CRAMER-RAO STUDY OF ONE-DIMENSIONAL SCATTERING SYSTEMS:

PART I: FORMULATION
Edwin A. Marengo, Maytee Zambrano-Nuñez, and David Brady
Department of Electrical and Computer Engineering
Northeastern University
Boston, Massachusetts, USA
email: [email protected]

ABSTRACT where the small scatterer can represent a local inhomogene-

A Cramer-Rao bound (CRB) study is developed in one- ity or iris, or a canonical monostatic or bistatic radar or lidar
dimensional (1D) space which sheds fundamental insight model with “in-line-of-sight” target. In the latter context
onto the information about multiply scattering point-like this model is suitable for (transversely) large, (longitudi-
scatterers that is contained in scattering field data corre- nally) thin planar targets that exhibit negligible variation
sponding to transmissive, reflective, and combined trans- of constitutive properties in the transverse coordinates and
missive plus reflective sensing geometries, and single- and non-negligible variation mostly in the direction of propa-
multi-frequency measurements. gation of the probing fields, as applicable to far-field trans-
missive or reflective tomographic measurements. The 1D
KEY WORDS geometry is also relevant to the forward and inverse scatter-
Cramer-Rao bound, scattering, estimation, inverse prob- ing problem for planar layered structures, which has been
lem, one-dimensional, Fisher information. the focus of a CRB study by Gustafsson and Nordebo [4]
whose connections to the present work are outlined in the
following.
1 Introduction
The use of Cramer-Rao analysis as a tool to assess
This two-part paper is concerned with the estimation- imaging systems is not new. It has been a topic of important
theoretic quantification of the “information content” about investigations in the past 40 years. Of much interest is the
a wave scatterer that is contained in scattered field data. work by Shahram and Milanfar [5] who focused on the rig-
The information content of scattered fields has been inves- orous information-theoretic determination of the resolution
tigated before by, e.g., Brancaccio et al. [1] and Pierri et al. of incoherent optical imaging systems by formulating a de-
[2] (see also the references therein), which emphasize the tection problem consisting of identifying single-source ver-
question of the number of degrees of freedom (NDF) of the sus two-source systems and who also studied the CRB in
linear mapping from the scattering potential function to the the estimation of source parameters. However, unlike this
scattering of the data or “essential dimensionality” of the seminal work, which focuses on an incoherent radiation
Born-approximated matrix. Our presentation uses a differ- system involving a linear signal model, we address a coher-
ent approach based on the fundamental Cramer-Rao bound ent scattering system where the signal model is nonlinear.
(CRB) [3], and addresses both Born-approximable condi- Of greater relevance to the present investigation are a num-
tions and the more general multiple scattering case, focus- ber of more recent papers by Shi and Nehorai [6], Sentenac
ing on canonical systems of multiply-scattering point-like et al. [7], and Simonetti et al. [8]. Shi and Nehorai [6] de-
scatterers. veloped a Cramer-Rao analysis of multiply-scattering point
In particular, we quantify information about point- targets in three-dimensional (3D) space and showed via ex-
like scatterers that is contained in scattering data, for haustive numerical computation several interesting effects
different sensing configurations and single- versus multi- such as the possibility of adding artificial scatterers to en-
frequency conditions, via the CRB benchmark, which mea- hance estimation of parameters associated to sought-after
sures under mild statistical conditions the lowest achiev- scatterers. The role of scatterers to enhance imaging has
able error in the unbiased estimation from noisy scattering also been studied in [9], and this issue is, in fact, part of
data of sought-after scattering parameters (in the present the broader area of radiation and imaging enhancements
context, the scattering strengths and positions of point-like via substrate media including metamaterials [10]. Refer-
scatterers). ence [6] also illustrated further the enhancement in imaging
To facilitate analytical and computational handling thanks to multiple scattering as a physical resource. The
of a variety of interesting questions, we choose to focus latter topic has also been studied in Sentenac et al. [7] and
only on the formally simple case of scatterers or inhomo- Simonetti et al. [8] via the CRB benchmark.
geneities in one-dimensional (1D) space. This particular In particular, recently there has been much interest
scenario simulates, e.g., a transmission line environment in addressing the resolution limits in imaging when mul-
tiple scattering is significant. This has been investigated tilayer media, in terms of resolution, as measured by the
in [6, 7, 8] in connection with the forward point of view, length of the grid cells adopted in inversions, versus estima-
where comparison is made between two different physical tion accuracy, associated to the inversion of the sought-after
scattering models, one where multiple scattering is negligi- constitutive property (permittivity) at each cell. The study
ble (Born approximation), applicable to weakly scattering considered time-domain reflective (R) and transmissive (T)
objects describable by the Born approximation, versus the data. Some of our results complement the pioneering work
more general one where multiple scattering is significant in [4]. For example, among other results, it was found that
and cannot be neglected. This has been investigated also transmissive data give a rank-1 Fisher information matrix,
in connection with the companion inverse point of view a result that is also obtained in the present paper. But we
or imaging from given data, where one compares algo- also consider complementary aspects not addressed in [4],
rithms valid only in the Born approximation versus more such as the Born approximation versus multiple scattering
general methods applicable in the exact scattering frame- comparison, more sensing configurations (in particular, be-
work [8, 11, 12]. It has been shown that so-called funda- sides R, T, and the one-sided reflective and transmissive
mental “resolution limits” in imaging such as the Rayleigh data (RT), we also explore two-sided reflective data plus
resolution limit hold only in the Born or linearizing ap- single-sided transmissive data (RRT), and two-sided reflec-
proximation regime, and that they do not generally hold tive and transmissive data (RRTT)), as well as the role of
under active inverse-scattering-based imaging if significant a priori information on the scatterer strengths or scatterer
multiple scattering is involved. Enhancement of resolution positions.
via multiple scattering is possible. However, clearly the The remainder of this two-part paper is organized as
enhancement may depend on the particular remote sens- follows. In the present, part I paper, we focus on the theo-
ing configuration and particular scattering parameters, and, retical formulation, and in the following, part II paper we
in fact, multiple scattering can either enhance or diminish focus on the computer simulation aspects and the interpre-
resolution [7]. Both possibilities are demonstrated in the tation of the results. Section 2 of this first paper develops
present study. the forward scattering analysis in 1D space. As is custom-
It is important to emphasize that the real physical arily done in imaging resolution studies, we focus on the
model is, of course, the multiple scattering one. The Born particular case of two point-like scatterers whose scattering
approximation is valid for weakly scattering objects only. parameters (scattering strengths, positions, scatterers sepa-
However, in comparing the two models one gains insight ration, etc.) one wishes to estimate from scattering data.
into the possible gain in resolution due to multiple scatter- Section 3 presents, for two point scatterers, and for single-
ing. Clearly if there are scattering configurations for which and multi-frequency cases, the CRB calculations for differ-
the Born approximation model outperforms the multiple ent scattering parameters of interest.
scattering one, this is clearly a fictitious and non-physically
important result, since in the same physical situation only 2 Scattering Formulation
the multiple scattering model is exact. Thus for fixed scat-
tering parameters, if it is found that the Born approxima- Consider in 1D free space, primary scalar sources (ρ) and
tion outperforms the multiple scattering model, this means scalar fields (incident fields) (ψ i ) related by the scalar
that in reality the Born-approximation resolution limit actu- Helmholtz equation
ally is too optimistic for those particular scattering parame-
 2 
ters, while if on the contrary the multiple scattering model ∂ 2
is better, then this means that the diffraction limit associ- +k ψi (x) = ρ(x) (1)
∂x2
ated with the Born approximation is in fact too pessimistic
and that better resolution (super-resolution) is possible. so that the incident fields
Thus in the present contribution we re-visit the use 
of CRB in imaging systems to further examine the role of ψi (x) = G0 (x, x )ρ(x )dx (2)
multiple scattering in subwavelength imaging. But besides
this issue, we are also interested in taking advantage of where the free space Green function (see [13], p. 912)
the derived CRB framework to address other open ques-
tions. One is the role of different measurement configu- i
G0 (x, x ) = − exp(ik|x − x |). (3)
rations for scatterer information extraction. They reduce 2k
to 3 canonical arrangements: reflective (R) system (like a It obeys
monostatic radar or lidar system), transmissive (T) system
 
