i

GPR FOR FAST PAVEMENT ASSESSMENT

May 2007

JHR 07-310 Project 00-2

Lanbo Liu

This research was sponsored by the Joint Highway Research Advisory Council (JHRAC) of the University of Connecticut and the Connecticut Department of Transportation and was carried out through the Connecticut Transportation Institute of the University of Connecticut. The contents of this report reflect the views of the authors who are responsible for the facts and accuracy of the data presented herein. The contents do not necessarily reflect the official views or policies of the University of Connecticut or the Connecticut Department of Transportation. This report does not constitute a standard, specification, or regulation.

ii

1. Report No. 2. Government Accession No. 3. Recipient’s Catalog No.

JHR 07-310 N/A 4. Title and Subtitle 5. Report Date

May 2007 6. Performing Organization Code

GPR for Fast Pavement Assessment

N/A 7. Author(s) 8. Performing Organization Report No.

Lanbo Liu JHR 07-310 9. Performing Organization Name and Address 10. Work Unit No. (TRAIS)

N/A 11. Contract or Grant No.

University of Connecticut Connecticut Transportation Institute Storrs, CT 06269-5202 N/A 12. Sponsoring Agency Name and Address 13. Type of Report and Period Covered

FINAL 14. Sponsoring Agency Code

Connecticut Department of Transportation 280 West Street Rocky Hill, CT 06067-0207

N/A 15. Supplementary Notes

This study was conducted under the Connecticut Cooperative Highway Research Program (http://www.cti.uconn.edu/crp_home.html). 16. Abstract

This report summarizes the findings of Project JHRAC 00-2 for studying the fast assessment of road pavement with the ground penetrating radar (GPR). The report contains four parts. The first two on actual GPR data acquisition and analysis: one on laboratory measurements and one on field surveys. The last two parts examine the GPR signal forward modeling and signal data processing. We conducted laboratory tests of engineered materials using the RAMAC radar system with the 1-GHz antenna. The experiments used geotechnical fundamental measurements as the known parameters and correlated with the electromagnetic (EM) parameters obtained by GPR measurements. The fundamental parameters include thickness, density, aggregate material ratio, and air void ratio, etc. The GPR experiments test the materials to get the dielectric constant that directly determines EM wave velocity. Asphalt tests have been conducted on 30 asphalt specimens. The asphalt specimens produced at the Connecticut Advanced Pavement Laboratory have a dimension of 15 cm in diameter with 11.5 cm in height. Serving as the strong reflector to mark the travel time, an aluminum plate is placed underneath the specimen, to allow an easier identification of the time travel to and reflected back from the plate. The calculated EM wave velocity is about 131 m/microsecond for most dry asphalt specimens. The corresponding dielectric constant is ranging from 4 to 5. We have also tested the response of the asphalt specimens in different ambient states (dry, saturated with water, and frozen). GPR data using the 1-GHz antenna were collected several times on a then-newly paved road segment on the Depot Campus of the University of Connecticut (UConn) under different meteorological conditions. The GPR data clearly showed signature changes of the asphalt response between dry and water-saturated conditions. The radar wave velocity reduced about 5% when the asphalt was water-saturated (after heavy rain, collected in the morning of September 20, 2000). This change makes a relative clearer identification. Meanwhile, the reflectivity increased 60%, making the reflection from the bottom of the base of the asphalt layer much more visible. To theoretically understand the effect of a thin dielectric layer (such as the hot mixed asphalt pavement) on radar signal propagation, a finite difference time domain (FDTD) forward modeling algorithm was developed. The simulation clearly demonstrated the wave guide effect of this surface thin layer. This simulation can be used as guidance for developing multi-channel GPR systems. To improve the resolution and preserve the penetration, if high-and low-frequency surveys are conducted simultaneously over the same road segment, it is possible to extrapolate the high frequency signal to a deeper depth numerically through a digital signal processing algorithm known as extrapolation with deterministic deconvolution (EDD) and consequently gain a higher resolution to a greater depth.

17. Key Words 18. Distribution Statement

ground penetrating radar (GPR), non-destructive testing, pavement assessment, GPR antenna, surface coupling antenna, horn antenna, dielectric permittivity, electromagnetic wave velocity, finite difference time domain method, wave guide

No restrictions. This document is available to the public through the National Technical Information Service Springfield, Virginia 22161

19. Security Classif. (of this report) 20. Security Classif. (of this page) 21. No. of Pages 22. Price

Unclassified Unclassified 76 N/A

Technical Report Documentation Page

Form DOT F 1700.7 (8-72) Reproduction of completed page authorized

iii

Acknowledgments

This work is sponsored by Connecticut Department of Transportation under Project JHRAC

00-2, “GPR for Fast Pavement Assessment: Experimental Tests.” The GPR equipment used in

acquiring the laboratory and field data was partly funded by the National Science Foundation

(NSF). The PI thanks Prof. Jack Stephens and Mr. James Mahoney for their assistance in

conducting the laboratory experiments, and Prof. Gregory Frantz for helping with sample

preparation.

iv

v

Table of Contents

Title Page i

Technical Report Documentation Page ii

Acknowledgments iii

Modern Metric Conversion Factors iv

Table of Contents v

1. Introduction 1

2. GPR laboratory tests of asphalt specimens in three states 2

2.1 Introduction 3

2.2 Laboratory measurements 4

2.3 Calculations of EM wave velocity and dielectric constant 7

2.4 Results discussion 10

2.5 Conclusions 15

3. GPR surveys on paved road for moisture content assessment 17

3.1 Introduction 17

3.2 Practical limitations on radar wave velocity measurements 18

3.3 Field measurements 19

3.4 Results discussion 21

3.5 Conclusions 22

4. Numerical modeling of GPR surveys over dielectric thin layers 23

4.1 Introduction 23

4.2 Summary of major features of guided wave propagation 25

4.3 GPR wide-offset field observations 29

4.4. Numerical simulations of TE and TM modes for field GPR data 30

vi

4.5 Results discussion 31

4.6 Conclusions 36

5. Mathematical treatment to improve the resolution and penetration 37

5.1 Introduction 37

5.2 Extrapolation by deterministic deconvolution 39

5.3 Applications to synthetic data 42

5.4 Applications to field GPR data 49

5.5 Conclusions 53

6. Conclusions and future work 54

6.1 A general summary 54

6.2 Future work 54

Disclaimer 55

References 56

Appendix: A list of publications on GPR authored by the PI and his research group 68

1

1. Introduction

A fast, economical, and accurate means of assessment of road pavement conditions, at both

the network and project levels, is of great significance to highway management and

maintenance. The emerging ground penetrating radar (GPR) technology has a natural application

as a supplementary device for other road survey techniques such as the road video and infrared

surveys. GPR can be used to identify valuable parameters that cannot be obtained from a visual

evaluation of the road surface. These parameters include pavement-layer thickness and,

potentially, damaged, or deteriorated subsurface areas. Moreover, it would be possible to obtain

an more accurate characterization of the pavement structure by combining the data provided by

such a non-destructive testing (NDT) system with that supplied by destructive testing (coring

and sampling). Over the last decade, these potential benefits have prompted transportation

agencies and private industry in North America, Europe, and some developing countries, to start

incorporating GPR technology into pavement-assessment engineering practice. However, the

most challenging aspects for using GPR are, inevitably, the precise identification of subsurface

targets and the fast reduction of a huge amount of data into a form that is readily interpretable to

highway engineers.

To solve these problems, automatic acquisition and interpretation of GPR data for pavement

assessment has recently attracted tremendous attention and research efforts. As a preparation

phase for future research on the application of GPR techniques for pavement and bridge

assessment, the Principal Investigator (PI) of this project conducted a feasibility investigation on

state-of-the-art use of ground penetrating radar to rapid pavement assessment (Liu, 1998).

This report summarizes the major research results conducted by Project JHRAC 00-2 in 6

chapters. As an overall introduction, this chapter lays out the structure of the report. Chapter 2

reports the results on GPR laboratory testing results conducted in the Connecticut Advanced

Pavement Laboratory (CAP Lab). Chapter 3 summarizes the major findings of GPR field tests

over a low traffic volume road over a long period of time under different meteorological

conditions. Chapter 4 reports a piece of theoretical work on numerical simulation of the GPR

response to a surface thin layer of dielectric materials. Chapter 5 presents a digital processing

algorithm to improve the resolution of GPR signals and preserve the penetration depth. Finally,

2

Chapter 6 provides an overall summarization and makes recommendations for using GPR as an

engineering tool for pavement assessment in the future.

2. GPR laboratory tests of asphalt specimens in three states

There are two major purposes for using ground penetrating radar (GPR) field survey in

pavement assessment. The first is for determining the thickness of the asphalt pavement; and the

second is for detecting subsurface deterioration. Determination of thickness needs to know the

velocity of radar waves, and deterioration between layers involves reflectivity of radar wave at the

interfaces. Both velocity and reflectivity are determined by the dielectric constant of the asphalt

pavement. This section of the report discusses the dielectric constant determination of asphalt

specimens in laboratory-controlled conditions. The experiment results may provide certain degree

of constraints for interpreting GPR field data for seeking thickness variation and subsurface

deterioration.

A series of laboratory experiments was conducted to determine the dielectric constant of hot

mix asphalt (HMA) pavement specimens in dry, water-saturated, and frozen states. We used the

RAMAC GPR system with the 1-GHz ground coupled antenna to measure the travel times for the

direct and reflected waves to calculate the radar wave velocity and dielectric constants of 30

specimens. Our major conclusions are: (1) In general, radar wave travels fastest in the asphalt

specimens under dry condition and slowest in water-saturated condition. When the water-saturated

specimens are frozen, radar wave velocity increases again to some degree, but slower than that in

dry condition; (2) Radar wave velocity increases slightly with the increase of void ratio for dry

samples. In contrast, it decreases significantly through water-saturated samples as their void content

increased. Correspondingly, the dielectric constant decreases noticeably with an increasing void

ratio in dry conditions, and increases appreciably in saturated conditions, in the range of void ratio

from 0.57% to 6%; (3) The changes of EM velocity and resulting dielectric constant for dry,

saturated, and frozen conditions can be reasonably predicted by the complex refractive index model

(CRIM) for porous media; (4) Variations in EM velocity and resulting dielectric constant with

respect to asphalt binder content implies that a lower value in asphalt binder content corresponds

with a higher void ratio; and (5) For a particular HMA specimen with a given void ratio, the

observed variations in EM velocity and resulting dielectric constant can be attributed to the changes

3

in moisture content under different physical states.

2.1. Introduction

One of the most important factors affecting the life span and deformation of the HMA

pavements is the void content expressed by the void ratio (Saarenketo, 1997). Void ratio denotes

the volumetric fraction of void (or pore) space relative to aggregate and asphalt binder of a

pavement. The void ratio is important because void space could be filled with air, water, ice, or any

fractional combination of these three depending on the temperature environment. Each of these

three void materials has different mechanical properties which may influence the HMA’s overall

strength, durability, and stability. For example, the presence of water in void space can significantly

change the shear modulus of an HMA pavement. In contrast, the presence of ice in the voids will

effectively change an HMA’s bulk modulus. Moreover, freeze-thaw processes lower the tensile

strength of an HMA pavement and facilitate tensile failure (cracks). Clearly, the amount of

moisture penetrating into an HMA pavement, whether from above or below, depends on the

HMA’s void ratio and connectivity, and can have detrimental effects on the durability of that

pavement. In the last decade, ground-penetrating radar (GPR) has emerged as one of the most

widely used non-destructive testing tools in assessing HMA pavements (Saarenketo, 1992; al-Qadi,

1992; Davis et al., 1994; Maser, 1996; Liu, 1998; Saarenketo and Scullion, 2000).

A rapid and reliable estimation of the void ratio of an HMA pavement is necessary to

understand the role of moisture content in HMA durability and stability. Conventionally, HMA

void ratio and moisture content have been assessed by direct excavation or drill sampling, or by

radioactive methods (Saarenketo, 1997), which are costly and invasive. In this context, if a robust

relationship between radar wave velocity, void ratio and moisture content can be established; non-

destructive testing using GPR can be an efficient complementary tool for HMA pavement

assessment.

GPR has been used in the field to determine subsurface soil moisture content below highway

pavement (Topp et al., 1980; Greaves et al., 1996; Berktold et al., 1998; al Hagrey and Muller,

2000; Charlton, 2000; Redman et al., 2002). In laboratory experiments, Saarenketo (1998) reported

the effects of water in determining the dielectric permittivity of clay and silty soils. Saarenketo and

Scullion (2000) reported the relationship between void content and dielectric constant for dry HMA

4

samples. Their studies resulted in the development of vehicles that can be used to assess pavement

properties of a newly placed mat by combining some GPR with infrared sensing technology with

the calibration from coring. Shang and Umana (1999) studied the dielectric constant and relaxation

time of asphalt pavement materials. In this paper, we are reporting the laboratory results of GPR on

prepared HMA specimens in three distinctive states (dry, water-saturated and frozen).

Using a radar system with 1-GHz antenna we tested HMA pavement materials in the CAP Lab

at the University of Connecticut. The experiments correlated the dielectric constant obtained by

GPR measurement with aggregate material composition, and void ratio of HMA specimens under

dry, water-saturated, and frozen states. Our objectives were to (1) determine the dielectric

properties of the asphalt specimens; (2) correlate the dielectric constants with void ratio and asphalt

binder content ratio; and (3) elucidate the evolution of the dielectric constant with respect to

simulated environmental states (dry, saturated and frozen).. These results can be used as a baseline

for GPR field surveys over road pavements in real world conditions, as well as for quality control

of pavement compaction assessment using radioactive methods.

2.2. Laboratory measurements

We tested 30 HMA specimens in dry, water-saturated, and frozen conditions. These cylindrical

asphalt specimens were produced at the CAP Lab using the applicable AASHTO standard at the

time TP4 for the fabrication of Superpave gyratory specimens. They are 15 cm in diameter, about

11.3 cm in height, and are composed of the same aggregate source materials. The GPR antenna was

set on top of a specimen, which, in turn, was set on a metal (aluminum) plate to gain total reflection

(Figure 2.1). The use of a metal plate facilitates an easy identification of the reflected waves by

providing perfect reflectivity.

5

HMA

air waveT R

H

air

metal plate

ground wave

reflectedwave

D

Figure 2.1. This figure depicts the EM wave's general paths as it travels through the specimen. Idealized near-surface, near-field propagation paths along the interface of the air and an HMA pavement layer (or HMA specimen) with a thickness of H. T: the location of the transmitter; R: the location of the receiver. The separation of T and R is D. The air wave travels from T to R above the surface at the fastest velocity c0 (speed of light). The ground wave travels from T to R below the surface at a velocity of c0/√εr. where εr is the dielectric constant of the HMA. The reflected wave has a longer ray path: it hits the HMA-base (here a metal-plate) interface, and then is reflected to the receiver.

