Subsurface Imaging with Ground Penetrating Radar Carey M. Rappaport CenSSIS Dept. Elect. and Comp. Engineering Northeastern University April 2011 © Carey Rappaport 2011 Propagation Characteristics in Real Soil •Concepts of dielectric constant, electrical conductivity •Velocity, attenuation, dispersion, reflection and refraction at interfaces •Moisture and density dependence •Nonmetallic target scattering in lossy media •Rough surface effects Wave and Helmholtz Equation: Lossy Media (Soil, Water, Tissue) The electric field for a wave traveling in linear, homogeneous, non-dispersive, and lossy medium is given by: 2E - E/ t - 2E/ t2 = 0 = conductivity (S/m), ranging from ~ 0 to 107 For time harmonic wave, the Helmholtz Equation remains: 2E + k2E=0 But the dispersion relation is modified by : k = [00 ’(1 - j tan)] = - j With Loss Tangent defined by: tan = / ( ’0) Electromagnetic Waves in Lossy Media k j ' 0 1 j Propagation (Wave) Number Slightly lossy medium ' 0 1 Very lossy medium ' 0 / ' 0 / 2 ' / c 0 / 2 ' ' 0 1 / 2 Velocity v / , 2 , Impedance = Slightly lossy medium Very lossy medium ' 0 '0 1 1 j ' 0 1 j '0 2 ' 0 skin depth 2 / (1 j) 2 1 0 j tan ' Propagation in Soil is Frequency Dependent Frequency f (1 MHz – 10 GHz) Dielectric constant ’ (1 – 25) Electrical conductivity (0.0001— 1) Wave Number, k (meters-1) E ( x, t ) e j[ ( f ) j ( f )] x e j 2ft 2 Ex 2 Ex 2 z Exact derivation of Wave Numbers in Lossy Media Starting from scalar Helmholtz Eqn. 2U 2 k U 0, U Ex , or E y , or Ez 2 z where the complex wave number is: 2 2 2 2 ' 1 j tan k 0 0 ' j 0 c Separate into real and imaginary components (k = – j ) 2 2 2 c 2 2 2 2 c 0 ' Solve for the quadratic equations for and 2 ' 1 1 c 2 ' 0 1 2 2 ' 1 1 c 2 ' 0 1 2 Decibel Scale The decibel (dB) is a logarithmic transformation of ratios of amplitudes or powers. A power ratio R corresponds to r = 10log10R (dB). An amplitude ratio R corresponds to 20log10R (dB). 1/10 power 1/2 power 10log10(1/10) 10log10(1/2) 1/10 amplitude 1/2 amplitude = -10 dB. = -3 dB. 20log10(1/10) = -20 dB. 20log10(1/2) = -6 dB. An intensity attenuation by a factor exp(-a) is equivalent to -4.3a dB . The decibel changes multiplication into addition When a wave is transmitted through a cascade of two media resulting in intensity reduction by factors R1 and R2, the overall reduction is a factor R = R1R2. The change in dB units is r = r1+ r2. If the rate of attenuation of a medium is a dB/m, a distance z (m), corresponds to attenuation of az (dB). Courtesy of B. Saleh, BU Logarithms Without Calculators • • • • • • • • Log 10 = 1.0 Log 1 = 0 Log 2 ~ 0.3 Log 5 = Log 10/2 = Log 10 – Log 2 = 0.7 Log 3 ~ Log 101/2 = ½ Log 10 = 0.5Log 4 = Log 22 = 2 Log 2 = 0.6 Log 6 = Log (2 X 3) = Log 2 + Log 3 = 0.8 Log 8 = Log 23 = 3 Log 2 = 0.9 Log10 e = 1/ Loge 10 = 1/2.302 Penetration Depth v. Frequency for Various Dielectric Materials Penetration Depth d10 (19%) (26%) = Distance for the power to drop by a factor of 10 (—10 dB) Wavelengths for Various Dielectric Materials Wavelength: = 2/ Fields for Different Soil Types f =2.5 GHz 1 Dry Sand 0 -1 10 5 10 15 20 YPG 0 -1 