(like a bistatic radar or lidar system), and combined trans- ∂2 2
missive and reflective setups (like a more general multi- + k G0 (x, x ) = δ(x − x ) (4)
∂x2
static system). The aspect of sensing configuration is also
treated by Gustafsson and Nordebo [4] who employed the and outgoing wave conditions at infinity. In these results,
CRB as a way to assess the ill-posedness of the electro- k = ω/c is the wavenumber of the field at angular oscilla-
magnetic inverse scattering problem for nonmagnetic mul- tion frequency ω where c is the free space speed of light.
 For more general media, the differential equation re- In our Cramer-Rao analysis we shall consider five dif-
lating sources ρ and (total) fields ψ t is ferent groupings of these data, for single frequency and the
 2  more general multi-frequency case:
∂ 2
+ k ψt (x) = V (x)ψt (x) + ρ(x) (5) • Case R: Reflective −, −.
∂x2 • Case T: Transmissive +, −.
• Case RT: Reflective −, −, and transmissive +, −.
where the scattering potential V = κ 2 (x) − k 2 where κ(x) • Case RRT: Reflective −, −, reflective +, +, and trans-
is the space-dependent wavenumber of the field in the (to- missive +, −.
tal) medium. • Case RRTT: Reflective −, −, reflective +, +, transmis-
For the particular case of point-like scatterers of in- sive +, −, and transmissive −, +.
dividual strengths or reflectivities τ m , m = 1, 2, ..., M ,
and positions X m , m = 1, 2, ..., M , we have V (x) =
M 2.2 The Forward Scattering Map in the Born Approx-
m=1 τm δ(x − Xm ). If scattering is elastic (non- imation
dissipative), then τm must obey the condition derived in
Appendix A. This condition will be used in selecting val- The scattered field (total minus incident field) for the above
ues for the computer study; however, we do not require it in experiments is defined by G − G 0 , which according to
the CRB calculations which hold for arbitrary τ m . The to- Eq.(6) gives: 
tal Green function G in the medium including the scatterers • K(−, −) = G(0,0)−G0(0,0) = dxG0(0,x)V (x)G(x,0) for
obeys the case “reflective −, −”. 
• K(+,+)=G(D,D)− G0(D,D)= dxG0(D,x)V (x)G(x,D)