The experimental setup for testing the specimens in saturated condition is shown in Figure 2.2.

Figure 2.2. GPR measurement setup for a test on the water-saturated HMA specimen using the 1-GHz antenna in the CAP Lab, University of Connecticut.

6

The asphalt binder grade (nomenclature for the binder grade is PG 64-28) used for producing

these specimens is identical for all samples. The asphalt binder content (AC %) in any one

specimen is given in terms of weight percentage. The samples consist of a range of proportions of

aggregates to form a void ratio that varies from 0.57% to 6.0%. The parameters of these 30

specimens are listed in Table 2.1.

The EM wave velocity of these newly produced HMA specimens were first tested in dry state.

Each specimen was then submerged into tap water in a sealed container and vacuum pumped, such

as is done for the Gmm (the max theoretical density of HMA test) to extract the air for 10 minutes to

reach water saturation. The samples were then submerged in a water tank to hold water-saturated

status until testing. These saturated specimens were measured with the 1-GHz GPR again to get the

water-saturated velocity. After this, the specimens were immediately refrigerated with a constant

temperature of –18 ºC for 20 hours to reach a frozen state. (Specifications for classic pavement

freezing-thawing experiments request the specimen to be frozen no less than 18 hours in one

freezing-thawing cycle). Finally, the frozen specimen was tested by the same procedure as was

done in dry conditions to get the EM velocity in frozen states. All tests were conducted at room

temperature in a time period as short as possible to prevent pore water dripping or thawing. For one

specimen under a given state, 10 GPR measurements were taken. The average of the 10

measurements on the reflected wave travel time was used to calculate the EM wave velocity for

that state.

Table 2.1 Basic parameters of the HMA specimens used in GPR laboratory experiments

Specimen Label

Height (cm)

Volume (cm3)

Dry Weight

(g)

Surface wet Weight

(g)

Submerged Weight (g)

10-min. vacuum

Weight (g)

Gmm (g/cm3)

AC (%)

Void Ratio (%)

A 11.31 1998.647 5062.8 5064.8 3095.8 5070.7 2.589 6.4 0.72 B 11.27 1991.578 5062.2 5066.4 3098.2 5067.2 2.589 6.4 0.57 C 11.25 1988.044 5054.1 5055.5 3092.2 5059.6 2.589 6.4 0.59 D 11.31 1998.647 5066.9 5070.6 3102.3 5082.6 2.626 5.4 2.02 E 11.26 1989.811 5008.9 5015.4 3055.3 5039.8 2.609 5.9 2.05 F 11.33 2002.181 5045.1 5051.7 3086.5 5063.4 2.609 5.9 1.57 G 11.41 2016.318 5053.8 5058.2 3093.1 5065.3 2.609 5.9 1.43 H 11.32 2000.414 5023.7 5029.9 3066.1 5063.3 2.626 5.4 2.63 I 11.36 2007.482 5062.5 5068.2 3086.6 5104.4 2.626 5.4 2.63

7

J 11.43 2019.852 5046.0 5054.5 3059.0 5116.0 2.626 5.4 3.79 K 11.51 2033.99 5036.8 5046.5 3079.7 5112.8 2.626 5.4 4.06 L 11.38 2011.017 5050.1 5062.7 3079.3 5105.7 2.626 5.4 2.92 M 11.35 2005.715 5058.7 5062.0 3093.6 5077.1 2.609 5.9 1.56 N 11.31 1998.647 5046.7 5050.8 3078.5 5077.1 2.609 5.9 1.85 O 11.36 2007.482 5073.7 5075.3 3097.8 5083.3 2.589 6.4 0.96 P 11.33 2002.181 5082.7 5084.4 3105.2 5089.6 2.589 6.4 0.85 Q 11.35 2005.715 5085.6 5087.0 3104.6 5093.3 2.589 6.4 0.94 R 11.11 1963.304 5010.9 5015.9 3092.0 5035.3 2.665 4.8 2.30 S 11.18 1975.674 5066.6 5055.4 3134.1 5078.7 2.665 4.8 1.87 T 11.07 1956.235 4997.2 5001.9 3085.4 5008.2 2.638 5.3 1.16 U 11.30 1996.88 5023.6 5039.7 3088.0 5068.3 2.661 4.6 2.92 V 11.24 1986.277 5023.4 5039.7 3088.0 5068.3 2.661 4.6 2.35 X 11.48 2028.688 4987.2 5000.2 3060.1 5099.5 2.71 4.0 6.00 Y 11.32 2000.414 5000.2 5005.7 3064.1 5067.7 2.71 4.0 5.50 Z 11.52 2035.757 5019.2 5044.2 3072.2 5109.2 2.71 4.0 5.78

AA 11.30 1996.88 5041.2 5044.3 3052.6 5052.1 2.637 5.5 1.18 AB 11.42 2018.085 5129.8 5135.6 3160.7 5154.0 2.637 5.5 1.54 AC 11.26 1989.811 5029.9 5039.5 3072.8 5061.7 2.637 5.5 1.46 AD 11.20 1979.208 5000.3 5005.6 3079.1 5030.2 2.645 5.1 1.89 AE 11.19 1977.441 5056.7 5061.0 3048.2 5073.4 2.645 5.1 1.35

* mmG is the maximum theoretical specific gravity.

2.3. Calculations of EM wave velocity and dielectric constant

We make three justifiable assumptions about the electromagnetic parameters of an HMA to

simplify the EM velocity calculation. The first is that velocity does not depend on the magnetic

permeability because the asphalt binder and most crushed stones used for making pavement

aggregates are essentially non-magnetic. Second, most of the minerals composing the aggregates,

such as quartz, mica, and feldspar, are good insulators. The asphalt binder can also be regarded as

insulation material. Since the void ratio of an HMA is small, even when the voids are fully

saturated with fresh water, the conductivity is still very small. This fact warrants us the third

assumption, that the imaginary part of the dielectric permittivity can be considered negligible.

Therefore, EM wave velocity only depends on the real part of the dielectric permittivity.

The electromagnetic (EM) velocity v can be measured in a number of ways. The easiest way is

to measure the time difference between the direct air wave traveling in air and the ground wave

8

traveling in the HMA (Figure 2.1). Unfortunately, for a near-offset (the separation distance between

the transmitter and the receiver is very short) survey, the arrival time difference is very small and

hard to measure accurately. Another way is the reflection method, in which the travel time

difference for the direct and reflected waves is the fundamental information for computing EM

wave velocity. As shown in Figure 2.1, the thickness of an HMA is comparable to GPR transmitter-

receiver separation. Taking this into account v was calculated by:

20

22

)(4

ttHDv

Δ++

=, (2.1)

where D is the separation between the transmitter and receiver; H is the thickness of the specimen;

t0 is the direct air wave travel time to the receiver, and �t is the time difference between the reflected

wave from the underlying metal plate and the direct air wave. All parameters on the right-hand side

of Eq. (2.1) are measurable in terms of either geometry or time. For example, the transmitter-

receiver separation is 0.1 m for the 1-GHz ground-coupled antenna of RAMAC/GPR manufactured

by MALA Geoscience, Inc., which was used in this study. The reason for choosing the ground-

coupled antenna is due to its smaller foot-print than the air-coupled horn antenna that is more

suitable for measuring the asphalt specimens with a diameter of 15 cm. The air-coupled horn

antenna has the advantage of rapid data acquisition that we do not need in this project for

measuring asphalt specimens in lab, but its larger foot print is definitely a disadvantage for

measuring small-sized asphalt specimens.

As argued before, due to the justified simplification, the velocity depends only on the real

part of the relative dielectric permittivity (the dielectric constant) through

2

222

vcorcv o

rr

o == εε ,

(2.2)

where �r is the dielectric constant; and co is the electromagnetic wave velocity in vacuum.

9

Figure 2.3. An example of raw waveform of GPR surveys over a dry asphalt specimen which was placed on an aluminum plate. The arrow with letter D indicates the arrival of the direct wave, while the arrow with letter R indicates the arrival of the reflected wave from the aluminum plate. The wave trains in the later times are multiple reflections irrelevant to travel time estimation.

10

Figure 2.4. Variations of EM wave velocity (a) and the dielectric constant (b) as a function of the void ratio obtained from GPR measurements of 30 HMA specimens for dry (Vd and Ed, with blue symbols and lines), water-saturated (Vw and Ew, with magenta symbols and lines), and frozen (Vz and Ez, with yellow symbols and lines) conditions.

EM wave velocity and the dielectric constant were calculated by this approach for all of the 30

pavement specimens. A typical time domain radar record is shown in Figure 2.3. The results of EM

velocity and the inferred dielectric constant as functions of the void ratio are shown in Figure 2.4

and discussed in detail in the next section.

2.4. Results discussion

To the first order, the results (Figure 2.4) clearly show that the zero-void velocity is about

130 m/�s and the zero-void dielectric constant is about 5.3. It implies that the HMA specimens

are good dielectric materials and quite transparent in an electromagnetic sense. When the void

ratio becomes larger, the EM velocity spreads out among dry, water-saturated, and frozen

conditions up to a value of 13% of the reference velocity (130 m/�s) when the void ratio is 6%.

11

The corresponding changes in the dielectric constant reach as large as 25%.

The CRIM model

For analyzing the EM velocity in a porous medium like an HMA, a good approximation for a

hybrid material model is a solid skeleton containing void space, with the void filled with two

kinds of fluid. A dielectric model formulation commonly used is the complex refraction index

model (CRIM) (e.g., Robert, 1998; al Hagrey and Muller, 2000):

agwb nSnSnn )1()1( −+−+= φφφ , (2.3)

where n=√�r is the refractive index, and sub-indices b, w, g, and a represent bulk, water,

aggregate, and air, respectively. Also, ��is the porosity, and S is the water saturation. The

porosity ���is related to the void ratio e through

ee+

=1

φ . (2.4)

In the low porosity regime (less than 7%), the porosity and void ratio are not significantly

distinguished from each other. We applied the CRIM model to analyze the observed velocity and

dielectric constant by using the following parameters. Because the dielectric constant is about

5.5-6.5 for aggregates and 2.6-2.8 for asphalt binder (Saarenketo, 1997), and the asphalt content

(AC%) is only about 5% of the total weight of a specimen, the effective dielectric constant of 5.3

was used for the solid matrix (aggregates bonded by asphalt binder). The dielectric constant is 81

for fresh water, 3.17 for fresh water ice (Arcone, 1984), and 1 for air. Replacing na by ni, the

refractive index of ice, Eq. (2.3) was used again to compute the bulk dielectric constant for voids

filled with ice-water mix. Using these values, the modeling results for the case of air-water mix

filling the void space are shown in Figure 2.5. The modeling results for the case of ice-water

mix filling the voids are shown in Figure 2.6.

12

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.0

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

(a) (b)

Porosity (%)Porosity (%)

Velo

city

(m/μ

s)

Die

lect

ricC

onst

ant

Figure 2.5. Prediction of EM wave velocity (a) and dielectric constant (b) variations as a function of porosity for water saturation changes from 0 to 1 by the CRIM model. The value of 0.80 best fits the observed results (Figure 2.4); it implies that the water-saturation reached about 80% after 10-min air-stripping pumping.

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.0

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

(a) (b)Porosity (%)Porosity (%)

Velo

city

(m/μ

s)

Die

lect

ricC

onst

ant

Figure 2.6. The CRIM predicted EM wave velocity (a) and dielectric constant (b) as a function of porosity for the ice-water mix case. In this case, the void space is filled with only ice and water, no air. Water saturation changes from 0 to 1 correspond to 100% frozen to unfrozen. The curve of water saturation of 0.2 best fits the observed results in frozen conditions (Figure 2.4); it implies that 20% of the pore water is still unfrozen after 20-hours freezing in a refrigerator.

The CRIM model appears to sufficiently interpret the observed results for the dry and water-

saturated conditions. When the void ratio increases, the velocity increases slightly for dry

conditions and decreases significantly for water-saturated conditions. This is simply due to the

fact that a larger fraction of air filling the voids leads to higher velocity, and a larger fraction of

pore water leads to lower velocity. The curve of S=0.8 may best fit the observed results for

water-saturated conditions. This implies that the pores of the HMA specimens were 80%

saturated by water, and 20% filled with air (Figure 2.5). This indicates that the 10 minutes of air-

13

stripping pumping in the vacuum process we used to prepare saturated samples were not long

enough to reach ideal 100% saturation. In frozen conditions, the CRIM model also predicts that

velocity should increase with increase of the void ratio, if there is less than about 15% void

space filled with water (Figure 2.5a). Actually, the observed trend for the frozen condition

(Figure 2.4) slightly decreases with increase of the void ratio, and implies that there is about

20% of the pore water remaining unfrozen or being thawed when we took the GPR

measurements.

Void ratio-asphalt content correlation

The relationships between the observed EM velocity and resulting dielectric constant with

respect to the asphalt content were also studied (Figure 2.7). The variation trends are opposite to

variations of the velocity and dielectric constant with respect to the void ratio (Figure 2.4). The

EM velocity and resulting dielectric constant for the dry and frozen conditions do not change at

all with variation in asphalt content: the asphalt bitumen is only less than 6.5 % in an HMA

specimen, and so has little effect upon the bulk dielectric constant. An inverse correlation was

also found between the void ratio and the asphalt content (Figure 2.8). This observed inverse

correlation implies that lower asphalt content may lead to a larger void ratio. Therefore, the

amount of asphalt binder applied to an HMA plays a significant role in optimizing void ratio to

construct a durable and stable road surface.

Moisture content predicted by the Topp model

Based on our fitting for the saturated condition, the moisture content varies from zero to

5.6% when the void ratio changes from zero to 7%, with 80% saturation. The EM velocity drops

from 130 m/�s to 113 m/�s (Figure 2.4a). According to the model (Figure 2.9) given by Topp et

al (1980) the moisture content is related with the dielectric constant through

362422 103.4105.51092.2103.5 rrrv εεεθ −−−− ×+×−×+×−= . (2.5)

Using the velocity-dielectric constant relationship given by Eq. (2.2), for the same range in

velocity changes, the predicted moisture content increase by the Topp model is about 4.2%.

Apparently, the observed trend agrees with the Topp model reasonably well; the Topp model

slightly under-estimated the moisture changes.

