HAL Id: jpa-00245397
https://hal.archives-ouvertes.fr/jpa-00245397
Submitted on 1 Jan 1985
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Inverse scattering on electromagnetic measurements in a stratified medium
D. Maystre, A. Roger, E. Toro
To cite this version:
D. Maystre, A. Roger, E. Toro. Inverse scattering on electromagnetic measurements in a stratified medium. Revue de Physique Appliquée, Société française de physique / EDP, 1985, 20 (12), pp.815- 821. �10.1051/rphysap:019850020012081500�. �jpa-00245397�
815-
REVUE DE PHYSIQUE APPLIQUÉE
Inverse scattering
onelectromagnetic
measurementsin
astratified medium
D. Maystre, A. Roger and E. Toro
Laboratoire d’Optique Electromagnétique (*), Faculté des Sciences et Techniques, Centre de St-Jérôme,
13397 Marseille Cedex 13, France
(Reçu le 6 décembre 1984, révisé le 2 juillet 1985, accepté le 2 août 1985 )
Résumé. 2014 Dans le cadre d’un modèle théorique monodimensionnel, nous étudions les possibilités de déduire
la permittivité d’un milieu stratifié à partir de mesures du champ électromagnétique rayonné par un émetteur se
déplaçant perpendiculairement aux interfaces. Des résultats très surprenants sont démontrés. Nous donnons une
méthode simple et très efficace que nous testons numériquement sur des mesures expérimentales simulées.
Abstract - In the framework of a theoretical one-dimensional model, we investigate the possibility of deducing
the permittivity of a stratified medium from measurements of the electromagnetic field radiated by an emitter
which moves in a direction perpendicular to the interfaces. Surprising results are demonstrated : a very efficient and simple method is given and numerically checked on simulated experimental data.
Revue Phys. Appl. 20 (1985) DÉCEMBRE 1985, PAGE 815
Classification
Physics Abstracts
03.40K - 06.30L - 91.35
1. Introductioa
The determination of the
complex permittivity
of astratified medium from measurements of an electro-
magnetic
field is atechnique currently
used inOptics
and
Electromagnetics.
InOptics,
the source isplaced
in the upper medium
(air)
and, for instance, measure-ments of the reflected field for different angles of
incidence may
permit
one to reconstruct apermittivity profile
[1]. Ingeophysics,
bore-hole measurements areimplemented by using
an emitter and receiversmoving
in the direction
perpendicular
to the surface, in orderto retrieve the
complex permittivity
of the formationsurrounding
the bore-hole [2-4]. In this paper, wemainly
deal with this secondproblem.
Our first aim is tostudy
the conditions under which one mayhope
toretrieve the
permittivity.
To this end, we define a onedimension
problem, simpler
than the situations en-countered in
practice,
but very close from a theoreticalpoint
of view. Thanks to thesimplicity
of this model,we are able to state some
surprising
conclusions about (*) E.R.A. au CNRS no 597.the
possibilities
of such atechnique.
Then, we presenta very efficient and
easily
implemented method. Un-fortunately,
thegeneralization
of this new method tothe three dimensional case is not
straigthforward,
atleast in the current
experimental
conditions, but wehope
to be able to achieve thisgeneralization
in the nearfuture.
2. Direct and inverse problems.
2.1 THE MODEL
(FIG.
1). - We consider a stratified mediumhaving
acomplex permittivity
e(z)piecewise
constant. An emitter E located at z = ( generates a field whose complex
amplitude
F (with timedepen-
dence in
e - irot)
satisfies the equation :F and
dF/dz being
continuous at the interfaces,k2(z) = k[ 8(z) being
apiecewise
constant function(ko
= 2NIÂO,
whereà
denotes thewavelength
in vacuum), and àbeing
the Dirac distribution. Of course,Article published online by EDP Sciences and available at http://dx.doi.org/10.1051/rphysap:019850020012081500
816
Fig. 1. - The model : the horizontal full lines represent the interfaces and the horizontal dashed lines represent the location of the emitter (E) and the two receivers (Ri and R2).
the function F must
satisfy
anoutgoing
wave conditionfor z - ± oo. In other words, F must decrease or pro- pagate towards z - oo (resp. z - - oo) in the upper
(resp.