M
for the case “reflective +, +”.
G(x,x ) − G0 (x,x ) = G0 (x,Xm )τm G(Xm ,x )
m=1
• K(+,−)=G(D,0)− G0(D,0)= dx G0(D,x)V (x)G(x,0) =
K(−,+) for the cases “transmissive +, −” or the reciprocal

M
= G(x,Xm )τm G0(Xm ,x ), equivalent, case “transmissive −, +”.
m=1 Let us emphasize next the particular case of two
(6) point-like scatterers of individual strengths or reflectivi-
ties τ1 and τ2 and positions X 1 and X2 , so that V (x) =
a result to be used later, with emphasis on the particular τ1 δ(x −X1)+τ2 δ(x −X2).
two-scatterer case. In the remainder of the paper it will In the Born approximation for which G  G 0 we then
be assumed that the scatterers are in the region (0, D), obtain from (3,6)
and that scattering experiments are done involving point  2
sources and receivers at x ≤ 0 and/or x ≥ D. The particu- KB (−,−) = −
i
exp(i2kX1) [τ1 +τ2 exp(i2kd)] (7)
lar case studies of interest are detailed next. 2k

where the target separation d = |X1 −X2 |.

2.1 Case Studies
We also obtain
For the purposes of our analysis, it will suffice without  2
i
loss of generality to assume the following scattering ex- KB (+,+)= − exp[i2k(D−X1 −d)][τ2+τ1 exp(i2kd)]
2k
periments (refer to figure 1): (8)
• “Reflective −, −”: a point source at x = 0 (ρ(x) = We also obtain
δ(x − XT )) and a point-receiver also at x = 0. The mea-
 2
sured scattered field for this experiment will be denoted i
K(−, −). This is a particular entry in the scattering ma- KB (+,−) = − exp(i2kD)(τ1 +τ2) = KB (−,+) (9)
2k
trix K.
• “Reflective +, +”: a point source at x = D and a point- which completes the picture of scattering by two point-like
receiver also at x = D. The entry in the scattering ma- scatterers within the Born approximation. Let us consider
trix K associated to this experiment will be denoted as next the more general multiple scattering case.
K(+, +).
• “Transmissive +, −”: a point source at x = 0 and a point- 2.3 The Forward Scattering Map Including Multiple
receiver at x = D. Due to reciprocity this experiment has Scattering
the same signal content as the reciprocal counterpart in-
volving x = D as source point and x = 0 as receiver point. The more general case of multiply scattering point targets
The entries in the scattering matrix K of these experiments can be readily handled by carefully following the succes-
will be denoted as K(+, −) and K(−, +). These entries sive multiple scattering events as is done, e.g., in [14],
are, apart from noise, identical. However, in the presence p. 220-223 (see, in particular, equations 5.67-d, 5.68 in
of noise there can be additional information in taking the [14]). (Remark: However, care must be exercised in not-
complementary entry (larger SNR thanks to more samples). ing that the electromagnetic discussion in [14] applies to a
slightly different physical situation, in particular, one of a case we are limited in our resolution ability under a sin-
three-layered medium, unlike the present medium formed gle frequency experiment (we cannot distinguish between
by free space as background and two point-like scatterers distances separated by half wavelength). (Remark: This
or reflectivities embedded to that background. Thus the re- problem does not hold in the 2D or 3D case since then
sults in [14] reduce to our results after certain substitutions: there is wave amplitude attenuation in propagation, which
in our case, scattering by the point-like inhomogeneities is is not present in the 1D case). However, this is an aliasing
isotropic, hence the reflectivity by coming from the right is issue, and it does not mean that we cannot estimate sub-
the same as by coming from the left, unlike in the layered wavelength scatterer separations. As we will show next, we
medium where the reflectivity varies in sign when applied can, in the multiple scattering signal model. But, subject to
to the opposite direction of travel. But besides this issue, the ambiguity whether the estimated distance is, say, d, or
the results in [14] and the present results outlined below are d + nλ/2, n = 1, 2, .... It is in the latter interpretation that
equivalent.) we will speak in the following about subwavelength imag-
One obtains the multiple scattering generalization of ing resolution (with the clear understanding that we are re-
(7) ferring to estimation of the “principal value” of d only). In
   fact, in our computer simulations we will from the onset
i (1+ τ̂1)2τ̂2 exp(i2kd) restrict attention to d ∈ (0, λ/2).
K(−, −)= − exp(i2kX1) τ̂1+ (10)
2k 1− τ̂1τ̂2 exp(i2kd) Our next goal is to quantify and better understand the
information content for a two-scatterer system character-
where we have introduced for notational compactness
ized by scattering strengths τ 1 and τ2 , and scatterer position
  X1 and target separation d, from different scattering data
i
τ̂m = − τm m = 1, 2. (11) sets. This will be done via the fundamental Cramer-Rao
2k
bound pertinent to the variance (error) of unbiased estima-
Similarly, the multiple scattering generalization of (8) is tors for the scattering parameters τ 1 , τ2 , d and X1 , under
found to be the two signal models above (Born versus multiple scat-
   tering), under different configurations (data sets R, T, RT,
i (1+ τ̂2)2τ̂1 exp(i2kd) RRT, RRTT), and for single versus multi-frequency data.
K(+,+)= − exp[i2k(D−X1−d)] τ̂2+
2k 1− τ̂1τ̂2 exp(i2kd)
(12)
Finally, the multiple scattering generalization of (9) is 3 Cramer-Rao Study of Two-Point-Scatterer
Systems
  
