<%BANNER%>

Mixed Topographic and Shallow Bathymetric Mapping via Multi-Stop Single Photoelectron Laser Ranging

Permanent Link: http://ufdc.ufl.edu/UFE0041426/00001

Material Information

Title: Mixed Topographic and Shallow Bathymetric Mapping via Multi-Stop Single Photoelectron Laser Ranging
Physical Description: 1 online resource (191 p.)
Language: english
Creator: Shrestha, Kristofer
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: lidar, mapping, photonics, ranging
Civil and Coastal Engineering -- Dissertations, Academic -- UF
Genre: Civil Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Airborne laser swath mapping (ALSM) systems collect thousands of range measurements per second, enabling efficient high-resolution surface mapping. Low signal-to-noise ratio (LSNR) detection techniques allow for implementation of light detection and ranging (LIDAR) instrumentation on platforms with prohibitive power, size, and weight restrictions. The viability of single-wavelength photon counting techniques applied to mixed target detection and shallow-water bathymetry has not been previously investigated in detail. The system design process for constructing a LSNR LIDAR system is described, with emphasis on component selection as related to theoretical performance. Cost-effective microchip lasers are now commercially available that produce sub-nanosecond pulse widths and stable peak energies ( > 1 microjoule at 10 kHz). In order to detect backscattered laser radiation at low power levels, detector elements must be capable of single photon sensitivity; a high gain (10^5), fast response (sub-250 ps rise time) detector is therefore central to photon counting operation. Micro-channel plate photomultiplier tubes (MCP-PMTs) are capable of attaining these specifications with simultaneous two-dimensional detection over an array. The University of Florida?s prototype LSNR LIDAR sensor -- the Coastal Area Tactical-mapping System (CATS) -- was constructed with these major components in mind. The instrument is designed to operate in a fixed-wing aircraft flying 600 m above ground level, producing 3 microjoule output pulses at 8 kHz to allow for 20 cm spatial resolution by scanning a 10x10 array of laser beamlets. Ground-based field testing results yield 10 cm range accuracy and small-scale feature identification at variable scan settings. Experimental data is presented establishing the ability to record multiple returns from the same laser pulse. Evaluation of system dead time shows that the minimum separation distance required to discriminate multiple scattering objects is 15 cm, indicating the possibility of seamless topographic / bathymetric mapping with no gap at the waterline. Data collected during tests display vertical structure in shallow water areas based on fixed threshold crossings at very low signal levels. A major concern for the binary detection strategy is dark / atmospheric / crosstalk noise producing false-positive measurements. Potential causes of image ghosting are described, and data collection / processing solutions based on local density estimation are explored. LIDAR depth estimates from airborne profiles are compared to on-site measurements, and near-shore submerged feature identification is presented. Analysis of CATS data confirms simulation estimates of decimeter-scale spatial resolution for hybrid topographic and shallow bathymetric mapping. Dark noise production and atmospheric scattering in daylight operation at maximum signal amplification does not significantly impede feature identification procedures. In order to combat the high variability in dynamic range, modifications to the detector structure may be necessary to reduce anomalous effects.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Kristofer Shrestha.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Hatfield, Kirk.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0041426:00001

Permanent Link: http://ufdc.ufl.edu/UFE0041426/00001

Material Information

Title: Mixed Topographic and Shallow Bathymetric Mapping via Multi-Stop Single Photoelectron Laser Ranging
Physical Description: 1 online resource (191 p.)
Language: english
Creator: Shrestha, Kristofer
Publisher: University of Florida
Place of Publication: Gainesville, Fla.
Publication Date: 2010

Subjects

Subjects / Keywords: lidar, mapping, photonics, ranging
Civil and Coastal Engineering -- Dissertations, Academic -- UF
Genre: Civil Engineering thesis, Ph.D.
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
born-digital   ( sobekcm )
Electronic Thesis or Dissertation

Notes

Abstract: Airborne laser swath mapping (ALSM) systems collect thousands of range measurements per second, enabling efficient high-resolution surface mapping. Low signal-to-noise ratio (LSNR) detection techniques allow for implementation of light detection and ranging (LIDAR) instrumentation on platforms with prohibitive power, size, and weight restrictions. The viability of single-wavelength photon counting techniques applied to mixed target detection and shallow-water bathymetry has not been previously investigated in detail. The system design process for constructing a LSNR LIDAR system is described, with emphasis on component selection as related to theoretical performance. Cost-effective microchip lasers are now commercially available that produce sub-nanosecond pulse widths and stable peak energies ( > 1 microjoule at 10 kHz). In order to detect backscattered laser radiation at low power levels, detector elements must be capable of single photon sensitivity; a high gain (10^5), fast response (sub-250 ps rise time) detector is therefore central to photon counting operation. Micro-channel plate photomultiplier tubes (MCP-PMTs) are capable of attaining these specifications with simultaneous two-dimensional detection over an array. The University of Florida?s prototype LSNR LIDAR sensor -- the Coastal Area Tactical-mapping System (CATS) -- was constructed with these major components in mind. The instrument is designed to operate in a fixed-wing aircraft flying 600 m above ground level, producing 3 microjoule output pulses at 8 kHz to allow for 20 cm spatial resolution by scanning a 10x10 array of laser beamlets. Ground-based field testing results yield 10 cm range accuracy and small-scale feature identification at variable scan settings. Experimental data is presented establishing the ability to record multiple returns from the same laser pulse. Evaluation of system dead time shows that the minimum separation distance required to discriminate multiple scattering objects is 15 cm, indicating the possibility of seamless topographic / bathymetric mapping with no gap at the waterline. Data collected during tests display vertical structure in shallow water areas based on fixed threshold crossings at very low signal levels. A major concern for the binary detection strategy is dark / atmospheric / crosstalk noise producing false-positive measurements. Potential causes of image ghosting are described, and data collection / processing solutions based on local density estimation are explored. LIDAR depth estimates from airborne profiles are compared to on-site measurements, and near-shore submerged feature identification is presented. Analysis of CATS data confirms simulation estimates of decimeter-scale spatial resolution for hybrid topographic and shallow bathymetric mapping. Dark noise production and atmospheric scattering in daylight operation at maximum signal amplification does not significantly impede feature identification procedures. In order to combat the high variability in dynamic range, modifications to the detector structure may be necessary to reduce anomalous effects.
General Note: In the series University of Florida Digital Collections.
General Note: Includes vita.
Bibliography: Includes bibliographical references.
Source of Description: Description based on online resource; title from PDF title page.
Source of Description: This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Statement of Responsibility: by Kristofer Shrestha.
Thesis: Thesis (Ph.D.)--University of Florida, 2010.
Local: Adviser: Hatfield, Kirk.

Record Information

Source Institution: UFRGP
Rights Management: Applicable rights reserved.
Classification: lcc - LD1780 2010
System ID: UFE0041426:00001


This item has the following downloads:


Full Text





MIXED TOPOGRAPHIC AND SHALLOW BATHYMETRIC MAPPING VIA
MULTI-STOP SINGLE PHOTOELECTRON LASER RANGING















By

KRISTOFER Y. SHRESTHA


A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL
OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE DEGREE OF
DOCTOR OF PHILOSOPHY

UNIVERSITY OF FLORIDA

2010

































2010 Kristofer Y. Shrestha

































To my mother, for instilling in me a determination bordering on stubbornness









ACKNOWLEDGMENTS

This research was made possible only with the valued advice and support of my

family, friends, fellow students and professors. I would like to thank Dr. Clint Slatton for

directing me into this line of work and for exhibiting continual patience in the face of very

difficult obstacles. Dr. Slatton continues to serve as my inspiration to work even harder

in times of tough circumstance. I would also like to thank Dr. Bill Carter for pushing me

to find the alternative solution in times where I genuinely believed success to be an

impossibility. His expertise and guidance in the field of low-signal laser ranging have

been critical to this work.

Next, I would like to express my appreciation for the support of my fellow

Geosensing students. This small piece of text is probably all that will ever document the

selfless field work of these students on my behalf, so I need to take this opportunity to

thank Tristan Cossio, Juan Fernandez, Pang-Wei Liu, Kitiipat Kampa, Pravesh Kumari,

Tolee Matamas, Mike Starek, Abhinav Singhania, and Bidiya Yadav. I will not forget the

freezing cold nights in Citra, the struggles at Lake Wauburg, or the countless other

times that they aided me in CATS testing. I would also like to thank the staff at the CE

Structures Lab / Plant Science Research for use of the their facilities, William Gavert /

Jacob Hwang for their support in diagnosing electronics issues, and Sidney Schofield

for his invaluable help in preparing the system for airborne operation.

Finally, I would like to express my deep gratitude to my father, Ramesh, and my

mother, Chew-Ying, for providing me with opportunities that few children get to have,

and to my sister, Veena, for all her love. Finally, I would like to thank Cynthia Gillikin for

her patience and support through this long process.









TABLE OF CONTENTS

page

A C KNOW LEDG M ENTS .......... ..................... ....... .. ......................................... 4

LIST O F TA B LES .......... ..... ..... .................. ............................................. ...... .. 9

LIST OF FIGURES.................................. ......... 10

A BST RA C T ............... ... ..... ......................................................... ...... 15

CHAPTER

1 INTR O D U CT IO N ............................................................................................. 17

Introduction to Rem ote Sensing................................................. .................... 17
Photogram m etry and LIDA R ...................................................................... 17
A dva nta ge s of L ID A R ........................................................................ 18
Applications of LIDAR ................................. ........................ .. ....... .. 18
Im proving Im age Fidelity ................................. .......................... ...... .. 19
LIDAR Point Clouds ............... ..... ..... ..... ..................... ..... 19
Existing Methods for Improving Feature Detection ........................................... 19
Existing Methods for Improving Data Quality...................... ........... 20
The LSNR Mapping Approach .................... ........ ............. .. ........... 21
Problems with High Energy / High Pulse Rate Designs ............... ............ 21
Motivations for a LSNR LIDAR System .................................................... 22
R e se a rc h A im s ....... ... ............... ....... ........................................... 2 4

2 LS N R C O M PO N E N T S ........................................................................ ......... 25

Commercial Sector Activities and Scientific Impact ..................... ............ 25
C O TS M market and C costs ......... ............. ................................ ............... 25
LSNR Concept Systems.............................. .................... 25
Overview of Lasers ............................ ......... .. ................ 27
Quantum Dynamics Fundamentals ........ .................... ..................... 27
C oherence ................................................................ 29
Three- and Four-Level Laser Models ...................... ........... ...... 30
O optical Resonators ........................ ... ......... ............... ............... 31
Q -S w itching .................. ............ .... .................... ............ 32
Microchip Laser Advancements ............................ ......... 34
Beam Quality Measures .... ........................... ... ........... ...... 36
Effects of Laser Parameters .... .... ......................................... ................. 36
Operating W avelength ............... ....... ............. .......... ............... 37
L a se r P u lse W id th ......... ......... ...................................................... 3 7
Pulse Energy ............... .................................... ............... 38
Selecting a LSNR Photodetector ...... .. .................................... .. ............... 40
Avalanche Photodiode ......... ...... ...................... 40









P hotom ultiplier T ube A dvantages............................................ .. .................. 41
C om binations ......... .. ............ .............. .... ................. ........... 42
Photodetector Im provem ents.............................. ........................... ............... 43
Photon Counting.......................................................... 43
Sem conductor Principles ........................................ ..... .......... .......... 44
Micro-channel Plate Photomultiplier Tube Concept.................... ........... 46
M CP-PMT Gain .............................................................. .......... ......... 46
Detection Efficiency and Pulse Height Distribution ........................................ 48
Variation in PMT Gain and Output Instability.................................... 50

3 LSNR PROTO TY PE DESIG N .......................................... ................................ 60

Photon Counting System Prototype: CATS .................................................... 60
G general Scanning P principles .................................... ...................................... 60
System Proposal ....................................................... ............ ...... 60
O p tic s ....................................................... ...... ...... ...... 6 1
M icrochip Laser ..................................................................... ....... 61
B e a m E x p a n d e r ................................................................. ................ 6 2
R isley P rism s .................................................... 6 3
Range Gate ...................................................................... ......... 64
Spatial and Spectral Filters ............................. ................. 64
Dove Prism ................................. ............ ....... ...... .............. 65
Telephoto Lens and Mechanical Shutter ...................................................... 66
M C P-PM T ...................................................................... ....... ........ 66
Multi-stop Event Timer.................................................... 67
Assembly and Footprint Dimensions ............. ............................... ............... 68

4 LSNR PERFORMANCE M ETRICS .................................... ........... .................. 75

Laser Ranging Concepts and Implementation ................................................... 75
Range Calculation Fundamentals ........................................... ............... 75
R ange D iscretization ................................ ......... ... .............. ........... 76
D ata P processing F ie lds.................................. ................. ............. 77
Ground-based Static Testing ...................... ......... ..................... 78
Experim ental Setup ............. .... ............. ....................... ...... 78
Range Accuracy ......................................................... ............... 79
M inim um S eparation D instance ................................................ .... ............... 79
Crosstalk Evaluation............................ ............... 82
Water Penetration Depth ..................... ......... ..................... 86
M ulti- vs. Single-Photon Sensitivity ........................................... ................. 88
Ground-based Scan Testing .............................. ................. ......................... 90
Horizontal Scan Building Features ...................... ....................... 90
Rotating Line Scan Building Features ............................. ............... 90
Rotating Line Scan Building Features (Partial Obscuration)....................... 91

5 LSNR AIRBORNE LASER BATHYMETRY ...................................................... 108



6









A LB B ackgro und ......... ......................................................... ..... ..... .... 108
Multiple Information Channels in Laser Ranging .................................... 108
Challenges Facing Single-Channel Detection .............. .......... ................. 110
Pulse W idth and Output Energy Considerations ............ .......................... 111
Contrast to W aveform Digitization ...................... .......... .................... 112
Experimental Setup ... .......................................... ............... 113
Small Structure Identification ........... ............. ... ............. ........... 114
Vertical Canopy Structure Identification......... ............. ...... ..... ............ 114
Optimizing System Performance ................................ 115
Data Artifacts ....................................................................................... ................... 116
PMT Afterpulsing ........................................ ............... 119
Impedance Matching ................. .............................. 120
Gain Block Overdrive ...................... ............... .... 123
Point Cloud Filtering Based on Density Estimates ............ ................ ........... 124
Fresh W ater Bottom Mapping .................. ....... ................. ........ ......... 125
Salt W ater Bottom M apping ............ ......................... .... .................. 126
Mapping Statistics and Discussion ................ ................. .............. 127

6 CONCLUSIONS AND FUTURE WORK .......... ........... ...................... 145

Scope of Research ........... ........ ........... ......................... 145
Design and Testing Overview ................ .... .......................... ........ ......... 145
F utu re W o rk ............ .......... ................. ....... ....................................... 15 3
C conclusions .............. ......... ............ ............ ........... ........... 155

APPENDIX

A DATA PROCESSING ELECTRONICS....................... ....... ................... 160

Ranging Receiver .. ................................. .... .. .............................. 160
Event Tim er C ircuitry ........................ ...................... .. ............................. 162

B SYSTEM TROUBLESHOOTING ......... ............ .................................. 169

Range Offset Error........................................... 169
Oscillator Variation........................................ 170
Scan Synchronization .............. ...... ................. ............... 171
Image Generation: Early Steps ..... ..................... ........ ........... 171
Point Cloud Results from a Building: Problems ............ ........................... 173
Shot-By-Shot Analysis of Early Data ....... .... ............ ...................... 173
D ata T im e T a g g ing ......... ............................ .................................... 17 5
Tim etag Theory Re-Evaluation............... ................. ................................... 175
Problems with Previous Timetag Implementation.............. ............... 176
Restructuring of Timetag Generation ....... .... .............. ............... .... .... 176
Testing of Tim etag Restructuring ......... ................................. .................. 177









