Download as pdf or txt
Download as pdf or txt
You are on page 1of 51

Ground Penetrating Radar (GPR)

Acquisition
Equipment
Antennas
tx rx

masonry bridge

Data

Ground Penetrating Radar (GPR)


Acquisition
Equipment
Antennas
tx rx

masonry bridge

Data

GPR videos
DATA ANALYSIS and INTERPRETATION

Diffractions
Antennas
Antennas Antennas Diffraction generated
txtx rxrx tx rx
by the pipe Distance

pipe
Time (two-way)
soil

ISMES test site


Raw data (RAMAC 250MHz)

Radar data processing


Filtering

Time calibration

Gain
Velocity analysis
Deconvolution
Topographic corrections
Trace alignment
Time to depth conversion
Background subtraction
NMO
for multi-fold data: Mute
Stack

Migration

Envelope extraction
BandPass Filter

x
Filtering
Data (t,x)
t

F.F.T.
x

Data (f,x)
f

x
filter (LP, HP, BP, Notch)
f

F.F.T. -1 x

t
Data (t,x)
Radar data processing
Filtering

Time calibration

Gain
Velocity analysis
Deconvolution
Topographic corrections
Trace alignment
Time to depth conversion
Background subtraction
NMO
for multi-fold data: Mute
Stack

Migration

Envelope extraction

Time calibration
Antennas
tx rx
Timing

tx
rx
trigger
soil pipe time

D
Time calibration
 Time calibration moves the time
scale so that zero time occurs
when the tx is triggered. Timing
 This time is obtained from the
data by picking event B and .......
tx
offset offset rx
t0  tB 
c tx rx trigger
time
with c = 30cm/ns

time zero
Time calibration

before calibration

time zero

after calibration
Radar data processing
Filtering

Time calibration

Gain
Velocity analysis
Deconvolution
Topographic corrections
Trace alignment
Time to depth conversion
Background subtraction
NMO
for multi-fold data: Mute
Stack

Migration

Envelope extraction

Gain
Aim: remove amplitude decay
Spherical divergence
Wavefront
Assumptions: of intensity E1
• homogeneous medium
• point source S
• no directivity r1
• no absorption Wavefront
r2
of intensity E2

ETOT   E  dS  4    r12  E1  4    r22  E2


S

2
E2  r1 
 
E1  r2 
Gain function
A2 r1 1 1
 A(r )  A(t )  G (t )  v  t
A1 r2 r v t
Divergence

G (t )  t

Time [ns]

Time [ns]

Gain is 20 dB/decade

Time [ns]

Divergence

Si(t)

kt

So(t)
Absorption

A(r )  A0  e  r α is the absorption coefficient

α is proportional to  and ’’


where  is conductivity and ’’ is the imaginary part of permittivity

α is higher at higher frequencies

Gain function: G (r )  e r G (t )  e vt

Absorption

G (t )  e k t

Time [ns]

Gain is given in
dB/ns or dB/m
(e.g., 0.05dB/ns).

Time [ns]

Time [ns]
Absorption

Si(t)

ekt

So(t)

Spherical Exponential Compensation (SEC)

Because of spherical divergence:


1
A(r ) 
r

Because of absorption:
A( r )  e   * r

SEC compensation:

G ( r )  r * e * r G (t )  v * t * e * v * t
Divergence + Absorption

Si(t)

ht ekt

So(t)

Gain

raw
data

after
SEC
Gain with AGC (Automatic) or PGC (Programmed)

Note. These methods apply a gain function computed on the data


regardless the phenomena producing the amplitude decay.

Both methods need a math operator to estimate the average amplitude of the
radar trace within a time window of proper length.

Most used operators:


N N
i i
1 1 2

2
mean 
( N  1)
 N
sj RMS 
N  1 j i  N
s 2j
j i 
2 2

where sj is the amplitude of the jth sample within a


time window of N+1 samples.

The time window must be properly long.


At least as long as the duration of the wavelet generated by the TX antenna.