14

Figure 2.7. EM wave velocity (a) and dielectric constant (b) variations for 30 HMA specimens as a function of asphalt binder content (AC%) in dry (Vd and Ed, with blue symbols and lines), water-saturated (Vw and Ew, with magenta symbols and lines), and frozen (Vz and Ez, with yellow symbols and lines) conditions.

15

AC = -0.3958VR + 6.3132R2 = 0.6503

3.0

3.5

4.0

4.5

5.0

5.5

6.0

6.5

7.0

0.00 2.00 4.00 6.00

Void Ratio (%)

Asp

halt

Con

tent

(%)

Figure 2.8. Correlation of void ratio and asphalt content extracted from 30 HMA specimens.

(a) (b)

Velocity (m/μs) Dielectric Constant

Moi

stur

eC

onte

nt(%

)

Moi

stur

eC

onte

nt(%

)

Figure 2.9. An empirical prediction (Topp et al., 1980) of the moisture content variations as functions of EM wave velocity (a), and the bulk dielectric constant (b).

2.5. Conclusions

The major conclusions from these experiments are (1) The EM wave velocity increases slightly

with the increase of void ratio for dry HMA specimens. In contrast, it decreases significantly by

increasing void ratio in water-saturated conditions. Correspondingly, with the increase of the void

ratio, the dielectric constant decreased slightly when the samples were dry, and increased

16

appreciably when the samples were water-saturated. (2) The changes of EM velocity and resulting

dielectric constant for dry, water-saturated, and frozen conditions can be sufficiently predicted by

the CRIM model for porous media. (3) A comparison of the observed and predicted EM velocity

and resulting dielectric constant variations in water-saturated conditions implies that some HMA

specimens did not reach full saturation when the GPR measurements were taken. (4) The observed

change of EM velocity and resulting dielectric constant in frozen conditions implies that water in

the HMA specimens were not completely frozen when the GPR measurements were taken. (5) The

variations in EM velocity and resulting dielectric constant with respect to asphalt binder content

ratio imply that low asphalt content corresponds with a high void ratio, so that in the lower end of

asphalt content EM velocity has the maximum fluctuation among different conditions. (6) The

observed variation in EM velocity and resulting dielectric constant was essentially affected by the

changes in moisture content, as predicted by the Topp model. With more GPR and other direct

measurements, GPR may prove to be an efficient way to non-destructively detect void ratio,

moisture content, and degree of pore water freezing for road pavements. This is of special interest

to engineers who are responsible for road inspection and maintenance in cold regions.

17

3. GPR surveys on paved road for moisture content assessment

A series measurement using the 1-GHz direct contact antenna was conducted on a segment

of low-volume traffic road on the Depot Campus of the University of Connecticut. The objective

of this measurement is to observe the difference of GPR signature on the meteorological effect

over different seasons, and to correlate the testing results obtained from the laboratory

experiment over the HMA specimens in three conditions described in the last chapter.

3.1. Introduction

As discussed in the introduction of the last chapter, sufficient durability and stability are the

two most wanted features for pavement using hot mix asphalt (HMA). The void ratio is one of

the key factors affecting the durability and stability of an HMA pavement. The reason of the

void ratio’s importance is obvious: void space could be filled with air, water, or ice (frozen

water), or any fractional combination of all of them in a low-temperature environment. This is of

particular importance in the New England area where freezing and thawing is the major player in

pavement degradation and deterioration. Each of these three constituents has different

mechanical properties. Thus, the presence of different amount of void constituents significantly

affects the overall physical behavior of an HMA pavement, in turn, influences the HMA’s

durability and stability. For example, the presence of water in void space, quantified by the

moisture content, can significantly change the shear modulus of an HMA pavement. In contrast,

the presence of frozen water (ice) in the voids will effectively change an HMA’s bulk modulus.

Moreover, the freezing-thawing process lowers the tensile strength of an HMA pavement and

facilitates tensile failure (cracks). Clearly, moisture penetrating into an HMA pavement, whether

from above or below, can have detrimental effects on the durability of that pavement. Reliable

estimation of the moisture content in an HMA pavement is the first step for thorough

understanding of the void-filling materials’ role in HMA’s durability and stability.

Conventionally, the information of an HMA’ void ratio and moisture content were assessed by

direct excavation method, like drilling samples, and radioactive methods in recent years

(Saarenketo, 1997). In this chapter a series of field experiments on correlating the moisture

content, and EM wave velocity for a given road segment whose void ratio is assumed to be

constant is discussed in details. The field data acquired over an entire annual cycle in spring,

summer, fall and winter are presented and compared to link the GPR velocity and the void-

18

filling materials.

3.2. Practical limitations on radar wave velocity measurements

For GPR measurements, the sampling rate is usually set to be about 10 times of the signal’s

central frequency. For example, the sampling frequency is 10.142 GHz for using the 1-GHz

antenna to collect data presented in this chapter. It is equivalent to a time domain sampling

interval of dt=0.0986 ns. For a trace or scan contains 480 samples, the total time window is

about 47.3 ns. For the 1-GHz monostatic antenna, the separation of the transmitter and the

receiver is fixed at 0.1 m. Thus, the difference in travel time of the air wave and the ground wave

is only

nscD

cD

cD

ttt rr

ag 434.0)13.5(3.01.0)1(

000

=−≅−=−=−=Δ εε

where tg and ta are the travel time of the ground wave, and the air wave, respectively. D is the

transmitter-receiver separation distance; co is the light speed in the air, and �r is the dielectric

constant of the HMA pavement. Given the typical nominal values for all the quantities involved,

the time difference value of 0.434 ns is about 4.4 sample intervals using the aforementioned

sampling rate. Moreover, the amplitude of the ground wave is √�r times large than the air wave

(Arcone, 1984). This is why the quantity of √�r is also known as the refraction index.

Meanwhile, the difference in travel times of the direct ground wave and the reflected wave (see

Figure 2.1) is

nsDHDcc

Dc

HDttt rrr

gr 9485.0)1.01.041.0(3.03.5)4(

)4( 222222

=−×+≅−+=−−+

=−=Δεεε

where tr is the travel time of the reflected wave from the pavement-base interface, and H is the

thickness of the pavement layer. This value of travel time difference is about 9.6 sample

intervals when using the same sampling rate mentioned above. The finite length of a source

19

impulse is another factor to mask accurate identification of wave phase arrivals. In the ideal

nominal case, the period of a GPR source wavelet with a central frequency of 1GHz is 1 ns. This

is equivalent to 10 samples in time domain. Antenna-ground coupling effect always shifts the

signal significantly to a lower frequency; the exact value depends on the electromagnetic

properties of the ground. This effect makes the period of the source wavelet even longer.

Clearly, identification of the arriving times of different wave types is the first step and

fundamental request to use GPR technique in pavement assessment correctly and fruitfully. It is

not a trivial task.

In the field data, the relative velocity changes associated with rainwater wetting in the HMA

can be up to 3-5%. This corresponds to 2-3 sample intervals’ change in GPR travel times. By

applying the Topp model (Topp, et al., 1980), this range in velocity variation suggests a moisture

change of 0.8-1.4%, and corresponds to the saturation change of 10-20% in an HMA with 7%

void ratio.

3.3. Field measurements

GPR surveys with the 1-GHz antenna were repeatedly carried out on an HMA paved, low

traffic volume road segment on the University of Connecticut’s Depot Campus on August 26,

2000, September 20, 2000, and May 21, 2002. The surveys were conducted along the edge of the

travel way, relatively close to the old highly distress pavement. The total length of this segment

under test is 23 meters. The spatial sampling interval of GPR scans (the distance between 2

adjacent time traces) is 0.05m; there are 450-460 scans in a GPR profile. Figure 3.1 shows the

typical survey setup. The field data clearly show the interface of the HMA pavement and the

base (Figure 3.2). The thickness of the HMA pavement layer varies from 0.05m in one end of the

survey profile, to 0.18 m in the other. The GPR profile shown as Figure 3.2a was acquired on

August 26, 2000, and the one shown as Figure 3.2b was acquired in the morning of September

20, 2000, after an overnight heavy rain. Though the surface of the asphalt layer might be

completely saturated with water, due to rapid run-off, the water table might not have significant

changes and was well beyond the detection depth of 1-GHz GPR, thus irrelevant to the GPR

surveys conducted at this site. Finally, Figure 3.2c shows the GPR profile surveyed on the same

segment on May 21, 2002. There was no rainfall directly before that date. Obviously, the

20

pavement-base interface was enhanced and became more continuous in the GPR profile when

the condition changes from dry to water-saturated. The reflectivity at this interface increased by

60%, as indicated by the amplitude changes in the single traces extracted at the horizontal

position x=18 m shown at the right side of the GPR profiles (indicated by the arrow points red

ellipse about 3 nanoseconds in travel time).

Figure 3.1. GPR field survey with the 1-GHz antenna on a newly HMA paved road segment on the University of Connecticut’s Depot Campus on August 26, 2000.

21

strong reflection

(c) May 21, 2002

(b) September 20, 2000 after rain

(a) August 26, 2000

Trav

elTi

me

(ns)

Trav

elTi

me

(ns)

Trav

elTi

me

(ns)

Figure 3.2. GPR profile acquired with the 1-GHz antenna on the HMA paved road segment shown in Figure 3.1. (a) data acquired on August 26, 2000; (b) data acquired on September 20, 2000 after overnight rain; (c) data acquired on May 21, 2002. The single traces for each survey date on the right are located at the distance of 18 meters in the profile. The reflectivity at the bottom of the pavement for September 20, 2000 was significantly increased (the red ellipse) implying strong presence of water in the base.

3.4. Results discussion

Careful examination of the reflector position of the pavement-base interface reveals a delay

of 2-3 sample intervals in the GPR profile in water-saturated conditions (Figure 3.2b), when

compared with GPR survey in dry conditions as shown in Figure 3.2a. This is equivalent to a

velocity change of 0.003-0.006 m/ns, about 2-5% relative change of a nominal dry velocity of

0.13 m/ns for the dry HMA pavement. By applying the Topp model, this variation range in

22

velocity implies a moisture content change of 0.8-1.4 %, and corresponds to a saturation change

of 10-20%, by using the CRIM model, in an HMA with 7% void ratio, a nominal value for HMA

pavement in natural conditions. Nevertheless, there is no direct measurement (e.g., coring) to

verify or validate these estimations. More coordinated work should be done in the future to make

this indirect method improve its applicability for road pavement assessment.

From the GPR profile we can see that it is still hard to accurately delineate the bottom of the

pavement layer. This is mainly due to that the period of the GPR signal generated by the 1-GHz

antenna is still too long. The wavelet of the direct wave masked the reflection from the pavement

bottom. In the future, an antenna with higher frequency (1.5 to 2 GHz) would be a better choice

for it will generate a shorter period wavelet and therefore with a higher resolution.

3.5. Conclusions

Field experiments using the 1-GHz GPR were conducted to study the effects of water content

on EM velocity and resulting dielectric constant under natural meteorological conditions. The

main conclusions are: (1) The changes of EM velocity and resulting dielectric constant for dry

and water-saturated conditions can be interpreted by the Topp model for accounting for the

water content variations. (2) The detected velocity change caused by rain water wetting results in

GPR signals velocity change of 0.003-0.006 m/ns, suggests the moisture content changes on the

order of 0.8-1.4% using the Topp model. Using GPR to determine subsurface moisture content

has been a sustainable interest in geology and geophysics (Berktold, et al., 1998; Charlton, 2000;

Greaves, et al., 1996; Haslund and Nost, 1998) at a larger scale than highway pavement. With

finer sampling intervals in time domain and an established baseline (obtained from direct depth

control, or repeated GPR surveys), GPR should also be capable to provide an efficient way to

detect moisture content variation, and pore water freezing for the purpose of HMA pavement

assessment.

23

4. Numerical modeling of GPR surveys over dielectric thin layers

The asphalt pavement is a typical dielectric thin layer for GPR surveys. Though monostatic

reflection mode survey is the most economical way to carry out GPR data acquisition, GPR

moveout survey, i.e., GPR data were acquired while the transmitter-receiver separation is

changing from short to wider distance, may reveal more information regarding subsurface

conditions. This chapter assesses GPR moveout survey performance in different near-surface

geological stratigraphy and different antenna orientations using 2-dimensional (2D) finite

difference time domain (FDTD) numerical simulations of field data. We first treat the simple

cases of radar pulses propagating along (a) the interface between two half-spaces (air/ice); and

(b) an ice thin layer wave-guide (air/ice/water) between two half-spaces. We then simulate four

more complex cases combining two radiation polarizations (TM and TE), and two geological

settings: a sandy/gravelly half-space overlain by a silty/clayey layer, and a silty/clayey half-

space overlain by a sandy/gravelly layer. Both cases are represented through different dielectric

constants. The results show that (1) more EM energy is radiated as an air wave for the TM mode,

and more EM energy will be sent into the ground when the TE mode is used, regardless of

stratigraphic sequence. (2) Where a gravelly sandy half-space overlain by a silty/clayey layer,

more EM energy will be trapped in the silty/clayey layer as the guided ground wave. (3) When

the TE mode is used there is much less air radiation for the case of silt overlying gravel than that

of silty/clayey half-space overlain by a gravel/sand layer. (4) For the stratigraphic sequence of a

sandy/gravelly half-space overlain by a silty/clayey layer, the TE mode fundamentally contains

only a ground wave, and the TM mode essentially contains only air wave energy. This implies

that for this case a far more complete separation of the air wave and the ground wave can be

reached. (5) Dispersion of phase and group velocities of the guided ground wave will be well

developed for the TE mode. These simulations imply that antenna polarization mode is an

important factor when using moveout surveys to study subsurface electromagnetic properties.

4.1. Introduction

The electromagnetic properties of the uppermost 5-10 meters of the solid earth play the most

important role when probing the earth with the electromagnetic devices like ground penetrating

radar (GPR). For many cases, the near-surface environment can be idealized as a dielectric thin

layer overlaying a half-space with different electromagnetic properties. The system of asphalt

24

pavement and the underlying sub-base makes a perfect example in the geotechnical world

(Saarenketo, 1997; Saarenketo and Scullion 2000). An ice sheet or soil layer forms other

examples (Arcone, 1984; Arcone et al., 2003). Though the thickness scale of these examples

appears to be quite different, for electromagnetic sounding purposes, they all can be viewed as a

wave-guide for electromagnetic probing at suitable, corresponding frequency ranges. When

performing varying-offset moveout GPR surveys of such strata, highly dispersive guided-wave

propagation can result, making analysis of the electromagnetic structure of the earth difficult.