LIST OF REFERENCES .................. .... ......... .............. 185

BIOGRAPHICAL SKETCH ............... ..... ....... .......... .......... 191









LIST OF TABLES


Table page

2-1 Sample system parameters for use in LIDAR link equation.............................. 55

4-1 Ranging trials for shots reporting 3+ returns................................................ 96

4-2 Effect of range bias offset on measurement error (in cm)................................ 96

4-3 Channel orientations for crosstalk calculations.................... ................. 98

5-1 Time Delay Reflectometry test data (courtesy of Fibertek Inc.)........................ 133

5-2 Comparison of LIDAR depth estimates and measurements............................. 139

5-3 Comparison of LIDAR platform distance estimates and measurements........... 142









LIST OF FIGURES


Figure page

2-1 Point clouds resulting from data collected by the DARPA Jigsaw. ..................... 51

2-2 Point clouds resulting from Jigsaw sensor data after filtering algorithms are
a p p lie d ............ .... ............................................... ............ .... ......... 5 1

2-3 Diagram of energy level transitions in four-level laser. ................. .............. 52

2-4 Maximum pulse repetition rate versus pulsewidth in SPMCL PQS microchip
la s e r [3 9 ] .......... .. ........ ......................................................... ... 5 2

2-5 Typical absorption coefficients of water as a function of wavelength [41].......... 53

2-6 Gaussian-shaped function generated in MATLAB to simulate a "long"
outgoing laser pulse. ............ .......... .. ....... ......... ...... ................. 53

2-7 Simulation of ideal water surface and ocean bottom reflections for a "long"
pulse width............................................ .......... 54

2-8 Gaussian-shaped function generated in MATLAB to simulate a "short"
outgoing laser pulse w idth. ...................................................... ............ ... 54

2-9 Simulation of ideal water surface and ocean bottom reflections for a "short"
o utg o ing p u lse w id th ............... ..................................................... 5 5

2-10 Data from NASA Goddard Space Center's microaltimeter instrument [23]........ 56

2-11 A typical return waveform from the Optech SHOALS bathymetric mapping
system, measured by a PMT detector [11]. ....... ...... ..................... ............... 56

2-12 Output pulses from a PMT at different light levels. ............................. .......... 57

2-13 PMT signal under varying light intensities .................................... ................ 57

2-14 Schematic structure of a MCP-PMT and the principle of multiplication [50]. ...... 57

2-15 Typical pulse height distribution plot (PHD) [50]............. ............................... 58

2-16 PHD plots depicting an increase in incident light level and supply voltage......... 58

2-17 Typical PMT plateau characteristics............................................. 59

2-18 Typical PMT output variation versus supply voltage ............... ............... 59

3-1 Block diagram of optical path layout. ....... ..................................... ............ 70









3-2 Detailed optical path with turning mirrors, mechanical shutter, receiver
electronics, and fiber optic detector .................................... .......... ............... 70

3-3 Photo of JDS Uniphase NanoPulse lasers. ....... ........ ..................... ............... 71

3-4 CATS beam polarization test. ................. ............... ............... 71

3-5 Isometric and cutout views of scanner head ............. .. .... .................. 72

3-6 Lab trials of scanner. .................. ............ ......... ............ ............... 72

3-7 Photo of Hamamatsu Electronics Inc. model R4110U-74-M100D MCP-PMT
and quantum efficiency curve.................. ................ ............... 72

3-8 Photos of the CATS and related components. ........................... ... ............... 73

3-9 Photo of several building structures in Lanham MD. ............. ................ 74

3-10 Footprint projected on a brick wall at a distance of 303.7570 m...................... 74

4-1 CATS static (no scan) returns analysis from a brick wall approximately 300 m
from the instrum ent........... ........... .......................... ....... ........ 93

4-2 Relative accumulated hit counts per channel for a CATS range calibration
te s t ...................................................................................................................... 9 4

4-3 CATS static footprint test setup ...... .............. ... ... ............... 94

4-4 Illustration of beamlet cross-section projected onto wood surfaces ............... 95

4-5 Calculation of reported separation for 19 cm distance between multiple
scattering surfaces (trial D)............... ..... ....... .............. ... ........... 95

4-6 Output profile of multi-anode MCP-PMT when a light spot of approximately
20 pm diameter is scanned across the photocathode active areas [50]............. 96

4-7 Contour map of beam energy for a single pixel ......................... ... ............. 97

4-8 CATS static footprint test setup.. .................. .......................... .............. 97

4-9 Geometric orientation of isolated pixels (shaded central four).............. ........... 98

4-10 Photos of CATS barrel testing. ................................. ............ ............... 98

4-11 Histograms illustrating accumulated signal events for a heavily turbid water-
sand solution. ............... .................. ........... ...... ......... ............. 99

4-12 Stacked histograms depicting accumulated atmospheric events per channel
at three different PMT supply voltages. ............... ....... ......... ......... 99









4-13 Histograms depicting the consistency of accumulated events through the
duration of an 8 ps range gate for three different PMT supply voltages. ......... 100

4-14 Point cloud (green) resulting from CATS horizontal scan after timetag
structure m modification .......................................... .............. .............. 100

4-15 Top view of point cloud generated from rotating line scan data at 2100 V. ...... 101

4-16 Isometric view of point cloud generated from rotating line scan data at 2100
V .............. ............................................. .......................... .. 1 0 1

4-17 Top view of point cloud generated from rotating line scan data at 2200 V. ...... 102

4-18 Isometric view of point cloud generated from rotating line scan data at 2200
V .............. ............................................. .......................... .. 1 0 2

4-19 Front view of point cloud generated from rotating line scan data at 2200 V,
zoom -in of tree structure .......................................................... .................. 103

4-20 Photograph (front view) of a building with a large number of nearby trees....... 103

4-21 Overhead view of large point cloud generated from the building with heavy
tree obstruction....................... ........ .......................... ......... ...... 104

4-22 Overhead or top view of 0-3 m data segment associated with building
location. .............. ...................................................................... ... .... 104

4-23 Isometric view of 0-3 m data segment to emphasize heights ........................ 105

4-24 Zoom-in on isometric view of 0-3 m data segment.. ...................... 106

4-25 Top and side view of 0-5 m data segment ................. .............. .............. 107

5-1 Photos of the CATS mounted in aircraft. ............... ............... ............. 129

5-2 Flight testing in Gainesville, FL over University of Florida campus............... 129

5-3 Small point cloud from forested region in Mine Run, VA ............................... 130

5-4 Large point cloud from forested region in Mine Run, VA. ............................ 130

5-5 Ghost images in raw data ........ ................. .................. ... 131

5-6 Vertical sliding window to evaluate magnitude of ghost responses ............... 131

5-7 Results from sliding w indow ............................................................ ............... 132

5-8 Graphical model depicting reflections from detector cable interfaces .............. 132









5-9 Large signal (110 mV) injected into gain block. ........ .... ........ ................. 134

5-10 Output of gain block from 110 mV signal............... ................... ................. 134

5-11 Large signal (200 mV) injected into gain block. ........ .... ........ ................. 135

5-12 Output of gain block from 200 mV signal .............. ................... ................... 135

5-13 Large signal (300 mV) injected into gain block. ................................................ 136

5-14 Output of gain block from 300 mV signal .............. ................... ................... 136

5-15 Result of first step of MATLAB local density filter for a single channel of raw
airborne profile data..... ................................... .......... 137

5-16 Result of second step of MATLAB local density filter for raw airborne profile
data. .................................. ............................ ....... .......... 137

5-17 Flight testing in Mine Run, VA over the Lake of the Woods, southwest............ 138

5-18 Flight testing in Mine Run, VA over the Lake of the Woods, northeast............. 140

5-19 Flight testing in St. Augustine, FL over docks on the intracoastal waterway .... 141

5-20 Flight testing in St. Augustine, FL over a sandbar on the intracoastal
w aterw ay. .............. ............ ... .......... ..... ................................ 143

5-21 Flight testing in St. Augustine, FL over the eastern coast.............................. 144

6-1 Flight testing of second generation LSNR LIDAR by Sigma Space
Corporation over US Naval Academy campus in Annapolis, MD [81].............. 158

6-2 Flight testing of second generation LSNR LIDAR by Sigma Space
Corporation in Chesapeake Bay off US Naval Academy Campus [81]............. 158

6-3 Flight testing of second generation LSNR LIDAR by Sigma Space
Corporation in Greenland. ............ ....................................................... 159

A-1 Receiver block diagram for n channels.................................. ............... 166

A-2 Data capture board block diagram ................ .............................................. 166

A-3 Range/control board block diagram ..................................... .............. 167

A -4 R ece iver com ponents ................................................................ ........... ........ 167

A-5 Assembled CATS system photo with timing electronics "box" outlined. ........... 168

B-1 Differential measurements of CATS scanner periods................................... 180









B-2 Time data file associated with possible scanner timetag errors...................... 180

B-3 Filtered differential measurements of CATS scanner periods ..................... 181

B-4 Point clouds created from preliminary CATS sensor data. ............................. 181

B-5 Illustration of scan period wandering. ................................... .......... ......... 182

B-6 Point cloud for horizontal profile scan of bank building.............. ........... 183

B-7 Effect of shot variation in generated image ....... ..................................... 183

B-8 FPGA block diagram showing redesign of CATS timetag structure to improve
calculation of start of scan periods ...... .................. ..... ............. 184









Abstract of Dissertation Presented to the Graduate School
of the University of Florida in Partial Fulfillment of the
Requirements for the Degree of Doctor of Philosophy

MIXED TOPOGRAPHIC AND SHALLOW BATHYMETRIC MAPPING VIA
MULTI-STOP SINGLE PHOTOELECTRON LASER RANGING

By

Kristofer Y. Shrestha

August 2010

Chair: Kirk Hatfield
Major: Civil Engineering

Airborne laser swath mapping (ALSM) systems collect thousands of range

measurements per second, enabling efficient high-resolution surface mapping. Low

signal-to-noise ratio (LSNR) detection techniques allow for implementation of light

detection and ranging (LIDAR) instrumentation on platforms with prohibitive power, size,

and weight restrictions. The viability of single-wavelength photon counting techniques

applied to mixed target detection and shallow-water bathymetry has not been previously

investigated in detail.

The system design process for constructing a LSNR LIDAR system is described,

with emphasis on component selection as related to theoretical performance. Cost-

effective microchip lasers are now commercially available that produce sub-nanosecond

pulse widths and stable peak energies (>1 pJ at 10 kHz). In order to detect

backscattered laser radiation at low power levels, detector elements must be capable of

single photon sensitivity; a high gain (105), fast response (sub-250 ps rise time) detector

is therefore central to photon counting operation. Micro-channel plate photomultiplier

tubes (MCP-PMTs) are capable of attaining these specifications with simultaneous two-

dimensional detection over an array.









The University of Florida's prototype LSNR LIDAR sensor the Coastal Area

Tactical-mapping System (CATS) was constructed with these major components in

mind. The instrument is designed to operate in a fixed-wing aircraft flying 600 m above

ground level, producing 3 pJ output pulses at 8 kHz to allow for 20 cm spatial resolution

by scanning a 10x10 array of laser beamlets. Ground-based field testing results yield

10 cm range accuracy and small-scale feature identification at variable scan settings.

Experimental data is presented establishing the ability to record multiple returns

from the same laser pulse. Evaluation of system dead time shows that the minimum

separation distance required to discriminate multiple scattering objects is 15 cm,

indicating the possibility of seamless topographic / bathymetric mapping with no gap at

the waterline. Data collected during tests display vertical structure in shallow water

areas based on fixed threshold crossings at very low signal levels. A major concern for

the binary detection strategy is dark / atmospheric / crosstalk noise producing false-

positive measurements. Potential causes of image ghosting are described, and data

collection / processing solutions based on local density estimation are explored. LIDAR

depth estimates from airborne profiles are compared to on-site measurements, and

near-shore submerged feature identification is presented.

Analysis of CATS data confirms simulation estimates of decimeter-scale spatial

resolution for hybrid topographic and shallow bathymetric mapping. Dark noise

production and atmospheric scattering in daylight operation at maximum signal

amplification does not significantly impede feature identification procedures. In order to

combat the high variability in dynamic range, modifications to the detector structure may

be necessary to reduce anomalous effects.









CHAPTER 1
INTRODUCTION

Introduction to Remote Sensing

Our surrounding environment is a diverse collection of refractive and reflective

media. The interaction of light with matter is of special interest for remote sensing, a

branch of science concerned with the acquisition of information about an object or

phenomenon without direct physical contact. Information about positions, sizes, and

interrelationships between objects can be concisely conveyed through imagery.

Because humans possess a high level of proficiency in deriving information from such

images, we experience little difficulty in interpreting even those scenes that are visually

complex. In fact, it is only when we attempt to replicate these capabilities using

computer programs that we realize how powerful our abilities are to derive this kind of

intricate information [1].

Photogrammetry and LIDAR

Photogrammetry is one method of obtaining reliable information about the

environment through processes of recording, measuring, and interpreting images and

patterns of electromagnetic radiant energy and other phenomena [2]. Recent advances

in technology have given rise to parallel methods of remote sensing, one example being

light detection and ranging (LIDAR). This acronym is commonly used to describe the

intersection of an inertial navigation system (INS), a light amplification through

stimulated emission of radiation (laser) device, and a global positioning system (GPS)

[3]. Precise timing of transmitted and received light is the focus of LIDAR instruments.

Given that light travels at a known velocity, time-of-flight observations allow for









measurement of propagation distances. Detailed geo-spatial information is obtained by

the combination of these observations and enhanced navigation tools.

Advantages of LIDAR

LIDAR offers advantages over photogrammetry which has caused airborne LIDAR

to emerge as an attractive alternative to traditional methods of large-scale geospatial

data capture [3]. Photogrammetric methods for digital terrain model (DTM) generation

can be labor intensive and time consuming. LIDAR allows for fast surveying of remote

areas because each data coordinate is individually geo-referenced; aerial triangulation

or orthorectification of data is not required. Because of a dependence on solar

illumination, aerial photography systems are limited to daytime operation and are

affected by weather. LIDAR systems can operate at night and through a variety of

weather conditions. In addition, LIDAR offers a distinct improvement over

photogrammetry in the collection of elevation data in heavily forested areas. Dense

canopy cover generally prevents retrieval of accurate terrain information by

photogrammetric sensors [3]. Airborne laser mapping sensors, on the other hand, can

be used to acquire 3D discrete point measurements of canopy structure and the

underlying surface topography with sub-meter resolution [4]. We will extend this

concept to airborne bathymetry in later chapters.

Applications of LIDAR

A typical airborne LIDAR sensor provides tens to hundreds of thousands of range

measurements per second, which can be used to construct a high resolution map of an

area. Airborne laser swath mapping (ALSM) is a proven method for obtaining high

quality three-dimensional mapping data [5], [6]. The impact of this technology is easily

seen from a militaristic viewpoint; improved target detection capabilities create









advantages for surveillance and reconnaissance of enemy operations. The

contributions are not necessarily limited to warfare, however-airborne mapping can

provide valuable information about rural or inaccessible areas, record observations of

delicate or highly sensitive terrain, and aid in hazardous environment mapping.

Improving Image Fidelity

LIDAR Point Clouds

The basic visualization of LIDAR data is a projection of three dimensional points

into a "point cloud." Features within the point cloud can be identified at a number of

scales. For example, large-scale topographic undulations can be mapped, while at the

same time small-scale road edges, roof outlines and tree locations can be pinpointed.

