microwave near-field imaging in real timeims.nus.edu.sg/events/2018/theo/files/natalia.pdf ·...
TRANSCRIPT
Microwave Near-field Imaging in Real TimeNatalia K. [email protected]
September 27, 2018
McMaster UniversityDepartment of Electrical and Computer EngineeringElectromagnetic Vision (EMVi) Research Laboratory
Hamilton, ON CANADA
Workshop on Qualitative and Quantitative Approaches to Inverse Scattering ProblemsInstitute for Mathematical Sciences, National University of Singapore
2
APPLICATIONS OF MICROWAVE NEAR-FIELD IMAGING
• penetration into optically obscured objects (clothing, walls, luggage, living tissue…) the lower the frequency the better the penetration frequency bands from 500 MHz into the mm-wave bands (≤300 GHz)
• compact relatively cheap electronics esp. in the low-GHz range
• diverse suite of image reconstruction methods
VARIOUS APPLICATIONS
whole body scanners nondestructive testing through-wall imaging
medical imaging underground radar
3
APPLICATIONS: LUGGAGE INSPECTION, NDT[Ghasr et al., “Wideband microwave camera for real-time 3-D imaging,” IEEE Trans. AP, 2017]
16×16 array
20 GHz to 30 GHz frequency range
video
[https://youtu.be/RE-PPXmtTeA]
Prof. Zoughi’s team at Missouri University of Science & Technology
4
APPLICATIONS: WHOLE BODY SCANNERS[Sheen et al., “Near-field three-dimensional radar imaging techniques and applications,” Applied Optics 2010]
40 GHz to 60 GHz (U band) cylindrical scan
10 GHz to 20 GHz polarimetric cylindrical scan
Pacific Northwest National Laboratory, Washington, USA
40 GHz to 60 GHz (U band) cylindrical scan
5
APPLICATIONS: MEDICAL IMAGING[Song et al., “Detectability of breast tumor by a hand-held impulse-radar detector: performance evaluation and pilot clinical study,” Nature Sci. Reports 2017]
Prof. Kikkawa’s team at Hiroshima University, Japan
6
MICROWAVE NEAR-FIELD IMAGING: COMMERCIAL GROWTH
• mm-wave whole-body imagers for airport security inspection ( > 30 GHz)
• through-wall and through-floor infrastructure inspection for contractors and home inspectors (UWB, 3 GHz to 8 GHz)
• numerous underground radar applications: detection of pipes, cables, tunnels, etc. ( < 3 GHz)
[video credit: https://walabot.com/diy]
[video credit: https://www.youtube.com/watch?v=_YZEa1hiGO0]
7
OUTLINE
specifics of near-field microwave imaging in
• data acquisition
• forward models
• inversion strategies (real-time)
examples and comparisons of methods
8
DATA ACQUISITION: SCANNING
Principle: N-D imaging (N=2,3) needs abundant and diverse N-D data sets
spatial scans – 1D (linear) or 2D (surface)• illuminate target from various angles• collect scattered signals at various angles/distances• acquisition surfaces – planar, cylindrical, spherical• mechanical vs. electronic scanning
bksk
sV
aS
Tx
RxRx
Rx
Rx
Rx
RxRx
mechanical scan
electronic switching