i (1+ τ̂1)(1+ τ̂2) The Cramer-Rao study is developed directly in the multi-
K(+,−) = − exp(i2kD) −1 =K(−,+)
2k 1− τ̂1τ̂2 exp(i2kd) frequency case under the assumption that the scattering
(13) strengths are frequency-independent (nondispersive me-
dia) for the frequency bands used in the interrogation ex-
2.4 Born Versus Multiple Scattering Cases periments, thus in the multi-frequency case we consider
τm (ω) = τm , m = 1, 2. The latter assumption simplifies
Some differences become obvious in terms of the informa- the analysis and applies to many practical situations. For
tion available in the multiple scattering case, due to greater the more general case of dispersive scatterers, see [4]. The
contrast giving rise to multiple scattering, versus in the single-frequency case is discussed as an important special
Born approximation case for weakly scattering targets. For case.
example, we see that while the Born approximation scat-
tering matrix entry K B (+, −) in (9) has no dependence on
3.1 Signal Model and Parameters
the target separation d, the multiple scattering counterpart
(13) does exhibit such a dependence. Thus in the Born ap- In practice, one collects noisy data which we take into ac-
proximation the transmissive experiment giving K B (+, −) count via the signal model:
provides no information about the target separation, but in-
stead gives us only (apart from a constant) the sum of the
(ξ) = K̄ + W
K (14)
two target strengths (corresponding to low-frequency in-
formation about the scattering potential as a whole). In where K̄ is the noise-free data vector formed by the avail-
contrast, in the more general multiply scattering case the able entries of the response matrix K, which depend on the
same transmissive experiment does provide such informa-

remote sensing configuration (R, T, RT, etc.), K(ξ) is the

tion thanks to multiple scattering events between the two noisy realization of that matrix, W is white Gaussian noise
scatterers. with variance σ02 (which is handled next as nuisance pa-
We also note, however, that the dependence on d rameter that needs to be estimated from the data), and ξ is
is of the form exp(i2kd) which implies that d and d + the vector of the scattering plus noise parameters that ones
nλ/2, n = 1, 2, ... (where λ = 2π/k) exhibit the same wishes to estimate from the data. We consider in this paper
value of K(+, −) so that even in the multiple scattering different vectors ξ corresponding to different combinations
of the following parameters: The position X 1 of target 1, where H denotes the conjugate transpose, and where the
2
the target separation d, the real and imaginary parts τ (r)m covariance matrix [3, pp. 501] C K
= σ0 I where I de-
and τ(i)m of target strength τ m , m = 1, 2, and the unknown notes the identity matrix, whose size is determined by the
noise variance σ02 . Thus we consider the case of no a priori length of the data vector K
which we will denote next as
knowledge, where the parameter vector L. Clearly L is equal to Nf for cases R and T, to 2N f for
 T case RT, to 3Nf for case RRT, and to 4N f for case RRTT.
ξ = X1 , d, τ(r)1 , τ(r)2 , τ(i)1 , τ(i)2 , σ02 (15) The CRB (CRB[ξ(i)]) is the lower bound for the variance
var[ξ(i)] of any unbiased estimator for the parameter ξ(i),
as well as the special cases where τ1 , τ2 are known (mod-
and is defined in terms of the Fisher information matrix
eling a priori knowledge of the scatterer’s materials), so
I(ξ) by [3, eq.3.20]
that  T
ξ = X1 , d, σ02 (16) var[ξ(i)] ≥ [I −1 (ξ)]i,i = CRB[ξ(i)]. (19)
and where the positions X 1 , X2 (and thus also d) are known
Under the parameter vector ξ in (15) which is of
(modeling a priori knowledge of the support of the scatter-
length 7, the Fisher information matrix associated to the
ers), so that
signal model Eq.(14) reduces to
ξ = [τ(r)1 , τ(r)2 , τ(i)1 , τ(i)2 , σ02 ]T . (17)  
L 2 ∂K
H (ξ) ∂ K
(ξ)
In the following we shall refer to the cases in (16) and (17) I(ξ) = δi,j δi,7 + 2 
σ04 σ0 ∂ξi ∂ξj
as “known material” and “known support”, respectively. ⎡ H ⎤