Analysis of post-processed data has yielded automatic methods for feature

segmentation and target identification [7], [8], [9]. Point density within the cloud

determines the spatial resolution at which slowly varying continuous features are

monitored and sharply varying discrete features are identified. The ability to plot more

points per square meter results in higher image fidelity.

Existing Methods for Improving Feature Detection

It is possible to merge LIDAR points with data from another type of sensor to

improve feature detection capabilities. For example, it has been shown that the fusion

of small-footprint LIDAR and multi-spectral data can help to estimate plot-level volume

and biomass in deciduous forests [10]. Gaps in point cloud data over water can be filled

with information from sonar, but usually at coarse intervals. It has also been shown that

the addition of high-resolution hyperspectral imagery aids in classification of benthic

LIDAR waveforms [11]. Data fusion and weighting of feature detection algorithms

continues to be a growing area of study in the field of remote sensing.









Preferably, however, a single sensor would be able to provide all the necessary

information and eliminate the need for data fusion considerations. For LIDAR alone,

increasing the laser pulse repetition rate decreases the time between illuminations of

ground footprints; thus subject to a static scan pattern (not dependent on laser rate) and

a consistent platform velocity, spacing between ground points is reduced and horizontal

resolution is improved. This approach eventually results in having more than one laser

pulse in flight at the same time, which can cause difficulties when the range to the

surface changes rapidly and unpredictably, as is often the case in mountainous terrain

or urban settings. The two-way travel time of the laser pulses set a maximum pulse rate

of 250 kHz at an average flying height of 600 m. If, however, a single laser beam is split

into an array of elements and transmitted at once, the spatial resolution is limited by the

number of array elements and the capability of the receiver to process returns over the

array simultaneously. A moderate laser pulse rate (<10 kHz) can then theoretically

provide contiguous coverage of the surface of the terrain at decimeter-scale resolution

on a single pass of an airborne LIDAR system [9].

Existing Methods for Improving Data Quality

Low-level signal classification is often difficult in LIDAR systems given practical

considerations for variations in surface reflectivity; diffuse reflections of laser light are

not always easily detectable. Characterization of system noise is important for

discrimination of true events, meaning that noise components should be kept to a

minimum where possible. Superior contrast between signal and noise may generally be

achieved by transmitting higher laser pulse energy. Providing large outgoing pulse

energy increases the probability that weak light-matter interactions will produce a

detectable response. Lasers operating at high energy can be paired with a variety of









detector components and still be able to "see" the desired target regardless of faint

backscattered radiation or low amplification. An assumption for this case is that the

paired photodetector can operate at an acceptable dynamic range; i.e., be able to

detect the faintest and strongest measured signals with equal accuracy. Another

assumption is that noise returns do not increase at the same rate as signal returns

given a more energetic pulse; otherwise false events would obscure the desired

response.

The LSNR Mapping Approach

Problems with High Energy / High Pulse Rate Designs

Meter-scale spatial resolution in airborne scanning laser altimeter systems can be

achieved by employing high energy / high pulse rate designs. Currently deployed

ALSM implementations are designed to produce a high signal-to-noise ratio (SNR);

typical specifications at standard altitudes of 500 to 1500 m are transmitted energy of

100 pJ per pulse and laser repetition frequency of 50 to 170 kHz [12]. These systems

are capable of generating thousands of return photons incident on the receiver for each

outgoing laser pulse. High SNR ensures that a system will be able to separate true

return signals from sources of spurious noise events such as atmospheric scattering

and detector dark noise, but may still contain erroneous points associated with multi-

path (sequential reflections from different surfaces prior to returning to the sensor).

High energy, high pulse rate systems are not always practically desirable because

of the size, complexity and cost of hardware. For example, attempting to integrate a

high energy LIDAR design into a small unmanned aerial vehicle (UAV) platform is

problematic. This type of platform operates on a fixed power budget, so introducing a

heavy payload impedes flight operations. Utilizing a high energy laser can shorten









mission duration by draining battery life; high power lasers require complex electronics

and sophisticated cooling systems. Bulky equipment is unsuitable for scalability with

low-power applications. UAVs are often intended for potentially "disposable" functions,

and the loss of an expensive laser mapping unit can be prohibitive. Operation in

populated areas also restricts maximum laser power and sensor altitude due to eye

safety requirements. Designs based on a high energy requirement per illuminated area

limit the size of the footprint, resulting in less than contiguous coverage of the terrain.

Motivations for a LSNR LIDAR System

Practical requirements such as those mentioned above have motivated alternate

methods to detect weak light-matter interactions. Current commercial terrain mapping

systems provide time-of-flight measurements for multiple reflections from each laser

shot. The event timer used to determine the time of flight of the recorded returns can

use various discriminator designs to analyze the photodetector voltage response,

including threshold, center of gravity, maximum, zero crossing of the second derivative,

and/or constant fraction [13]. Classically a constant fraction discriminator has been

used to obtain first / last responses related to separate reflecting objects in complex

waveforms [14]. This approach may limit the minimum distance between two

distinguishable objects in the surveyed scene because lasers of relatively high energy

(greater than 100 pJ per pulse) generally have pulse widths of a few nanoseconds or

longer.

It has been suggested that a conventional single element detector approach

operating under high SNR does not make the most efficient use of available laser

photons [15]. This idea has shown merit in low-SNR (LSNR) LIDAR experiments on the

Multi-kiloHertz Micro-Laser Altimeter (MMLA), developed at the NASA Goddard Space









Flight Center under the Instrument Incubator Program [16]. It is estimated that the

surface sampling rate can be increased by over three orders of magnitude for a given

laser power/telescope aperture product by emitting the available photons in a high

frequency train of low energy pulses, as opposed to a low frequency train of high energy

pulses, and by employing photon-counting detector arrays with multi-channel timing

receivers [16].

In theory, surface mapping efficiency can be significantly improved by using a

LSNR LIDAR paradigm based on lower energy and shorter pulse lasers [15]. Small-

footprint LIDAR are intended to illuminate only a few surfaces within the footprint,

yielding distinct return pulses corresponding to separate reflecting objects [17]. LSNR

concept systems such as the Remote Ultra-Low Light Imaging system developed at the

Los Alamos National Laboratory [18], the Jigsaw sensor created under the DARPA

program at the Massachusetts Institute of Technology Lincoln Laboratory [19], and a

second generation 3D imaging and polarimetric LIDAR developed by Sigma Space

Corporation [20] have established LSNR LIDAR as an effective approach to terrain

mapping.

Though these ventures have shown the promise of the LSNR technique, they still

represent a limited number in comparison to the many HSNR parallels. In the context of

modern technology advancements, this has left significant gray areas in a

comprehensive understanding of LSNR signal analysis. Simulation can be a powerful

tool in this case, as a modeling capability can analyze performance as a function of

system specifications and mission environment. Current mathematical models support









the idea that a single-wavelength LSNR mapping instrument can provide hybrid

topographic and bathymetric mapping in shallow coastal areas [21].

Research Aims

The goal of this work is to move beyond simulation and explore the viability of

topographic and shallow bathymetric mapping via a single wavelength (532 nm), low

altitude (500 m), low energy (3 pJ), short pulse (500 ps) laser ranging system. A

prototype LSNR system has been developed by the University of Florida for this

purpose, and detailed analysis of sensor data is shown here. In Chapter 2 of this work,

the fundamentals of selecting an appropriate illumination source and photodetector are

presented. In Chapter 3, the specifics related to the University of Florida's LSNR

prototype design are listed. In Chapter 4, analyses of static experiments yield

performance metrics associated with LSNR feature detection. In Chapter 5, analyses of

airborne experiments explore the sensor's ability to provide shallow water bathymetry in

a variety of water types. Chapter 6 presents conclusions about the viability of the

technique and suggestions for improvement.









CHAPTER 2
LSNR COMPONENTS

Commercial Sector Activities and Scientific Impact

COTS Market and Costs

Prior to 1995, laser altimetry was generally conducted using custom-designed

sensors operated by research groups or built by commercial survey firms to exploit

niche markets. The majority of these efforts were based on single prototype designs,

limiting their ability to create and service a broad, sustainable, sector-wide demand for

the technology. Since 1995, however, a commercial off-the-shelf (COTS) instrument

market has developed which has removed many of the constraints for incorporating

laser altimetry into operations [22]. These conditions have allowed for new types of

lasers and photodetectors to become available without prohibitive cost to a variety of

research organizations.

LSNR Concept Systems

The Satellite Laser Ranging Station (SLR2000) [15] was one of the earliest

experimental platforms to demonstrate LSNR potential. The SLR2000 employed

passively Q-switched microlasers with output energies of 135 pJ and rates below 5 kHz

to achieve mm-level ranging precision at satellite altitudes of 20,000 km. Using a sub-

nanosecond pulse width, a laser beam divergence of 10 arcseconds, and a receive

aperture of 76 cm resulted in a signal strength of less than one photoelectron per laser

shot. Target discrimination was made possible by utilizing a range receiver able to

identify and extract single photon laser echoes from solar background noise through

Poisson statistical analysis of the received photon distribution. The Post Detection









Poisson Filtering (PDPF) technique has been extended to the MMLA sensor for signal

extraction in an airborne scanning altimeter capable of daylight operations [23].

The Remote Ultra-Low Light Imaging (RULLI) system also responds to individual

photons using a low power pulsed laser and Pulse Absolute Timing (PAT) channels for

time discrimination in the front-end electronics [18]. Three dimensional imaging was

accomplished with a vertical resolution of 5 cm. The detector is made up of a micro-

channel-plate (MCP) image intensifier front end with high gain for photon counting and a

wire-would cross delay line (CDL) anode. A stack of three MCPs allow for electronic

gain up to 107. Each PAT channel is designed around a number of internal circuits in a

hybrid electronics module called a time interval meter (TIM). This includes a constant

fraction discriminator (CFD) front end, a dual interpolator stage, and a ten bit time to

digital converter. This allows for detection of signals with amplitude variation of 40 dB

while maintaining a time walk of less than 100 ps.

The Jigsaw system uses arrays of single photon sensitivity avalanche photodiodes

(APD) for integration onto various surveillance platforms [19]. The sensor was also

tested in an indoor environment for use in surface reconnaissance experiments. An

example of raw data from the Jigsaw system is shown in Fig. 2-1. The "noisy" nature of

these images illustrates the challenges associated with LSNR LIDAR detection, as most

of the points originate from avalanche photodiode dark counts. These challenges can

be met through post-processing techniques, as researchers have created a suite of

novel algorithms to perform noise reduction, ground-plane identification, detector

response deconvolution, and illumination pattern renormalization [24]. Results of these

efforts are shown in Fig. 2-2.









Selection of a well-paired illumination source and photodetector can improve SNR

so that data processing presents less of an obstacle. The following sections describe

how to select components with appropriate characteristics to maximize photon-counting

performance. A brief history of lasers is presented to introduce microchip laser design

parameters. We will then transition into classic detector pairings and the photon

counting concept.

Overview of Lasers

Quantum Dynamics Fundamentals

Lasing is made possible by the creation of population inversion in a group of

atoms or molecules. This phenomenon is best introduced with quantum mechanics. In

the simplest form, a group of N atoms can exist in two basic states or energy levels: E,

the ground state, and E2, the excited state, with E2 > E,. The number of atoms in the

ground and excited states are given by N, and N2, respectively. The total number of

atoms in the system is N, where N = N, +N2. The energy difference between the two

states is given by A E 2 E,. This difference determines the characteristic frequency

f12 of light which will interact with the atoms, following the expression [25],

AE=hf2,,, (2-1)

where h is Planck's constant (6.6261 x10-34 J-s). Thermodynamics shows us that for a

system in thermal equilibrium, the ratio of a number of atoms in each state is given by

the Boltzmann distribution,

N, (E2E1) (2-2)
S kT
N1









where T is the thermodynamic temperature of the group of atoms and k is Boltzmann's

constant (1.3807 x 10-23 J.K1). The ratio of the atom populations at room temperature

(T~ 300 K) for an energy difference delta E that corresponds to visible light (f~ 500

Terahertz) can then be calculated; A E = 2.07 eV, and kT= 0.026 eV. Since this ratio

forms the argument for the negative exponential and A E >> kT, it follows that the ratio

N
N2 is very small; i.e., there are very few atoms in the excited state for a system in
N1

thermal equilibrium. The Boltzmann distribution dictates that in this case, even at

infinite temperature, the population N2 will not exceed N1. Achieving population

inversion thus requires pushing the system into a non-equilibrated state.

Methods of achieving population inversion are based on precise manipulation of

optical properties. Light can interact with a system of atoms in the following three ways:

absorption, spontaneous emission, and stimulated emission. In the first, photons of

frequency f12 may be absorbed by ground state atoms, causing them to transition to the

higher energy level. The probability of occurrence is proportional to incident light

intensity and N,. In the second, atoms in the excited state may also relax back to the

ground state, and in the process release photons of frequency f21; this is called

spontaneous emission.

The final interaction, stimulated emission, is the most important for the optical

amplification process which makes lasing possible. An atom in the excited state may be

perturbed by the passage of a photon with frequency f21 (exactly corresponding to the

energy gap A E), causing a transition to the ground state and the emission of a second

photon at the same frequency. Albert Einstein showed that the base probability of a









photon causing stimulated emission in a single atom is equal to the probability of a

photon being absorbed by an atom in the ground state [26]. Consequently, when the

number of atoms in the excited and ground states are equal, the rate of absorption is

equal to the rate of stimulated emission for a given radiation density.

Coherence

The stimulated transition process is the sum of two separate processes. The

lower energy level contains atoms that are stimulated by the applied signal to make

upward transitions, while the upper energy level contains atoms that are stimulated by

the applied signal to make downward transitions. The net flow of atoms is always from

the higher population level to the lower population level. If the lower level is more

populated, the signal is attenuated. If the higher level is more populated, the signal is

amplified. The stimulated transition process results from a resonant response of the

atomic wave function, or of the atomic charge cloud in each individual atom, to the

applied signal. The internal induced oscillation or dipole response that is produced in

each atom is stimulated by the applied signal [27].

Atomic oscillations follow the driving optical signal in full coherence, meaning that

they follow the same amplitude and phase. An amplifying laser medium produces a

linear reproduction of the input signal and correspondingly any amplitude or phase

modulation present in the signal except that a) the output signal is increased in

magnitude; b) the signal modulation may be decreased in bandwidth because of the

finite bandwidth of the atomic response; and c) the signal generally has a small amount

of spontaneous emission noise added to it [27].









Three- and Four-Level Laser Models

It has been shown that the ideal two-level system is not enough to achieve

population inversion. A more realistic approach considers three levels: E,, E2, and E3,

with E3 < E2 < E. Initially the majority of the atoms are in the ground state E,. These


atoms transition to E3 by absorbing radiation at a frequency f3 = This is
h-(E3-E,)

often referred to as the "pump transition." For a medium suitable for laser operation, it

is required that these excited atoms quickly decay to E2. Spontaneous emission may

occur as a result, but in practice the E3 to E2 transition is radiationless; energy is

transferred to vibrational motion of the host material surrounding the atoms without the

generation of a photon.

Correspondingly, an atom may decay from E2 to E, and release a photon of

frequency f12. This is the intended laser transition, and it is achieved by ensuring that

the lifetime of the transition t21 is much longer than the lifetime of the transition t32; often

called a favorable lifetime ratio. Over time, the population N3 in this case will become

very small, and a population of excited atoms will accumulate at E2. Optical

amplification then can be achieved with the population inversion N2 > N .