speed low HIGHcomplexity LOW highflexibility in adjusting scan parameters
GREAT limited
frequency/temporal sweeps
polarization diversity
9
DATA ACQUISITION: SPATIAL SAMPLING
spatial data diversity• each sample must add independent information – improves uniqueness• linearly dependent data may lead to ill-posed inversion problems• over-sampling: pros and cons
minmax , ,
4sinx yλζ ζ ζ
α∆ ≤ ∆ ≈ ≡
xz
aperture edges
A,maxα
scatteringobject
AΩantennabeam
A,max Amin( , / 2)α α= Ω
• effective near-field wavelengthsmay be shorter than b2 / kλ π=
eff,min max,
2
x ykπλ =
2D( , ) ( , )x yS k k S x y=
• staying below but close to the maximum spatial sampling step ensures diversity
10
DATA ACQUISITION: FREQUENCY SAMPLING
frequency data diversity in frequency-sweep measurements
• stay below but close to the maximum frequency sampling step• ensures that back-scattered signals (after IFFT) from all targets
at distances ≤ Rmax do not overlap
bmax
max max
12 4
vf fT R
∆ ≤ ∆ = ≈
OUT
TxRx
maxR
antennaarray radar
1s
is
Ns
1is +
to SPU
11
DATA ACQUISITION: TEMPORAL SAMPLING
• temporal data diversity in time-domain measurements
stay below but close to the maximum time sampling step ensures that all frequency components of the pulsed signals are fully used (Nyquist)
minmax
max
12 2
Tt tf
∆ ≤ ∆ ≈ =
s
0
known
tot
constant
sc
s
inc (2
)i (( ))ri
i Vkik k d
a aS ε εω ′′∆ ′= ⋅′∫∫∫ E rr rE r
12
FORWARD MODEL OF SCATTERING: S-PARAMETER DATA EQUATION[Nikolova et al., APS-URSI 2016][Beaverstone et al., IEEE Trans. MTT, 2017]
Green’s vectorfunction
data
• scattering from penetrable objects (isotropic scatterer)
complexpermittivitycontrast
total internal field due to Tx antenna
tot ( ) :k ′E ris
-th portk
ks
-th port
i
ka
ib
ˆ ku
ˆ iu
Tx
Rx
OUT
sV
inc ( ) :i ′E r incident internal field due to Rx antenna if it were to transmit
p, 1, ,i k N=
total number of experiments: 2pN
reciprocity: 2p p( ) / 2N N+
total internal field
,b( ) ( ) ( )r r rε ε ε′ ′ ′∆ = −r r r
13
FORWARD MODEL: BORN’S LINEARIZING APPROXIMATION
• total field is generally unknown and depends on contrast: data equation is nonlinear in the unknown contrast
• Born’s approximation of the total internal field linearizes the data equation:
( ) nctot i; ( )) (k kε ′≈′ ′∆E Er rr
( )s
tosc n t0 i ci ( )2
( ) ; ( )iik Vi kr rkS d
a aωε ε ε′ ′ ′∆ ∆′ ′= ⋅∫∫∫ r rr rE Er
Green’s function
data contrast(to be found) total internal field
( )tot ; ( )k ε′ ′∆E r r
s
inc inc
resolved
s
nt kernel(assume known)
c ( ) ( ) ( )ik V ir kS dε ′⇒ ≈ ′∆ ′ ′⋅∫∫∫ E r E r rr
• main challenge of near-field imaging with linear inversion methods: incident-field distributions in antennas’ near zone are difficult to model
14
• object is in the near-field region of antennas
2A,max
far2Dr D
λ< ≈
any one of the conditions below implies near-field (short-range) imaging