∂K

H ∂ K

The three cases in (15, 16, 17) are studied next for both ∂X1 ∂X1 . . . ∂τ(i)2 ∂X1 0
∂K ∂K

single- and multi-frequency problems.  ⎢ .. .. .. ⎥

2 ⎢ .. ⎥
Consider first the single-frequency case. Then for the
L ⎢ . . . .⎥
= δi,j δi,7 + 2⎢ ⎥(20)
different remote sensing cases, the noise-free data vector σ04 σ0 ⎢∂ K

H


H

.. ⎥
⎣ ∂X ∂τ . . . ∂τ ∂τ . ⎦
∂ K ∂ K ∂ K
K̄ is as follows (with T denoting transpose): 1 (i)2 (i)2 (i)2
0 ... ... 0
• Case R: K̄ = [K(−, −)]
This holds for both Born and multiple scattering cases. The
• Case T: K̄ = [K(+, −)] results for the parameter vectors ξ in (16) and (17) are of
the same general form, and are constructed by suitable re-
• Case RT: K̄ = [K(−, −)K(+, −)]T duction of the Fisher information matrix above, thus, for
• Case RRT: K̄ = [K(−, −)K(+, −) K(+, +)]T example, for the ξ vector in (16) (“known material” case)
one obtains
• Case RRTT: K̄ = [K(−, −)K(+, −)K(−, +)K(+, +)]T ⎡ H ⎤
 ∂K
∂K
∂K

H ∂ K
0
2 ⎢ ∂X1H ∂X1

0⎥
L ∂d ∂X1
If data are gathered for a number N f of frequencies I(ξ) =
∂K
δi,j δi,3 + 2⎣ ∂ K
∂K
H ∂ K ⎦ (21)
4
ω1 , ..., ωNf , then in place of the scalar K(−, −) we have σ0 σ0 ∂X1 ∂d ∂d ∂d
0 0 0
the vector [K(−, −; ω 1) ... K(−, −; ωNf )]T , and so on.
Then the noise-free data vector for each case is as follows: Let us highlight salient aspects of each case (R, T,
RT, etc., and Born versus multiple scattering) separately.
• Case R: K̄ = K̄−,−,N
T
≡ [K(−, −; ω1 ) ... K(−, −; ωNf ]T
f
Rather than providing the analytical expressions for the
• Case T: K̄ = K̄+,−,N
T
≡ [K(+, −; ω1) ...K(+, −; ωNf ]T Fisher information matrix for each case, which would ac-
f
count for a lengthy discussion, we limit to highlighting the
• Case RT: K̄ = [K−,−,Nf K+,−,Nf ]T general structure and rank of this matrix for each case. Ta-
bles of the rank of the Fisher information matrix for repre-
• Case RRT: K̄ = [K−,−,Nf K+,−,Nf K+,+,Nf ]T (where sentative cases of interest are provided in Tables 1, 2, and
K+,+,Nf is the “+,+” analog of the “-,-” quantity 3, for the parameter vectors in Eq.(15), (16) and (17), re-
K−,−,Nf defined above). spectively.