The three-level framework was the basis for many early lasers, including the first

ruby laser as implemented by Theodore Maiman in 1960 [28]. In the context of modern

lasers the three-level system is inefficient because the medium must be very strongly

pumped in order to achieve inversion. Most lasers now operate via a four-level process.









A four-level laser energy level diagram is shown in Fig. 2-3. The fast, radiationless

transition R, from the pump band to E3 the upper laser level results in the

population N4 being rapidly depleted. The same is true for atoms at the lower laser

level E2; they will make a quick transition to the ground state E, via Rb. The laser

medium is selected such that the transition of atoms from E3 to E2 is the intended laser

transition. A four-level system is much more efficient than a three-level, however,

because any significant population accumulating at E3 produces a population inversion

with respect to the negligible population N2. Optical amplification is thus possible for

the frequency f32.

Optical Resonators

Now that laser operation has been introduced from the quantum perspective, the

device characteristics behind resonance or amplification can be described. While at

Columbia University in 1959, graduate student Gordon Gould introduced the idea of the

open resonator [29] and its potential applications. An optical cavity or resonator is an

arrangement of mirrors that forms a standing light wave. The cavity surrounds the gain

medium and provides feedback for laser light. Multiple reflections of light from mirrors

at either end of the resonance chamber produce standing waves for certain resonance

frequencies.

Certain patterns of standing waves are sustained via constructive interference,

called the modes of the laser, while others are suppressed due to destructive

interference. There are two types of modes. Longitudinal modes refer to the

wavelengths of the wave reinforced by constructive interference after many reflections

from the cavity mirrors, and transverse modes refer to the electromagnetic field pattern









measured in a plane perpendicular to the beam propagation direction. The most basic

transverse mode is the Gaussian energy distribution.

The simplest optical cavity is created by two facing plane (flat) mirrors, commonly

called the Fabry-Perot arrangement. The difficulty in using flat mirrors comes in the

alignment; even a few arc-seconds of error can cause the intracavity beam to leak out

the sides of the cavity. This problem becomes much less significant when dealing with

short mirror separation distances (L < 1 cm) [30] however, allowing for their use in

microchip and semiconductor lasers.

Q-Switching

Pulsed lasers are particularly important for laser ranging applications, as the pulse

duration and energy are of prime relevance to a system's timing and detection

capabilities. "Q-switching" is a technique for increasing the energy of short pulses by

modulating the intracavity losses and the "Q-factor" of the laser resonator. Q-factor, or

quality factor, is a term borrowed from electronic circuits (for example, describing

inductive-capacitive elements in the context of a continuously operating oscillator). The

term was adapted for optical resonators to measure the strength of the oscillation

damping in the optical cavity [31]. Manipulation of the Q-factor allows for generation of

"giant" pulses [32], referring to a large amount of energy output in a short amount of

time.

Q-switching is achieved by putting a variable attenuator inside the laser's optical

resonator. When the attenuator is functioning, light which leaves the gain medium does

not return, and lasing cannot begin. The attenuation in the cavity corresponds to a

decrease in the Q-factor of the optical resonator. A high Q-factor corresponds to low









resonator losses per round trip, and vice versa. The variable attenuator is commonly

called a "Q-switch" when used for this purpose.

Initially the laser medium is pumped while the Q-switch is set to prevent feedback

of light into the gain medium (producing an optical resonator with low Q). This produces

a population inversion, but laser operation cannot yet occur since there is no feedback

from the resonator. Since the rate of stimulated emission is dependent on the amount

of light entering the medium, the amount of energy stored in the gain medium increases

as the medium is pumped. Due to losses from spontaneous emission and other

processes, after a certain time the stored energy will reach some maximum level; the

medium is said to be "gain saturated." At this point, the Q-switch device is quickly

changed from low to high Q, allowing feedback and the process of optical amplification

by stimulated emission to begin. Because of the large amount of energy already stored

in the gain medium, the intensity of light in the laser resonator builds up very quickly;

this also causes the energy stored in the medium to be depleted almost as quickly. The

net result is a short pulse of light output from the laser, known as a giant pulse, which

may have very high peak intensity.

The Q-factor can be reduced via both active and passive methods; the output of

an energy pulse in this manner is called Q-switching. For active Q-switching, losses in

the cavity are typically modulated via acousto- or electro-optics. A pulse is formed after

an electrical trigger signal is received. With a sufficiently fast trigger, pulse widths on

the order of the cavity round-trip transit time can be achieved [33]. These are often

referred to as "cavity-dumped lasers." Mechanical switches such as a Lummer-

Gehrcke plate in conjunction with a rotating prism also been used [34], but are less









common. The idea of using a saturable absorber to achieve passive Q-switching (PQS)

was presented in 1965 by Szabo and Stein [35], along with derivation of rate equations

and numerical analysis.

Microchip Laser Advancements

To obtain closed-form solutions for key laser parameters such as output energy

and pulsewidth in modern lasers, one can solve the three coupled differential equations

which describe PQS. Optimization of the amplification system depends on the

combination of characteristics related to the laser gain medium, absorber medium, and

pump intensity. The Lagrange multiplier technique introduced by Degnan [36] can

determine the optimum mirror reflectivites and unsaturated absorber transmissions as a

function of two dimensionless variables: z, the ratio of the logarithmic round-trip small

signal gain to the dissipative optical loss; and a, the saturation energy density of the

amplifying medium divided by the saturation energy of the absorber. These theoretical

calculations were shown to be in agreement with experimental results, and independent

optimization theories for rapid PQS lasers were subsequently formed [37].

The early part of the 21st century gave rise to significant advancements in

continuous wave (CW) pumped PQS microchip lasers, with emphasis on large-

amplitude repetitive behavior of PQS in solid-state devices [38]. Numerical and analytic

solutions were presented to describe the gain medium as an ideal four-level atom and

the saturable absorber as a two level atomic system. The region of interest in this case

is where the saturation intensity for the saturable absorber is larger than that for the

gain, and when both decay rates are fast. Numerical bifurcation studies and asymptotic

analysis show [38] that PQS is possible if these restrictions are met, and microchip laser

design equations can be derived that are valid whenever the gain is depleted and the









saturable absorber is bleached. When applied to pulsing in a microchip laser with a

neodymium-doped yttrium aluminum garnet (Nd:YAG) gain medium and chromium-

doped yttrium aluminum garnet (Cr4+:YAG) saturable absorber, theoretical results were

shown to be within 10% of the experimental values [38].

Studies such as these formed the theoretical basis for microchip laser

development. For the PQS Nd:YAG / Cr4+:YAG laser specifically, it has been shown

that the maximum repetition rate is correlated with the pulse width; specifically, it

decreases as the pulse width gets shorter [39]. If the absorption cross section of the

saturable absorber is large in comparison to the gain cross section of the gain medium,

relatively few stimulated photons are necessary to rapidly bleach the absorber.

Absorption in Cr4+:YAG occurs at 808 nm the wavelength used to pump the Nd:YAG

gain medium which can cause partial bleaching of absorption. This results in a

smaller absorption coefficient at the oscillating wavelength and a smaller effective

Cr4+:YAG fill factor (the normalized fraction of the filled cavity) for the laser.

The fraction of the laser cavity filled with Nd:YAG decreases as the Cr4+ fill factor

increases, meaning that the Nd:YAG must be more highly inverted to reach threshold.

More heat is thus deposited per unit length of active material. The gain cross section

decreases and a greater inversion is required as a result of the increase in thermal

energy. Positive feedback occurs, leading to a well-defined threshold for thermal

runaway [39]. The maximum pulse repetition rate is correlated with the Cr4+ fill factor,

meaning that the maximum repetition rate and pulse width are also correlated.

Experimental results for several SPMCL microchip lasers used in the Zayhowski study

are shown in Fig. 2-4.









Beam Quality Measures

A common measure of beam quality is the M2 factor. This is defined as the

product of the beam radius (measured at the beam waist) and the beam divergence

half-angle (measured in the far field) divided by -, according to ISO Standard 11146


[40],

M2/ (2-3)
8-


A laser beam is often said to be diffraction-limited by M2 times, as a diffraction-

limited Gaussian beam has a M2 factor of 1. For a given beam divergence angle,

which is limited by the numerical aperture of the focusing lens, the M2 factor limits the

degree to which the beam can be focused. Together with the optical power, the beam

quality factor determines the radiance of a laser beam. The degree of linear

polarization is normally quantified by the polarization extinction ratio (PER). This is

defined as the ratio of the optical powers in the two polarization directions.

Effects of Laser Parameters

Ideally, a LIDAR system would be built with a custom laser through which pulse

outputs are directly controlled. Pulse repetition frequency (PRF), energy, and width

could then be changed to suit the particular application. While this kind of direct control

provides inherent operating advantages related to quality monitoring, designs driven by

free-running lasers allow the use of cheaper COTS components. The tradeoff is that

system operations may then be affected by small per-pulse variations (if not directly

monitored). The capability to swap in cheaply available lasers operating at different

PRF, however, allows for such systems to approximate the versatility of a more









expensive directly controlled laser. Additional diagnostic operations may prove critical

for ensuring accuracy; without direct control of the laser, generating a time history of

events related to laser fire is essential to limiting errors in point cloud generation. For

example, one way to monitor PRF is to "tag" each laser pulse with GPS times. By

similarly tagging certain scanner characteristics, mechanical components can be

simultaneously monitored for stability.

Operating Wavelength

An imaging LIDAR is an active remote sensing instrument, meaning that it

illuminates the investigated area with its own energy source (via a collimated beam of

light) rather than relying on passive solar illumination. The selection of the specific

wavelength and amount of output energy per laser pulse has direct relation to the

sensing capabilities of the system. The propagation characteristics of electromagnetic

radiation through a given medium are dependent on the incident wavelength. For

example, visible green light (532 nm) exhibits low attenuation through water [41],

making it the preferred wavelength for bathymetric applications. Fig. 2-5 shows the

spectral dependence of typical water absorption coefficients.

Laser Pulse Width

The width of the outgoing laser pulse has particular importance for a system

operating at low output energies. Bathymetry is one area where this can be

demonstrated, because discrimination between sea surface and sea bottom returns is

of prime concern. For this application, it is necessary that light-matter interactions

produce sufficiently short return pulses to differentiate the two responses. It is possible

to produce feature classification given that these requirements are met, as is illustrated

in the following simulations.









MATLAB was used to convolve the Gaussian function shown in Fig. 2-6 with two

discrete Dirac delta functions. This mimics two "ideal" reflections i.e., no contributions

due to roughness or slope of a laser pulse from a sea surface and bottom. The height

of the two discrete elements was set to 0.3 and 0.7 to produce more contrast in

convolution operation result. The separation of the two elements was varied from 1 to 6

m to simulate different water depths. The resulting responses for convolution

operations of the 10 ns FWHM (3 m in air) pulse in Fig. 2-7 shows how differentiation

between surface and bottom is only possible at the larger depths, i.e. 4 m and larger.

An outgoing laser pulse width of 480 ps (Fig. 2-8) 1/20 of the previous value -

requires much less depth to produce significant differentiation between surface and

bottom signals. Simulating a 1 m depth, which produced no temporal separation of

signal events in the previous simulation, now shows a clear division. 480 ps

corresponds to a distance of 0.144 m in air. The two signals are resolvable at twice this

value, or 0.3 m (Fig. 2-9).

Pulse Energy

If we consider outgoing laser pulse energy as an accumulation of highly coherent

individual photons, the ideal photodetector pairing to sense backscattered radiation

would be sensitive to even a single returning photon. As will be described in the

following section, detector improvements have brought this ideal closer to the realm of

possibility. In order to estimate the output energy required for single photon detection,

the signal strength is estimated as a function of transmitter, receiver, and environment

characteristics. The uncertainty of an observation can be measured as a function of the

various errors sources [42]. Even for a radiation field of constant intensity, the number

of photons arriving at the photodetector for a given time interval is inherently uncertain









due to the discrete nature of the incident light, charge carriers, and the interaction of

light with the photodetector [43]. The relationship between the number of photons

incident on the detector and the number of emitted photoelectrons (p.e.) can be

expressed via a Poisson distribution. The probability of n, detected p.e. is therefore,

n e n (2-4)
P(n,,n,) = n, ,
n, !

where n, is the expected number of p.e. recorded by the receiver. The probability of

registering at least one signal event can then be calculated as a function of n, [44]. For

example, it is expected that for a signal level of one p.e., there is a 63.21% chance that

there will be a signal event registered. The LIDAR link equation relates basic elements

to estimate backscattered signal strength. A modified form of this equation yields the

following expression for transmitted energy per laser pulse [15],

hv. n, (2-5)
AE,
nhnqn r p, cos(a) r [exp(-fe ,R)i[


where nh is the hologram (in the case of an altered beam) efficiency, nq is the detector

quantum efficiency, n, is the receiver optical efficiency, h is Planck's constant, v is the

photon frequency in hertz, p, is the wavelength-dependent surface coefficient, a is the

local incidence angle on the surface, A, is the collecting area of the receiver aperture,

Be, is the atmospheric extinction coefficient per meter, and R is the range to surface.

Table 2-1 shows sample system parameters for moderate laser output power,

modest telescope aperture, and average efficiencies for commercially available

materials. The sample values indicate that we expect 20% transmission loss from the









hologram creating the beamlet pattern; 28% quantum efficiency in the detector; 60%

loss due to spatial and spectral filters; 15% reflectivity for typical ground soil at 532 nm;

a 6.5 cm diameter telescope; and 600 m flying height. To have a 63.21% chance of

detecting one signal event, the transmitted energy E, should be larger than 13.5 nJ.

Goddard Space Center test flights of a scanning laser altimeter aboard a NASA P3B

aircraft has shown potential for instruments operating at low pulse power and PRF [16].

Fig. 2-10 describes the test parameters and ranging results with respect to buildings,

tree canopy heights, and shallow water bathymetry.

Selecting a LSNR Photodetector

Operating at low pulse energy and thereby LSNR requires a detector with high

gain capabilities. There are several types of light sensors commonly used for LIDAR

mapping.

Avalanche Photodiode

One type of sensor classically used to convert weak light backscatter into analog

electrical signals is the avalanche photodiode (APD). APDs are high-speed (up to 250

ps rise time) photodiodes using an internal gain mechanism that functions by applying a

reverse voltage. When light enters a photodiode, electron-hole pairs are generated if

the incident energy is higher than the band gap energy. The drift speed of these

electron-hole pairs or carriers depends on the electric field strength. As these carriers

collide with the crystal lattice, ionization in which new electron-hole pairs are generated

takes place. This starts a chain reaction referred to as the avalanche multiplication of

photocurrent [45].









The internal gain effect of an APD is normally between 100 and 500 [Perkin-Elmer

2006]. It is possible to achieve very high gain 104 to 106 by operating an APD in

"Geiger mode," which refers to a sensor biased above its breakdown voltage. An APD

will normally conduct a large current when biased above breakdown, but if this current is

limited to less than the APD latch current, there is a strong statistical probability that the

current will fluctuate to zero in the multiplication region. The APD will then remain in a

nonconductive state until an avalanche pulse is triggered by either a bulk (dark current)

or photo-generated carrier. By keeping the number of bulk pulses low, the APD is then

used to count individual current pulses from incident photons. Minimization of the dark

current is a significant factor for this type of APD operation, and is reduced by cooling.

APDs are commercially available that span the spectral range from 300 to 1700

nm [46]. Silicon (Si) APDs can be used from 300 to 1100 nm, germanium (Ge) between

800 and 1600 nm, and indium gallium arsenide (InGaAs) from 900 to 1700 nm.