• distance from object to antennas ≤ object’s size OUT,maxr D≤
• distance from object to antennas ≤ wavelength
r λ≤
• implication: resolvent kernel depends on incident fields which do not conform to analytical free-space far-zone propagation models, e.g.,
biinc ˆ( ) ( , )
k reGr
θ ϕ−
′E r p
Goran M Djuknic - US Patent 6657596, commons.wikimedia.org/w/index.php?curid=20417988
not valid
FORWARD MODEL: FAR-ZONE vs. NEAR-ZONE ANTENNA FIELDS
15
LINEARIZED FORWARD MODEL: TIME DOMAIN
• Born’s linearizing approximation is applied in the same way
time-domain (pulsed-radar waveforms)
s Rx Tx
inc i
,T
scR R
( ; ;T
ncxx x
)x ( )( , ; ) ( )
V ts u dt hκ
′′′∗ ′′ ≈ ∫∫∫
r r rrr r r
data(known)
contrast(unknown) impulse response of
Rx incident field(assumed known)
2b
( ) r
vεκ ∆′ =r
[Nikolova, Introduction to Microwave Imaging, 2017]
Tx incident field(assumed known)
frequency-domain (S-parameters)
s
inc insc c( )( ) ( )ik V r kiS dε ′⋅′ ′′≈ ∆∫∫∫ rE E rr r
16
LINEARIZED FORWARD MODEL: BORN vs. RYTOV DATA APPROXIMATION
• Born scattered-field data approximation
• Rytov scattered-field data approximation
sa
sc tot inc inc inca ( )
data calibration step
( ) ( ) ( ) ( )ik ik ik r i kVSS S S S dε
∈ ′ ′ ′ ′∈ ≈ − ≈ ∆ ⋅ ∫∫∫r
r r E r E r r
sa
totsc inc inc inc
a inc
data calibration step
( ) ln ( ) ( ) ( )ikik ik r i kV
ik S
SS S S dS
ε∈
′ ′ ′ ′∈ ≈ ≈ ∆ ⋅
∫∫∫
r
r r E r E r r
[Tajik et al., JPIER- B, 2017][Shumakov et al., Trans. MTT, 2018]
17
FORWARD MODEL: APPROXIMATIONS
total internal field approximation (Born)• limitations on both size and contrast of the scatterer
2 2 2s b s( ) 1,a k k V− ∈r r
bksV
s ( ')k r s ( ')k r
a
• if OUT violates limit: image contains artifacts which reflect differences between and rather than contrasttot
Tx ( )′E rincTx ( )′E r
[Nikolova, Introduction to Microwave Imaging, 2017]
data approximation
s b a2 ( ) ,a k k Sπ− < ∈r r• Born – limitation on both size and contrast:
• Born – neglects multiple scattering & mutual coupling between antennas and OUT
• Rytov – limitation on contrast only: ( )2 2 2s ab b/ 1, k k k S− < ∈r
• expect trouble in areas of strong multiple scattering & mutual coupling
18
LINEAR INVERSION METHODS: THE ENGINES OF REAL-TIME IMAGING
s
incs incc ( ( )) ( )ik V kriS dε ′′ ′≈ ⋅ ′∆∫∫∫ E r rE rr
principle: linearize forward model & solve resulting linear system of equations
data(known)
contrast(unknown)
normalized incident fields(assumed known)
• contrast and compare with nonlinear inversion methods principle: solve nonlinear forward-model equations for BOTH contrast andtotal field using nonlinear optimization and/or iterative methods
( )s
tincsc to( ( )) ; ( )r rki iVkS dε ε′ ⋅′ ′ ∆ ′≈ ′∆∫∫∫ r rrE rr E
data(known)
contrast(unknown)
Green's function(assumed known)
total internal field(unknown)
Maxwell’s equations
19
RECONSTRUCTION WITH FREQUENCY-SWEEP DATA: HOLOGRAPHY