= K(−, −)+W , where in general
• Case R: In this case K
• Case RRTT: K̄ = [K−,−,Nf K+,−,Nf K−,+,Nf K+,+,Nf ]T
K(−, −) is given by (10) while in the Born approximation
it is approximated by K B (−, −) in (7). Consider the sin-
3.2 Fisher Information Matrix and CRB
gle frequency case N f = 1. In this case the Fisher infor-
The corresponding Fisher information matrix I(ξ) is given mation matrix has at most rank 3 so that in the no a priori
by [3, eq.15.52] knowledge and “known support” cases the Fisher informa-
 tion matrix is singular and it is not possible to estimate all
∂CK
(ξ) −1 ∂CK
(ξ)
−1
I(ξ)i,j = tr C (ξ) C (ξ) the desired parameters from the data. However, the “known

K ∂ξi

K ∂ξj material” case which has only 3 unknown parameters ad-

 H
∂K
(ξ) −1 ∂ K(ξ)
mits a non-singular Fisher information matrix, so that the
+2 C (ξ) (18)
∂ξi

K ∂ξj CRB can be studied in that case. The rank is 3 since the
 Table 3. Rank of the 5 × 5 Fisher informa-
tion matrix under single-frequency (SF) and multi-
frequency (MF) conditions for parameter vector ξ =
 T
τ(r)1 , τ(r)2 , τ(i)1 , τ(i)2 , σ02 . Case A: τ1 = −0.5 + i0.5,
τ2 = −0.5+i0.5. Case B: τ1 = −0.5+i0.5, τ2 = −0.9+i0.3.
Table 1. Rank of the 7 × 7 Fisher informa-
tion matrix under single-frequency (SF) and multi- Cases A, SF B, SF A, MF B, MF
frequency (MF) conditions for parameter vector ξ = BR 3 3 5 5
 T
X1 , d, τ(r)1 , τ(r)2 , τ(i)1 , τ(i)2 , σ02 . Case A: τ1 = −0.5+ MR 3 3 5 5
i0.5, τ2 = −0.5 + i0.5. Case B: τ1 = −0.5 + i0.5, τ2 = BT 3 3 3 3
−0.9+i0.3. MT 3 3 5 5
BRT 5 5 5 5
Cases A, SF B, SF B, MF(2 freq.) B, MF (3 or more)
MRT 5 5 5 5
BR 3 3 5 7
BRRT 5 5 5 5
MR 3 3 5 7
MRRT 5 5 5 5
BT 3 3 3 3
BRRTT 5 5 5 5
MT 3 3 6 6
MRRTT 5 5 5 5
BRT 5 5 7 7
MRT 5 5 7 7
BRRT 6 7 7 7
MRRT 6 7 7 7 signal is complex-valued so that K(−, −) contains two in-
BRRTT 6 7 7 7 dependent data points and the noise is independent from
MRRTT 6 7 7 7 the signal which gives up to 2 scattering system features
plus 1 noise feature (σ 02 ) that can be estimated from the
data, hence the rank 3. In the multi-frequency case this re-
flective configuration achieves a full-rank, invertible Fisher
information matrix (meaning that all parameters can be es-
timated with finite error) with at least 3 different data fre-
quencies for the no a priori knowledge case and 2 different
frequencies for the “known support” case. The rank re-
mains 3 even for multiple frequencies in the “known mate-
rial” case. This was validated with several numerical exam-
ples during the course of this investigation (refer to Tables
1, 2, and 3).
• Case T: Consider first the single-frequency regime, in the
Table 2. Rank of the 3 × 3 Fisher information matrix un-
Born approximation. From (9), clearly the signal does not
der single-frequency (SF) and multi-frequency (MF) con-
= 0 = ∂K

 T depend on X 1 or d, i.e., ∂X ∂d , so that the general

∂K
ditions for parameter vector ξ = X1 , d, σ02 . Case A: 1

τ1 = −0.5+i0.5, τ2 = −0.5+i0.5. Case B: τ1 = −0.5+i0.5, expression (20) takes the following reduced form
τ2 = −0.9+i0.3. ⎡ ⎤
0. . . . . . . . . . . . 0
⎢0. . . . . . . . . . . . 0⎥
Cases A, SF B, SF A, MF B, MF ⎢ ⎥
 ⎢0 0 ∂K
H ∂K
. . . ∂K
H ∂K
0⎥
BR 3 3 3 3 1 2 ⎢⎢ ∂τ(i)2 ∂τ(r)1 ⎥
∂τ(r)1 ∂τ(r)1
⎥
MR 3 3 3 3 I[(ξ)]i,j = 4 δi,j δi,7 + 2⎢.. .. .. .. .. ..⎥ (22)
σ0 σ0 ⎢. . . . . .⎥
BT 1 1 1 1 ⎢ ⎥
⎢.. .. ∂K
..⎥.
⎣. .
∂K
. . . ∂K
∂K
H H
MT 2 2 2 2 ∂τ(r)1 ∂τ(i)2 ∂τ(i)2 ∂τ(i)2
⎦
BRT 3 3 3 3 0. . . . . . . . . . . . 0
MRT 3 3 3 3
BRRT 3 3 3 3 which means that, as outlined earlier, there is no informa-
MRRT 3 3 3 3 tion in the data about the target positions. As illustrated in
BRRTT 3 3 3 3 Tables 1 and 3, the rank is 3 in the no a priori knowledge
MRRTT 3 3 3 3 and “known support” cases. But in the “known material”
case the rank is 1. These results hold for both single and
multiple frequencies. The rank 1 is associated only to the
noise since there is no information in the T data about the
target positions. The information content associated to the
transmissive experiment is only on the scattering strengths,
and we will see that transmissive data in combination with
reflective data (e.g., in the RRT and RRTT cases to be stud- The rank for the no a priori knowledge case is now
ied in further detail next) does enhance the estimation per- at most 7, and this value is achieved if τ 1 = τ2 . However
formance pertinent to the scattering strengths. for τ1 = τ2 the rank was found to be 6 in the single fre-
In the more general multiple scattering regime ad- quency case (refer to Table 1). But for multiple frequency
dressed in (13) the data still does not depend on X 1 (hence one recovers the rank 7.
∂K(+,−)
∂X1 = 0) but unlike the Born approximation the data For the “known material” and “known support” cases
now depends on the target separation d, so that (in con- the Fisher information matrix is full rank, as is illustrated in
trast with (22)) the Fisher information has the more general Tables 2 and 3, hence all parameters can be estimated with
structure finite error. These results hold, for both single and multi-
⎡ ⎤ ple frequency conditions, and for both Born and multiple
0 ... ... ... 0
⎢.. ∂ K scattering models.
⎢.
H ∂K