10 5 10 15 20 5 10 15 20 0 -1 10 A.P. Hill 0 -1 10 5 10 15 20 5 10 Distance (cm) 15 20 0 -1 Saturated Sand 0 Bosnian (Alicia); 25% moisture Exercise: Microwave Penetration in Soil Determine the loss in dB for a wave at 300 GHz penetrating 1.0 mm into uniform soil and then reflecting back out for a) Yuma and b) AP Hill Soil Hint: Extrapolate the loss curves from previous slide. Extrapolated Penetration Depths at 300 GHz (Terahertz range) Return signal power (in dB) from a radar source incident on a metallic target buried a depth D in lossy soil: -20 D/d d=Penetration Depth Radar Return (dB) (D = 1 mm) Yuma PG 55.7 cm -0.036 Dry Sand 4.57 cm -0.44 Wet Sand 0.31 cm -6.5 Bosnian soil 54.3 m -368 A P Hill 40.0 m -500 Soil Type Wire on Flat Ground: Bosnian Soil 26% Moisture E-field parallel to wire H-field parallel to wire Difference Wire on Rough Ground: Bosnian Soil 26% Moisture E-field parallel to wire (Ez) (Ez) no wire Modeling Soil Media for Electromagnetic Wave Propagation • Type of models • Simulated wave response Summary of Dielectric Mixing Models Source: Kansas Geological Survey, 2001 Category Method Types Advantages Disadvantages References Phenomenological Relate frequency dependent behavior to characteristic relaxation times. Cole-Cole; Debye, Lorentz - Component properties/geometry relationships unnecessary - Dependent on frequencyspecific parameters. Powers, 1997; Ulaby 1986; Wang, 1980. Volumetric Relate bulk dielectric properties of a mixture to the dielectric properties of its constituents. ComplexRefractive Index (CRI); Arithmetic average; Harmonic average; LicheteneckerRother; - Volumetric data relatively easy to obtain. - Do not account for microgeometry of components, -Do not account for electrochemical interaction between components. Alharthi 1987; Birchak 1974; Knoll, 1996; Lange, 1983; Lichtenecker 1931; Roth 1990; Wharton 1980. Empirical and Semiempirical Mathematical relationship between dielectric and other measurable properties. Logarithmic; Polynomial. - Easy to develop quantitative relationships, -Able to handle complex materials in models. - No physical justification for the relationship, -Valid only for the specific data used to develop the relation may not be applicable to other data sets. Dobson 1985; Olhoeft 1975; Topp 1980; Wang 1980. Effective medium Compute dielectric properties by successive substitutions. Bruggeman-HanaiSen (BHS) - Accurate for known geometries. - Cumbersome to implement, - Must choose number of inputs, initial material, and order and shape of replacement material. Sen 1981; Ulaby 1986. Fourier Transform • Short pulse in time transforms into broadband frequency signal Y( f ) y(t ) exp( j 2 f t ) dt y (t ) Y ( f ) exp( j 2 f t ) df t • Long pulse in time transforms into narrow frequency signal 1/t t f Temporal Dispersion • Pulses in time are composed of many frequencies (Fourier relationship) • Most real material has frequency-dependent dielectric parameters • If material has constant loss, it is strongly dispersive 0 ' j / 0 • Each frequency component travels at a different velocity and with a different decay