Here we examine the nature of this propagation with theoretical simulations, which are powerful

and economical alternatives to field experiments, especially for environments in which real data

acquisition is prohibited or too expensive. Moreover, numerical simulation assists designing of

data acquisition geometry and procedures to make observations more effective and productive,

and allows many stratigraphic combinations to be tested.

Annan (1973) systematically summarized the mathematical development in electromagnetic

interferometry in a near-surface environment. He used complex integrals (e.g., Brekhovskikh,

1960) and normal mode analysis (e.g., Budden, 1961). Existing laboratory and field observations

(e.g., Rossiter et al., 1973; Annan et al., 1975; Arcone, 1984; Arcone et al., 1998) indicate that the

ground wave can be manifested by modal propagation over several tens of meters at lower

frequencies (100 MHz or less) when wide angle offset soundings are performed to measure ground

wave speeds. Nevertheless, small scale, natural near-surface layering can severely limit this path,

as we will see here.

This chapter investigates the wave-guide effects of a surface layer on electromagnetic wave

propagation using the finite difference time domain (FDTD) simulation technique. We first

present FDTD simulation results for half-space cases to demonstrate the physical fidelity of

FDTD numerical simulation. Then we treat the modal propagation case of a thin layer of ice

between two half-spaces (air, and water). Finally, the results of electromagnetic wave

propagation in TE and TM modes at pulse central frequencies of 400 and 65 MHz are discussed.

We simulated field data acquired from Fort Richardson, Alaska (AK) (a lower dielectric constant

sandy gravel half-space overlain by a higher dielectric constant sandy silt layer), and Hanover,

New Hampshire (NH) (a sandy silt half-space overlain by sandy gravel layer). The field data and

synthetic results were analyzed with time-frequency analysis with instantaneous parameters to

25

extract guided-wave group velocity dispersion characteristics. We conclude that the simulation

results reasonably resemble fundamental wave propagation features in the field data and are

helpful to interpret field observations.

4.2. Summary of major features of guided wave propagation

We first define TE and TM modes used in this chapter (Figure 4.1). The TE mode is the

‘transverse-electric mode with respect to the x-direction, which is the direction of the GPR profile,

and can be expressed by the 2-dimensional Maxwell’s equation with in-plane magnetic field

components Hx, and Hy and out-of-plane electric field component Ez such that

,

,

,

zszxyz

zy

zx

EJy

Hx

Ht

Ex

Et

HyE

tH

σε

μ

μ

−−∂∂

−∂∂

=∂∂

∂∂

=∂

∂∂∂

−=∂∂

(4.1)

where ���� and � are dielectric permittivity, magnetic permeability, and electrical conductivity,

respectively; and Jsz is an electrical current source function in the z-direction. This case is

equivalent to the TMz mode (the transverse-magnetic mode with respect to the z-direction) as

defined in another convention, for example, by Taflove and Hagness (2000). Similarly, the TM

mode is the ‘transverse magnetic mode with respect to the x-axis’ that has in-plane electric field

and transverse magnetic field expressed by the 2D Maxwell’s equation in TM mode such that

,

,

,

szyxz

yzy

xzx

Mx

Ey

Et

H

Ex

Ht

E

Ey

Ht

E

−∂∂

−∂∂

=∂∂

−∂∂

−=∂∂

−∂∂

=∂∂

μ

σε

σε

(4.2)

where Msz is a magnetic dipole source function in the z-direction. GPR data can be acquired in

either a TE or TM mode, as depicted in Figure 4.1. We adopted a right-hand coordinate system

with the x-direction as the strike of the GPR profile, the y-direction is vertical and downward, and

the z-direction is perpendicular to the plane coinciding with the GPR profile. By using 2D

26

modeling we assume there are no variation in both material properties and electromagnetic fields in

the z-direction. Horizontally polarized antennas with the parallel configuration facing each other

communicate TE waves (upper panel of Figure 4.1); whereas horizontally polarized antennas with a

serial configuration communicate TM waves (lower panel of Figure 4.1).

Figure 4.1. Sketch of TE and TM modes with corresponding antenna orientations. The positivex-direction is that of wave propagation from the transmitter to the receiver; along the GPRprofile. The y-axis is vertical downward; and the z-direction is perpendicular to the x-y plane, in accordance with the right-hand-rule.

As indicated by Annan (1973), the propagation of radar waves along the interface between two

dielectric half-spaces (e.g., air and the ground) can be idealized by four types of propagation

(Figure 4.2). Wave A is the fundamental spherical air wave propagates through the air at the free-

space velocity c; wave B is the counterpart of A in the ground known as the ground wave. If the

ground is homogeneous with refractive index, n (n=√�), then ground wave B propagates at the

velocity of c/n and is n times stronger than the air wave. These two waves are matched at the air-

ground interface by two other waves in order to maintain field continuity across the boundary. In

air, wave C is an evanescent wave (also known as the inhomogeneous wave) that decays

exponentially with height (see Figure 4.3) and propagates horizontally at the speed c/n to match the

ground wave B. In the ground, wave D, known as the head wave (also named as surface wave, or

27

lateral wave), propagates downward at the critical angle )/1(sin 1 nc−=θ with a planar phase front,

but with a horizontal velocity of c to match the air wave A. In the near field, waves B and D are

indistinguishable. The two waves are well separated with increasing propagation distance into the

far field.

Figure 4.2. Idealized near-surface, near field propagation paths along the interface of the freespace and a dielectric half-space (ground) for a TE mode antenna orientation ((a), modifiedfrom Annan, 1973). S: the location of the TE mode source; A: the wavefront of the air wave; B: the wavefront of the ground wave; C: representation of the inhomogeneous evanescent airwave matching the ground wave B; and D: the wavefront of the head wave (or the so-called lateral wave) in the ground matching the spherical air wave; θc: the critical angle. The snapshot at the time of 25 ns after the firing of the TE source from FDTD simulation (b)clearly shows the different paths illustrated in (a). The modeling was carried out at theinterface of the free space and a dielectric half-space with dielectric constant of 3.17 (for freshwater ice). The source is a Ricker wavelet with central frequency of 200 MHz. The fastdecay of the evanescent wave C with respect to height above the ground is illustrated in Fig.3. Note the reversal of phases between the air wave (leading half-cycle is blue) and the ground wave (leading half-cycle is red).

The situation can change radically when the ground is layered. Depending on the relative values

of refractive index n of each layer, guided modes can propagate in the layers and either vastly

extend or vastly limit the range of the ground wave. To illustrate this a thin layer of ice between the

28

Figure 4.3. (a) Time domain synthetic records of the Ez electric field along x-axis at the surface; (b) The amplitudes of the Ez electric field at different elevation at elapsed time of 25 ns. Theevanescent wave decays exponentially with increasing elevation above the surface, and is in phase with the ground wave below the surface; whereas the air wave and the ground wave haveopposite phases. The total length of the profile is 19.25 m (350 x 0.055m).

Figure 4.4. The FDTD time domain wide-angle offset simulation of the Ez electric field for the TE mode generated by a 200-MHz Ricker wavelet and received on the surface of a 40-cm thick ice layer of an air-ice-water system. Note the continual excitation of refracted air waves.

29

free half-space and the half-space of water was simulated with FDTD. Figure 4.4 presents the

FDTD synthetic wide-angle reflection and refraction (WARR) GPR profile of the Ez field for a

TE mode generated by a 200-MHz Ricker wavelet recorded along the surface of a 40-cm thick

ice layer of an air-ice-water system. The ice layer has a dielectric constant of 3.17 and the cold,

fresh water has a dielectric constant of 87. The profile shows shingling associated with wave

mode dispersion and simulates field data presented by Arcone (1984, Figure 10) for a lake ice

cover in mid-winter.

4.3. GPR wide-offset field observations

Wide offset GPR surveys were conducted at two sites with path lengths up to tens of meters,

and with a quasi-infinite medium to eliminate the reflections that occur from lateral boundaries

(e.g., walls) in scale models. The first site is on Fort Richardson, near Anchorage, Alaska, where a

surface layer of wet, sandy silt with an average thickness of about 34 cm has a slower speed than

the underlying drier strata of sands and gravel (Arcone et al., 2003). The second site is in Hanover,

New Hampshire, where a thin layer of drier, gravelly sand that varies from 16 to 20 cm in thickness

overlies sandy silt of glacial Lake Hitchcock. At each site we established a profile test line along

which we dug several holes to aid profile interpretation. At the Hanover site we used 400-MHz

antennas to acquire wide-angle, variable offset profiles (upper panel of Figure 4.5) and an 800-

MHz antenna set to record reflection profiles (antennas at constant separation, lower panel of

Figure 4.5). The wide-angle profiles allow us to look at the modes propagating in the layer; while

the reflection profile allows us to interpret stratigraphic structure of the ground. At Fort Richardson,

we acquired a 100-MHz wide-angle reflection and refraction (WARR) profile in TE mode (shown

later in Figure 4.8a). At each site our structural studies include documentation of layer depth,

general sediment type, water content, and mineralogy in order to determine what attenuation

mechanisms may be present. In addition, GPR reflection profiles compensated for the difficulty of

complete excavation in order to document near surface layering.

30

Figure 4.5. Varying offset, 400-MHz Wide-angle (top) and constant-offset, 800-MHz reflection (bottom) profiles recorded along a single test line in Hanover, NH. Left panel of the wide-angle offset profiles is acquired in TE mode and the right panel is in TM mode with transmitter placed at the 0-m distance of the reflection profile. Arrow indicates the reflection from the bottom of the wave-guide layer. Dielectric constant of lower layer was interpreted from hyperbolic diffraction, that of upper layer from direct observation of depth, and the time delays.

4.4. Numerical simulations of TE and TM modes for field GPR data

Our numerical simulation algorithm adapts the finite difference time domain (FDTD) method

(e.g., Taflove and Hagness, 2000), based on Yee’s original staggered grid (Yee, 1966), with a

perfectly matched layer as the absorption boundary condition to truncate outbound waves

(Berenger, 1994). Based upon the field observations the electromagnetic parameters for soil

materials and stratigraphic geometry at the two sites are listed in Table 4.1. Since the main concern

is the generation and propagation of the guided ground wave caused by the surface layer, sub-strata

were simplified as a uniform half-space with constant electromagnetic properties. The conductivity

values were measured directly at Fort Richardson (Arcone and Delaney, 2002) and are estimated at

Hanover. The quantity h in Table 4.1 is layer thickness.

31

Table 4.1. Electromagnetic material properties and stratigraphy at the two sites under study

Site Fort Richardson site, AK Hanover site, NH

Free space � =1.0, � =0.0 S/m � =1.0, � =0.0 S/m

Surface layer � = 22.7, � = 0.0005 S/m

h =0.34 m

� = 8.0, � = 0.0004 S/m

h =0.20 m

Sub-strata � = 10.6, ��= 0.0002 S/m �= 12.0, � = 0.002 S/m

For all materials the relative magnetic permeability is equal to unity. We used a central

frequency of 400 MHz for the source wavelet for both Hanover and Fort Richardson sites to

carry out FDTD simulations in TE and TM modes; and also a 65 MHz source wavelet

propagating in TE mode for Fort Richardson for comparison with existing WARR GPR data.

The 400-MHz, variable offset GPR simulated syntheses for the two sites at Hanover, NH and

Fort Richardson, AK are shown in Figure 4.6. The time step in the simulation is 0.0667 ns, a

value that sufficiently satisfies the Courant stability condition (Taflove and Hagness, 2000). The

total time window is 100 ns. Figs. 6a and 6c present the transverse horizontal electric component

Ez for the TE mode; while Figure 4.6b and 6d present the in-plane, horizontal electric field Ex for

the TM mode. With similar arrangement, the wave field snapshots for different modes and

different geological stratigraphies are shown as Figure 4.7. Each panel contains the snapshot at

13.33 ns elapsed time for the corresponding electric field and magnetic field.

4.5. Results discussion

When there is total reflection at the interfaces there is no loss of energy other than that due to

geometric spreading of the waves, and the propagation is thus modal. This occurs when the upper

and lower layers have higher wave speeds than the waveguide layer (e.g., Fort Richardson site).

The total reflection occurs at a critical angle �c = sin-1 (n1/nm), where m is either 0 (for air) or 2 (for

sub-strata). For the opposite case where n0 or n2 is greater than n1, transmission occurs through the

interface and energy is continually lost during propagation. For example, FDTD simulations

predict that a much stronger TE guided ground wave occurs at Fort Richardson site (Figure 4.6c)

32

than at the Hanover site (Figure 4.6a) where total reflection does not occur on the bottom interface

of the thin layer. For the TM mode, FDTD simulations also predict that a much stronger guided

wave travels at a group velocity between that of the air wave and the ground wave, with a relatively

low attenuation rate, in the low-velocity thin layer (the Fort Richardson case, Figure 4.6d) than in

the Hanover site wave guide (Figure 4.6b). It is clear from Figure 4.7 that the TE mode generates

more ground waves; whereas the TM model generates more air waves, with a stronger head wave

matched in the thin layer, regardless of near-surface stratigraphic sequence. For the TE mode, at the

same elapsed time, the case of sandy layer overlaying silty sub-strata (Hanover, NH) has a larger

propagating range, while the case of a silty layer overlaying gravelly sub-strata has more electric

energy being trapped in the surface thin layer.

(a) (b)

(c) (d)

Fig. 4.6. Simulated 400-MHz wide-angle GPR profiles of (a) the Ez component in the TE mode; (b) the Ex component in the TM mode, for the Hanover site; and (c) the Ez component in the TE mode and (d) the Ex component in the TM mode, for the case of Fort Richardson site.

33

Figure 4.7. The simulated 400-MHz GPR wave propagation snapshot in (a) TE mode and (b) TMmode at time step 200 (elapsed time of 13.33 ns) for the Hanover site, and in (c) TE mode and(d) TM mode for the Fort Richardson site.

Where the upper layer is electrically thin (comparable to an in-situ wavelength or shorter) and

34

has a faster speed than the lower sub-strata, such as a drier soil over a wetter soil (e.g., the case in

Hanover, NH), or an ice layer over water (Figure 4.4), or a frozen soil over thawed soil, wave

energy continually transmits into the lower layer at a rate that can only be predicted by full wave

modal theory.