• reflected signals: S11, S22• transmitted signals: S21, S12
type of response
number of values Sik(xꞌ,yꞌ)
co-pol X-X 4xNω
co-pol Y-Y 4xNω
cross-pol X-Y 4xNω
cross-pol Y-X 4xNω
TOTAL 16xNω = NT
number of responses acquired at each position
x
y
D
z
co-pol
cross-pol
#1#2
OUT
holography refers to reconstruction methods that use both the magnitude and phase of the scattered waves recorded at a surface to produce a 3D image in a single inversion step
Tx
Rx
for a pair of antennas
20
HOLOGRAPHY: RESOLVENT KERNEL
s
inc inc,Rx ,Tx ( , , ; , , ; )
sc ( , , ,)( , , ; ) 1, , Tx y zV r x y zS x y z dx dy dz Nx y zξ ξ ξ ω
ω ξε′ ′ ′
′ ′ ′ ′ ′ ′≈ =⋅∆ ∫∫∫ E E
approximate (Born) resolvent kernel ( ; ; )ξ ω′r r
• assume kernel is translationally invariant in x and y (background is uniform or layered)
0,
Then ( , , ; , , ; )( , ; ; )
x y z x y zx x y y z
ξ
ξ
ωω
′ ′ ′ =′ ′ ′− −
0,Let ( , ; ; )( , , ;0,0, ; )
x y zx y z z
ξ
ξ
ωω
′ ′ ′′ ′ ′≡
s
sc0,( , , ; ) ( , , ) ( , ; ; )rV
S x y z x y z x x y y z dx dy dzξ ξω ε ω′ ′ ′ ′ ′ ′ ′ ′ ′⇒ ≈ ∆ − −∫∫∫
antennas at aperture center
x
z(0,0)
Tx 0z = Rxz D=
( , )x y
inc0,Tx ( , , )x y z′ ′ ′E
inc0,Tx ( , , )x x y y z′ ′ ′− −E
21
HOLOGRAPHY: RESOLVENT KERNEL, examples
• examples: analytical kernels used with far-zone reflection data 2 2 2
bi2 ( ) ( ) ( ')inc incRx Rx Rx( , , ; , , ; ) k x x y y z zx y z x y z eξ ω ′ ′− − + − + −′ ′ ′ = ⋅E E
2 2 2bi2 ( ) ( ) ( ')
Rx 2 2 2( , , ; , , ; )( ) ( ) ( ')
k x x y y z zex y z x y zx x y y z zξ ω
′ ′− − + − + −
′ ′ ′′ ′− + − + −
inc inc,Rx ,Tx( )ξ ξ=E E
plane waves:
spherical waves:
(2) 2 20 (2 ), ( ) ( ) ,H k x x y y z z constρ ρ ′ ′ ′= − + − = =cylindrical waves:
• far-field analytical kernels do not work well with near-field data
• kernels computed from simulated incident-field distributions suffer from modeling errors [Amineh et al, Trans. AP, 2011]
• near-field kernels are best determined through measuring the system PSF[Savelyev&Yarovoy, EuRAD 2012][Amineh et al., IEEE Trans. Instr.&Meas., 2015]
22
HOLOGRAPHY: MEASURING THE POINT-SPREAD FUNCTION (PSF)
• PSF is the system response to a point scatterer
scattering probe
• relating PSF to kernel
0 , ( , ,PSF0,
; )PSF ( , , ; , ;0 )
z x y zx y zzξ
ξ ωω
′ ≡′
Let
probe at centerof plane z const′=
( )0, 0 , ,sp sp( , ; ; ) PSF ( , , ; ) /z rx y z x y zξ ξω ω ε′′ = − − ∆ ΩThen
• PSF-based kernels enable quantitative imaging in real time
23
DATA EQUATION OF HOLOGRAPHY IN TERMS OF PSF
sc0 ,
,sp sp
2D convolution
1( , , ; ) PSF ( ;( , , ,) )zr z y x
rS x y z x x y yy d y dzx z x dξ ξ ωεωε
′′ ′ ′
′ ′ ′ ′ ′≈ ⋅ − −∆ Ω
′ ′ ′∆∫ ∫∫
• in real space
• in Fourier (or k) space
sc0 ,
,sp sp( , ; ; () PSF ;, ( ,; ) )x zx y xy yzr
Fx yS k k z kk k dz zk ξξ ω ωε
′′
∆ ′′ ′∆ ′≈ ⋅
∆ Ω ∫
2DFT ( , , )r x y zε ′ ′ ′∆
• system of equations to solve at each spectral position