rate • Amplitude of each frequency component lessens by a different amount with distance • Because of dispersion, multiplication in frequency domain becomes temporal convolution in the time domain D( ) ( ) E ( ) D(t ) (t ) * E (t ) Dispersion of a Pulse 3 Fourier Components of Pulse at t0 Same components at t>t0 • Each component travels at a different velocity (dispersion) • Amplitude of component lessens in time (loss) Modeling Dispersion for Easy Transformation to Time Domain Standard (2nd Order) Debye Model: simple form for complex permittivity, easily transformed to time domain differential equation N 0 ' 0 v p 1 For 2 and 0 A2 / 2 0 ' 0 Ap 1 j p N=2 A1 1 j 1 j 0 Lorentz Model: 2nd order when N = 1 N 0 ' 0 p 1 Ap 2 2 2 j p 0 [Cole-Cole Model is more accurate, not easily converted to time domain] 0 ' 0 's ' 1 j Conversion of Dispersion Models to Time Domain Replace by D/E and multiply through by denominator D j 0 1 j1 E j 0 1 j1 0 ' 0 j 0 A1 1 j1 Convert to time domain with Debye j t 2 D D 2E E 1 2 0 ' 1 2 0 ' 0 A1 1 E t t t 0 t 0 D 2 20 2 j p E 2 20 2 j p 0 ' 0 A1 Convert to time domain with j t 2 2D D E E 2 2 2 D ' 2 ' ' A1 E p 0 p 0 0 2 2 0 0 t t t Lorentz t Modeling Dispersion for Easy Transformation to Time Domain Z-Transform model keeps real permittivity constant, and matches conductivity to measured values in terms of Z-1 [4 Zero Model] ’ = Constant, Z 1/ 2 b0 b1 Z 1 b2 Z 2 b3 Z 3 1 a1 Z 1 ,Z = e jt Since Z-1 transforms to unit time delay, application to FDTD is simple D( Z ) Av E ( Z ) J (Z ) (Z ) E (Z ) D(t ) Av E (t ) t 3t J (t ) a1 J (t ) 2 2 b0 E (t ) b1 E (t t ) b2 E (t 2t ) b3 E (t 3t ) Dielectric Constant and Conductivity for Puerto Rican Clay Loam (1.2 g/cc) -1 9 10 Rappaport Debye data 8 -2 7 5% 10 10% 6 2.5% ’ 10% 5% -3 5 10 4 3 2.5% -4 0 500 Frequency (MHz) 1000 10 0 500 Frequency (MHz) 1000 Real and Imaginary Wave Number for Puerto Rican Clay Loam (1.2g/cc) 60 0 Rappaport Debye data 50 10% -0.5 2.5% 5% - (1/m) -1 (1/m) 40 5% -1.5 30 2.5% 20 -2 10 -2.5 0 7 7.5 8 8.5 9 Log Frequency 9.5 -3 7 10% 7.5 8 8.5 9 Log Frequency 9.5 Wave Propagation Variation as a Function of Clay Loam Moisture Rough Surface Test Geometry 40 Transmitter 30 Receiver Depth (cm) 20 10 0 -10 -20 Non-Metallic Mine -30 -40 60 80 100 120 140 Transverse Position (cm) 160 180 Non-Metallic Mine Scattered Field 10 cm Deep - Smooth Surface 0.4 ------- ++++++ oooooo xxxxxx Relative Amplitude 0.3 Air Dry Sand Non-Dispersive Loam 20% moisture Dispersive Loam 20% moisture 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 0 1000 2000 3000 4000 5000 6000 Time (ps) 7000 8000 9000 10000 Non-Metallic Mine Scattered Field (about 10 cm burial) - Rough Surface 0.4 ------- ++++++ oooooo xxxxxx Relative Amplitude 0.3 Air Dry Sand Non-Dispersive Loam 20% moisture Dispersive Loam 20% moisture 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 0 1000 2000 3000 4000 5000 6000 Time (ps) 7000 8000 9000 10000 Relative Amplitude Non-Metallic Mine Scattered Field 10 cm depth a) Flat Surface, b) Rough Surface 0.1 0.05 0 Relative Amplitude -0.05 -0.1 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 