Theory and observations (Annan et al., 1975; Hoeskstra and Delaney, 1974, Arcone and

Delaney, 2002) have proved that when the upper thin layer has a slower speed than the sub-strata,

modes will propagate at angles beyond the critical angle for both interfaces and the upper thin layer

becomes a refractive waveguide, just like an optical fiber. This occurs when a wetter soil overlies a

drier soil, or when an unfrozen soil sits on top of permafrost, or bedrock. This is clearly shown by

the field observations and FDTD numerical simulation as shown in Figure 4.8. The shingling of the

waveforms is clearly associated with dispersive propagation velocity. The numerical simulation

(Figure 4.8b) qualitatively resembles the observation shown in the field WARR profile. More

quantitative analysis to exploit the data for more propagation characteristics using the instantaneous

amplitude and the instantaneous frequency (Taner et al., 1979; Barnes, 1991; Liu and Oristaglio,

1998) is shown in Figure 4.9. In Figure 4.9a the distribution of the instantaneous amplitude is

shown as a function of the corresponding instantaneous frequency at the same travel time and

location. The distribution should, and does resemble the Fourier transform (FT) results with a

central frequency of about 65 MHz. Much more data points are close to the central frequency,

which is not so apparent by using the FT method. Direct examination of the waveforms of Figure

4.8a shows lower frequency ground wave trains arrive before the higher frequency ones (Arcone et

al., 2003) to mark apparent velocity dispersion. Figure 4.9b gives a different view of the velocity

dispersion phenomenon by plotting the instantaneous amplitude as a function of the instantaneous

frequency and the group velocity. This is similar to the technique used by Parra (1996). The group

velocity was inferred from the phase velocity (determined by the phase travel time and receiver

location). Though the velocity is obviously dispersive, no clear Airy phase (Arcone, 1984) can be

identified based on this analysis. This may possibly be caused by the silty surface layer being too

thin, or, the bandwidth of the GPR signal being not wide enough.

35

Figure 4.8. The observed 100-MHz WARR GPR profile in TE mode at Fort Richardson, Alaska(a); and the corresponding FDTD simulated synthesis with a source impulse of Ricker wavelet at65-MHz central frequency (b). The field data were generated by nominally rated “100-MHz” antenna, whose ground-loaded value determined from near-field coupling was 65 MHz (after Arcone et al., 2003).

Figure 4.9. (a) Amplitude spectrum extracted from the instantaneous parameter analysis of the65-MHz Fort Richardson WARR data. It resembles the spectral analysis given by conventionalFourier transform with the central frequency at about 65 MHz; (b) Dispersion relationship derived by the instantaneous parameter analysis from 65-MHz field data. The wave propagation is dispersive with obvious frequency dependency, but no obvious Airy phase can be found. Thefrequency-velocity-amplitude technique used here follows Parra (1996).

36

4.6. Conclusions

Using finite difference time domain technique we have modeled four cases: a combination of 2

situations of geology (a gravelly sandy half-space overlain by a sandy silt layer, and a sandy silt

half-space overlain by gravelly sand half-space) and 2 transverse modes of antenna radiation (TM

and TE). The main conclusions are summarized as follows.

More EM energy is radiated as an air wave for the TM mode; and more EM energy will be sent

into the ground for the TE mode, regardless of geological setting (Figure 4.6). Meanwhile, in a

geological setting where a lower � gravelly sand half-space is overlain by a higher � sandy silt layer

(the case of Fort Richardson), more EM energy will be trapped in the sandy silt layer as a TE mode

in the ground wave guide and very little energy will occur as air radiation, when compared with the

same radiation mode in the case of a higher � sandy silt half-space overlain by a layer of lower �

gravelly sand. For the geological setting of a lower � gravelly sand half-space overlain by a higher �

sandy silt layer, the TE mode mainly contains a ground wave and the TM mode mainly contains air

wave energy, which is traveling at a phase velocity of c and a lower group velocity, as evidenced by

the apparent shingling of the energy arrivals (Figure 4.6d). This implies that for this case a far more

complete and long lasting separation of the air wave and the ground wave can be reached.

It appears that for this common case of an electrically thin drier layer over wetter sediments

that both TE and TM waves are highly suppressed when compared with the case of a thin wetter

layer over drier and coarser sediments (e.g., Fort Richardson), especially for TM waves. For

higher frequency antennas, the guided ground wave modes may propagate better, but then strong

attenuation properties (water relaxation and scattering) intrinsic to the layer itself will also

become a factor in suppressing the ground modes. These properties will also be a factor for

thicker layers, which will offer more path length.

37

5. Mathematical treatment to improve the resolution and penetration

At a given location if both high- and low-frequency GPR data were available, a synthetic

GPR image with deep-penetration and high-resolution can be reconstructed by a signal

processing procedure known as extrapolation with deterministic deconvolution (EDD). There are

two fundamental assumptions to make EDD work. First, the correlation (a transfer function or

filtering function) between the high- and low-frequency data at one location can be

deterministically extracted. Second, this transfer function is spatially stationary so that it can be

applied to another location. Based on the principle of applying EDD to seismic data, which is for

generating the synthetic well logs from surface seismic data, we have extended EDD treatment

to GPR data for the purpose of recovering the high-resolution contents at greater depths. The

compensation of the frequency-dependent absorption loss is based on the constant Q model. This

chapter presents the preliminary test results with numerical synthetic time records. EDD with the

synthetic data appears to be promising. More tests need to be done before EDD can be routinely

applied to GPR field data. For a successful application of EDD to real world data, a careful pre-

processing of the field data is far more critical than the EDD algorithm itself.

5.1. Introduction

The tradeoff between penetration and resolution of a given radar pulse is a well-known

phenomenon to the GPR user community. It is trivial to honor the fact that objects far away

cannot be seen as clearly as those closer to an observer. Likewise, GPR images become blurry or

lose resolution at greater depths, this is caused by the fact that the high-frequency content of the

signal suffers more severe loss than low-frequency content. However, it has always been

desirable that GPR have the ability to detect targets at greater depths and with higher resolution.

In general, it is hard to obtain a subsurface image with deep-penetration and high-resolution

simultaneously. However, if both high- and low-frequency data were acquired at the same

location, we may be able to take the advantages of the relative deeper penetration of the low-

frequency signal and the relatively higher resolution of the high-frequency signal. By taking

advantage of the deep penetration of the low-frequency data and the higher resolution of the

high-frequency data simultaneously, it was a common practice in the last two decades to patch

the high-frequency GPR profile with shorter time window at shallow depth onto the low-

38

frequency GPR profile with longer time window. Clearly, the desire to get deep-penetration and

high-resolution simultaneously can hardly be fulfilled with any rigidly configured physical

approach. The only possible solution is seeing if digital signal processing techniques can lend us

some help.

Deterministic deconvolution was originally used to increase signal resolution in seismic

exploration (Oldenburg, et al., 1981; Treitel and Lines, 1982) then adopted to enhance GPR

reflection profiles (Xia, et al., 2004). Digital signal analysis recognizes the low- and high-

frequency GPR responses at a particular location as the convolutions of the low- and high-

frequency source wavelets with the material properties (medium imperfection shown as

attenuation and impedance contrast shown as reflectivity), respectively (Liu and Oristaglio,

1998; Treitel and Robinson, 1966; Liu, et al., 1998). If both low- and high-frequency responses

at a shallower (for surface GPR) or closer (for borehole radar) location and low-frequency

response for deeper or farther locations are available, one can infer the high-frequency response

at the deeper or farther locations. The information about the high-frequency response for the

deeper/farther locations is extrapolated by extending the relationship of high- and low-

frequency correspondence at shallower parts into greater depths. In certain circumstances, this

extension can be achieved through the extrapolation by deterministic deconvolution (EDD), as

has been down in the treatment of seismic data (Zhou, et al., 2001).

The procedure is explained in detail in the following sections with mathematical derivations

at the first, followed by applying EDD to both synthetic and field data to verify the effectiveness

of this method. The examples showed that EDD has restored some high frequency contents in

later times (greater depth). However, EDD is more effective at improving the resolution of

synthetic data than the real field GPR data, simply because synthetic data is noise-free and real-

world field data contains random noise caused by variety of sources and system errors due to

differences in system response for antennae working at different central frequencies. This

chapter concludes with a discussion on future research for this topic.

39

5.2. Extrapolation by deterministic deconvolution

Subsurface structure responses to the impinging radar source are recorded by the receiving

antenna as time-domain traces. And these recorded responses R(t) can be regarded as the

convolution among the GPR source signal wavelet W(t), the subsurface structures S(t), in terms

of the reflectivity, and the material’s intrinsic attenuation A(t). Given this simple analytic model,

the recorded responses Rl(t) for the low- frequency, and Rh(t) for the high-frequency, can be

expressed as the convolutions of the source wavelet Wl(t) and Wh(t) and the material structure

S(t), respectively, and Al(t) and Ah(t), the material’s attenuation properties for low- and high-

frequency GPR signal, respectively, in the following two equations:

)()(*)()()()(*)()(tAtStWtR

tAtStWtR

hhh

lll

∗=∗= (5.1)

where the asterisk symbol ‘*’ denotes convolution. The structure function, S(t), is the same for

both high- and low-frequency signals anywhere in the subsurface. Combining the two equations

in eq. (5.1), S(t) cancels out and we end up with a direct relationship between Rl(t) and Rh(t)

)()()( tRtGtR lh ∗= (5.2)

where G(t) is the lump-sum transfer function that defined as the ratio of the convolutions

between the source wavelet and attenuation for the high- and low-frequency signals;

)()()()()(

tAtWtAtWtG

ll

hh

∗∗

= (5.3)

the attenuation operator can be simplified with a reasonable theoretical model which

characterizes the medium’s intrinsic attenuation property. One such model is the well-known

constant Q model (Liu, et al., 1998), which defines a linear dependence of the attenuation to

signal frequency, i.e., �=�0f, so the attenuation operator becomes:

ftt eetA 0)( αα −− ==

and the ratio of the attenuation operators for high- and low-frequency GPR signals is simply:

40

fttff

l

h eetAtA

lh Δ−−− == 00 )(

)()( αα

Thus, the lump-sum transfer function can be expressed as

ftft

l

h

ll

hh etgetWtW

tAtWtAtWtG Δ−Δ− ∗=∗=

∗∗

= 00 )()()(

)()()()()( αα (5.4)

where g(t) is the transfer function between the high- and the low-frequency source wavelets.

Apparently, g(t) was originally defined as the spectral ratio in frequency domain. It can also be

elucidated as the cross-correlation of the high- and low-frequency time sequences in time

domain by the following derivations. From eq. (5.2)

ftllh etRtgtRtGtR Δ−∗∗=∗= 0)()()()()( α (5.5)

It is straightforward to invert eq. (5.5) and get the transfer function g(t) by

)](),([)()( 0 tRtRdeconvetgtG lhft =∗= Δ−α (5.6)

where function deconv[Rh(t), Rl(t)] means de-convolving Rl(t) out of Rh(t) and

ftlh etRtRdeconvtg Δ∗= 0)](),([)( α (5.7)

Due to the assumed spatial invariability of g(t), the predicted high frequency signal response

at a different location, Rh2(t), can be obtained by convolving g(t) with the low frequency signal

response Rl2(t) measured at location 2. The mathematical expression can be described by

)()()( 22 tRtgtR lh ∗= (5.8)

With this process, the ‘extrapolated’ Rh2(t) has inherited both advantageous properties of the

different frequencies, high resolution and deep penetration. The inversion of g(t) is a

deconvolution that is a deterministic process (eq.(5.7)), such that the high frequency signal

response, Rh2(t), is extrapolated from the measured low frequency response, Rl2(t) (eq.(5.8)).

Therefore, this method is regarded as the extrapolation by deterministic deconvolution (EDD).

41

A simple and effective way to find g(t) is to use a linear optimum discrete-time filtering

method. This chapter uses the widely adapted Wiener filtering technique. The low-frequency

signal, Rl(t), can be regarded as the input, g(t) is the filter system and the filter response or

output is Rh(t), so that

k)-g(t(k)R = g(t)*(t)R = (t)RN

0kllh ∑

=

⋅ (5.9)

where N is the length of the discrete input signal. The principle of orthogonality (Haykin, 2002)

specifies the necessary and sufficient conditions for the optimum operation of the filter as

NitRigtRktREN

ilhl …==−⋅−− ∑

=

0,1,2,k ,0))]()()()(([0

(5.10)

where g(i) is the i-th coefficient in the impulse response of the optimum filter. Expanding this

equation and rearranging terms produces,

0,1,2,...Nk ),()()(0

=−=−⋅∑=

kpkirigN

i (5.11)

where )]()([)( itRktREkir ll −⋅−=− is the auto-correlation of the input; while

)]()([)( tRktREkp hl ⋅−=− is the cross-correlation between the filter input, Rl(t-k), and the

desired response, Rh(t), for a time step lag (k). Let R denote the N-by-N autocorrelation matrix

of Rl(t), and P denotes the cross-correlation vector between the input of the filter and the desired

output, simplifying eq. (5.11),

PgR =⋅ (5.12)

This is the so-called Wiener-Hopf equation (Haykin, 2002). The filter vector, g(t), can now be

readily calculated by the inverse of the matrix R times P,

PR g -1 ⋅= (5.13)

Using g(t) in eq. (5.8), we are able to calculate the high-frequency extrapolation at any location

given the low-frequency response at those locations.

The resolution enhancement using EDD is illustrated with a single trace of real GPR data

collected at one location using both high- and low-frequency records (Figure 5.1). The top panel

42

of Figure 5.1 is a snippet of 100-MHz GPR data, the low-frequency sequence, and the second

panel is the 400-MHz data, the high-frequency sequence. The recording time window for high-

frequency data is usually shorter than the low-frequency counterpart. The transfer function

derived from the records shown in the top two panels is presented as the third panel. Apparently,

the effective length of g(t) is limited by the length of the high-frequency data. The bottom panel

shows the EDD synthesis of the high-frequency record. Comparison of the EDD synthesis and

the original high-frequency data in the earlier time shows a reasonable agreement between these

two sequences. This agreement is the base to entrust the EDD extrapolation for later times

(corresponding to greater depth). The approach and effectiveness of EDD are further examined

in the following sections when it is applied to synthetic and real GPR data.