H ∂ K
..⎥⎥ • Case RRTT: Since the signals K(+, −) and K(−, +)
⎢ ∂d ∂d . . . ∂τ(i)2 ∂d .⎥
∂K

Nf 2 ⎢ .. .. .. .. ..⎥ are identical, one expects that the additional data entry
I [(ξ)]i,j = 4 δi,j δi,7 + 2⎢
⎢ . . . .
⎥
.⎥ (23)
σ0 σ0 ⎢ will provide only marginal extra information about the
⎢.. ∂ K

H


H

..⎥
⎥ scattering parameters relative to case RRT. This will be
⎣. ∂K
... ∂K ∂K
.⎦
∂d ∂τ(i)2 ∂τ(i)2 ∂τ(i)2
illustrated in the computer simulations section. In this
0 ... ... ... 0
case the Fisher information matrix is of the general form
In this case the rank is 3 in the no a priori knowledge and Eq.(20) with L = 4N f and, in the Born approxima-
 
“known support” cases, and it is 2 in the “known material” tion model ∂ K
= ∂KB (−,−) 0 0 ∂KB (+,+) T and a
case for a single frequency. The latter is an enhancement ∂d ∂d ∂d
similar expression for X 1 . In the multiple scattering
due to multiple scattering, relative to the Born model where
=
the rank is only 1. model, there is greater dependence on d, and ∂X ∂K
 T 1

Note that for the three situations associated to the pa- ∂KB (−,−)
0 0 ∂KB (+,+)
. The Fisher information
rameter vector ξ, the Fisher information matrix is singu- ∂X1 ∂X1
matrix is full rank in this case (see Tables 1, 2, and 3), ex-
lar under the transmissive data set alone, for a single fre-
cept in the single frequency case when τ 1 = τ2 .
quency. In the multi-frequency case, the rank is at most
6 for the no a priori knowledge case and remains 2 in the
“known material” case. However, in the “known support” 4 Conclusion
case the rank is 5 (full rank), consequently finite CRB for
the five parameters (strengths and noise) can be computed This first paper of the present two-part paper developed
in this case. the Born approximation and multiple scattering scattering
• Case RT: The Fisher information matrix for this case has models in 1D space, as two different signals models for the
the general form Eq.(20) with L = 2N f . In the single- subsequent CRB analysis. The general Fisher information
frequency regime, under the Born approximation, ∂∂d
=
K matrix and CRB for different single- and multi-frequency
 T cases and sensing configurations were developed and dis-
∂KB (−,−)
0 and a similar expression for X 1 . Under
∂d cussed. The companion part II paper presents the respec-
multiple scattering, there is greater dependence on d but tive computer illustration along with further discussion.
 
∂K
= ∂K(−,−) 0 T .
∂X1 ∂X1
In this case the rank of the Fisher information ma- Acknowledgments
trix is 5 in the no a priori knowledge and “known sup-
port” cases, and 3 for the “known material” case for single This work was supported by the United States National Sci-
frequency case. Therefore in the case RT, and under the ence Foundation under grant number 0746310.
“known material” and “known support” cases the Fisher
information matrix is invertible (hence all the parameters E-mail address: [email protected].
can be estimated with finite error). In the multi-frequency
case the rank of the Fisher information matrix increases for
the no a priori knowledge case, and the Fisher information Appendix A
matrix becomes invertible for at least two different data fre-
quencies. It is shown next that, under energy-conservative, elastic
• Case RRT: The Fisher information matrix has the gen- scattering, the scattering strengths τ m obey the condition
eral form Eq.(20) with L = 3N f . In the single-