InGasAs APDs are typically available with low noise current, extended spectral

response to 1700 nm, and high frequency bandwidth for a given active area. Ge APDs

are useful for applications under high electro-magnetic interference, where amplifier

noise is significantly higher than APD noise [45]. APDs with high gain and fast rise

times, however, often exhibit slow relative decays (10-100 ns), however, causing range

tails as shown in the Jigsaw sensor [24].

Photomultiplier Tube Advantages

The photomultiplier tube (PMT) is another type of light sensor that provides very

high internal gains similar to those of a Geiger mode APD. Generally PMTs are able to

achieve higher internal gains up to 107 than APDs with superior recovery times.

They also, however, require voltages from 1 to 3 kV, are more fragile, and more









sensitive to stray magnetic fields. For system designs intending to amplify single

photoelectron events, PMTs can be attractive as long as they remain robust under

normal system operation.

Combinations

A LIDAR system can use a combination of an APD and a PMT in order to

compensate for varying environmental response. In the bathymetric case, for example,

deep water causes rapid decay in the power of the received signal. Depending on the

water depth and clarity, there can be more than six orders of magnitude dynamic range

between strong water interface returns and weak bottom returns [47]. This difference

must be handled by the detector without anomalous effects and compressed into the

useful input range of the detector, which is typically only two or three orders of

magnitude.

The use of a PMT or high-gain APD alone for bathymetric mapping can cause

several problems related to the strong interface reflections that occur frequently near

nadir. Even with the tube gated off, high incident light levels can cause a build-up of

space charge around the photocathode [47]. This can affect output signal levels,

linearity, and bandwidth. The result is a decrease in performance and may even be

damaging to the tube. There is also experimental evidence of "afterpulses" at a delayed

time after the impulse as a result of overly strong input signals [48]. This idea will be

explored in greater depth in a subsequent chapter. Unexpectedly large surface returns

may also create detector saturation, and corresponding distortion in the output

waveform [49].

The U.S. Army Corps of Engineers' Scanning Hydrographic Operational Laser

Survey (SHOALS) system uses an APD to measure the bottom response in depths









shallower than 14 m, and a PMT for depths deeper than 8 m. A measure of the

returned power is recorded via waveform digitization. The received power and resulting

waveform is a function of the transmitter and receiver characteristics, target laser cross

section, and atmospheric effects, and target reflectance [11]. Fig. 2-11 shows how this

information can be analyzed to produce classification of the surface return, volume

backscatter, and bottom return. Analyzing responses from two types of detectors can

thus improve characterization of benthic information.

Ideally, a single highly sensitive detector would be able to handle difficult

classification problems such as the aforementioned issues associated with LIDAR

bathymetry without negative consequences. If this were possible, the need for multiple

sensors or waveform digitization could be eliminated, allowing for a more simplistic

system structure with low overhead.

Photodetector Improvements

We now present an overview of recent photodetector improvements in order to

estimate the capabilities of a highly sensitive detector when paired with a low energy

laser.

Photon Counting

Large pulse energy lasers produce enough light backscatter from illuminated

targets to generate multiple photoelectric events at the detector. For highly sensitive

detectors, this is referred to the multiple photoelectron state. As shown in Fig. 2-12 [50],

repeated photoelectron emission in this state creates output pulse intervals so narrow

that they overlap each other. The combination of these overlapped pulses results in a

signal of consistent amplitude with high-frequency components, or Fig. 2-13(a) [51].

The output signal is effectively the mean value of this waveform containing the AC









component. When the incident light level drops to a fraction of its previous level as in

Fig. 2-13(b), the AC component becomes more significant. In a very low incident light

state as in Fig. 2-13(c), the individual discrete pulses can be analyzed in a digital mode.

Recent advancements in PMT design have led to improved precision, sensitivity,

and robustness in commercially available units. Commercial and developmental single-

photon sensitivity detectors now exist which are capable of providing centimeter level

ranging resolution [15]. Typical photocathode / dynode / anode configurations had

previously been singular; new units are available in multi-channel arrays. These

configurations allow for the receiver to angularly resolve the source of a single photon

event within the receiver field of view [15]. If the time resolution of the PMT is

sufficiently fast that the average time intervals between signal pulses are wide in

comparison, then individual photons can be "counted" and the receiver is then capable

of "photon counting," or single photon sensitivity. It can be argued that because the

generation of a photoelectron in this configuration is essentially a binary event given

proper discriminators, the photon counting method is superior to analog signal

measurement in terms of detection efficiency and stability.

Semiconductor Principles

Semiconductor PMTs operate on the principle of the external photoelectric effect,

by which photoelectrons are emitted into a vacuum from a material called the

photocathode [52]. The wavelength of maximum response and long-wavelength cutoff

are determined by the combination of alkali metals used for the photocathode and its

fabrication process. In a semiconductor band model, there exists a forbidden band gap

or energy gap (EG) that cannot be occupied by electrons. The electron affinity (EA) is

an interval between the conduction band and the vacuum level barrier (vacuum level).









The work function (y/) is the energy difference from the Fermi level and the vacuum

level. When photons strike a photocathode, electrons in the valence band absorb

energy and become excited, diffusing toward the photocathode surface. If the diffused

electrons have enough energy to overcome the vacuum level barrier, they are emitted

into the vacuum as photoelectrons. The quantum efficiency q(v) is the ratio of output

electrons to incident photons, given by,

P, R (2-6)


* R : reflection coefficient

* k : full absorption coefficient of photons

S P : probability that light absorption may excite electrons to a level greater
than vacuum level

* L : mean escape length of excited electrons

* Ps : probability that electrons reaching the photocathode surface may be
released into the vacuum

* v : frequency of light

In the above equation, if an appropriate material is chosen to determine

parameters R, k, and P,, the factors that dominate the quantum efficiency will be L

(mean escape length of excited electrons) and Ps (probability that electrons may be

released into the vacuum). L becomes larger by use of a better crystal and Ps greatly

depends on electron affinity (EA).

If a surface layer of electropositive material such as Cs20 is applied to a

photocathode, a depletion layer is formed, causing the band structure to be bent

downward. This bending can make the electron affinity negative. This state is called









NEA (negative electron affinity). The NEA effect increases the probability (Ps) that the

electrons reaching the photocathode surface may be emitted into the vacuum. In

particular, it enhances the quantum efficiency at long wavelengths with lower excitation

energy. In addition, it lengthens the mean escape distance (L) of excited electrons due

to the depletion layer [50].

Micro-channel Plate Photomultiplier Tube Concept

PMTs normally operate via a chain of dynode stages. After a photoelectric event,

each dynode stage increases electron production by a certain degree; the product of

these combined amplifications results in the overall gain. MCP-PMTs are different in

structure and operation from conventional discrete dynodes. An MCP consists of

millions of very-thin, conductive glass capillaries (4 to 25 microns in diameter) fused

together and sliced into a thin plate. The two surfaces of the plate are covered with thin

metal electrodes, and a voltage is applied to create an electric field in the direction of

the channel axes. Electrons emitted by the photocathode of the PMT enter the

capillaries. After a wall-electron collision, secondary electrons are accelerated by the

electric field and travel along parabolic trajectories determined by their initial ejection

velocity. As a result of repeated secondary electron emissions, the electron current

increases exponentially as the electrons travel through the micro-channel-plate [51].

The schematic structure of the micro-channel plate is shown in Fig. 2-14 along with the

principle of multiplication.

MCP-PMT Gain

L
The gain of a MCP-PMT is determined by the length-to-diameter ratio a = of a
d

channel, and the secondary emission characteristics gain factorG. The gain factor is a









product of the inherent characteristics of the channel wall material. The overall gain is

approximated as,

u = exp(G a). (2-7)

In general, it is possible to obtain a higher gain pu by increasing which also

moves the gain rising point to the higher voltage side. This is not without upper bound,

however, because at some point the noise begins to increase significantly due to ion

feedback events. An upper threshold for a =40 is usually selected so that a single MCP

provides a gain of 104 at 1 kV supply voltage [51]. A two-stage MCP-PMT can then

provide high enough gain to enable photon counting measurements, i.e. from 105 to 106

Employing thousands of independent amplifying micro-channels allows for a MCP-

PMT to avoid geometric saturation effects that can hinder more conventional high gain,

segmented anode, metal dynode chain photomultipliers. In the latter, any noise

photons that hit a particular portion of the photocathode can uniformly deplete the gain

in the corresponding anode. In an MCP-PMT, however, a noise photon can drain a

particular micro-channel (or set of micro-channels) without affecting the others in a

given anode.

When a single photon enters a MCP-PMT, the photocathode converts it into an

electron which travels to the anode while being multiplied. The transit time of the

resulting electron bunch differs depending on the input photon. The distribution of this

transit time is referred to as the transit time spread (TTS). In most photomultiplier

tubes, the electron transit time is dependent on the position of photocathode

illumination. When the entire photocathode is uniformly illuminated, the difference in

transit time with respect to position is referred to as the cathode transit time difference









(CTTD). The TTS is usually affected by the CTTD, but in the case of proximity-focused

MCP-PMTs, it has little effect [51].

During the multiplication process from cathode to anode in a metal dynode chain

PMT, the signal pulse can be broadened due to the emission-angle distribution and

initial-velocity distribution of photoelectrons and secondary electrons, as well as the

effect of the focusing lens. MCP-PMTs can largely ignore this effect because a strong

electric field is applied nearly parallel from the photocathode to MCPin and MCPout to

anode. Since the plate takes the place of a conventional dynode structure, the electron

transit time in the secondary electron multiplication process is also very short, allowing

for TTS up to 25 ps in Hamamatsu MCP-PMTs [51].

MCP-PMTs are available in small packages, with 100 anode configurations less

than 6 x 6 x 3.5 cm. In magnetic fields parallel to the tube axis, the tube can be stably

operated for fields up to 2.0 T. For fields perpendicular to the axis, the output drops

drastically if fields exceed 0.07 T. Current micro-channel plate implementations are

able to offer compact size, fast time response, and two-dimensional detection with high

spatial resolution, and stable operation in high magnetic fields [51].

Detection Efficiency and Pulse Height Distribution

The effective capability of a PMT to collect photoelectrons emitted from the

photocathode is called the collection efficiency (E). This value, combined with the

photocathode quantum efficiency (q), directly affects the measured signal. The ratio of

the number of output pulses (Nd) to the number of incident photons (N,) is called the

detection efficiency [51],









Nd (2-8)
NP

The output pulse height can vary depending on fluctuations in the secondary

electron emission ratio. In order to gauge the time-varying response of a PMT, it is

useful to consider a pulse height distribution (PHD) plot. This plot is a histogram of

accumulated pulse heights given a series of test data, shown in Fig. 2-15(A) [51]. A

typical PHD signal-to-noise comparison is shown in Fig. 2-15(B). Note that the noise

line exists even with no illumination of the photocathode; these events result from

thermionic emissions (thermal or dark noise events).

When the incident light level is increased, the amplitude of the signal + noise

increases as shown in Fig. 2-16(A). Increasing the supply voltage causes the PHD

curve to be stretched and flattened as shown in Fig. 2-16(B). These images illustrate

how detection efficiency is affected by threshold level. For example, the first valley in

the PHD waveform is usually selected as the lower level discriminator (LLD), and a

point midway down the falling edge of the waveform is selected for the upper level

discriminator (ULD) [50]. Since a large portion of the noise often occurs at lower pulse

heights, an appropriate setting can allow for filtering of thermal (dark noise) events. If,

for instance, the LLD is selected far to the right of the waveform peak, the majority of

signal events may not be passed once the output pulses are brought into a binary

signal.

Experimenting with LLD settings other than the initial valley is not without merit,

however, depending on the application. If a large amount of noise is causing signal

classification problems in data post-processing, one solution may be to increase the

LLD to a point where only multi-photon events have a high probability of being detected.









Re-collecting the data within a multi-photon state rather than a single-photon state can

have advantages for cleaning up very noisy signal as long as the signal provides a

strong response. Decreasing the PMT supply voltage can have the same effect; both

values have significant impact on the overall detection efficiency.

Variation in PMT Gain and Output Instability

The effect of changing the supply voltage can be seen by counting the number of

pulses with the LLD fixed. Although the analog gain increases exponentially with supply

voltage, the number of counted pulses remains fairly constant for a certain range. This

range is referred to as the plateau characteristic, and analyzing the associated plot

indicates an appropriate region for setting the supply voltage. An example of a plateau

characteristic is shown in Fig. 2-17, which also illustrates why the photon counting

method is relatively insensitive to small-scale supply voltage fluctuations. The signal to

noise ratio remains flat throughout the plateau range. Even when the gain changes

significantly as a result of supply voltage instability (Fig. 2-18), the photon counting

mode remains stable with respect to count rate [50].
























A B


Figure 2-1. Point clouds resulting from data collected by the DARPA Jigsaw program
sensor at MIT Lincoln Laboratory. A) Raw Jigsaw data colored according to
range relative to the sensor. B) View from inside the raw point cloud of
genuine signal and background noise [24].


Figure 2-2. Point clouds resulting from Jigsaw sensor data after filtering algorithms are
applied. Following preliminary cleaning of the 3D image, human subjects and
test-range objects are much more apparent than in the raw input [24].











E4, N4
Ra (raditationles,
Sransilion)
E3, N3 -- -- x -

P (pump L (
*rarisiTiorn I trar


E2, N2 -- -
Rb (radiationless
transition)
E1, N1 tnin-

Figure 2-3. Diagram of energy level transitions in four-level laser.


18

16

? 14

_ 12

10

8

6

4

2


I I I I
With Endcaps
V Without Endcaps






*
*
U-




200 250 300 350 400 4
Pump Width (ps)


Figure 2-4. Maximum pulse
laser [39].


repetition rate versus pulsewidth in SPMCL PQS microchip


















MS.Theis, 1981 lOnm
0l0nm
Ipm
lOpm
lOpm
Imm
10mm
100mm
Im



OMLC lOOm

Wavelength (nm)


10,000


S1,000


o 100


10


1OOnm lum IOum
Wavelength (nm)


- Illm
lOpm




IOO1m


1mm


I 2 3 5 T 10 20
Wavelength (pm)


C

Figure 2-5. Typical absorption coefficients of water as a function of wavelength [41]. A)

Segelstein, 1981; B) Hale and Querry, 1973; and C) Wieliczka, Weng and

Querry, 1989.


Outgoing Laser Pulse
0 3 5 : ---- c ---- ---- [ ---- ---- L ---- ---- c ---- I--I I
FWHM
., meters


Figure 2-6. Gaussian-shaped function generated in MATLAB to simulate a "long"

outgoing laser pulse. This plot corresponds to a "long" system pulse width of

10 ns at full width half maximum (FWHM), or 3 m (in air).


OOMLC


\ -









r Hale & Querry,
Appl Opt, 12,555(1973)


lOOpm

Imm

10mm
100mm m
'r




Im

10m


100pm


025


02


015


01


nn -


4 -3 -2 -1 0
-ete-s


1 2 3 4

























O surface / bottom separation 1 meter
2

0L.Z Z .--
0 20 40 60 80 100 12e
Reflectors
3-

2


S 200 400 600 800 1000 1201
Response


A

Simple Conpluton Idea Water Surface and Ocean Bottom Reflections





0 200 400 600 800 1000 1200
Outgoing Pulse

| o su40ace / bo.. om separatIon 4 mIers



0 50 100 150 200 250 300 350 400 450
14


12 -
0


C D

Figure 2-7. Simulation of ideal water surface and ocean bottom reflections for a "long"

pulse width (10 ns). Three subplots are shown for the outgoing pulse, "ideal"

reflectors, and resulting response. Depth is varied for four plots. A) Depth of

1 m. B) Depth of 3 m. C) Depth of 4 m. D) Depth of 6 m.