lower) medium. Two receiversRi
andR2
arelocated at z = ri and z = r2l
with r,
= 03B6 +/1’
r2 = ( +
12, 12
>h
> 0.Throughout
this paper, we supposethat 4
and12
are fixed, while ( (and conse-quently
relatedquantities
such as ri, r2,(ri
+r2)/2,
etc...) is variedIn practice, the theoretical model we are dealing with
can be connected to the 2D model described in [2]
in the
following
way : the field F(z)corresponds
to the0 component
EfJ(z, po)
of the electric field radiated at a distance po of the Oz axisby
a currentloop
withaxis Oz and radius po, the receivers
Ri
andR2 being loops
of radius poplaced
at distancesh
and12
fromthe emitter such that :
Under these conditions, the electric field
Ee(z, po)
onthe receivers satisfies an
equation
close toequation
(1).2.2 THEORY OF THE DIRECT PROBLEM. - Here, we
consider the
following
« direct »problem : k2(z) being
known, calculate F(z).Now, we shall recall a classical
expression
forF(z) [5],
which isnothing
but the Green function associated with the differentialequation (1).
We assume that thestratified medium is composed
by
a finite number N + 1 ofhomogeneous
materials, numbered from the bottom to the top. Wecall si (i
E (1,N))
the ordinateof the ith interface
(between
the ith and (i + 1)thmaterials) and ki
and ei the values of k(z) and e(z) inthe ith material. We define the two functions
Hp(z)
and
H.(z)
continuous, with continuous first derivative,satisfying
the homogeneous waveequation :
Hp satisfying
anoutgoing
wave condition for z- + oo(upper
medium) andH.
the same condition forz = - oo (lower medium).
Taking
into account theform of the field in
exp(ikN + 1
z) and exp( -ik1
z) inthese two extreme media, the
outgoing
wave conditioncan be written in the mathematical form :
So,
Hp,.(z),
which satisfies a linearhomogeneous
differential
equation
of the second kind and theboundary
condition (3) or (3’), is not unique, twoarbitrary
solutionsbeing proportional.
Finally,
let us define the two functions :and the wronskian :
which is
independent of z (this
result may be establishedby multiplying
Eq. (2)by Hm,p
and bysubtracting
thetwo
equations
so obtained, the final result, i.e.Hp
Hm -Hm HP
= 0,showing
that W’ =0).
An
elementary
calculation shows that :and it is to be noticed that the value of F(z) so obtained
does not
depend
on the determination ofHp
andHm
which have been chosen.
Finally,
the function F(z), whichdepends
on theparameter "
can be calculated from the two functionsHP(z)
andHm(z) independent
of 03B6. This propertyconsiderably simplifies
the numericalapplication.
2. 3 NUMERICAL SOLUTION OF THE DIRECT PROBLEM. -
This
problem
reduces to the calculation of the functionHP(z)
andH.(z)
defined in section 2.2.For the sake of
simplicity,
weonly
describe thecomputation of HP(z).
Inthe jth
material, the field canbe written in the form :
a / and a Ç being complex
numbers.In the
following,
we shall describe the fieldby
thevalues
V + (z)
and V -(z)
of each term in theright
handside
of(7).
Thus, inthe jth
material :with
and V(z) will denote the vector
( V + (z),
V -(z)).
It must be noticed that in
the jth
material, V(z) and V(z’) are linkedby
the relation :Q
being
a 2 x 2 matrix givenby :
The unknows in our numerical
implementation
are thevectors
An
andBn, respectively
equal to the limit ofV(z)
above and below the nth interface(Fig.
2) :From (8), it turns out that, in the nth material,
An - 1
canbe deduced
from Bn by
the relation (Fig. 2) :with
Furthermore, a
straightforward
calculation shows that thecontinuity
of F anddF/dz
entails the matrix relation(Fig.
2) :Fig. 2. - Determination of the function
H,(z)
when N = 3.the transmission matrix
Tn
beinggiven by :
with
It is worth
noting
that the inverse ofTn
is obtained veryeasily by replacing kn by kn+ 1
andconversely.
Equation (12)
allows us togeneralize equation (8)
and to state the
following
fundamental lemma : Lemma : The vector V(z) at any ordinate may be known as soon as its valueV(zo)
for an arbitraryordinate zo is
given.
The matrix relation between V(z)and
V(zo)
may be obtained byusing
(8) or (11) insideeach material
separating
thepoints
z and zo, andequation (12)
at each interface between thesepoints.