( )( )0 ,
1( ) PSF ( )( ; )
zNmmzn
nf zS ξξ ′
=
′≈ ∑κ κκ
1, ,1, , T
m NN
ωξ==
( , )x yk k=κ
,sp sp( ; ) ( ; )n
n nr
x Fyz zf zε′ ′ ′∆ ∆ ∆
= ⋅∆ Ω
′ ′κ κ
vΩ
24
ADVANTAGES OF SOLVING IN FOURIER SPACE: DIVIDE AND CONQUER
( )( )0 ,
1( ; )( ) PSF ( )
zNmmzij ij
nij nfS z ξξ ′
=
′≈ ∑κ κκ
1, ,1, , T
m NN
ωξ==
( , )1, , ; 1, ,
ij x yx y
i k j ki N j N
= ∆ ∆= =κ
[ ] [ 1] [ 1]( ) ( ) ( )T z z Tij N N N ij N ij N Nω ω× × ×⋅ =A κ f κ d κ
(e.g. 60 5)T zN N Nω × ×
• we solve (Nx⸱Ny) such systems (on the order of 104 to 105)
• the size of each system is small:
• typical execution times: 2 to 3 seconds on a laptop using Matlab
• solution is orders of magnitude faster than solving in real space: 6 7
v v7 8
where 10 to 1010 to 10
D x y z
D x y T
N N N N N NN N N N Nω
× ==
25
FINAL STEP: BACK TO REAL SPACE
,sp sp 12D( , , ) ( ; ) , 1, ,r
r n n zn
x y z f z n Nx y zε
ε −∆ Ω′ ′ ′ ′∆ = =
′ ′ ′∆ ∆ ∆κ
• at each plane along range (z' = const)
,b( , , ) ( , , )r n r r nx y z x y zε ε ε′ ′ ′ ′ ′ ′= + ∆
5mm250MHz
x yf
∆ = ∆ =∆ =
26
EXAMPLE: METALLIC TARGETS IN AIR
X-band (WR90) open-end waveguides ( 6.56GHz)cf ≈
f (GHz) λ (mm) Dfar (mm)3 100 12.5
8.2 37 3420 15 83
[photo credit: Justin McCombe]
[Amineh et al., Trans. Instr.&Meas., 2015]
SP
27
METALLIC TARGETS IN AIR – RESULTS WITH SIMULATED KERNELS[Amineh et al., Trans. Instr.&Meas., 2015]
28
METALLIC TARGETS IN AIR – RESULTS WITH MEASURED KERNELS (PSF)
expected spatial resolution
depth:10 mmzδ ≈
lateral:, 4 mmx yδ ≈
[Amineh et al., Trans. Instr.&Meas., 2015]
29
EXAMPLE: IMAGING TISSUE
[photo credit: Daniel Tajik]Tissue Color
HighlightRelative Permittivity Averaged over 3 to 8 GHz
Chicken Wing NABone 21 10iSkin 13 6iMuscle 45 23iPeanut Butter & Jam 7 3i
Carbon Rubber 10 3i
[Tajik et al., JPIER-B 2017]
3mm100MHz
[3,8] GHz
x yf
f
∆ = ∆ =∆ =∈
a layer in the OUT
30
EXAMPLE: IMAGING TISSUE
[photo credit: Daniel Tajik]Tissue Color
HighlightRelative Permittivity Averaged over 3 to 8 GHz
Chicken Wing NABone 21 10iSkin 13 6iMuscle 45 23iPeanut Butter & Jam 7 3i
Carbon Rubber 10 3i
[Tajik et al., JPIER-B 2017]
3mm100MHz
[3,8] GHz
x yf
f
∆ = ∆ =∆ =∈
a layer in the OUT
31
EXAMPLE: ACQUIRING THE PSF[Tajik et al., JPIER-B 2017]
[photo credit: Daniel Tajik]
calibration object with small scattering probe at center: ,SP 18 i0, radius 5 mm, height 10 mmrε ≈ −
32
EXAMPLE: IMAGING TISSUE[Tajik et al., JPIER-B 2017][ EuCAP 2018]
rε ′ rε ′′
33
QUALITATIVE IMAGING WITH SENSITIVITY MAPS[Tu et al., Inv. Problems, 2015]• reconstruction formula
a
sc
incinc tot
( , )1( , )
( , )( )
( )
TN
S
S
SD S S d d
ξ
ξξ ξω ω
ξω
ωω
ε
∗
∈=
≈−
∂ ′ = − ⋅ ′∂
∑∫ ∫∫r r
r
rr r
r
• sensitivity map: 3D image of Fréchet derivative of ℓ2 norm of the differences of all total and incident responses
a
2tot inc21
( ) [ ( )] 0.5 ( , ) [ , ; ( )]TN
SD F S S d dξ ξω
ξε ω ω ε ω
∈=