AGC (Automatic Gain Control) – with non overlapping windows

Radar trace

Average amplitude

Gain function

Radar trace after AGC


AGC (Automatic Gain Control) – with overlapping windows

N
n
1 2

or
An   sk
N  1 k n N
2

N
n
1 2
An  
N  1 k n  N
sk2
2

sn s ~
sn
~
sn  n
An

Gain

raw
data

after
AGC
(10ns)
Gain

Too long
after
AGC
(30ns)

Too short
after
AGC
(3ns)

PGC (Programmed Gain Control)

A unique gain function is computed for the whole GPR profile.

N
n
M
1 2

or
An    si ,k
M ( N  1) i 1 k n  N
2

N
n
M
1 2
An   
M ( N  1) i 1 k n N
si2,k
2

sn s ~
sn
~
sn  n
An
Gain

raw
data

after
PGC
(7ns)

Radar data processing


Filtering

Time calibration

Gain
Velocity analysis
Deconvolution
Topographic corrections
Trace alignment
Time to depth conversion
Background subtraction
NMO
for multi-fold data: Mute
Stack

Migration

Envelope extraction
Raw data (2-6GHz)
sand box Average trace
0
surface

5
target
(metal cup
filled with
plastic)
10

15

Before deconvolution After deconvolution


1GHz data on a mine test site (JRC-Ispra)
Before deconvolution After deconvolution

For details on deconvolution, see seismic data processing

Radar data processing


Filtering

Time calibration

Gain
Velocity analysis
Deconvolution
Topographic corrections
Trace alignment
Time to depth conversion
Background subtraction
NMO
for multi-fold data: Mute
Stack

Migration

Envelope extraction
Trace Alignment
Before After

Algorithm based on crosscorrelation between neighbouring traces

Trace Alignment
Before After

1GHz profile on a stone wall


Shift calculation for trace alignment
Trace k Trace k+1
t
xcorr =

Radar data processing


Filtering

Time calibration

Gain
Velocity analysis
Deconvolution
Topographic corrections
Trace alignment
Time to depth conversion
Background subtraction
NMO
for multi-fold data: Mute
Stack

Migration

Envelope extraction
Background Subtraction
SUBTRACTION
TRACE ALIGNMENT AVERAGE TRACE
OF AVERAGE TRACE
0

10

20
AVERAGE TRACE
time [ns]

30

40
Raw data after trace alignment
50

60
100 200 300 400 500

0 -
trace number

10 10

20 20
time [ns]
time [ns]

30

40
= 30

40
Average trace section Subtraction
50 50

60 60
100 200 300 400 500 100 200 300 400 500
trace number trace number

Background Subtraction

SUBTRACTION
TRACE ALIGNMENT AVERAGE TRACE
OF AVERAGE TRACE

for more effective subtraction

AMPLITUDE EQUALIZATION JITTER COMPENSATION


Background Subtraction
Raw data
0

10

20
time [ns]

30

40
Subtraction
50 0

60 10
100 200 300 400 500
20

time [ns]
trace number

30
More effective subtraction
40
0
50
10
60
20 100 200 300 400 500
time [ns]

trace number
30

40

50

60
100 200 300 400 500
trace number

Background Subtraction x
by FK filter Data (t,x)

F.F.T 2D

f
Data (f,k)
k

Dip filter (fan filter, velocity filter) f

k
F.F.T -1 2D x

t
Data (t,x)
ns/trace
s s

12-MHz Sinusoids
MHz MHz

  Wavenumber (cycles/m)
(adapted from Yilmaz, 2001)

ns/trace
s s

48-MHz Sinusoids
MHz MHz

  Wavenumber (cycles/m)
Aliasing
(adapted from Yilmaz, 2001)
ns/trace
s s

MHz MHz

  Wavenumber (cycles/m)
(adapted from Yilmaz, 2001)

s s

MHz MHz

x=1m -0.5 0 0.5 x=1m -0.5 0 0.5


Wavenumber Wavenumber
(cycles/m) (cycles/m)
(adapted from Yilmaz, 2001)
s s

MHz MHz

x=1m -0.5 0 0.5 128m -1 0 1


Wavenumber Wavenumber
(cycles/m) (cycles/m)
(adapted from Yilmaz, 2001)

Background Subtraction by FK filter

2D
FFT

=
2D
IFFT
Examples of FK filtering
2GHz antenna (IDS)

Acquisitions on a stone block to detect possible defects


(fractures, cavities, ...)