5.3. Applications to synthetic data

To test the effectiveness of EDD for improving GPR data resolution, synthetic time traces

are computed from a 2-dimensional (2D) finite-difference time-domain numerical modeling

technique (Liu and Arcone, 2003). A five layer model was created and a metal cylinder was

positioned in the center of the third layer (Figure 5.2). The model consists of 600 grid cells, 300

of which are in the horizontal or x-direction and 200 grid cells in the vertical or y-direction. All

the four sides of the model domain are bounded by an 8-grid cell thick perfectly match layer

(PML) absorption boundary (Berenger, 1994). All grid cells have a uniform size of 0.02 m on a

side. Thus, the modeled physical domain is 6m x 4m. The material properties of different layers,

from top to bottom, are:

1) A strip of air at the top of the model with the dielectric constant of vacuum and zero

conductivity (�r = 1, and ��= 0.0 S/m);

2) A 0.4 m thick surface layer with �r = 9, and ��= 0.0 S/m;

3) A 1.6-m thick dielectric layer with �r = 16, and ��= 0.0 S/m;

4) An electrically conductive cylinder (�r = 25, and ��= 107 S/m) with a radius of 0.5 m

centered in the third layer;

43

Figure 5.1. EDD illustration using a single trace from a piece of real data. From top to bottom: 1) the low- frequency record; 2) the high-frequency record; 3) the transfer function; and 4) the extrapolated high-frequency syntheses by EDD.

44

5) Another thin layer with the same thickness and property of Layer 2;

6) A dielectric half-space with the same material property of Layer 3.

Figure 5.2. The model used to generate the synthetic GPR time records for evaluating the EDD method. The dielectric constant of each layer is labeled as �� The unit of the conductivity � is Siemens per meter (S/m).

The synthetic time sequences are generated with a source impulse whose shape is a Ricker

wavelet with the central frequency of 100- and 400-MHz, respectively. A time step of 0.026667

nanoseconds (ns) is used for both high- and low-frequency time sequences, a value fulfills the

stability condition for FDTD modeling (Schroder and Scott, 2002) when the grid size is 0.02 m.

The time step adapted in the forward model is about 10 times shorter than the sampling rate used

in acquiring 400-MHz GPR field data. For example, the sampling frequency of 4141 MHz

corresponds to a sampling interval of 0.2415 ns. Thus, it is necessary to re-sample the synthetic

time traces to a longer time interval to reduce the redundancy and increase the efficiency of EDD

calculation. For the numerical modeling, the total simulated time is 66 ns (2500 time steps),

which allows enough time for the reflected waves from the bottom interface to travel back to the

receiving points at the air-ground surface. The sources and receivers were placed one grid above

the air-ground interface to mimic the monostatic reflection profiling. A total of 51 time traces

were generated with this model with a 0.4-m source-receiver separation. The profile extended 5

meters in the x-direction with 0.1-m spacing between two adjacent traces.

45

The evaluation of EDD was conducted following procedures described in Section II with

results shown in Figures 5.3, 5.4 and 5.5. The top panels of Figures 5.3, 5.4 and 5.5 are the

original low-frequency (100 MHz) records. The second panels are the original high-frequency

(400 MHz) records. The third panels are the calculated transfer function, g(t). Finally, the bottom

panels are the extrapolated high-frequency traces with EDD.

0 1 2 3 4 5

0

20

40

60

Low-frequency (100MHZ) records

Tim

e(n

s)

0 1 2 3 4 5

0

20

40

60

High-frequency (400MHZ) Records

Tim

e(n

s)

0 1 2 3 4 5

0

20

40

60

Transfer Functions

Tim

e(n

s)

0 1 2 3 4 5

0

20

40

60

High-frequency Syntheses by EDD

Distance (meter)

Tim

e(n

s)

Figure 5.3. The EDD extrapolation for the synthetic records generated with the model shown in Figure 5.2. From top to bottom: 1) the low- frequency record; 2) the high-frequency record; 3) the transfer function; and 4) the extrapolated high-frequency syntheses by EDD. The recording time windows are 66 ns for both high- and low-frequency synthetic data.

46

0 1 2 3 4 5

0

20

40

60

Low-frequency (100MHZ) recordsT

ime(

ns)

0 1 2 3 4 5

0

20

40

60

High-frequency (400MHZ) Records

Tim

e(n

s)

0 1 2 3 4 5

0

20

40

60

Transfer Functions

Tim

e(n

s)

0 1 2 3 4 5

0

20

40

60

EDD with only High-frequency Channel 11

Distance (meter)

Tim

e(n

s)

(by Ch. 11)

Figure 5.4. Same as Figure 5.3 with the exception of only trace No. 11 of the high-frequency record 400 MHz record is used for calculating the transfer function g(t) in EDD extrapolation.

47

0 1 2 3 4 5

0

20

40

60

Low-frequency (100MHZ) records

Tim

e(n

s)

0 1 2 3 4 5

0

20

40

60

High-frequency (400MHZ) Records

Tim

e(n

s)

0 1 2 3 4 5

0

20

40

60

Transfer Functions

Tim

e(n

s)

0 1 2 3 4 5

0

20

40

60

High-frequency Syntheses by EDD

Distance (meter)

Tim

e(n

s)

Figure 5.5. Same as Figure 5.3 with the exception of a shorter time window (45 ns) is used for the high-frequency 400 MHz record for calculating the transfer function g(t) in EDD extrapolation.

48

Data in Figure 5.3 shows both the low- and high-frequency traces which are recorded with a

time window of 66 ns. Each pair of the low- and high-frequency traces at one location generates

one unique transfer function. A comparison of the EDD extrapolated high-frequency signal and

the original 400 MHz traces shows that the extrapolation works properly: the extrapolated and

the original simulated traces are literally identical, as they should be, since no extrapolation has

been conducted here at all. The high-frequency information is available in all space where low-

frequency information available. This step was designed to verify that the EDD algorithm can

work properly for a monostatic GPR profile.

To test algorithm’s capability of extrapolation in the lateral direction, we assumed that the

high-frequency 400 MHz data with the full length of the time window is available at only one

location, Trace No. 11. EDD was carried out by using g(t) calculated from Trace No. 11 alone

and applied to all locations on the low-frequency profile to get the extrapolated high-frequency

syntheses. The extrapolation result is shown in Figure 5.4. Obviously, by losing the correlation

constraints at all other locations, the image formed by the extrapolated traces is much more

smeared. Nevertheless, the dominant frequency of the traces is still high and close enough to the

original high-frequency record, and the reflection from the lower thin layer is clearly seen.

For surface GPR, extrapolation into greater depths is the main purpose to run EDD. We

tested the capability of EDD by shortening the window of the 400-MHz record to 40 ns then

carried out the EDD extrapolation again. The results are shown in Figure 5.5. Even with

truncated 400-MHz data, the EDD algorithm was still able to show the reflection from the thin

layer which is beyond the cut-off window for the 400-MHz data.

To examine how much high-frequency contents has been recovered, we turn to an

examination of the spectra from the original and EDD extrapolated high-frequency traces. The

comparison of the spectra from the EDD extrapolated syntheses and the original records

corresponding to the time domain records (Figure 5.5) are shown as Figure 5.6. It is clear that

EDD has recovered the high-frequency lose in deeper part effectively with the comparison of the

original high-frequency records (the second panel from the top) and the EDD extrapolated

syntheses (the bottom panel). From Figure 5.6 we can see that the peak value of the amplitude is

about 100 MHz in the spectra of the low-frequency records, while it is corresponding to about

49

400 MHz for the high-frequency records. As expected, the high-frequency records have a wider

frequency bandwidth than the low frequency records. The spectra of g(t) and EDD have

maximum amplitude at about 400 MHz. The spectra of the high-frequency syntheses by EDD

resembled the original simulated high-frequency records. This good agreement between the

original and EDD data can be explained by the fact the simulated synthetic data is noise-free and

has close cross-correlation at all the locations. The comparison of the four spectra plots show

that the extrapolated high-frequency records retained the same resolution as the original high-

frequency records.

5.4. Applications to field GPR data

For serving the ultimate objective of boosting up the resolution of the field GPR data, EDD

was further examined by being applied to a set of the 100-MHz and 400-MHz field data

collected in the town of New Milford, Connecticut, for highway engineering purposes (Liu,

2004). Both the 100- and 400-MHz data sets were collected along a single survey line with 10-

cm trace spacing. The soil at the site is very sandy/gravelly with a great depth of penetration.

Both data sets show a strong reflector at the two-way travel time 100 ns. The time window

length of the 100-MHz data is 400 ns (only 250 ns is used for EDD analysis), while the length

for the 400-MHz data is 105 ns. Figure 5.7 shows the original data and the EDD processing

results. Though the EDD extrapolation has not reached the level of effectiveness shown in the

synthetic data, it does catch the major features of the 100-MHz data beyond 105 ns and boost up

them to a shorter period (higher frequency).

50

0 1 2 3 4 5

0

200

400

600

800

Spectra of Low-frequency (100MHZ) records

Fre

qu

ency

(MH

z)

0 1 2 3 4 5

0

200

400

600

800

Spectra of High-frequency (400MHZ) Records

Fre

qu

ency

(MH

z)

0 1 2 3 4 5

0

200

400

600

800

Spectra of Transfer Functions

Fre

qu

ency

(MH

z)

0 1 2 3 4 5

0

200

400

600

800

Spectra of High-frequency Syntheses by EDD

Distance (meter)

Fre

qu

ency

(MH

z)

Figure 5.6. From top to bottom: the spectra of the 100-MHz records, 400-MHz records, the transfer function g(t), and the EDD extrapolated syntheses.

51

0 2 4 6 8 10 12 14 16

0

50

100

150

200

250

Low-frequency (100MHZ) recordsT

ime(

ns)

0 2 4 6 8 10 12 14 16

0

50

100

150

200

250

High-frequency (400MHZ) Records

Tim

e(n

s)

0 2 4 6 8 10 12 14 16

0

50

100

150

200

250

Transfer Functions

Tim

e(n

s)

0 2 4 6 8 10 12 14 16

0

50

100

150

200

250

Hi-frequency Syntheses by EDD

Distance (meter)

Tim

e(n

s)

Figure 5.7. EDD extrapolation of field data acquired at the New Milford, CT site (Liu, 2004).

52

0 2 4 6 8 10 12 14 16

0

100

200

300

400

500

Spectra of Low-frequency (100MHZ) records

Fre

qu

ency

(MH

z)

0 2 4 6 8 10 12 14 16

0

100

200

300

400

500

Spectra of High-frequency (400MHZ) Records

Fre

qu

ency

(MH

z)

0 2 4 6 8 10 12 14 16

0

100

200

300

400

500

Spectra of Transfer Functions

Fre

qu

ency

(MH

z)

0 2 4 6 8 10 12 14 16

0

100

200

300

400

500

Spectra of High-frequency Syntheses by EDD

Distance (meter)

Fre

qu

ency

(MH

z)

Figure 5.8. From top to bottom: the spectra of original 100-MHz and 400-MHz field data, the transfer function g(t), and the EDD extrapolated high-frequency Syntheses for the New Milford data set.

53

The comparison of the spectra for the field data are shown in Figure 5.8. The extrapolated

high-frequency syntheses by EDD have the same resolution as the original 400-MHz data. The

centroid of the spectra of the field data has a downshift to lower frequencies for both the 100-

and the 400-MHz data, and it is more obvious for the 400-MHz data. The coupling effect of the

antenna to the ground (Liu, et al., 1998; Zhou, et al., 2001) is the main cause for this shift. The

spectra in Figure 5.8 are relatively rough in comparison with the spectra shown in Figure 5.6,

due mainly to the existence of signal noise caused by real-world conditions. When comparing

with the application to the synthetic data (Figures 5.3 – 5.6), without surprise, the application to

the real field data shows a moderate success, caused mainly by the uncertainties and errors

existing in both the high- and low-frequency data from various sources. The degree of

correlation has been degraded.

5.5. Conclusions

In this chapter we described the extrapolation with deterministic deconvolution (EDD)

method to enhance signal resolution in later times for a GPR profile. The EDD approach has

been explained with the assistance of a number of examples using both synthetic and field GPR

data. Application of EDD to synthetic data demonstrates that it can work effectively as the

method designed for. The application of EDD to real GPR data achieved certain effectiveness

with moderate success. Apparently, we need carry out further theoretical development before

applying EDD to real field data. A sequence of carefully designed procedures of field data

acquisition and data pre-processing underlies the effectiveness of the EDD method.

54

6. Conclusions and Future Work

6.1 A general summary

This report presents a scope of research on GPR pavement assessment covering (1) the

laboratory experiments on HMA specimens in dry, water-saturated and frozen conditions; (2) a

series of GPR field measurements in different meteorological conditions for assessing the water

content in paved road; (3) a numerical simulation of GPR response to a surface dielectric thin

layer with the finite difference time domain method; and finally, (4) a digital signal processing

algorithm to improve the resolution of GPR signals. As a non-destructive testing tool for

pavement assessment, GPR proves an approach supplementary to the traditional direct and

impact engineering tests for HMA pavement properties.

With the combination of laboratory test results and field observations, GPR is capable of

providing indirect estimate on void ratio of the pavement as well as the state of pore fluid in the

voids. This information is critical for extending the durability and stability of HMA paved road,

especially when the information is in-situ, and real-time (or close to real-time). Calibration of

field data with laboratory test results of the HMA specimens with the same composition may

substantially improve the accuracy of pavement parameter estimations.

6.2 Future work

GPR technique for pavement assessment is becoming more and more mature in the last

several years. A number of commercially available systems have been emerging to the market.

Advancement in electronic and computer engineering technologies will continue the reduction of

GPR hardware cost. Numerous theoretical and analytical studies, including results presented in

this report, could be acting as the technical support for GPR pavement assessment in production

mode at both the project and network levels. Advances in hardware and software will make GPR

technique affordable to most state DOTs for paved road status monitoring at the network level

and quality assessment and control at project level. Close interactions among DOT laboratories,

project managers, and academic researchers will push the GPR technique to a new level of

engineering application and make it more attractive to pavement engineers. These interactions

55

can be hosted by state DOT laboratories or DOT sponsored pavement research centers in

academic institutions.

Disclaimer: The contents of this report reflect the views of the author who is responsible for

the facts and accuracy of the data presented herein. The contents do not necessarily reflect the

official views or policies of the Connecticut Department of Transportation or the Federal

Highway Administration. The report does not constitute a standard, specification, or regulation.

Neither the United State Government nor the State of Connecticut endorses products or

manufacturers. Trade marks or manufacturer names appear herein only because they are

considered essential to the objective of this document.

56

References

Annan, A.P., 1973, Radio interferometry depth sounding, Part I-Theoretical discussion,

Geophysics, 38, 557-580.

Annan, A.P., Waller, W.M., Strangway, D.W., Rossiter, J.R., Redman, J.D., and Watts, R.D., 1975,

The electromagnetic response of a low-loss, 2-layer dielectric earth for horizontal electric

dipole excitation, Geophysics, 40, 285-298.

Annan, A.P., 1996, Transmission dispersion and GPR, J. Environ. & Engineer. Geophysics, 0,

125-136.