Finally,
it must be remarked thatAN + 1
= 4, due tothe
outgoing
wave condition onHp.
Now, if thecoefficient
AN+ 1
isgiven,
we know the vectorA.+,
=(A:+1,
0) therefore the lemma shows thatV(z)
can be deducedlinearly
fromAN + 1.
SinceAN + 1 only depends
on amultiplicative
constantsA: + l’
V(z) and
HP(z)
dependlinearly
onAN + 1
as well and weknow that the value of F(z)
given by (6)
isindependent
of this
multiplicative
constant which may be chosen tobe
equal
tounity. Figure
2 outlines the determination of theAi and Bj
fromAN+ 1 -
2.4 DEFINITION OF THE INVERSE PROBLEM. - Let S
and tl
be thequantities
definedby :
So
S(q)
contains both the ratio of theamplitude
andthe
phase
difference of the measurementsimplemented by
the two receiversR,
andR2.
The inverse
problem
westudy
is thefollowing : knowing S( 11)
for a set ofvalues l1i of 11
in arithmeticprogression, find the
piecewise
constant functionk2(z).
Of course, the choice of thisparticular
inverseproblem
isinspirated by
the situation in actual expe- rimental borehole measurements.3.
Some pmperties of the solutions
of the directproblenl
3.1 THE INVARIANCE PROPERTY. - Let us state the
following
property called « invariance property » : If the locations ofR,
andR2
are fixed, the signalS( 11)
onlydepends
on thepermittivity profile
of thestratified medium above the first receiver
Rl.
As aconsequence, neither the
permittivity
e(z) belowRl,
nor the location of the source
(provided
it remains belowRl)
has any influence on thesignal.
818
To demonstrate this
surprising
property, we firstuse
equation
(6) to express the signalS( ,,).
Sincer2 > r, > 03B6, it turns out that :
Obviously,
S(il)
does notdepend
on the location of the emitter E. Furthermore, the lemma shows that V(z)and thus the
signal S,
can be deduced fromAN11
= (1, 0)using
the linear relations (12) and(8)
in the
region
located betweenSN
and z(Fig.
2). Sincethese linear relations
only require
theknowledge
ofe(z)
between z and oo, thepermittivity
for z ri has no influence onS( 11).
3.2 THE EQUIVALENCE PROPERTY. - Let e
(z)
be afictitious
permittivity profile,
andS( 11)
the correspon-ding signal
which would be measured in the stratified mediumof permittivity (z).
Theequivalence
propertycan be stated in the
following
manner :For any value zo, there exists a fictitious stratified structure of
permittivity É (z)
constant for z > zo and «equivalent »
to the actual stratified structure,i.e.
having
thefollowing properties :
ef
being
a well chosen complex number,The demonstration of this property is a direct conse-
quence of the lemma : if
17(z)
denotes the vectorV(z)
of the fictitious stratified structure, it sufices to choose ef in order to have
V(zo)
=V(zo).
This willentail that
17(z)
= V(z) for z Zo(for
z > zo, thepermittivity
of the two structures are not identical and thus theequality
non longer holds). If thepoint
zo is located in the nth material of the actual structure, the vectorV(zo)
will be deduced from the vectorÂn
ofthe fictitious structure
by
a relation similar to(12) :
+n
and1 being
obtained from(14)
and(14’) replacing kn + 1 bY kf
=ko
Eg. Because of theoutgoing
wave condition,Ãn = (Â., 0)
and thequestion
whicharises is the
following :
is itpossible
to findÃn+
andkf
in such a way that
V(zo)
takes agiven
valueV(zo) ?
A
straightforward
calculationusing equation (17)
shows that it suffices to
choose kf
such that :3.3 CONSEQUENCES OF THE INVARIANCE AND EQUIVA-
LENCE PROPERTIES ON THE INVERSE PROBLEM. - Let us
suppose that a set of measurements of
S(r¡)
has beenimplemented,
Zm;n been the smallest value of ri and zmax the greatest valueof r2 during
the measurement Adirect consequence of the two
properties
established in thepreceding
section is that these measurements would be unchanged if the part of the structure located below zm;n were modified in anarbitrary
manner, or ifthe part of the structure located above zmax were
replaced by
an homogeneous mediumhaving
a wellchosen
permittivity
03B5f. So, it is obvious that the retrievalof
e(z) outside the interval(zm;n, zmax)
is quiteimpossible.