profile trajectory
187cm

Examples of FK filtering
Raw data

2D
FFT

After background subtraction

=
2D
IFFT
Examples of FK filter
After PGC

2D
FFT

After subtraction of side scattering

=
2D
IFFT

Radar data processing


Filtering

Time calibration

Gain
Velocity analysis
Deconvolution
Topographic corrections
Trace alignment
Time to depth conversion
Background subtraction
NMO
for multi-fold data: Mute
Stack

Migration

Envelope extraction
Velocity analysis on scattered signals
(monostatic system)
x
0 x
z
z2  x2 z
V

Scatterer in [0,z]

0 x

t0
t t

Diffraction hyperbola

2 z2  x2 4z2 4 x2 2 4x2 2z
t t2   2 t 2  t0  where t0 
v v2 v v2 v

Velocity Analysis

change to

change velocity
Velocity Analysis

too low too high

correct correct

when velocity varies with depth

z1 v1
z2 v2

zn vn

z
diffractions are nearly hyperbolic

4x2 2 zi
t0  i i  i
2
t 2  t0  2 where
vrms vi

vrms 
v  i i
2
i


and

i i
Velocity Analysis

11.2 8.3

7.3

Radar data processing


Filtering

Time calibration

Gain
Velocity analysis
Deconvolution
Topographic corrections
Trace alignment
Time to depth conversion
Background subtraction
NMO
for multi-fold data: Mute
Stack

Migration

Envelope extraction
Topographic corrections
Example
(200MHz data)

topography

before
topographic
corrections

after
topographic
corrections

Topographic corrections
highest
EM
E elevation
[m]
E(x) v average velocity
profile
elevation

x [m]

Topographic corrections are static corrections calculated as:

2[ E M  E ( x )]
t c ( x )  t ( x )  t ( x ) with t ( x ) 
v
data after topographic
corrections

Radar data processing


Filtering

Time calibration

Gain
Velocity analysis
Deconvolution
Topographic corrections
Trace alignment
Time to depth conversion
Background subtraction
NMO
for multi-fold data: Mute
Stack

Migration

Envelope extraction
Monostatic system

v = const when velocity is constant

t
z (t )  v
2

no trace distortion
t z

v = v(t) when velocity varies with depth

1 t
2 0
z (t )  v ( )d

trace distortion
t z

Bistatic system
when velocity is constant

tx rx
h
z t t2
v = const
v z 2 (t )  v 2  h2
2 4

t2
z (t )  v 2  h2
4
t stretching

when t is large (i.e., z >>h)


t
z (t )  v
2
stretching is negligible
Example (250MHz bistatic antenna with h=60cm)

Data in x,t coordinates

After time to depth conversion stretching

Radar data processing


Filtering

Time calibration

Gain
Velocity analysis
Deconvolution
Topographic corrections
Trace alignment
Time to depth conversion
Background subtraction
NMO
for multi-fold data: Mute
Stack

Migration

Envelope extraction
Migration

Problem energy reflected by a curved or dip surface


generates a distorted image of the reflector

energy scattered by a small anomaly generates


a distorted image of the scattering target

Analogy a non flat mirror generates distorted images


of the reflected objects

Aim to correct the image distortion of the reflected


or diffracted targets

Why migration ?
model space data space

A B x
A B x

z tB t
tA
v

C
tA = 2AC/v
D
tB = 2BC/v
Why migration ?
model space data space

A B x
A B x

z tB t
tA
v

after migration

C
tA = 2AC/v
tB = 2BC/v

Migration by semicircle superposition


(monostatic system)

data space after time-to-depth conversion

v 2t 2 2 2
x z
Z 4
Migration by semicircle superposition
(monostatic system)

The method of semicircle superposition consists in


distributing any signal along the locus of points where it
might come from as a result of a normal incidence reflection

Method distributing any sample along its semicircle


and summing the overlapping contributions

Note the method requires the knowledge of the


velocity field

Migration

before migration

after migration
Migration

before migration

after migration

Migration

before migration

after migration
Diffraction summation (or Kirchhoff method)
(monostatic system)
x
0 x
z
z
V

Scatterer in [0,z]

0 x

t0
t t

Diffraction hyperbola

2 4x2
t  t0  2
2

Diffraction summation (or Kirchhoff method)


(monostatic system)

Kirchhoff method consists in collapsing each diffraction


hyperbola to its origin

Method summing in the diffraction origin all the


contributions distributed along the
hyperbolic equation

Note the method requires the knowledge of the


velocity field

after migration
Migration (Focusing)

x=0

Migration

Antenna position x

What about a reflecting interface ?