Annan, A. P., J. L. Davis, and D. Gendzwill, 1988, Radar sounding in potash mines,

Saskatchewan, Canada, Geophysics, 53, 1556-1564.

Arcone, S.A., 1984, Field observations of electromagnetic pulse propagation in dielectric slabs,

Geophysics, 49, 1763-1773.

Arcone, S.A., and Delaney A.J., 2002, A field study of GPR attenuation rates in natural and

contaminated silt, in Proc. 9th Int. Conf. GPR, SPIE 4758, 302-307.

Arcone, S.A., Lawson, D.E., Delaney A.J., Strasser, J.C., and Strasser, J.D., 1998, Ground-

penetrating radar reflection profiling of groundwater and bedrock in an area of discontinuous

permafrost, Geophysics, 63, 1573-1584.

Arcone, S. A., P. Peapples, and L. Liu, 2003, Propagation of a ground-penetrating radar (GPR)

pulse in a thin-surface waveguide, Geophysics, 68, 1922-1933.

Attoh-Okine, N., 1995, Use of artificial neural networks in ground penetrating radar applications

in pavement evaluation and assessment, Proceedings of International Symposium on Non-

Destructive Testing in Civil engineering, pp. 93-99.

Bano, M., 1996, Constant dielectric losses of ground penetrating radar waves, Geophys. J. Int.,

124, 279-288.

Balanis, C. A., 1989, Advanced Engineering Electromagnetics, 981 pp. John Wiley & Sons, New

York.

57

Barnes, A.E., 1991, Instantaneous frequency and amplitude at the envelope peak of a constant-

phase wavelet, Geophysics, 56, 1058-1060.

Berenger, J.P., 1994, A perfectly matched layer for the absorption of electromagnetic waves, J.

Computational Physics, 114, 185-200.

Berktold, A., K. G. Wollny, and H. Alstetter, Subsurface Moisture Determination with the

Ground Wave of GPR. Proceedings of the 7th International Conference on GPR, Lawrence,

Kansas, May 1998.

Bethel, R., and R. Rahikka, 1987, An optimum first-order time delay tracker, IEEE Trans.

Aerospace and Electronic Systems, AES-23, 718-725.

Bethel, R., and G. Paras, 1994, A PDF multitarget tracker, IEEE Trans. Aerospace and

Electronic Systems, AES-30, 386-403.

Black, K., and P. Kopac, 1992, The application of ground-penetrating radar in highway

engineering, Public Roads, 56, 96-103.

Brekhovskikh, L.M., 1960, Waves in layered media,, Academic Press, New York, NY.

Budden, K.G., 1961, The wave-guide mode theory of wave propagation,, Prentice Hall, Englewood

Cliffs, New Jersey.

Brzostowski, M. and G. McMechan, 1992, 3-D tomographic imaging of near-surface seismic

velocity and attenuation, Geophysics, 57, 396-403.

Cai, J., and G. A. McMechan, 1995, Ray-based synthesis of bistatic ground-penetrating radar

profiles, Geophysics, 60, 87-96.

Charlton, M. Small Scale Soil-Moisture Variability Estimated Using Ground Penetrating Radar.

Proceedings of the 8th International Conference on GPR, Gold Coast, Australia, 2000.

Cheng, D. K, 1983, Field and wave electromagnetics, Addison-Wesley Publ. Co., pp. 317-322.

Chung, T., C. R. Carter, T. Masliwec, and D. Manning, 1994, Impulse radar evaluation of

concrete, asphalt, and waterproofing membrane, IEEE Transactions on Aerospace and

Electronic Systems, 30, 404-414.

58

Collins, M. E., and J. A. Doolittle, 1987, Using ground-penetrating radar to study soil

microvariability, Soil Sci. Soc. Am. J., 51, 491-493.

Daniels, J. J., R. Roberts, and M. Vendl, 1995, Ground penetrating radar for the detection of

liquid contaminants, J. Appl. Geophys., 33, 195-207.

Daniels, J., 1988, Locating caves, tunnels and mines, The Leading Edge, 7, 32-37.

Davis, J. L., and A. P. Annan, 1989, Ground-penetrating radar for high-resolution mapping of

soil and rock stratigraphy, Geophys. Prosp., 37, 531-551.

Davis, J.L., C.B. Dawley, D.E. Mesher and J.R. Rossiter, 1994, Quantitative Measurement of

Pavement Structures Using Radar, Presented at the 5th International Conference on Ground

Penetrating Radar, June 12-16, 1994.

Dawley, C.B., and D.E. Mesher, 1994, Characterization of Multi-Layer Pavement Structures

Using a New GPR Technology, Presented at the Innovative Design and Construction Session

of the 1994 International Road Federation Conference and Exposition, Calgary, Alberta.

Endres, A., and R. Knight, 1991, The effects of pore-scale fluid distribution on the physical

properties of partially saturated sandstones, J. Appl. Phys., 69, 1091-1098.

Endres, A., and J.D. Redman, 1996, Modeling the electrical properties of porous rocks and soils

containing immiscible contaminants, Jour. Environ. & Engineer. Geophys., 0, 105-112.

Feng, S., and P.N. Sen, 1985, Geometrical model of conductive and dielectric properties of

partially saturated rocks, Jour. Appl. Phys., 58, 3236-3243.

Fisher, E., G. A. McMechan, and A. P. Annan, 1992a, Acquisition and processing of wide-

aperture ground-penetrating radar data, Geophysics, 57, 495-504.

Fisher, E., G. A. McMechan, A. P. Annan, and S. W. Cosway, 1992b, Examples of reverse-time

migration of single-channel ground-penetrating radar profiles, Geophysics, 57, 577-586.

Gazdag, J., 1981, Modeling of the acoustic wave equation with transform methods, Geophysics,

46, 854-859.

59

Goodman, D., 1994, Ground-penetrating radar simulation in engineering and archaeology,

Geophysics, 59, 224-232.

Greaves, R. J., D. P. Lesmes, M. L. Lee, and M. N. Toksoz, 1996, Velocity Variations and Water

Content Estimated from Multi-Offset Ground-Penetrating Radar. Geophysics, 61, 683-695.

Greeuw, G., J.W. de Feijter, and A. Kathage, 1992, Field and laboratory tests on line scatterers.

Proceedings of the 4th International Conference on GPR, pp. 111-118.

Grumman, D.L., and J. J. Daniels, 1996, Experiments on the detection of organic contaminants

in the vadose zone, Jour. Environ. & Engineer. Geophys., 0, 31-38.

al Hagrey, S., and Muller, C., 2000, GPR study of pore water content and salinity in sand, Geophys.

Prospect., 48, 63-85.

Hammond, W. R., and Sprenke, K. F., 1991, Radar detection of subglacial sulfides, Geophysics,

56, 870-873.

Harrington, R. F., 1961, Time-harmonic electromagnetic fields, McGraw-Hill Book Co., pp. 1-

12.

Haslund, E., and B. Nost, 1998, Determination of Porosity and Formation Factor of Water-

Saturated Porous Specimens from Dielectric Dispersion Measurements. Geophysics, 63, 149-

153.

Haykin, S., 2002, Adaptive Filter Theory, (4th Edition), Prentice Hall Information and System

Sciences Series, 920 pp.

Hoeskstra, P., and Delaney, A.J., 1974, Dielectric properties of soils at UHF and microwave

frequencies, J. Geophys. Res., 79, 1699-1708.

Hohmann, G. W., 1987, Numerical modeling for electromagnetic methods of geophysics, in M.

N. Nabighian, Ed., Electromagnetic methods in applied geophysics-theory, Vol. 1, Soc. Expl.

Geophys., pp. 313-363.

Hugenschmidt, J., M. Partl, H. de Witte, and C. Uri, GPR inspection of a mountain motorway-a

case study, Proceedings of the 6th International Conference on Ground Penetrating Radar,

pp. 365-370, Sendai, Japan, Sept. 30-Oct. 3, 1996.

60

Huston, D., J. Hu, K. Maser, W. Weedon, and C. Adam, 2000, GIMA ground penetrating radar

system for monitoring concrete bridge decks, Jour. Appl. Geophys., 43, 139-146.

Huston, D., K. Maser, J. Hu, W. Weedon, and C. Adam, 1998, Bridge deck evaluation with

ground penetrating radar, Proceedings of the 7th International Conference on Ground

Penetrating Radar, pp. 595-599.

Johansson, E.M., and J.E. Mast, 1994, Imaging Algorithms for Synthetic Aperture Ultra-Wide

band Radar, Engineering Research, Development, and Technology, Lawrence Livermore

National Laboratory, Livermore, California, UCRL-53868-93.

Joncsher, A. K., 1977, The ‘universal’ dielectric response, Nature, 267, 673-679.

Joncsher, A. K., 1978, Low-frequency dispersion in carrier-dominant dielectrics, Philos. Mag.

B., 38, 587-601.

de Jongh, R., A. Yarovoy, L. Ligthart, I. Kaploun, and A. Schukin, 1998, Design and analysis of

new GPR antenna concepts, Proceedings of the 7th International Conference on Ground

Penetrating Radar, pp. 81-86.

Kjartansson, E., 1979, Constant Q–wave propagation and attenuation, J. Geophys. Res., 84,

4137–4748.

Kong, F.-N., H. Westerdahl, and L. Gelius, 1998, A very wide bandwidth step frequency GPR

for testing concrete re-bars, Proceedings of the 7th International Conference on Ground

Penetrating Radar, pp. 455-460.

Kosloff, D. D., and E. Baysal, 1982, Forward modeling by a Fourier method, Geophysics, 47,

1402-1412.

Kung, K-J. S., and Lu, Z-B., 1993, Using ground-penetrating radar to detect layers of

discontinuous dielectric constant, Soil Sci. Soc. Am. J., 57, 335-340.

Kuo, J. T., and V-H. Cho, 1980, Transient time-domain electromagnetics, Geophysics, 45, 271-

291.

Liu, L. J. W. Lane, and Y. Quan, 1998, Radar attenuation tomography using the centroid

61

frequency down-shift method, Journal of Applied Geophysics, 40, 105-116.

Liu, L. and Oristaglio, M., 1998, Instantaneous parameter analysis using wavelet transform, in

Proc. 7th Int. Conf. on GPR, 219-224.

Liu, L., and S.A. Arcone, 2002, Numerical simulation of near-surface GPR in TE and TM modes,

in Proc. 9th Int. Conf. GPR, SPIE 4758, 273-278.

Liu, L. and Arcone, S. A., 2003, Numerical simulation of the wave-guide effect of the near-

surface thin layer on radar wave propagation, J. Environ. Engineer. Geophys. 8, 133--141.

Liu, L., 2004, Detection of possible old buried graves in the Pickett District Cemetery with

ground penetrating radar (GPR) in New Milford, Connecticut, Technical Report to

Connecticut Department of Transportation, 23 pp.

Maser, K. R., 1994, Ground penetrating radar survey of pavement thickness on MN/ROAD

sections, INFRASENSE, Inc. Report No. MN/RC-95/06 to Minnesota Department of

Transportation, Office of Minnesota Road Research.

Maser, K. R., 1996, Condition assessment of transportation infrastructure using ground-penetrating

radar, J. Infrastructure Systems, 2, 94-101.

Mathews, S., A. Goodier, and S. Massey, 1998, Permittivity measurements and analytical

dielectric modeling of plain structural concretes, Proceedings of the 7th International

Conference on Ground Penetrating Radar, pp. 363-368.

McCann, D. M., P. D. Jackson, and P. J. Fenning, 1988, Comparison of the seismic and ground

probing radar methods in geological surveying, IEE Proc. Part F, Vol. 135, pp. 380-390.

Mesher, D.E., C.B. Dawley, J.L. Davis, and J.R. Rossiter, 1995, Evaluation of a New Ground

Penetrating Radar Technology to Quantify Pavement Structures, Presented at the 74th

Annual Meeting of the Transportation Research Board, Washington, D.C.

Mesher, D.E., C.B. Dawley, and B. Pulles, 1996, A comprehensive radar hardware, interpretation

software and survey methodology paradigm for bridge deck assessment, Proceedings of the

6th International Conference on Ground Penetrating Radar, pp. 353-358.

62

Mesher, D.E., 1992, Artificial Intelligence Applied to Bridge Deck Radar Interpretation, Ph.D

dissertation, McMaster University, Hamilton, Ontario, Canada.

Mesher, D.E., and W.F.S. Poehlman, 1992, Interpretation of pulsed radar backscatter waveforms

using a knowledge based system, presentation at AIENG'92, Applications of Artificial

Intelligence in Engineering, University of Waterloo, Ontario, Canada, July 14-17, 1992.

Mesher, D.E., and W.F.S. Poehlman, 1988, Impulse Radar Ground Profiling System, Presented

at Telecommunication Research Institute of Ontario (TRIO) Expert System Group Report,

McMaster University.

Molyneaux, T. S. Millard, J. Bungey, and J. Zhou, 1995, Radar assessment of structural concrete

using neural network, Non-Destructive Testing and Evaluation International, 28, 281-288.

Nelson, S.D., 1994, Electromagnetic Modeling for Target-Rich Embedded Environments,

Engineering Research, Development, and Technology, Lawrence Livermore National

Laboratory, Livermore, California, UCRL-53868-93.

Noon, D., 1996, Stepped-frequency radar design and signal processing enhances ground

penetrating radar performance, PhD Thesis, University of Queensland, Australia.

Noon, D., G. Stickley, and D. Longstaff, 1996, A frequency independent characterization of GPR

penetration and resolution performance, Proceedings of the 6th International Conference on

Ground Penetrating Radar, pp. 329-334.

Oldenburg, D. W., Levy, S., and Whittall, K. P., 1981, Wavelet estimation and deconvolution,

Geophysics 46, 1528--1542.

Pagnoni, T., 1996, An automated radar system for non destructive testing of bridge and highway

pavements, Proceedings of the 6th International Conference on Ground Penetrating Radar,

pp. 359-364.

Parra, J. O., 1996, Guided seismic waves in layered poroviscoelastic media for continuity logging

applications: model studies, Geophys. Prospect., 44, 403-425.

Patel, D. L., 1993, Desert Storm soil properties and mine detectors, Technical Report to Unite

State Army, USA-BRDEC-TR//2537, 80 pp.

63

Peacock, J. H., 1997, Dynamically phased radar: a new technique for geotechnical

investigations, SEG Expanded Abstracts, pp. 753-756.

Powers, M. H., 1995, Dispersive Ground Penetrating Radar Modeling In 2d. PhD Thesis,

Colorado School of Mines, Golden, CO, USA.