Outgoing Laser Pulse
FWHM
4 1441

















08 06 04 02 02 04 06 08 1
meters


Figure 2-8. Gaussian-shaped function generated in MATLAB to simulate a "short"

outgoing laser pulse width. This plot corresponds to 480 ps at FWHM, or

0.1441 m (in air).


4 suace / bottom separation pulse width (3 meters)
2

0 50 00 5Outgong P ul0 00 50 se






















6 -p r
2
4

0 200 400 600 800 1000 1200 140(


B

Simple Conlut on Ideal Water Surface and Ocean Bottom Refections






o 200 400 600 800 1000 120C
Outgoing Pulse

6
i 0 surface / bottom separation 6 meters
2






















o 0 /00 .-e / 50 200 250 300 35(t
2









A
Simpe Conuton dea Water Surae and Ocean Bottom Re ect on




0 50 100 150 200 250
Outgoing Pulse

aebottomeparatn 2 meters
12


C D
Figure 2-9. Simulation of ideal water surface and ocean bottom reflections for a "short"
outgoing pulse width (480 ps). Three subplots are shown for the outgoing
pulse, reflectors, and response. Depth is varied for four plots. A) Depth of 1
m. B) Depth of 0.144 m. C) Depth of 0.2 m. D) Depth of 0.3 m.


Table 2-1. Sample system parameters for use in LIDAR link equation.

nh nq n h v p, a A, Be, R

0.8 0.28 0.4 6.63 *1034J-S 5.64.1014 HZ 0.15 50 3.3.10-3 M2 0.297.10-3 M1 600 M


Ssui ace, bottom sepaatln 0 meter












* Engineering Flight Parameters
NASA P-3 Aircraft. Wallops I light Center
Locale: C'hincotcague. VA & Chesapeake Ba%
Flight Alliludes: 3.5 to 6.7 km (11.000 to 22.000 ft)
Early afternoon (maximum solar background)
Laser I nerg: < 2 pJ 'a 532 nm
Laser Repetition Rate: 3.8 kliz
Laser 'Power: -7 mniW
ELIecti\c Ielescope Diameter: 14 cm
Mean Signal Strcntllh per .aser ire: .88 pe



A


so 100 16O
Along Track Ostarce (mnters)


Tree Canopy Heights


Ground


One Story
Strrclture


0- Two Story
S Dwelling Tool Shed


10
a A -
S 3 310 320 3 3 340
Along Track Distance (m)


Buildings and Trees

B
10-



. Bay Surface

I
-10-
Bottom
0 10 100 150 200
Along Track Oitance (meters)


Shallow Water Bathymetry


Figure 2-10. Data from NASA Goddard Space Center's microaltimeter instrument. A)
Engineering flight parameters. Sample profiling data is shown from B) small
buildings and trees in the town of Chincoteague, VA, from 6.7 km altitude, C)
tree lines and underlying terrain from 3.4 km altitude; and D) shallow water
bathymetry over the Chesapeake Bay from 3.4 km altitude [23].


Figure 2-11. A typical return waveform from the Optech SHOALS bathymetric mapping
system, measured by a PMT detector [11].


PMT Waveform
$Vrlce Reohn
700 V
600
500 \"
E 400
300 --
200
100
0
0 50 100 150 200 250 300
Time











HIGHER LIGHT LEVEL (Multiple Photoelectron State)
IIIllllll 11111111 1111111l l1111111 ARRIVALOF PHOTONS
1 1 1 111111 II PHOTOELECTRON
EMISSION
Sfm sSIGNAL OUTPUT (PULSES)

SIGNAL OUTPUT
(PULSE OVER LAPPED)
Ml-TIME
LOWER LIGHT LEVEL (Single Photoelectron State)
I I I I I II I ARRIVALOFPHOTONS
S I I PHOTOELECTRON
EMISSION

S A A A SIGNAL OUTPUT
T (DISCRETE PULSES)
ie TIME

Figure 2-12. Output pulses from a PMT at
different light levels.


(a) HIGH (b) LOW (c) VERY LOW

Figure 2-13. PMT signal under varying
light intensities.


THIMfiMoEA
(a) Schematic structure of an MCP (b) Principle of multiplication
Figure 2-14. Schematic structure of a MCP-PMT and the principle of multiplication [50].












FREQUENCY
:LOW :HIGH :LOW


PULSE HEIGHT (AMOUNT OF CHARGE)


SIGNAL+NOISE


L
PULSE HEIGHT
D


A
Figure 2-15. Typical pulse height distribution plot (PHD) [50]. A) The histogram
variation is shown mapped to pulse heights from a train of incoming pulses.
B) Noise and signal are compared in a PHD plot.


PULSE HEIGHT


A


INCREASE IN


PULSE HEIGHT


Figure 2-16. PHD plots depicting an increase in incident light level (A) and supply
voltage (B).













7000

6000


5000

4000
i-
z
M 3000
0

2000


1000

0


S 096 0.98 1-00 1.02 1.04 1.06 1.08 1.10
RELATIVE SUPPLY VOLTAGE

Figure 2-18. Typical PMT output variation
versus supply voltage.


SIGNAL+NOISE


S/N RATIO



PLATEAU RANG




HIGH VOLTAGE SETTING

NOISE -'
yy ,-i----T----ry--r-- ""___


700 800 900 1000 1100 1200 1300 1400 1500
PHOTOMULTIPLIER TUBE SUPPLY VOLTAGE (V)

Figure 2-17. Typical PMT plateau
characteristics.









CHAPTER 3
LSNR PROTOTYPE DESIGN

Photon Counting System Prototype: CATS

Building upon the major concepts of Chapter 2, we now present a design for the

optical framework and receiver package to enable single-photon laser ranging.

General Scanning Principles

Laser mapping systems typically employ a scanning mechanism to direct the path

of outgoing laser pulses. The repetition rates (5 to 10 kHz) of commercially available

micro-lasers combined with the flight speeds of commercial aircraft (60 m/s) make it

necessary to illuminate a "patch" of terrain a few meters on a side to achieve contiguous

coverage of the terrain on a single pass. Expanding the size of the ground footprint can

be achieved by using a multi-element design. Instead of transmitting a single high

energy beam, the outgoing laser pulse can be separated into many low-energy

beamlets via holographic expansion to sustain coverage requirements. The result is

nearly overlapping footprints with small (decimeter) inter-footprint pixel spacing.

System Proposal

The availability of a cost-effective microchip laser and high gain MCP-PMT served

as a basis for the initial design of the University of Florida's Coastal Area Tactical-

mapping System (CATS). The CATS was built through a partnership with Fibertek, Inc.

and Sigma Space Corporation. This LSNR prototype instrument operates at low power

and low laser pulse rate while attempting to achieve contiguous ground coverage at a

spatial resolution of a few decimeters. In order to meet these goals, a receiver

processes returns at the single photoelectron level resulting from an array of outgoing

light pixels. While the main application is the mapping of coastal areas, including the









shallow depth shore, single photoelectron laser altimetry has demonstrated that CATS

could have advantages in developing high resolution digital topographic databases,

monitoring biomass and vegetation heights, sea and lake levels, ice sheet thickness,

etc. [23].

Optics

Fig. 3-1 shows a block diagram of the CATS optical path elements as they relate

to the transmit and receive paths. Fig. 3-2 shows a more detailed image with the

turning mirrors, mechanical shutter, receiver electronics, and fiber optic detector.

Several tip-tilt mirrors deflect the beam at right angles in order to reduce the overall

extent of the optical path. The chosen mirror coatings have 99.5% reflectivity at a

wavelength of 532 nm.

Microchip Laser

In the CATS, a NanoPulse class IIIB microchip laser manufactured by JDS

Uniphase produces high repetition rates, high peak power, and short pulses. The laser

cavity consists of a Nd:YAG gain medium bonded to a YAG-saturable absorber.

Pulsing is achieved through PQS. The cavity mirrors are vapor-deposited directly on

each side of the crystal to form a monolithic oscillator that is end pumped by a CW

diode laser. The NanoPulse laser product line shown in Fig. 3-3 generates 1064 nm

output and can be optimized for either 50 mW of average power or 8 pJ pulse energy.

In either case, the output is a pulse train of high peak power, sub-1 ns, near-diffraction

limited pulses. The beam diameter is 0.2 mm with 2 mrad divergence. These lasers

are capable operating at temperatures from 0 to 500 C at the base plate, and can be

stored at temperatures from -200 to 600 C. The power stability over 6 hours was









measured to be 3%, the ellipticity ratio is 1.1:1, and the warm-up time is approximately

10 minutes.

A variation of the NanoPulse laser, called the NanoGreen, utilizes a potassium-

titanyl-phosphate (KTP) frequency doubler at the same output as the crystal that is used

for the NanoPulse to produce 532 nm pulses. An IR blocking filter is added to remove

residual 1064 nm light. The PER is greater than 100:1 and the M2 mode approximately

1.2 for a Gaussian beam profile. Field tests for beam polarization are shown in Fig. 3-4.

The NanoGreen can be specified for up to 15 mW of average power or 3 pJ of pulse

energy. The entire laser assembly is contained within a compact 15 x 4 x 4 cm package.

Each manufactured laser has slightly different output characteristics. For the

operational CATS laser, a thermal head sensor has measured 27 mW at the direct laser

output (7.76 kHz), which corresponds to 3.48 pJ / pulse.

Beam Expander

Pulsed light passes through a beam expander, decreasing the beam spot size at

large distances. There are two main types of beam expanders: Keplerian and Galilean.

In the latter, the combination of a positive and negative focal length lens provide

magnification of the beam diameter and then reduction of the divergence. Keplerian

types achieve the same effect of producing a larger but highly collimated beam size, but

with two positive focal length lenses. Contrasting the two, a Galilean design contains no

internal focus point and the overall length of the Galilean as compared to a similar

power Keplerian is smaller. In the CATS, a plano-concave and achromatic lens

produce 10x total magnification; this reduces the beam divergence to 0.2 mrad.

BK-7 lens material with magnesium fluoride (MgF2) coating is chosen for its

compatibility with visible light applications, along with an anti-reflection coating. After









the beam expander, each pulse is passed through a holographic diffraction element that

separates the beam into a 10x10 square array of beamlets rotated 450. The efficiency

of the hologram is approximately 80%. A lens in the transmit path is shimmed to

increase the amount of defocus and resulting angular divergence of laser spots in the


far field. The spot sizes then have 77% linear fill factor at the 2 beam contour, or 46%
e

area fill factor for 95% of the total energy. Outgoing laser beamlets pass through a

shared afocal element resulting in a total magnification of 4x.

Risley Prisms

Laser light is deflected out of the shared afocal element through two Risley prisms,

shown as cross-hatched elements in Fig. 3-5. Each Risley prism is simply an optical

wedge V-coated for 99.5% light transmission. The refractive properties of this material

combined with the wedge surface angle deflect the beam through a predicted path.

BK7 material is selected with a refractive index of 1.52, wedge angle of 13.580, diameter

of 88 mm, and central thickness of 22.5 mm. Each incremental wedge rotation is slaved

directly to the leading edge of a synchronization pulse associated with the fire of the

laser. The wedges are set to rotate a constant angular amount between

synchronization pulses, and can be independently controlled to move at varying speeds

in either direction. Software routines created prior to system operation determine the

frequency of the two rotations.

This allows for the array of laser beamlets to be deflected in a variety of repeatable

patterns. The line scan in Fig. 3-6(A) closely approximates the familiar raster or

sawtooth scan, with platform motion responsible for creating the along-track

displacement. Raster scans are estimated to produce a maximum 225 m swath width









at an altitude of 600 m. An example of a different scan pattern available with the Risley

prism design is shown in Fig. 3-6(B). By slightly offsetting the rotation speed of one of

the prisms, the line scan can be made to rotate over time. This type of scan results in

much denser coverage of points near nadir. Different types of patterns can provide

enhanced coverage for specific applications. A test of time deviation of the nearest

laser synchronization pulse from the wedge reference pulse indicates that the pulse

position on the ground deviates by approximately 0.07 pixels from scan to scan; an

acceptable margin.

Range Gate

The system is operated with a range gate of 100 to 150 m. The supply voltage on

the tube provides full amplification only after a set amount of time has passed (related to

the range gate setting); meaning that the power is temporally stepped in relation to the

laser pulse train. This allows for the outgoing laser pulse to be sufficiently far away from

the output window as to minimize electronic noise variables associated with laser fire. It

also eliminates atmospheric backscatter immediately outside the aircraft and ensures

that strongly reflective targets near the output window (atmospheric reflections, birds)

do not produce a response. This helps to prolong the life of the photomultiplier tube.

Spatial and Spectral Filters

Resulting backscattered radiation passes back through the shared afocal element.

Dust particles residing on optical surfaces may produce scattered laser light resulting in

interference patterns. These patterns can produce contrasting background light that

interferes with desired signal information. A simple way to deal with this interference is

through a tight spatial filter. The idea here is to reduce the receiver field of view to the

minimum value which encompasses the transmit beam divergence and potential









pointing variations and/or co-alignment errors with the receiver [15]. Scattered light can

be blocked by centering a small aperture at the focal spot of the direct beam, while

allowing the direct beam to pass through. The resulting cone of light has a smooth

irradiance distribution and can be refocused to form a collimated beam that is almost

equally smooth.

The laser returns are also passed through a spectral filter to further improve return

signal quality before reaching the timing electronics. For systems operating at low

signal to noise ratios, it is necessary that filtering operations be precise. In the CATS,

an interference filter accepting wavelengths centered at 532.1 (+/- 0.05) nm with FWHM

0.25 (+/- 0.05) nm is used. The filter has a temperature coefficient of 0.002 nm / C. To

further reduce scattered light reaching the active area of the PMT, a central mask is

placed on the spectral filter to block out on-axis backscatter.

Dove Prism

The pattern of filtered beamlets is rotated 450 by a Dove prism. This optical

element is made from truncated right-angle prisms such that a beam of light entering

one of the sloped prism faces undergoes total internal reflection (TIR) from inside of the

bottom face and emerges from the opposite sloped face in flipped image orientation

[53]. By rotating the prism about its longitudinal axis, the transmitted image rotates at

twice the rate of the prisms; meaning that they can rotate a beam of light by an arbitrary

angle (450 in this case). The losses occurring as a result of the entrance and exit faces

are reduced to a small fraction of a percent by using a dielectric V-coating for both

surfaces.









Telephoto Lens and Mechanical Shutter

The resulting pixels are refocused by a telephoto lens before being imaged onto

separate cathodes of a 10x10 segmented anode photomultiplier tube. Detected signals

are sent to a 100 channel multi-stop ranging receiver. A mechanical shutter before the

telephoto lens ensures that the photosensitive material is not subject to stray light

during system transport and setup. The scanner window is also tilted relative to the

optical axis to minimize reflection of photon back into the receiver. The receiver field of

view (FOV) is a 3.5 m square. This is a larger aperture than the diameter of the

transmitted bundle of beamlets.

MCP-PMT

Because a very low amount of reflected radiation is expected from each laser

beamlet, a detector must be employed with high enough gain to amplify single photon

events. In order to achieve 105 gains with 200 ps rise time, 25 ps transit time spread

(TTS), and 200 ps dead time specifications, a micro-channel plate photomultiplier tube

(MCP-PMT) was selected. This device is capable of simultaneously detecting

independent events from a 10x10 pixel array, allowing for two-dimensional resolution.

The MCP-PMT provides exceptional gain while maintaining compact size.