′ ′ ′= = −∑∫ ∫∫rr r r r r rJ
( )( )
( )( )
Re ( )( )
Im ( )( )
mm
mm
dFDd
dFDd
ε
ε
′ =′ ′
′ = −′′ ′
rr
rr
→ indicates where contrast in ε' exists
→ indicates where contrast in ε" exists
adjoint sensitivity formula using simulated incident fields
34
SENSITIVITY MAPS USING MEASURED PSFs: THE NEAR-FIELD CASE
tot incinc inc sc,sp
sp sp
( , ) ( , )( , ) ( , ) PSF ( , ; )( ) ( ) ( )
S SS S ξ ξξ ξ ξω ωω ω ωε ε ε ε
′−∂ ∆≈ ≈ =
′ ′ ′∂ ∆ ∆ ∆
r rr r r rr r r
• from simulated incident fields to measured PSFs
scattering probe at center of z' plane
sc sc,0PSF ( , , , ; , , ) PSF ( , , , ; )x y z x y z x x y y z zξ ξω ω′ ′ ′ ′ ′ ′= − −
scattering probe at r'
uniform background along x and y
35
• sensitivity reconstruction formula with PSFs: scattered-power maps (SPM)
a
sc scsp ,0
1( ) ( ) ( , ) PSF ( , ; )
TN
SD M S d dξ ξω
ξε ω ω ω
∗
∈=
′ ′ ′−∆ ⋅ = = ⋅ ∑∫ ∫∫rr r r r r r
• SPM reconstruction formula with planar scanning
( )a
s( )
1
sc c( , , , ) PSF ,( , , ) , , ;T
S
NmM S x y z x x y y z dx y z dz dx yξω
ξξ ωω ω
=
∗ ′ ′ ′⋅ − − ′ ′ ′ = ∑ ∫∫ ∫
sc sccross-correlation of response and PSF in ( , )S x yξ ξ
[Shumakov et al., IEEE Trans. MTT, 2018][Nikolova, Introduction to Microwave Imaging, 2017]
SENSITIVITY MAPS USING MEASURED PSFs: THE NEAR-FIELD CASE – 2
• reconstruction is practically instantaneous – no systems of equations are solved
• effort shifted to near-field system calibration – measuring PSFs or simulating incident fields to compute response sensitivity inc ( , ) / ( )Sξ ω ε ′∂ ∂r r
36
SCATTERED-POWER MAPS: SIMULATION EXAMPLE (Altair FEKO)
simulation of data acquisitionsimulation of PSF acquisition
sample PSF: S11 at 4 GHz MAG/PHASE
r,b 1.0ε =r,sp 1.1ε = r,b 1.0ε =
r,OUT 1.2ε =
min
max
3 GHz16 GHz
1 GHz
fff
==
∆ =
37
SCATTERED-POWER MAPS: SIMULATION EXAMPLE (F SHAPE)
• blurring typical for cross-correlation methods – diffraction limit, limited number of responses
38
QUANTITATIVE REAL-TIME IMAGING WITH SPM
• requires measured PSFs – they scale accurately with the scattering-probe contrast
• reconstruction solves the linear problem
sr sp@
r,sp sp(( ) ( ))1
VM M dε
ε ′′′ ′ ′′=Ω
′∆∆
′∫∫∫ rr r rr
qualitative image (SPM) of OUT
qualitative image (SPM) of scattering probe
unknown contrast
• real-time solution via inversion in Fourier space (similar to holography)
sp@(0,0, )r,sp sp
convolution in ( , )
r1( , , ) ( , , )( , , ) z
z y x
x y
M x y z M x x y y zx y z dx dy dzε
ε ′′′′ ′′ ′′
′ ′ ′ ′ ′′ ′ ′′ ′ ′′ ′′ ′′= ⋅ − −∆ Ω
′′ ′′ ′′∆∫ ∫ ∫
[Shumakov et al., IEEE Trans. MTT, 2018][Tu et al., Inv. Problems, 2015]
39
QUANTITATIVE REAL-TIME IMAGING WITH SPM: FOURIER SPACE SOLUTION
vsp@(0,0, )
1r,sp sp( , , ) ( , , ), 1, ,( , , )
z
q
N
x y p z x y zx qq
y pM k k z Mf k kk z p Nk zε =
Ω= ⋅ =∆ Ω ∑
• small square system of equations to solve at each spectral position ( , )x yk k=κ( ) ( ) ( )=κ κ κM x m