This
surprising
property has a fundamentalimpor-
tance for the solution of the inverse problem.
4. The
approximate
method of the« homogeneous
material » (HM).
4. 1 PRINCIPLE OF THR METHOD. - This method lies on
assumptions
similar to those of the method described in[2].
If the structure is assumed to behomogeneous
from - oo to + oo, which
k(z)
=kl,
the functionHP(z)
reduces to anexponential
function :thus
and it turns out that
ki
can be deduced from one mea- surement :So, for each measurement of
S(tl),
the aboveassumption
enables one to associate a valuee( 11)
=kîlk2 using (19).
It is worthnoting
that the determi- nationof ki
from(19)
is notunique,
unless the distance12
-li
between the two receivers is chosen in sucha way that :
Under these circumstances, Log
(S( 11))
will be chosen in order to have animaginary
part less than 2 x.4.2 NUMERICAL RESULTS. 2013 We consider in
figure
3a stratified structure whose
permittivity
isgiven by
adashed line. We have
computed
thecorresponding
value of
S( il) using
the numerical method described inparagraph
2, then theapproximate
value ofe( q) given by (19) (full
line inFig. 3).
The main conclusion is that thisapproximate
method is able togive
accuratevalue of
a(z),
except at thevicinity
of an interface.More
precisely,
the lemma and elementary conside-rations lead us to state the
following
rule of thumb.The determination of
e(z)
from the method of thehomogeneous
mediumgives
accurate results as soonas the two
following
conditions are satisfied :-
R,
andR2
are in the same material,- The distance d between
R2
and the closest interface above it is greater thand0 ~ 2/Im (ki), ki being
the wave number in the material. Thisexplains
the failure of the method in the second material from the left.
Fig. 3. - Numerical computation of e(z) from the HM method with A. = 12, li = 0.635, l2 = 1.27, d = 0.3175.
--- Actual value of Re
(e(z)}
and Im(e(z)};
cor- responding results obtained from the HM methodIn conclusion, this method is very
simple
toimple-
ment but can
provide
erroneous conclusions since the presence of o horns » in thevicinity
of the interfaces could beinterpreted
as the consequence of the pre-sence of fictitious
layers
which, in fact, do not existThis
phenomenon
becomes greater as theresistivity
in the material increases.
5. A new method : the
locally
homogeneous material (LHM) method.5. 1 PRINCIPLE OF THE METHOD. - From
equations (7)
and
(15),
we deduce thatS( il)
can beexpressed
in theform :
provided
both RI andR2
are in thejth
material.In order to
simplify
the aboveexpression,
we define :where à denotes the
distance r2 - rl
between the tworeceivers, and :
After
straightforward
calculation we get :From the above
equation,
it turns out that thesignal S( r¡)
which is acomplex
number,depends
on twounknown
complex numbers kj
(ory)
and R,provided
the two receivers are in the same material. Of course, it is
quite impossible
to deduce these two unknowns from one measurement. On the other hand, it can beconjectured
that two successive measurements, fortwo different values of il, will
provide
enough infor-mations to retrieve R and
ki, provided
these twomeasurements are such that the receivers remain in the same material. Under these conditions,
denoting by Si
andS2
thesignals S(1’1)
andS(r¡ + d)
obtainedsuccessively,
we obtain the twoequations :
where
Now, the
problem
is thefollowing : Si
andS2 being
known,fînd k, (y
and àdepending
on theonly
variablekj).
With this aim, we first eliminate Rexp(-
2ikj ri)
between
(23)
and(24)
and, afterelementary
calcula- tions, we obtain the fundamentalequation :
This
equation
contains the two unknowncomplex
numbers y and à which are, in fact, linked to the
only
unknown kj (see equations
(21) and(25)),
and thusverify
theequation :
Thus we have to solve the system of two non linear
equations (26)
and(27)
with two unknowns y and ô.These two unknowns must be less than
unity
inmodulus
since kj
has apositive imaginary
part in(21)
and
(25),
due to the losses in the material.Before
solving
this system of equations, it is to be noticed that the determination of y and b does notimply
aunique
determination ofkj.