Pratt, D., and M. Sansalone, 1992, Impact-echo signal interpretation using artificial intelligence,

ACI Materials Journal, 89, 178-187.

al-Qadi, I. L., 1992, The penetration of electromagnetic waves into hot-mix asphalt, in Proceedings

of Nondestructive Evaluation of Civil Structures and Materials, Boulder, CO, 195-209.

Quan, Y., and J.M. Harris, 1997, Seismic attenuation tomography using the frequency shift

method, Geophysics, 62, 895-905.

Redman, D., Davis, J.L., Galagedara, L.W., and Parkin, G.W., 2002, Field studies of GPR air

launched surface reflectivity measurements of soil water content, in Proc. IX Int. Conf. Ground

Penetrating Radar, Santa Barbara, CA, SPIE 4758, 156-161.

Robert, A., 1998, Dielectric permittivity of concrete between 50 MHz and 1 GHz and GPR

measurements for building materials evaluation, J. Appl. Geophys. 40, 119-138.

Roberts, R. L., and J. J. Daniels, 1997, Modeling near-field GPR in three dimensions using the

FDTD method, Geophysics, 62, 1114-1126.

Roberts, R. L., and D. Petroy, 1996, Semi-automatic processing of GPR data collected over

pavement, Proceedings of the 6th International Conference on Ground Penetrating Radar, pp.

347-352.

Rossiter, J.R., LaTorraca, G.A., Annan, A.P., Strangway, D.W., and Simmons, G., 1973, Radio

interferometry depth sounding: Part II-Experimental results, Geophysics, 38, 581-599.

Rossiter, J.R., Strangway, D.W., Annan, A.P., Watts, R.D., and Redman, J.D., 1975, Detection of

thin layers by radio interferometry, Geophysics, 40, 299-308.

Saarenketo, T., 1992, Ground-penetrating radar applications in road design and construction in

Finnish Lapland, Geological Survey of Finland, Special Paper 15, 161-167.

64

Saarenketo, T., 1997, Using ground-penetrating radar and dielectric probe measurements in

pavement density quality control, Transportation Research Record, 1575, 34-41.

Saarenketo, T. and P. Roimela, 1998, Ground penetrating radar technique in asphalt pavement

density quality control, Proceedings of the 7th International Conference on Ground

Penetrating Radar, pp. 461-465.

Saarenketo, T. and T. Scullion, 1996, Laboratory and GPR tests to evaluate electrical and

mechanical properties of Texas and Finnish base course aggregates, Proceedings of the 6th

International Conference on Ground Penetrating Radar, pp. 477-482.

Saarenketo, T., and T. Scullion, 2000, Road evaluation with ground penetrating radar, J. Appl.

Geophys., 43, 119-138.

Sato, M., T. Ohkubo, and H. Niitsuma, 1995, Cross-polarization borehole radar measurements

with a slot antenna, J. Appl. Geophys., 33, 53-61.

Sato, M., and R. Thierbach, 1991, Analysis of a borehole radar in cross-hole mode, IEEE Trans.

Geoscience & Remote Sensing, 29, 899-904.

Schroder, C.T. and Scott, W. R., 2002, On the stability of the FDTD algorithm for elastic media

at a material interface, IEEE Trans. Geoscience & Remote Sensing, 40, 474--481.

Scullion, T., C. H. Lau, and T. Saarenketo, 1996, Performance specifications of ground

penetrating radar, Proceedings of the 6th International Conference on Ground Penetrating

Radar, pp. 341-346.

Sen, P.N., C. Scala, and M.H. Cohen, 1981, A self-similar model for sedimentary rocks with

application to the dielectric constant of fused glass beads, Geophysics, 46, 781-795.

Shang, J. Q., and Umana, J. A., 1999, Dielectric constant and relaxation time of asphalt pavement

materials, J. Infrastructure Systems, 5, 135-142.

Shaw, M., T. Molyneaux, S. Millard, J. Bungey, and M. Taylor, Automatic analysis of GPR

scans on concrete structures, Proceedings of the 7th International Conference on Ground

Penetrating Radar, pp. 449-453.

Shlager, K. L. et. al., 1994, Optimization of bow-tie antennas for pulse radiation, IEEE

Transactions on Antenna and Propagation, Vol. AP-39, pp 410-413.

65

Sheriff, R. E., and L. P. Geldart, 1995, Exploration Seismology, 592 pp, Cambridge University

Press.

Simone, A., U. Spagnolini, G. Gentili, and V. Rampa, 1998, Electromagnetic inversion and

interface tracking: system calibration and application, Proceedings of the 7th International

Conference on Ground Penetrating Radar, pp. 607-612.

Smith, S. S., 1995, Detecting pavement deterioration with subsurface interface radar, Sensors,

29-40.

Sokolnikoff, I. F., and R. M. Redheffer, 1966, Mathematics of physics and modern engineering,

McGraw-Hill Book Co., pp. 485-487.

Spagnolini, U., 1996, Joint approach of EM inversion and multi-layer detection/tracking for

pavement profiling, Proceedings of the 6th International Conference on Ground Penetrating

Radar, pp. 235-240.

Spagnolini, U., 1997, Permittivity measurements of multilayered media with monostatic pulse

radar, IEEE Trans. Geosci. and Remote Sensing, 35, 454-463.

Stickley, G., D. Noon, D., M. Chernniakov, and D. Longstaff, 1998, Current development status

of a gated stepped-frequency GPR, Proc. 7th Int. Conf. GPR, pp. 311-315.

Stutzman, W. L., and G. A. Thiele, 1981, Antenna theory and design, John Wiley & Sons, Inc.,

pp. 260-305.

Taflove, A., and Hagness, S.C., 2000, Computational Electromagnetics, Artech House, Boston,

Massachusetts.

Taner, M. T., Koehler, F. and Sheriff, R. E., 1979, Complex seismic trace analysis, Geophysics, 44,

1041-1063.

Telford, W. M., L. P. Geldart, and R. E. Sheriff, 1990, Applied Geophysics, 2nd edition,

Cambridge University Press, 770 pp., Cambridge, UK.

Topp, G. C., Davis, J. L., and Annan, A. P., 1980, Electromagnetic determination of soil water

content: measurements in coaxial transmission lines, Water Resource Research, 16, 574-582.

66

Treitel, S., and Robinson, E. A., 1966, The design of high-resolution digital filters, IEEE Trans.

Geosci. Electron. GE-4, 25--38.

Treitel, S., and L. R. Lines,, 1982, Linear inverse theory and deconvolution, Geophysics 47,

1153--1159.

Turner, G., 1994, Subsurface radar propagation deconvolution, Geophysics, 59, 215-223.

Turner, G., and A. F. Siggins, 1994, Constant Q attenuation of subsurface radar, Geophysics, 59,

1192-1200.

Ulriksen, C., 1982, Application of impulse radar to civil engineering, PhD Thesis, Department of

Engineering Geology, Lund University of Technology.

U.S. Department of Transportation, Federal Highway Administration, 1990, Our Nations

Highways, Selected Facts and Figures, Publ. No. FHWA-PL-90-024.

Wakita, Y., and Y. Yamaguchi, Estimation of the soil permittivity and conductivity by a GPR

antenna, Proceedings of the 6th International Conference on Ground Penetrating Radar, pp.

123-127, Sendai, Japan, Sept. 30-Oct. 3, 1996.

Warhus, J.P., J.M. Hernandez, S.D. Nelson, E.M. Johansson, and H. Lee, 1993, Ground

Penetrating, Imaging Radar for Bridge Inspection, Engineering Research, Development, and

Technology, Lawrence Livermore National Laboratory, Livermore, California, UCRL-

53868-92.

Weil, G. J., R. J. Graf, and L. M. Forister, 1994, Investigations of hazardous waste sites using

thermal IR and ground penetrating radar, Photo. Eng. Rem. Sen., 60, 990-1005.

Xia, J., Weis, T., Franseen, E., Miller, R. and Weis, T. V., 2004, Application of deterministic

deconvolution of ground-penetrating radar data in a study of carbonate strata, J. Appl.

Geophys. 56, 213--229.

Yee, K. S., 1966, Numerical solution of initial boundary problems involving Maxwell’s equations

in isotropic media, IEEE Trans. Antennas and Propagation, 14, 302-307.

Young, R.A., and J. Sun, 1996, 3D ground penetrating radar imaging of a shallow aquifer at Hill

Air Force Base, Utah, J. Environ. Eng. Geophys., 1, 97-108.

67

Zhou, C., L. Liu, and J.W. Lane, 2001, Nonlinear inversion of borehole-radar tomography data

to reconstruct velocity and attenuation distribution in earth materials, J. Appl. Geophys. 47,

271-284.

Zhou, H.-W., Al-Rufaii, K., and Byun, J., Retrieval of high-resolution components by

deterministic deconvolution: A field example, in Proc. 71st Int’l Exposit. AnnuMeeting Soc.

Expl. Geophysics, San Antonio, TX, September 9-14, 2001.

68

Appendix: A list of publications on GPR authored by the PI and his research group

Arcone, S. A., P. Peapples, and L. Liu, Propagation of a ground-penetrating radar (GPR) pulse in

a thin-surface waveguide, Geophysics, 68, 1922-1933, 2003.

Buursink, M., The application of ground penetrating radar (GPR) to the characterization of

fractures in crystalline bedrock, MS thesis, 163 pp. Department of Geology and Geophysics,

University of Connecticut, May, 1998.

Kochiss, C. Numerical and Physical Modeling of the Effects of Temperature Change on Ground

Penetrating Radar Signals, MS Thesis, Department of Geology and Geophysics, University of

Connecticut, 82 pp., December 2004.

Kochiss, C., and L. Liu, Numerical and physical modeling of the effects of temperature on

ground penetrating radar signals, EOS, Trans. American Geophysical Union, Joint Assembly

Supplement, abstract no. NS22A-05, 85(176), JA275, 2004.

Liu, L., The ground penetrating radar (GPR) survey at the future site of the Lebanon Historical

Society Museum, Connecticut, Report to the Archaeology Research Specialists, Inc. 19pp,

1996.

Liu, L., State-of-the-art rapid non-destructive pavement assessment: Ground penetrating radar

(GPR) in monostatic mode, Report to the Joint Highway Research Advisory Council,

Connecticut Department of Transportation, 1998.

Liu, L., Detection of possible old buried graves in the Pickett District Cemetery with ground

penetrating radar (GPR) in New Milford, Connecticut, Technical Report to Connecticut

Department of Transportation, 23 pp., 2004.

Liu, L., Fracture characterization using borehole radar: Numerical modeling, Water, Air, and

Soil Pollution, 2006 (in press).

Liu, L., and S. Arcone, Propagation of radar pulses from a horizontal dipole in variable dielectric

ground: A numerical approach, Subsurface Sensing Technologies and Applications: An

International Journal, 6, 5-24, 2005.

Liu, L. and S. A. Arcone, Numerical simulation of the wave-guide effect of the near-surface thin

layer on radar wave propagation, J. Environ. Engineer. Geophys. 8, 133—141, 2003.

69

Liu, L., and T. Guo, Determining the condition of hot mix asphalt specimens in dry, water-

saturated, and frozen conditions using GPR, Journal of Environmental & Engineering

Geophysics, 8, 143-149, 2003.

Liu, L., T. M. Habashy, and M. L. Oristaglio, Imaging the shape of a two-dimensional

cylindrical inclusion near a plane surface by electromagnetic wave scattering, in the

Proceedings of the Progress in Electromagnetics Research Symposium (PIERS ‘97),

Cambridge, Massachusetts, July 7-11, p. 40, 1997.

Liu, L. J. W. Lane, and Y. Quan, Radar attenuation tomography using the centroid frequency

down-shift method, Journal of Applied Geophysics, 40, 105-116, 1998.

Liu, L., and Y. Li, Identification of liquefaction and deformation features using ground

penetrating radar in the New Madrid seismic zone, USA, Journal of Applied Geophysics, 47,

199-215, 2001.

Liu, L. and M. Oristaglio, GPR signal analysis: instantaneous parameter estimation using the

wavelet transform, Proceedings of the 7th International Conference on Ground Penetrating

Radar (GPR ‘98), pp. 219-223, Lawrence, Kansas, May 27-30, 1998.

Liu, L., and Y. Quan, GPR wave attenuation tomography using the frequency shift method, in

the Proceedings of the 6th International Conference on Ground Penetrating Radar, pp. 335-

340, Sendai, Japan, Sept. 30-Oct. 3, 1996.

Liu, L., and Y. Quan, GPR attenuation tomography for detecting DNALs, in the Proceedings of

the Symposium on the Application of Geophysics to Engineering and Environmental

Problems (SAGEEP ‘97), pp. 241-251, Reno, Nevada, March 23-26, 1997.

Liu, L., C. Zhou, F. P. Haeni, and J. W. Lane, GPR attenuation tomography using frequency shift

method: Consideration of non-linear frequency dependence of attenuation, SEG Expanded

Abstracts, SEG 67th Annual Meeting, pp. 442-445, Dallas, Texas, November 2-7, 1997.

Liu, L., C. Zhou, and L. Xiao, Imaging the interior of the Nathan Hale Monument in Coventry,

Connecticut by GPR attenuation tomography, in the Proceedings of the 7th International

Conference on Ground Penetrating Radar (GPR ‘98), pp. 775-778, Lawrence, Kansas, May

27-30, 1998.

70

Steinen, R. P., L. Liu, and T. Guo, Anatomy of an ice-contact ridge (ice channel/esker?),

revealed by GPR and confirmed by excavation: Jordan Ridge, North Windham, CT,

Geological Society of America, Northeastern Section, the 36th Annual Meeting (March 12-

14, 2001), Burlington, Vermont.

Xiao, L., L. Liu, and V. Cormier, Three-dimensional finite-difference time-domain solution to

Maxwell’s equations using pseudo-spectral method, Proceedings of the 7th International

Conference on Ground Penetrating Radar (GPR ‘98), pp. 585-589, Lawrence, Kansas, May

27-30, 1998.

Zhou, C., and L. Liu, Multi-frequency radar diffraction tomography using quasi-linear

approximation, Proceedings of the 7th International Conference on Ground Penetrating

Radar (GPR ‘98), pp. 303-307, Lawrence, Kansas, May 27-30, 1998.

Zhou, C., and L. Liu, Diffraction tomography using the modified quasi-linear approximation,

IEEE Transactions on Geoscience and Remote Sensing, 38, 404-415, 2000.

Zhou, C., L. Liu, and J.W. Lane, Nonlinear inversion of borehole-radar tomography data to

reconstruct velocity and attenuation distribution in earth materials, Journal of Applied

Geophysics, 47, 271-284, 2001.