( ) 1( , ) ( , )z
TNf z f z = κ κ κx
( ) 1( , ) ( , )z
TNM z M z = κ κ κm
1
1
sp@(0,0, ) 1 sp@(0,0, ) 1
( )
sp@(0,0, ) sp@(0,0, )
( , ) ( , )
( , ) ( , )
Nz
z N zz
z z
z N z N
M z M z
M z M z
=
κ
κ κ
κ κM
• final step: back to (x,y) space
,sp sp 12D
v( , , ) ( ; ) , 1, ,r
r n n zx y z f z n Nε
ε −∆ Ω′ ′ ′ ′∆ = =
Ωκ
40
QUANTITATIVE SPM: SIMULATION EXAMPLE (F SHAPE)
Re rε Im rε
[Nikolova, Introduction to Microwave Imaging, 2017]
41
QUANTITATIVE SPM: EXPERIMENTAL EXAMPLE
5 cm thick carbon-rubber sample
2nd layer
4th layer
sheet with scattering probe
,b 10 i5rε ≈ −
Re rε
Im rε
[Shumakov et al., IEEE Trans. MTT, 2018]
42
TIME DOMAIN FORWARD MODEL WITH PSF
4th layer
• linearized time-domain resolvent kernel
s
sc inc incRx Tx
( ; , ), (( ))
V ts t h h w dκ
′ ′′ ′≈ ∗′ ∗ ∫∫∫
r rr r r
incTx( )u ′′
inc incRx Rx Tx Tx( ; , ) ( ; , ) ( ; , ) ( )t h t h t w t′ ′ ′ ′′⇒ = ∗ ∗r r r r r r
• resolvent kernel and PSF all over again: measure response with each antenna pair and position of a point scatterer at
Rx Tx
inc inc0 Rx Tx sp sp Rx Tx ( 0; , , )
sp sp 0 Rx Tx
PSF ( , , )( , , )
tt h h w
tκκ
′= ′′= Ω ∗ ∗
= Ωr r r
r rr r
[Nikolova, Introduction to Microwave Imaging, 2017]
0′ =r
2b
( ) r
vεκ ∆′ =r
43
TIME DOMAIN FORWARD MODEL WITH PSF – 2
• assuming uniform background: ( )Rx Tx 0 Rx Tx( , ; ) , ; ( )t t t′ ′≈ − ∆r r r r r r
0′ ≠r
• example: analytical far-zone kernelTx Rx 2
Rx Tx 0 Rx Tx 0b b
( , ; ) PSF ( , , ) /t t t t rv v
δ′′ ′ − −
− − −
rr r r r
r r r r
some reference time
implied in r
44
DELAY AND SUM (DAS): THE CROSS-CORRELATION EXPLANATION
• cross-correlation – a measure of signal similarity
a
s a
,01
,0 ,01
( , ) PSF ( , , ) ( , )
( ) PSF ( , , ) PSF ( , , )
T
T
N
S
N
V S
X t t s t d
t t d d
ξ ξξ
ξ ξξ
κ
∈=
′′∈ ∈=
′ ′= ⊗
′′ ′ ′′ ′′⋅ ⊗
∑∫∫
∑∫∫∫ ∫∫
r
r r
r r r r r
r r r r r r r
steering filter for antenna pair at r toward voxel r'
• with large number of responses, X(r',t) ~ κ(r') as autocorrelation term dominates r" integral
a,0 ,0
1autocorrelation term: ( ) PSF ( , , ) PSF ( , , )
TN
St t dξ ξ
ξκ
∈=
′ ′ ′⋅ ⊗∑∫∫rr r r r r r
a,0 ,0
1cross-correlation terms: ( ) PSF ( , , ) PSF ( , , )
TN
St t dξ ξ
ξκ
∈=
′′ ′ ′′⋅ ⊗∑∫∫rr r r r r r
signal processing
45
DAS: SIGNAL-FLOW SCHEMATIC
• image generation: plot the “intensity” distribution max 20
( ) ( , )T
tI X t dt
=′ ′= ∫r r
46
DAS: CONCEPTUAL EXAMPLE
47
DAS: CONCEPTUAL EXAMPLE – 2
( , )X t ′r 2( ) ( , )t
I X t dt′ ′= ∫r r
plot image
48
CONCLUDING REMARKS
• we have just grazed the surface of an extensive subject
• real-time microwave imaging is rapidly growing and developing
hardware – antennas & RF/radar electronics
calibration methods
inversion methods
49
50
IMAGE SPATIAL RESOLUTION – WHAT TO EXPECT
• lateral (or cross-range) resolutioneff,min
, max,4 2x y
x ykλ πδ ≥ =
[Nikolova, Introduction to Microwave Imaging, 2017]
• depth (or range) resolution