= [Re(τm )]2 + Re(τm ) + [Im(τm )]2 = 0 (24)
frequency regime, under the Born approximation, ∂∂d K
 T
∂KB (−,−)
0 ∂KB∂d(+,+)
and a similar expression for where Re and Im denote real and imaginary part, respec-
∂d
tively. To arrive at this result, one notes that in the present
X1 . In the multiple scattering model there is greater de-
  scalar wave context, the quantity that is analogous to the
pendence on d but ∂ K
= ∂KB (−,−) 0 ∂KB (+,+) T .
∂X1 ∂X1 ∂X1 Poynting vector for electromagnetic waves is the energy
flux vector ([15], p. 717) which in the present 1D frame- [8] F. Simonetti, M. Fleming and E.A. Marengo, “An il-
work can be expressed in terms of the field ψ(x) as lustration of the role of multiple scattering in sub-
  wavelength imaging from far-field measurements”, J.
∗ ∂ Opt. Soc. Am. A, Vol. 25, pp. 292-303, 2008.
F (x) = βIm ψ (x) ψ(x) (25)
∂x
[9] M. Cheney and R.J. Bonneau, “Imaging that exploits
where β is a positive constant. Consider a point target hav- multiple scattering from point scatterers”, Inverse
ing strength τ . Let exp(ikx) be the incident field, so that Problems, Vol. 20, pp. 1691-1711, 2004.
the field to the left of the target is exp(ikx) + τ exp(−ikx)
[10] N. Engheta, A. Alu, R.W. Ziolkowski, and A. Er-
while the field to the right of the target is (1 + τ ) exp(ikx).
entok, in Metamaterials: Physics and Engineering
Then the energy flux to the left is F (x < 0) = βk(1−|τ | 2 ),
Explorations, ed. N. Engheta and R.W. Ziolkowski,
while the energy flux to the right is F (x > 0) = βk[1 +
IEEE/Wiley, 2006, chapter 2, pp. 43-85.
2Re(τ ) + |τ |2 ]. In elastic scattering the flux remains con-
tinuous at the scattering interface so that k(1 − |τ | 2 ) = [11] K. Belkebir, P.C. Chaumet, and A. Sentenac, “In-
k[1 + 2Re(τ ) + |τ |2 ], which gives (24), as desired. Fur- fluence of multiple scattering on three-dimensional
thermore, it can be shown that, consequently, the most gen- imaging with optical diffraction tomography”, J. Opt.
eral scattering strength τ which gives rise to elastic (non- Soc. Am. A, Vol. 23, pp. 586-595, 2006.
dissipative) scattering is of the form
[12] F. Simonetti, “Multiple scattering: The key to unravel
τ = − cos(θ) exp(iθ) (26) the subwavelength world from the far-field pattern of
a scattered wave”, Phys. Rev. E, Vol. 73, 036619-1
where θ ∈ [−π/2, π/2]. (2006).
[13] G. Arfken, Mathematical Methods for Physicists,
References third ed., Academic Press, 1985.

[1] A. Brancaccio, G. Leone, and R. Pierri, “Information [14] C.A. Balanis, Advanced Engineering Electromagnet-
content of Born scattered fields: results in the cir- ics, Wiley, New York, 1989.
cular cylindrical case”, J. Opt. Soc. Am. A, Vol. 15, [15] M. Born and E. Wolf, Principles of Optics, 7th ed.,
pp. 1909-1917, 1998. Cambridge University Press, 1999.
[2] R. Pierri, R. Persico, and R. Bernini, “Information
content of the Born field scattered by an embedded
slab: multifrequency, multiview, and multifrequency-
multiview cases”, J. Opt. Soc. Am. A, Vol. 16,
pp. 2392-2399, 1999.

[3] S. Kay, Fundamentals of Signal Processing: Estima-

tion Theory, Wood Cliffs, NJ: Prentice Hall, 1993.

[4] M. Gustafsson and S. Nordebo, “Cramér-Rao lower

bounds for inverse scattering problems of multilayer
structures”, Inverse Problems, Vol. 22, pp. 1359-
1380, 2006.

[5] M. Shahram and P. Milanfar, “Imaging below the

diffraction limit: A statistical analysis”, IEEE Trans.
Image Processing, Vol. 13, pp. 677-689, 2004.

[6] G. Shi and A. Nehorai, “Cramer-Rao bound analy-

sis on multiple scattering in multistatic point-scatterer
estimation”, IEEE Trans. Signal Processsing, Vol. 55,
pp. 2840-2850, 2007.

[7] A. Sentenac, C.A. Guerin, P.C. Chaumet, F. Drsek,

H. Giovannini, N. Bertaux, and M. Holschneider, “In-
fluence of multiple scattering on the resolution of an
imaging system: A Cramer-Rao analysis”, Opt. Ex-
press, Vol. 15, pp. 1340-1347, 2007.