In the CATS, a 10x10 segmented anode array built by Hamamatsu Electronics Inc

(model number R4110U-74-M100D), shown in Fig. 3-7(A), is capable of sensing single

photon events. The aluminum housing of the MCP-PMT measures 5.330 x 5.330 x

3.175 cm. Three pins on the narrow end of the housing fit into a PCB that controls the

high voltage applied to the tube. Detector signals are routed through a PCB attached to

the back of the housing. Taking into account the aluminum mounts for these

components, the combined volume of the detector module is less than 1000 cm3. A









GaAsP (gallium-arsenide-phosphide) crystal activated with cesium is used as a

transmission type photocathode, meaning that photoelectrons are emitted in the same

direction as the incident light. This type of photocathode has high quantum efficiency

(QE) in the visible region [51], with a plateau curve from 300 to 850 nm. The QE plot for

the R4110U is shown in Fig. 3-7(B). The nominal photocathode uniformity, MCP gain

uniformity, and crosstalk values are 76%, 70%, and 3.8% respectively. The dark noise

is expected to be less than 270 counts per second at 4.5-105 gain.

Multi-stop Event Timer

Accurately detecting scattered light returns using single photon events can be

complex. The negative impact of reduced signal levels relative to conventional

altimeters is largely offset by the corresponding reduction in the detection threshold to

about one photoelectron. Instrument sensitivity in this case is maintained or even

enhanced [15]. The caveat to higher sensitivity is the increased presence of noise

effects. Reduction of the detection threshold to approximately one photoelectron allows

for robust detection at lower light levels. However, even with a range gate, narrow

spectral filter, and tight spatial filter, typically there are sufficient noise events from

atmospheric backscattering (especially in areas with atmospheric haze or fog) to

warrant the use of a multiple stop event timer.

Noise events can be separated from terrain or vegetation returns because they are

expected to be randomly distributed in time, or temporallyy incoherent," in comparison to

temporallyy coherent" signals which rise above the noise floor [15]. This permits single

photon detection in daylight with wide range gates as well as multiple returns per pixel

per laser fire from volumetric scatterers such as tree canopies or turbid water columns.









More information on the processing electronics associated with multiple-stop event

timing is found in the Appendices.

Multi-stop LIDAR sensors are capable of providing information about biomass in

forest understories related to sunlight distribution [11], surveying the sea surface / floor

in navigation channels, ports and harbors, coral reefs, beaches, and shorelines [11],

[47], and classification between spatially mixed topographic features in urban and

suburban areas [5]. Man-made objects that would be obscured to pilots in mission fly-

overs such as tanks concealed in heavy brush or camouflage netting are able to be

identified via observations of spatial structure in LIDAR point clouds [54].

Assembly and Footprint Dimensions

Fig. 3-8(A) shows the CATS sensor mounted on a tripod with laptop and power

supply connections supported by an equipment rack. A close-up of the optics, enclosed

scanner, and electronics cube contained within the black-colored metal assembly is

shown in Fig. 3-8(B). Fig. 3-8(C) shows a zoom-in on the communications ports for

laser and scanner operations. Fig. 3-8(D) shows communications ports for laser,

timetag outputs, and range data throughput. Fig. 3-8(E) / (F) show the front / back of

the scanner interface module.

Flying at nominal altitude (600 m), the system is expected to produce a horizontal

spatial resolution of 20 cm at nadir. Fig. 3-9 shows a photo of several buildings in

Lanham, MD. These structures were used to verify footprint dimensions. The diamond-

shaped footprint is projected onto a brick wall in Fig. 3-10.

Manual measurements of the footprint projected onto the brick surface at night

produced a side length of 0.99 to 1.01 m, an average of 1.00 m. The center-to-center

spot spacing was 9.52 to 10.16 cm, an average of 9.84 cm. The spot size diameter was









4.45 to 5.08 cm, an average of 4.77 cm. In addition to a CATS distance measurement,

the range to the wall was measured using an Optech ILRIS laser mapping sensor, and

by Google Earth satellite photography. The average distance to the brick wall by these

three independent measures was 303.76 m. Extrapolating these results to 600 m, the

average side length would be 1.97 m, the center-to-center spot spacing would be 19.44

cm, and the spot size diameter would be 9.42 cm.













transmit path


receive path


Figure 3-1. Block diagram of optical path layout.


Tip / Tilt
Mirror A


Sensor Window c


Risley Prism
Scanner






Annular Mirror


Holographic
Diffraction
Beam Element


Mirror


Shared
Afocal


Tip /Tilt
L Mirror


SSpatial
Filter

Spectral
Filter


Telephoto


Figure 3-2. Detailed optical path with turning mirrors, mechanical shutter, receiver
electronics, and fiber optic detector.






















Figure 3-3. Photo of JDS Uniphase NanoPulse lasers.


















no film film: film:
min transmission max transmission

A B C
Figure 3-4. CATS beam polarization test. A) Effect of rotating linear polarizing film
square is shown without attenuation, B) perpendicular to polarization axis and
C) parallel to polarization axis.




























Figure 3-5. Isometric and cutout views of scanner head. The scanner window is tilted
relative to the optical axis to minimize reflection of photon back into the
receiver.
















A B
Figure 3-6. Lab trials of scanner, showing A) line and B) rotating line scans.


1000


E

10

ii 5
||

Q


200 300 400 500 600
Wavelength [nm]


700 800


Figure 3-7. A) Photo of Hamamatsu Electronics Inc.
PMT and B) quantum efficiency curve.


model R4110U-74-M100D MCP-



















































E F


Figure 3-8. Photos of the CATS and related components. A) View of sensor mounted
on tripod with laptop and power supply components. B) Output window,
enclosed scanner, electronics cube, enclosed optical components. C)
Communications ports for laser and scanner operations. D) Communications
ports for laser, timetag outputs, range data throughput. E) Front and F) back
of scanner interface module.



















Figure 3-9. Photo of several building structures in Lanham MD, where tests were
conducted to observe the footprint of the CATS sensor. A brick wall and bank overhang
are indicated.












A B C
Figure 3-10. Footprint projected on a brick wall at a distance of 303.7570 m. A) Side
view. B) Front view. C) Close up of four pixels.









CHAPTER 4
LSNR PERFORMANCE METRICS

Experimental results from the CATS prototype have yielded proof-of-concept

LSNR operation and techniques for improving data processing related to system effects.

Because the CATS is able to process multiple returns and has the ability to change

sensitivity level between datasets, its penetration and noise reduction capabilities were

assessed in a variety of different test setups.

There are a number of situations that can create difficulties for classic first / last

return differentiation techniques. For example, Jutzi and Stilla describe sloped surfaces

within the footprint of the beam leading to shifted mean distances, different elevated

areas occurring at building walls leading to dilation and erosion, slightly different

elevations leading to range superimposition, and small objects distributed randomly

about a mean elevation value leading to a widened, low intensity pulse that is

problematic for a fixed threshold analog system [55]. The ability of post-processing

methods to overcome these difficulties in the LSNR case is dependent on measurable

system properties. A number of experimental configurations were designed to

investigate performance metrics for the CATS related to minimum separation distance,

water penetration depth, and atmospheric backscatter [56].

Laser Ranging Concepts and Implementation

Range Calculation Fundamentals

In order to determine the range to a target, the two-way propagation delay of a

laser pulse is measured. The ability of a ranging receiver to record multiple unique

events, i.e. multiple canopy returns, is dependent on the length of the transmitted laser

pulse. The pulse length contributes to a sensor's effective "blind zone," which refers to









a region after a registered return in which no subsequent returns can be detected [57].

The blind zone extent is determined largely by the light propagation speed c and the

pulse length r. In terms of estimated terrain resolution, this creates a distance below

the initially reflecting object in which no other objects can be detected. For a 10 ns

pulse length, the blind zone distance is,

cr = 10 10 9(s) 3 108(m/s) = 300(cm). (4-1)

The CATS sensor utilizes a short pulse length in comparison; approximately 480

picoseconds FWHM, which results in

cr = 480 .10 12(S) *3 -10'(m/s) = 14.4(cm). (4-2)

Considering only pulse width restrictions, the instrument should be able to

discriminate a distinct second object if it is 14.4 cm away from the first.

The overall resolution capability of a LIDAR system is not solely a function of

transmitted pulse length, but rather the product of simultaneously occurring factors.

Photodetector recovery time is another important element to consider. If the detector is

not capable of a recovery faster than the amount of time related to the blind zone

distance, the former will constrain the system specifications. The CATS photodetector,

for example, has a recovery time of approximately 210 ps, which ensures that it is not

the limiting factor.

Range Discretization

In the CATS, the laser pulse propagation delay is measured in discrete time

intervals called range bins. A registered return is assigned a range bin number (RBN)

based on the measured time since the transmission of the laser pulse. One of the









factors in the final range uncertainty is the length of the range bins used. This bin

length is related to the effective range resolution.

To complement the sub-nanosecond pulse length of outgoing laser pulses, the

internal system clock is set to 2 GHz. The range bin length is 1 / 2 GHz, or 0.5 ns.

Accounting for the two-way travel time of the laser pulse, this makes the range

resolution for consecutive single-photon events,

0.5-10 9(s) 3.108(m/s) (4-3)
tre = = 7.5(cm).

In order to eliminate noise returns associated with backscatter as the laser pulse is

just exiting the output window and to protect the PMT from being overexposed to

returns from close proximity targets, a range gate is used to temporally filter responses.

The range gate setting (RGS) is an offset to two-way travel time beyond the range gate

value, in nanoseconds. The total two-way travel time and the related range are,

t2- = RGS 1- 109 + RBN- 0.5 -109 (s), (4-4)

r RGS--10-9 +RBN 0.5-10-9
-way 23-108 (m).
l aWY 2

Data Processing Fields

CATS range information is bundled into a binary file format with the .SPR

extension. The raw files contain the following information about recorded returns:

* shot number
* channel number
* hit number
* range information
* time tag information

The first four fields are self-explanatory for the stream of data being recorded in a

multi-element, multi-stop laser ranging system. The goal of the fifth field global









positioning system (GPS) time tag information is to reconcile the position of the

scanner at any given time in relation to the temporal sequence of laser shots in a data

set and Risley prism wedge index marks. Incorporating this data field allows for

analysis of potential data flow problems, e.g. if the laser fire slaving system happens to

miss synchronization pulses, or if scanner rotation is malfunctioning. Once correctly

implemented, time tag information greatly aids in post-processing and potential

troubleshooting. This idea is further developed in Appendix B.

Ground-based Static Testing

Experimental Setup

Practical testing of the instrument was conducted on the rooftop of a building in

Lanham, MD-approximately 15 m tall. The sensor was mounted on a tripod with

associated power and communications components (Fig. 3-8). The output window was

pointed at buildings, vegetation, and other targets 200 to 800 m in the far field.

Initial data collections focused mainly on static pointing of the footprint, to measure

standard deviation of measured ranges. Several attempts were made to range to

nearby buildings. To establish static straightforward orientation of the laser footprint,

the two prisms were set 1800 out of phase so that the laser would be pointed straight

along the optical axis of the transmit / receive path. C++ range processing code and

MATLAB graphical capabilities were utilized to give a semi 'real-time' look at the limited

data runs made on the rooftop in hopes that testing modifications could be made

without incurring the delay of post-processing steps. PMT control software also

provided a simple graphical output showing accumulated channel returns in a 10x10

intensity grid every second.









Range Accuracy

Ranging to a brick wall provided a calibration test for the consistency of returns per

channel. Coding techniques for ranging are further explained in Appendix B, and the

significance of a slight calculation error and oscillator variation is presented. The range

gate value was incorrectly entered into software routines for the initial processing, so 40

m was later added to results to determine absolute range values. Relative

measurements are still applicable, however. Hits were filtered for returns from 264 to

266 m to discard any noise returns not associated with scattering off the brick. The

associated statistics and images are shown in Fig. 4-1.

At one standard deviation from the mean, 99% of the recorded range data falls

within the 10 to 20 cm level. The outlier at channel 33 only registered 6 hits, a very

small sample in relation to other channels. The distribution of channel hits, on the other

hand, showed significant channel-to-channel deviation. The images in Fig. 4-2 show

the orientation of the channels in the footprint, and the relative hit counts related to

footprint geometry. In Fig. 4-2(A), hit counts are associated with height are plotted to

emphasize variation in response levels. In Fig. 4-2(B), colors associated with these

heights are more easily identified for spatial trends. Of note is that the bottom left row

and far right row of channels are where most of the low hit count problems occur; this

was theorized to be related to a problem aligning the reflected beamlets to the PMT

anodes. Re-alignment procedures improved later ground-based data collections.

Minimum Separation Distance

In the case of a partially occluded laser beam, a multiple-stop receiver allows for

recognition of multiple objects within the same laser shot. Given the 250 ps MCP-PMT

recovery time and 0.5 ns (1 / 2 GHz) timing bins, it is expected that the "dead time" -









time before a subsequent multiple return can be registered will be dominated by the

480 ps laser pulse width. More specifically, the system dead time should theoretically

be limited primarily by the pulse width (14.4 cm in the atmosphere). Experiments were

conducted to verify this value; the minimum separation distance (MSD) required to

discriminate multiple scattering objects. Measurement of effective dead time

establishes a boundary for sensor penetration.

A controlled experimental setup was developed for analysis of MSD in the

situation where multiple hits would be necessary for target discrimination. The system

was mounted on a tripod and pointed at the structures shown in Fig. 4-3. A 2 cm thick

piece of sanded plywood was placed 1 m below a mirror approximately 560 m from the

CATS to act as a calibration surface. At this distance, the 10x10 array of laser beamlets

produced a rotated square footprint measuring approximately 2 m on one side. A

central section of the large diamond-shaped array (shown reflecting off the chain-link

fence in the background of Fig. 4-3) was intercepted by a cardboard mask. A square

hole in the mask allowed transmission of only four beamlets to a 450 turning mirror

located behind the cardboard. The mirror redirected the beamlets from the horizontal

propagation direction to the vertical, and onto reflecting surfaces below.

The cardboard mask assisted in channel discrimination. Returns from the mask

were filtered out temporallyy), allowing for post-processing analysis to more easily

identify desired signal responses. Half of a beamlet cross-section was projected onto

the wood calibration surface, and the other half projected at a varying distance above

on a similarly reflective wood sheet. Fig. 4-4 illustrates the intercepted surfaces.









Seven ranging trials (A-G) were conducted at various separation distances (Actual

Separation column in Table 4-1). Data was recorded for 10,000 laser shots (1.25 sec)

in each trial. Returns were filtered for a coarse 5 m window around the central

response to include contributions from atmospheric returns near the primary reflecting

surface. The return signal at the calibration surface (0 cm separation) was

approximately two range bins wide-as expected for 0.5 ns timing bins, 480 ps pulse

width, and a flat optical cross-section. Data was thus retained for laser shots reporting

three or more total intra-shot returns, as hits beyond the second return were potentially

secondary surface returns. For each laser shot, the differential range bin value (RBV)

between the first and last hit was calculated to identify the secondary surface distance,

shown in Fig. 4-5 plots. The analysis process for trial D is described here. In Fig. 4-

5(A), large (>10 RBV) data values occur sparsely, meaning that they represent

atmospheric returns. These peaks resulting from 3+ return shots account for less than

5% of the total data in trial D. Conversely, small (<5 RBV) data values occur frequently

and at a consistent level, meaning that they represent secondary surface returns.

In Fig. 4-5(B), the differentials were accumulated into a total count per range bin

distance (AD) and displayed in a histogram. The third column (Reported Separation) in

Table 4-1 was generated from the peak histogram value in each trial. The fourth

column in Table 4-1 shows the error in secondary surface measurement for each trial.