7000 8000 9000 10000 0.06 Non-Dispersive Loam 20% moisture Dispersive Loam 20% moisture 0.04 0.02 0 -0.02 -0.04 0 1000 2000 3000 4000 5000 6000 Time (ps) Shape Determination of Buried Non-Metallic Targets, Multiple Single-Frequency Observations Sandy soil: s = 2.5, s = 0.01 Target: m = 2.9, m = 0.004 Circular Target Air Air 20 cm d 11.28 cm Soil Square Target 20 cm d 10 cm Soil 10 cm 60 cm 80 cm 60 cm 80 cm Different Buried Test Target Shapes Height (cm) Square Circle Diamond 0 0 0 -5 -5 -5 -10 -10 -10 -15 -15 -15 -20 -10 -20 -10 -20 -10 -5 0 5 10 Star 0 -5 -10 -10 -15 -15 -5 0 5 10 -20 -10 0 5 10 5 10 Blob 0 -5 -20 -10 -5 -5 0 Horizontal Position (cm) -5 0 5 10 Scattered Field - Real Part Height (cm) Square Circle Diamond 20 20 20 0 0 0 -20 -20 -20 -40 -40 -40 -60 -40 -20 0 Star 20 40 -60 -40 -20 0 Blob 20 40 -60 -40 0.04 0 0 -20 -20 -40 -40 0.02 0 -0.02 -0.04 -60 -40 -20 0 20 40 -60 -40 -20 0 0 20 40 0.06 20 20 -20 20 40 Horizontal Position (cm) -0.06 500 MHz, depth = 5 cm Scattered Field - Real Part Square Height (cm) 20 Circle 20 20 0 0 0 -20 -20 -20 -40 -40 -40 -60 -40 -20 0 20 40 Star 20 -60 -40 -20 0 20 Diamond 40 -60 -40 -20 -20 0 -40 -40 -0.1 0 20 40 -60 -40 20 40 0.2 0 -20 0 Blob 20 0 -60 -40 -20 0.1 -20 0 20 40 Horizontal Position (cm) -0.2 1000 MHz, depth = 5 cm Surface Field - Magnitude 500 MHz, depth = 5cm 0.04 1000 MHz, depth = 5cm 0.05 square circle 0.04 0.035 diamond Intensity star blob 0.03 square 0.03 circle 0.02 diamond 0.025 0.01 star blob 0.02 -40 0 -20 0 20 40 -40 -20 Horizontal Position (cm) 0 20 40 Scattered Field - Aspect Ratio Dependence Height (cm) 20 0 Circle, r = 5.64 cm 10 x 10 cm 20 -20 0 -20 -40 -40 -60 -40 -20 0 20 40 -60 -40 -20 0 20 7.5 x 13.3 cm 20 40 5 x 20 cm 20 0.04 0.02 0 -0.02 -0.04 -0.06 -0.08 20 0 0 0 -20 -20 -20 -40 -40 -40 -60 -40 -20 0 20 40 -60 -40 -20 0 20 20 13.3 x 7.5 cm 20 x 5 cm 20 40 0 0 0 -20 -20 -20 -40 -40 -40 -60 -40 -20 0 20 40 = 2.5, = 0.01 freq = 500 MHz depth = 5 cm 2.5 x 40 cm -60 -40 -20 0 20 Sandy Soil 20 40 40 x 2.5 cm -60 -60 -40 -20 0 20 40 -40 -20 0 Horizontal Position (cm) 20 40 Distinguishing Shapes of 3D Buried Objects under Rough Surfaces: Geometry Point Source 10 cm Rough Surface 5 cm Soil Mine 4 cm 10 cm Total Ex Field from an x-Directed Point Source, with a Buried Non-Metallic Square Target Total Ex Field from x-Directed Point Source, with a Buried Non-Metallic Square Target (back view) Comparison of Total Ex Field for Buried NonMetallic Square and Circular Targets Comparison of Scattered Ex Field for Buried Non-Metallic Square and Circular Targets Soil Packing Affects Greatly Scattering: 3D FDFD with Short Cylindrical Target Relative Height 30 TNT in 26% moist Bosnian soil at 960 MHz Surface Scattering Clutter Increases with Frequency. Example: 4 GPR Freq., PRCL 10% moisture, 1.4 g/cc -5 Depth (cm) 0 Air 5 10 15 Soil Non-Metallic Target 20 25 30 35 -20 -15 -10 -5 0 5 10 15 Transverse Position (cm) 20 Display Format for each of Four Frequencies Mine scattered field: smooth surface Scattered field: rough surface with mine Scattered field: rough surface only Mine scattered