eff,min bmax2 2zz
vBk
λ πδ ≥ = ≈
• wide viewing angles are critically important: wide-beam antennas, large scanned apertures
51
APPLICATIONS: NONDESTRUCTIVE TESTING[Sheen et al., “Near-field three-dimensional radar imaging techniques and applications,” Applied Optics 2010]
X-BAND (8 TO 12 GHZ) SCANNER WITH 16×16 ELECTRONICALLY SWITCHED ARRAY: ABSORBER INSPECTION
16×16 array
Pacific Northwest National Laboratory, Washington, USA
52
APPLICATIONS: THROUGH-WALL IMAGING[Depatla et al., “Robotic through-wall imaging,” IEEE A&P Mag. 2017]
Prof. Mostofi’s team at the University of California Santa Barbara
https://www.youtube.com/watch?v=THu3ZvAHI9A
53
LIMITATIONS OF BORN’S APPROXIMATION: TOTAL INTERNAL FIELD
• limitations on both the size and the contrast of the scatterer
2 2 2s b s( ) 1,a k k V− ∈r r
bksV
s ( ')k r s ( ')k r
a
• if OUT violates the limits: images contain artifacts which reflect differences between and rather than contrast
totTx ( )′E r
incTx ( )′E r
EXAMPLE: TOTAL VS. INCIDENT INTERNAL FIELD
0 10 20 30 40 50 60
10
15
20
25
30
mag
nitu
de
0 10 20 30 40 50 60
position inside scatterer (mm)
-200
-100
0
100
200
phas
e (d
eg)
r = 1.0
r = 1.1
r = 4.0
60 MM THICK DIELECTRIC SLAB IN AIR
[Nikolova, Introduction to Microwave Imaging, 2017]
1 GHzf =
1 GHzf =
54
NUMERICAL ASPECTS OF THE SOLUTION IN FOURIER SPACE
( )( )0 ,
1( ; )( ) PSF ( )
zNmmzij ij n ij
nf zS ξξ κ κκ ′
=
≈ ′∑
1, ,1, , T
m NN
ωξ==
( , )1, , ; 1, ,
ij x yx y
i k j ki N j Nκ = ∆ ∆= =
( ) ( ) ( )ij ij ijκ κ κ⋅ =A f d
(1) ( )1 1 1
( ) ( ) ( ) , ( ) ( ) ( ) , 1, ,T T
TT NT T Tij ij ij ij ij ij TN N N N
S S Nω
ω ωξ ξ ξκ κ κ κ κ κ ξ
× × = = = d d d d
1 1( ) [ ( ; ) ( ; )]z zT
ij ij ij N Nf z f zκ κ κ ×′ ′=f
the data vector:
the contrast vector:
the system (PSF) matrix:
solve (Nx⸱Ny) such systemsof size: T zN N Nω ×
1
1
(1) (1)0 , 0 ,1
( ) ( )0 , 0 ,
( ) PSF ( ) PSF ( )( ) where ( )
( ) PSF ( ) PSF ( )
Nz
TNz
z zij ij ij
ij ijN NN ij z zij ijω ω
ξ ξ
ξ
ξ ξ
κ κ κκ κ
κ κ κ
′ ′
′ ′
= =
AA A
A
55
SYNTHETIC FOCUSING – 2
56
THE FORWARD MODEL OF HOLOGRAPHY
s
inc inc,Rx ,Tx ( , , ;
sc
data, , ; )
( , , 1; , ,, , ,)) (r x y Tz xV y zx dx dyS x dz Nzy z yξ ξ ωξε ξω
′ ′ ′ ′ ′′ ′ ′≈ = ′⋅∆∫∫∫ E E
• the S-parameter forward model
• ξ (response type) replaces (i,k)
response1 S11
2 S22
3 S12 = S21
ξ ( , )i k(1,1)(2, 2)
(1,2) or (2,1)3TN =
TYPICAL RESPONSE TYPES
57
THE FORWARD MODEL OF HOLOGRAPHY: PLANAR SCAN
s
inc inc,Rx ,Tx ( , , ;
sc
data, , ; )
( , , , 1, ,,; , )) (r x y Tz xV y zx dx dyS x dz Nzy z yξ ξ ξ ω
ε ξω′ ′ ′
′ ′′ ′ ′≈ = ′⋅∆∫∫∫ E E
• the S-parameter forward model
• during scan, the Tx & Rx antennas in each ξ-th experiment are fixed wrt each other:if rRx is known then rTx is known, e.g., Rx Tx( , , ) and ( , , )x y z x y z D≡ ≡ −r r
position of Tx/Rx antenna pair
x
y
D
z
co-pol
cross-pol
#1#2
OUTz z=