A reflecting interface is the superposition of an infinite number of scattering points

Few scattering points

Many scattering points


Dip reflector

Dip reflector

slope before migration

slope after migration


Example (100MHz data from a granite quarry)
Before migration

After migration

(Seol et al., 2004, JEEG)

Example

Before migration

After migration (v = 14cm/ns)


What happens with a wrong velocity ?

10 cm/ns (undermigration)

20 cm/ns (overmigration)

Velocity Analysis through focusing analysis


Migration by ellipse superposition
(bistatic system)

2h
Tx Rx
0

z
x,z

Migration ellipse
1 1
t ( x  h) 2  z 2  ( x  h) 2  z 2
v v

Migration by diffraction summation


(bistatic system)
2h
Tx Rx Tx Rx Tx Rx
x
0 x
z
V
(0,z)
Scatterer
0 x

Diffraction curve

1 1
t ( x  h) 2  z 2  ( x  h) 2  z 2
v v
Migrated results are often presented by extracting the envelope

Trace amplitude

Complex trace amplitude (i.e., envelope)

2D migration in non homogeneous media


Example 2 4x2
t 2  t0  2
vrms
3D Migration (monostatic system)
by diffraction summation

Migration

2 4x2 4 y2
t 2  t0  
v2 v2

or by sphere superposition

3D Migration (bistatic system)


by diffraction summation
2h
0 x
y Tx Rx
y
x z

V (x,y,0)
(0,0,z)
Scatterer

1 1
t ( x  h)2  y 2  z 2  ( x  h)2  y 2  z 2
v v

or by ellipsoid superposition
Example
BAR DETECTION IN A CONCRETE FLOOR
Raw data 2D processed data

3D processed data

Example
SUBSURFACE UTILITY DETECTION:
REAL CASE
3D migration in non homogeneous media
Example
4x2 4 y2
2
GPR for humanitarian demining
t  t0  2  2
2

vrms vrms

30 [cm/ns]
Vrms
air
surface

t0 soil
vrms (t0 ) 
0
v 2 ( )d Vrms vs. depth
t0
0
d

2dz
where d 
v( z )

Diffraction aperture

2D experiments 3D experiments
tx rx

The aperture of the diffractions is influenced by the antenna


footprint (larger at higher depth) and by soil absorption
Diffraction aperture
Example

Aperture of migration operator

zero offset section

Reducing the aperture,


migration is less time
consuming and less noisy
but

diffraction summation
large aperture migration

semicircle superposition

small aperture migration


diffraction summation

semicircle superposition

(Yilmaz, 2001)
Spatial aliasing problems
surface

reflector

no aliasing

Spatial aliasing problems


surface

reflector

with aliasing
Spatial aliasing problems
DISTANCE [cm]

Input data 2D spectrum

DEPTH [cm] at v=8 [cm/ns]

FREQUENCY [MHz]
TIME [ns]

KX [1/cm]
DISTANCE [cm]

2D spectrum
DEPTH [cm] at v=8 [cm/ns]

FREQUENCY [MHz]
TIME [ns]

Migrated data
KX [1/cm]

Spatial aliasing problems


DISTANCE [cm]

Input data 2D spectrum


DEPTH [cm] at v=8 [cm/ns]

FREQUENCY [MHz]
TIME [ns]

KX [1/cm]
DISTANCE [cm]

2D spectrum
DEPTH [cm] at v=8 [cm/ns]

FREQUENCY [MHz]
TIME [ns]

Migrated data
KX [1/cm]
Spatial aliasing problems
DISTANCE [cm]

Input data 2D spectrum

DEPTH [cm] at v=8 [cm/ns]

FREQUENCY [MHz]
TIME [ns]

KX [1/cm]
DISTANCE [cm]

2D spectrum
DEPTH [cm] at v=8 [cm/ns]

FREQUENCY [MHz]
TIME [ns]

Migrated data
KX [1/cm]

You might also like