field: rough surface 480 MHz Depth (cm) 0.1 10 0 20 -0.1 30 -0.2 -20 -10 0 10 20 Scattered field: rough surface with mine 0.2 0 0.1 10 0 20 -0.1 30 -0.2 -20 -10 0 10 20 Transverse Position (cm) Transverse Position (cm) Scattered field: rough surface only Mine scattered field: rough surface 10 0 20 -0.1 30 -0.2 -20 -10 0 10 20 Transverse Position (cm) 0 Depth (cm) Depth (cm) 0.1 0.2 0.1 10 0 20 -0.1 30 -0.2 -20 -10 0 10 20 Transverse Position (cm) Amplitude Relative to Incident 0 Amplitude Relative to Incident 0.2 Amplitude Relative to Incident 0 Amplitude Relative to Incident 0.2 Depth (cm) Mine scattered field: smooth surface 960 MHz Scattered field: rough surface with mine 0.1 10 0 20 -0.1 30 -20 -10 0 10 20 -0.2 0 Depth (cm) Depth (cm) 0 0.2 0.1 10 0 20 -0.1 30 -20 Transverse Position (cm) 10 0 20 -0.1 30 10 20 Transverse Position (cm) 20 -0.2 -0.2 Mine scattered field: rough surface 0.2 0 Depth (cm) Depth (cm) 0.1 0 10 0.1 10 0 20 -0.1 30 -20 -10 0 10 20 Transverse Position (cm) -0.2 Amplitude Relative to Incident 0 Amplitude Relative to Incident 0.2 -10 0 Transverse Position (cm) Scattered field: rough surface only -20 -10 Amplitude Relative to Incident 0.2 Amplitude Relative to Incident Mine scattered field: smooth surface 1920 MHz Mine scattered field: smooth surface Scattered field: rough surface with mine 10 0 20 0 Depth (cm) Depth (cm) 0.05 0.1 0.05 10 0 20 -0.05 30 -10 0 10 20 -0.1 30 -20 Transverse Position (cm) -10 0 10 20 0.05 10 0 20 Mine scattered field: rough surface 0.1 0 0.05 10 0 20 -0.05 30 -10 0 10 20 Transverse Position (cm) -0.1 Amplitude Relative to Incident 0 Amplitude Relative to Incident 0.1 -20 -0.1 Transverse Position (cm) Scattered field: rough surface only Depth (cm) -0.05 Depth (cm) -20 Amplitude Relative to Incident 0 Amplitude Relative to Incident 0.1 -0.05 30 -20 -10 0 10 20 Transverse Position (cm) -0.1 3840 MHz Mine scattered field: smooth surfaceScattered field: rough surface with mine 0 20 30 -20 -10 0 10 20 0 Depth (cm) Depth (cm) 10 0.05 10 0 20 30 -0.05 -20 Transverse Position (cm) -10 0 10 20 Mine scattered field: rough surface 0 -20 -10 0 10 20 Transverse Position (cm) -0.05 0 Depth (cm) Depth (cm) 10 0.05 10 0 20 30 -20 -10 0 10 20 Transverse Position (cm) Amplitude Relative to Incident 0 Amplitude Relative to Incident 0.05 30 -0.05 Transverse Position (cm) Scattered field: rough surface only 20 Amplitude Relative to Incident 0 Amplitude Relative to Incident 0.05 -0.05 Short Pulse GPR Interaction with Rough, Dispersive Ground / Mine Air Soil PMN-1A Non-Metallic AP Mine Geometry From MineFacts, version 1.2, National Ground Intelligence Center Snapshot of Total Time Domain EField (with Target) Air Soil Snapshot of Background Time Domain E-Field Air Soil Snapshot of Scattered Time Domain E-Field (Mine Only) Air Mine Soil Effect of Rough Ground of Bistatic GPR Signals Height (cm) Transmitter 12 Receiver - 0 -12 -48 -24 0 24 Signal Amplitude Rough Ground (cm) 0.5 0.5 Mean 48 Height variation h= 6cm Correlation distance between surface peaks lc= 15cm 0 0.0 -0.5 -0.5 0 0.0 100 100 200 200 Time Step (t = 20ps) 300 300 Rough