlag, 1992. [23] I. Catapano, L. Crocco, and T. Isernia, “A simple two-dimensional
[2] L. B. Conyers and D. Goodman, Ground Penetrating Radar—An Intro- inversion technique for imaging homogeneous targets in stratified
duction for Archaelogists. Lanham, MD: Alta Mira, 1997, division of media,” Radio Sci., vol. 39, no. RS1012, 2004.
Sage Publications. [24] H. Brunzell, “Detection of shallowly buried objects using impulse
[3] D. J. Daniels, Ground Penetrating Radar, 2nd ed. London, U.K.: The radar,” IEEE Trans. Geosc. Remote Sensing, vol. 37, no. 2, pp.
Institution of Electrical Engineers, 2004. 875–886, Mar. 1999.
[4] W. C. Chew, Waves and Fields in Inhomogeneous Media. Piscat-
away, NJ: IEEE Press, 1995.
[5] G. Leone and F. Soldovieri, “Analysis of the distorted Born approxi-
mation for subsurface reconstruction: Truncation and uncertainties ef- Raffaele Persico was born in Naples, Italy, on May
fects,” IEEE Trans. Geosci. Remote Sensing, vol. 41, no. 1, pp. 66–74, 15, 1969. He received the Electronic Engineering
Jan. 2003. degree (summa cum laude) from the University of
[6] R. Persico, R. Bernini, and F. Soldovieri, “The role of the measure- Naples Federico II in 1996 and the Ph.D. degree
ment configuration in inverse scattering from buried objects under the from the Second University of Naples in 1999.
Born approximation,” IEEE Trans. Antennas Propag., vol. 53, no. 6, Previously, he was a Research Scientist in the Con-
pp. 1875–1886, Jun. 2005. sortium for Advanced Research in Remote Sensing
[7] F. Soldovieri, R. Persico, and G. Leone, “Effect of source and receiver Systems CO.RI.S.T.A. In December 2001, he joined
radiation characteristics in subsurface prospecting within the distorted the Institute for the Electromagnetic Sensing of the
Born approximation,” Radio Sci., vol. 40, no. RS3006, May 2005. Environment (I.R.E.A.), which is part of the Italian
National Council of Researches (C.N.R.). His main
[8] P. Meincke, “Linear GPR inversion for lossy soil and a planar air-soil
interests are linear and non linear inverse scattering problems, especially in re-
interface,” IEEE Trans. Geosci. Remote Sensing, vol. 39, no. 12, pp.
lationship with GPR data processing and GPR systems.
2713–2721, Dec. 2001.
Dr. Persico is a member of Centro di Chimica, Fisica Geologia e Ingegneria
[9] I. Akduman, “An inverse scattering problem related to buried cylin- (CICFGIA), Lecce, Italy, (Dipartimento di Scienza dei Materiali), and of the
drical bodies illuminated by Gaussian beams,” Inverse Problems, vol. Association of Italian-German Studies, Merano, Italy. He is a Reviewer for the
10, no. 2, pp. 213–226, Apr. 1994. IEEE and other international journals.
[10] M. Slaney, A. C. Kak, and L. E. Larsen, “Limitations of imaging with
first order diffraction tomography,” IEEE Trans. Microwave Theory
Tech., vol. 32, pp. 860–874, 1984.
[11] L. Gurel and U. Oguz, “Three-Dimensional FDTD modeling of a
Francesco Soldovieri received the Laurea degree
ground penetrating radar,” IEEE Trans. Geosci. Remote Sensing, vol.
in electronic engineering from the University of
38, no. 4, pp. 1513–1521, Jul. 2000.
Salerno, Salerno, Italy, in 1992 and the Ph.D. degree
[12] L. Gurel and U. Oguz, “Optimization of the transmitter-receiver sepa-
in electronic engineering from the University of
ration in the ground penetrating radar,” IEEE Trans. Antennas Propag., Naples “Federico II,” Naples, in 1996.
vol. 51, no. 3, pp. 362–370, Mar. 2003. In 1993, he joined the Electromagnetic Research
[13] L. Chommeloux, C. Pichot, and J. C. Bolomey, “Electromagnetic Group of the University of Naples and in 1998/1999
modeling for microwave imaging of cylindrical buried inhomo- he held a Postdoctoral Fellowship at the same Uni-
geneities,” IEEE Trans. Microw. Theory Tech., vol. MTT-34, no. 10, versity. From 1999 to 2000, he collaborated with the
pp. 1064–1076, Oct. 1986. Research Group on Applied Electromagnetism at the
[14] A. J. Devaney, “Diffraction tomographic reconstruction from intensity Second University of Naples. Since 2001, he has been
data,” IEEE Trans. Antennas Propag., vol. 1, no. 2, pp. 221–228, Apr. a Researcher at the Institute for the Electromagnetic Sensing of the Environment
1992. of the Italian National Research Council (IREA-CNR). His main scientific inter-
[15] M. Testorf and M. Fiddy, “Imaging from real scattered field data using ests include electromagnetic diagnostics, inverse scattering, GPR applications,
a linear spectral estimation technique,” Inverse Problems, vol. 17, pp. and antenna diagnostics.
1645–1658, 2001. Dr. Soldovieri was awarded the 1999 Honorable Mention for the H.A.
[16] A. J. Devaney and M. Dennison, “Inverse scattering in inhomogeneous Wheeler Applications Prize Paper Award of the IEEE Antennas and Propaga-
background media,” Inverse Problems, vol. 19, pp. 855–870, 2003. tion Society.

You might also like