In order to forcethe
uniqueness,
we must assume, for instance, that thephase
ofexp(ikj
a) = y ispositive
and less than 2 03C0, whichimplies
that :It is very
important
to notice that this conditionimplies
theuniqueness
of the determination of à820
from y in (27). Indeed, this condition can be written in the form :
If the phase of y is
positive
and less than 2 x, Log y isunique,
thus ô isunique.
So,(28)
enables us towrite (26) and (27) in the form of a non linear
equation :
where the
phase
ofy"la
is obtainedby multiplying by
2
dla
thephase
of y (between 0 and 2 03C0).It is not our purpose to
study equation (29)
in detailand to show the
uniqueness
of the solution whose modulus is less thanunity.
We have verified this property in the cases wheredla
= 1, 0.5, 0.25 where(29)
becomes analgebraic equation.
For instance,when
dla
= 0.5 (29) becomes :where y = 1 is a solution whose modulus is equal to 1
and must be
rejected. Eliminating
this solution leads to ourequation
of the second order :The
product
of the two solutionsof (30)
isequal
to 1, and this entails theuniqueness
of the solution whose modulus is less than 1.When the solution
of (29)
is known for agiven
valueof
dla,
it can be known for close values. It suffices to take theLogarithmic
derivative of (29) to deduce thevariation
03B403B3
from the variation ofdla.
A
general
means forsolving (29)
would be to searchfor the root of this
equation
in thecomplex plane using
an iterative process
(Newton
method forinstance).
A first estimate of the solution would be obtained
by considering
first the known solution in the casewhere
(29)
can be solvedanalytically (dla
= 0.5 forinstance),
thenby modifying slowly dla
in order tofollow the solution of the transcendental
equation.
We have not
implemented
thisgeneral
method.Another means tp find y in the
general
case is toconsider two successive
equations (29)
withsignals Sl’ S2, S3, assuming
that these three measurements have beperformed
in the same material.Eliminating y2d/a
between these twoequations
leads to thealgebraic equations :
Fig. 4. - The same as figure 3, but with the LHM method and using equation (29).
Fig. 5. - The same as figure 3, but with the LHM method and using equation (30).
Eliminating
the solutions y = ± 1 in the aboveequation
leads to theequation
of the second order :with
The solution is the root
of (31)
whose modulus is less thanunity
(theproduct
of the two roots is equal tounity).
5. 2 NUMERICAL RESULTS. 2013
Figure
4 shows the nume-rical result obtained
using
equation (29) fordla
= 0.5.Our theoretical
predictions
arefully
confirmed : thepermittivity
isrigorously
given when the tworeceivers are in the same material. The
superiority
of this new method is all the more obvious since in the
region
where the two receivers are not in the same material, the curve of thepredicted permittivity
in general does not oscillate. It is not so for the curvein figure 5 obtained from
equation (31),
which exhibits large horns. At présent, we are unable to understand thesuperiority
ofequation (29)
onequation (31)
from a numerical
point
of view. We conjecture thatthe
origine
of thissurprising
phenomenon may be found in the variationalproperties
of the two formulae.6. Conclusion.
Our
simple
model has allowed us to state importantconclusions. The LHM method is much more precise
and
powerful
than the HM method which assumesthe material to be homogeneous. Moreover, we can
conjecture
from the rules of invariance andequi-
valence that the determination of the
permittivity
of astratified structure outside the domain of measurement in the two-dimension
problem
is very diSicult, eventhough
thesimple
rules of the one dimensionalproblem
cannot begeneralized
to this more difficultproblom.
A generalization of the LHM method to thetwo-dimension
problem
is notstraightforward
Thisis one of the aims of our future work.
References
[1] ROGER, A., MAYSTRE, D. and CADILHAC, M., J. Optics
9 (1978) 83.
[2] HUCHITAL, G. S., HUTIN, R., THORAVAL, Y. and CLARK, B., paper presented at the 56th Annual fall techni-
cal conference and exhibition of the S.P.E. of
AIME, San Antonio, October, paper S.P.E.10988
(1981).
[3] Yu, J. S., REARDON, P. C. and LYSNE, P. C., IEEE Trans.
A. P., AP-31 (1983) 397.
[4] TABBARA, W., LESSELIER, D. and FALCHETTI, F., Digest,
IEEE 1983 Symposium on Antennas Propag.,
Houston (1983).
[5] ROARCH, G. F., Green’s functions, Chap. 1 (Van Nostrand Reinhold) 1970.