Ground Clutter Signal Characterization • Signals from rough ground vary considerably – Pulse shape depends on roughness and TR position – Peak depends on particular TR position – Overall amplitude varies • Monte Carlo simulation can model following relevant features – 2D FDTD model – Real measured impulse GPR excitation and dispersive soil – 500 different rough surface realizations Monte Carlo Analysis • Run many simulations • Vary each run – Change geometry – Change signal • Compute statistics – Mean values – Standard deviations • Conclude “typical” behavior – Determine likelihood of given test • Set threshold and count number of occurrences of detection or false alarm --> ROC curve Computational Geometry Transmitter Receiver 24.5 cm Z = 28cm Z=0 Z = depth mine soil L = 294 cm 2 0 -2 -4 Relative Amplitude 4 Impulse Ground Penetrating Radar Specifications 0 100 200 Time Step (t=20ps) 300 Original Signal Averages Obscure Mine Signal No Mine 500 computed signals Mine Physics-based Signal Processing flowchart Raw Signals Cross-correlate with reference Shift and scale raw signals and take average Shifting Scaling Compute different velocity in soil, shift to line up the target feature Subtract shifted and scaled average from each raw signal Ground Clutter Signal Removal No Mine Mine Realigning Signals to Presumed Mine Position No Mine Mine Average Mine Scattered Signals h=3cm lc=10cm lc=3cm h=1cm ROC Curves for Mismatched Target Depths 8.5 9.8 h= 1cm lc= 10cm 4.8 3.6 6.1 2.4 Trail depth=8.5cm test depth= 2.4cm 3.6cm 4.8cm 6.1cm 8.5cm 9.8cm Soil Moisture Change with Wetting Water movement in a vertical column of a medium is described by the advectiondispersion equation in the z-direction, as: d d d d ( D ( ) ) K ( ) dt dz dz dz Where: = moisture content z = depth [L] D = dispersion coefficient of water [L/t2] K = hydraulic conductivity [ L/t] t = time [t] Time Response Due to Saturating Soil Surface Moisture Profile Ksat = 0.2 cm/min Moisture Content (%) 0.25 0.1 Minute 1 Minute 2 Minutes 3 Minutes 4 Minutes 0.20 0.15 0.10 0.05 0.00 0 10 20 30 40 Depth into Soil (cm) 50 60 Rough Surface with Buried Non Metallic Mine and Point Source Geometry -100 -80 -60 Air -40 Source -20 Ground Surface 0 20 40 60 -100 Non-Metallic Target Soil with Varying Moisture Content -80 -60 -40 -20 0 20 40 Testing geometry 60 80 100 Planar Ground Surface, 5% Uniform Moisture Planar Ground Surface, 20% Uniform Moisture Planar Ground Surface, 5 - 20% Moisture Profile Rough Ground Surface, 5% Uniform Moisture Rough Ground Surface, 20% Uniform Moisture Rough Ground Surface, 5 - 20% Moisture Profile Summary • Realistic soil media complicates the sensing of subsurface objects – Loss affects penetration depth and makes surface clutter more dominant – Rough interfaces produce additive uncertain clutter and distort transmitted signals – Moisture variations cause huge propagation differences • Small contrast differences makes detection/ imaging more challenging • Shapes of underground dielectric object are hard to distinguish • Multistatic wideband GPR can provide much more information than monostatic