Results from trial A (zero separation) are included to show that a single target produces

a signal response less than three returns wide. The 2.6% of shots reporting multiple

hits with 3 or more returns in this trial are attributed to atmospheric scattering.

Increasing the separation of the two surfaces beyond 10 cm, as in trial B, begins to









produce significant (non-atmospheric) multiple returns. Because the resolution limit of

the timing electronics is 2 GHz, the smallest surface that can be detected is the nearest

range bin edge, or 15 cm resolution.

Though the timing oscillator is expected to be stable about 2 GHz, it is possible

that the central frequency drifts over time with extended system usage. The

significance of potential drift is further explored in Appendix B. In order to evaluate this

empirically, the error values in Table 4-1 were calculated for known target separations.

To determine a systemic range bias, the effect of imposing a constant offset on range

estimation was tested. Table 4-2 shows analysis of four values of range bias in each

trial-5, 4, 3, and 2 cm. The values in each column are calculated by taking the

absolute value of the Reported Separation (RS) and Actual Separation (AS) difference

in each trial. The average error for each offset is then calculated.

Analysis of distributed returns allows for identification of ground reflections under

dense foliage and penetration past the water surface. Given CATS receiver

specifications, dead time is theoretically limited primarily by laser pulse width.

Assuming that the PMT is not saturated by the return pulse and that the intercepted

cross-section is not vertically elongated, experiments show that a secondary surface

can be detected at a distance of 15 cm from the first. Range errors in secondary

surface identification for six separate trials were within 7.5 cm, or within the timing

resolution limit of the system. Systemic bias testing showed that an offset of 4 cm

minimized the average error in secondary surface measurement (down to 1.3 cm).

Crosstalk Evaluation

In the CATS, backscattered radiation from a 10x10 square of beamlets is imaged

onto an equally spaced grid of photocathode material. Photoelectric events amplified









from each grid section are collected by separate anodes. Because all of the sub-

beamlets are transmitted at once, the MCP-PMT must be capable of simultaneous

sensing over each element in its grid. It is expected that each anode response will

provide unique XY position information about intra-footprint responses. The spatial

resolution of the system with regards to a single photon event is dependent on its ability

to accurately register independent responses from individual anodes. A concern in

adopting this type of detector structure is the influence of crosstalk on recorded

observations.

Multi-anode PMT designs are commercially available with the following geometric

plate layout: squares of photosensitive material associated with each anode are

separated by narrow borders of non-conductive material in a single plate. Fig. 4-6

shows the output profile of a multi-anode MCP-PMT when a light spot of approximately

20 pm diameter is scanned across the plate.

Light incident on a singular portion of the photocathode material may produce a

response at an adjacent anode-crosstalk between anodes. The overlapping regions of

the waveforms in Fig. 4-6 are crosstalk areas. The amount of crosstalk between

adjacent anodes of the photomultiplier tube contributes to the uncertainty of resolving a

single return at the specified spatial resolution. This can result from broadening of

electron flow when light is converted into electrons, or incident light spread within the

faceplate. Testing of two types of Hamamatsu R5900 series photomultiplier tubes

showed fairly low (~5%) nearest-neighbor crosstalk and even lower (1%) next-nearest-

neighbor measurements [58]. These experiments indicated that crosstalk is mainly due

to the spreading of optical photons in the entrance window; the M64 PMT used had a









1.3 mm thick glass entrance window. Reduction in window thickness to 0.8 mm further

reduced inter-channel crosstalk. In comparison, the R4110U model PMT used in the

CATS system has a 4.77 mm thick window; the observed crosstalk for this device may

thus be higher than 5%.

The contributions from overlapping waveforms can be mitigated by restricting the

area of focused energy. In the CATS, a lens in the transmit path is shimmed to

increase the amount of defocus and resulting angular divergence of laser spots in the

far field. The spot sizes then have 77% linear fill factor, or 46% area fill factor. A typical

CATS pixel energy distribution is shown in Fig. 4-7. The teal contour line (and

superimposed dotted circle) corresponds to 50% of the peak amplitude (FWHM). The

blue line is the 20 contour, or 95% of the total energy. Focusing backscattered energy

to a smaller area on the active photocathode has the potential to reduce crosstalk, but

can also reduce the effective lifetime of the tube.

In order to practically evaluate crosstalk contributions in the CATS, the setup

pictured in Fig. 4-8 was created. Four pixels of the footprint were masked to create

temporal spacing in receiver returns between pixels outside the central square. Pixels

19, 20, 23, and 24 were isolated for this experiment. For each laser pulse, returns were

filtered temporally for a 5 ns window centered on the wood signal response. This

separates any possible stray returns from the mask. Of the 8 surrounding pixels, 5

channels are considered to get a measure of pixel-to-pixel crosstalk. For example,

channels 18, 22, 30, 31, and 32 are considered in the channel 19 crosstalk analyses.

The contributions from these channels are labeled as a18-a32 in Table 4-3 to express

geometric relation within the CATS footprint (Fig. 4-9).









An experiment was first performed for multi-photon sensitivity (2300 V supply), or

approximately 2-105 gain. The dark noise count at this gain is expected to be 119

counts per second (cps). Over 10,000 shots or 1.3 seconds of data, on average 25,000

returns were registered for each signal pixel (2-3 hits per shot). Each surrounding non-

signal pixel registered on average 150 returns. This is close to the dark noise

specification at this gain. Hits from channels a18- a32 were accumulated and divided by

the number of total non-signal pixels (5). This value was 0.17% of the accumulated hits

from channel 19 (100% level). Average crosstalk at the other three channels was

calculated as the following: channel 20- 0.62%, channel 23- 0.22%, and channel 24-

0.65%. For multi-photon sensitivity, crosstalk between receiver channels is small

relative to MCP-PMT design specifications (3.80%).

The same analysis was performed for single photon sensitivity (2500 V supply), or

approximately 8-105 gain. Over 10,000 shots or 1.3 seconds of data, on average

45,000 returns were registered for each signal pixel (4-5 hits per shot). Each

surrounding non-signal pixel registered on average 5,000 returns. This is an order of

magnitude larger than the expected dark noise level at this gain, so channel crosstalk is

assumed to have significant effect here. The accumulated hits from a18- a32 in this

case were on average 12.22% of the accumulated hits from channel 19 (100% level).

Crosstalk at the other three channels was calculated as the following: channel 14-

7.24%; channel 22- 15.39%; and channel 23- 9.96%. For single-photon sensitivity, the

measured crosstalk is in some cases nearly four times as large as PMT specifications

(3.80%).









Because of the large disparity between values for example, channel 22 exhibited

more than two times the amount of crosstalk than channel 14 it was expected that

sub-pixel backscatter was not properly aligned to the PMT grid. Even a very slight (0.1

mm) offset in the tube alignment can cause one channel in the grid to receive

significantly more light than a neighboring channel. Re-alignment procedures were

carried out to ensure that the backscatter was being uniformly projected on the grid

before conducting airborne testing.

Water Penetration Depth

The potential for bathymetric mapping of near-shore waters was evaluated by

recording ranges through water columns of variable depth. To gauge the signal

response in a wide breadth of bathymetric scenarios, the optical clarity of tested waters

was also altered.

A 55 gallon, 90 cm tall barrel was placed below a turning mirror so that four pixels

of the CATS footprint were distributed down into the barrel (Fig. 4-10(A)). With the

barrel filled with 90 cm of distilled water, the barrel bottom (painted white for optimal

reflectivity) was clearly visible to the human eye. A 1800 backscatter turbidimeter

(infrared) reported an average value of 1.5 nephelometric turbidity units (NTU) for this

sample. The barrel was also filled with natural water transported from the Gulf Coast.

This salt water was skimmed off the top of a near shore area. The turbidity of the

sample was measured at 18.6 NTU. The white barrel bottom was no longer visible to

the eye at a depth of 43 cm.

Measurements were performed at night to aid in positioning of the 10x10 footprint

on the cardboard mask / large turning mirror shown in Fig. 4-8. During the day, the

laser reflections were sufficiently weak at 600 m that they were not visible to the naked









eye; meaning that precise positioning of a single channel down through a water column,

for example, was not possible. Operating at night did have the benefit of reducing

potential solar events in these trials, but previous experiments have shown that daytime

operation has minimal effect on SNR for wood surfaces at 550 m.

Because the location of the barrel bottom was precisely known 60 cm water

column depth for this experiment, a narrow range window corresponding to 3 bins

around the expected response was chosen. This made it possible to accumulate a

history of returns related exclusively to the barrel bottom response, rather than including

sporadic returns from the water column and potential noise events (by using a wide

range window). For each laser shot in the accumulated return history, data from a

selected channel corresponding to a beamlet projected down into the barrel was

considered only within the narrow range window. For example, if on average 70% of

the transmitted laser shots registered one or more signal events from the barrel bottom,

this means that 300 out of a total 1000 accumulated laser shots did not register at least

one hit within the selected window.

For 60 cm of this clear water, on average 97% of the transmitted laser shots

registered one or more signal events from the barrel bottom. Estimating the beam

attenuation coefficient from this data and extrapolating the signal intensity to further

depths, it is expected that approximately 90% of the incoming beamlets would register a

signal event at 5 m. Likewise, at a depth of 50 cm in coastal water, on average 12% of

shots registered at least one signal event. Extrapolation of experimental results to

further depths leads to the suggestion that this is a sufficient signal level to produce

meter-scale shallow (2 to 5 m depth) bathymetry data [56].









A 22.7 kg bag of sand was poured into distilled water. Data was collected while

silica particulate was settling down into the water column. This was a heavily turbid

suspension; visibility to the eye was approximately 7.5 cm. Histograms from a data

window of 10,000 laser shots are shown in Fig. 4-11 for three different supply voltages.

The blue and red signals are spaced approximately 15 range bins peak-to-peak; close

to the height of the water column after accounting for the change in the index of

refraction. The blue signal is thus associated with the water surface return, the green

from the middle of the column, and the red from the bottom of the barrel. A plot of all

four channels merged together (left image) and a single channel (right image) are

shown to illustrate response consistency.

Although detection of small low reflectivity targets submerged in highly turbid

waters may be difficult, the density of data points from the sea bottom given 20 cm pixel

spacing should be sufficient to establish near-shore depths (up to 5 m) at coarser

spatial resolution (1 m scale). For example, if the signal strength at 5 m degrades such

that 4% of the pixels within a footprint produce ocean bottom returns, a 600 m flying

height, 8 kHz laser rep rate, and 60 m/s platform velocity is enough to reconstruct the

digital terrain model at 1 m grid spacing. Partially submerged targets with intercepted

cross-sections smaller than 0.25 m2, however, may not be detected. Additionally, for

the heavily turbid case, the multi-photon state (2300 V or 2.5.105) was not sufficient for

feature identification. Extraction of the bottom signal in a heavily turbid suspension

necessitated maximum MCP-PMT gain (2500 V or 8.105).

Multi- vs. Single-Photon Sensitivity

Varying levels of photomultiplier tube sensitivity were assessed to weigh noise

production versus variations in supply voltage. For example, Fig. 4-12 displays stacked









histograms of accumulated atmospheric events per channel for a 1.25 second data

capture. Changing the supply voltage from a multi-photon state (blue) to a single-

photon state (red) drastically affected the recorded number of atmospheric scattering

events across most channels.

Atmospheric backscatter as a result of laser light propagating through an 8 [is

range gate and sky irradiance was measured. The sensor output window was pointed

towards the sky on a clear day and returns recorded for the duration of the range gate.

This procedure was repeated for three different PMT voltages, or gain settings of

2.5.105, 3.5-105, and 8-105. Fig 4-13(A) displays results from all channels combined,

and Fig. 4-13(B) displays results from one channel. Blue colored bins correspond to

events early on in the gate, and red to events near the end. There were significantly

more events recorded early in the range gate at the single photon level; the theorized

cause was a switching problem related to the high voltage supply. The majority of the

gate showed a fairly constant level of returns, however.

The contributions from laser light / diffuse solar irradiance scattering by

atmospheric particulate and direct solar irradiance were significantly increased for the

central 32 channels as gain increased toward a single photoelectron threshold. This

was an expected result as atmospheric events are generally single photon events, and

reaching a single photon threshold required full gain (8-105). On average, the quantity

of recorded returns increased by a factor of about 70 by increasing the gain from

2.5-105to 8.105.

The CATS offers the capability to change the PMT supply voltage between

experimental collections. Operating at multi-photon sensitivity may have value for









situations in which high ambient noise precludes single-photon sensitivity, as man-made

objects such as buildings features may be easily distinguishable even at lower gain

(2.5.105). This idea is explored in the following sections describing scan results. It is

expected that operating at high gain (8-105) will produce the best fine-scale resolution

and foliage/water penetration, but large amounts of noise can obscure some features.

Decreasing PMT gain may then provide higher contrast between signal and noise for

highly reflective features.

Ground-based Scan Testing

Horizontal Scan Building Features

Early attempts at creating a three-dimensional image from range and timetag

information are chronicled in Appendix B. Diagnostic signal analysis allowed for

identification of mechanical problems such as a cracked encoder plate. After rectifying

these issues, point cloud generation aided by diagnostic tools based on time histories

has produced significantly cleaner point clouds. With scan motion set to project a

horizontal line pattern, the CATS was run for 30 seconds. Post-processing of returns

yielded the point cloud shown in the top left of Fig. 4-14. The footprint shape reflected

off the main scattering surface is prominently shown across the solid brick building face.

Metal light poles located approximately 10 m in front of the building are also discernable

from the building front. Noise events occurring throughout the scan are easily

separated from the bulk of signal returns.

Rotating Line Scan Building Features

Data from a rotating line scan (shown in Fig. 3-6) was collected to produce a larger

three-dimensional view of the same building. A top view of the resulting point cloud in

Fig. 4-15 is shown for 9 seconds of scan data (approximately 72000 laser shots). An









isometric view of the same scene is shown in Fig. 4-16. This data was collected with a

2100 V PMT supply voltage.

Raising the voltage level on the PMT to 2200 V produced the expected result of a

larger variety of distinguishable fine features at the cost of more noise returns. This is

shown in Figs. 4-17 and 4-18. Tree structure is also visible in Fig. 4-19. During this

data collection, there were semi-occlusions as a result of leaves; points on the building

face were discernable through the spaces between tree trunks / branches.

Rotating Line Scan Building Features (Partial Obscuration)

Test scans of a building with heavy vegetation obstructions occluding the building

(Fig. 4-20) yielded the large point cloud shown in Fig. 4-21. A top view of the cloud is

plotted here using the QTModeler software package and colored according to elevation.

To get a better idea of the sensor response from building faces and objects directly in

front of the building, points were filtered for a 3 m elevation window in this region. The

results, shown in Figs. 4-22 and 4-23, depict features more clearly. Here we see range

"tails" associated with strong responses from highly reflective surfaces. While these can

pose a serious problem for automated imagery generation and display, such effects

may be beneficially exploited. Isolated dark noise points do not exhibit the same type of

geometrically organized response. The deconvolution algorithm employed in the MIT

Jigsaw system, for example, eliminates these responses while simultaneously

enhancing SNR [54].

Fig. 4-24(A) shows a photo near the building and further zoom-in on the isometric

cloud view. Here we see small bushes, narrow tree trunks, and a white stone bench. In

Fig. 4-24(B), the white stone surface is the most highly reflective and produces the most

returns; the geometry of the small top arch is discernable. The dark tree bark is also









highly reflective. There is a significant difference between the number of the returns

from the trunks and the dense small bushes. The system demonstrates in this case the

potential to separate the trunk and foliage features.

Expanding the points filter to 5 m elevation shows penetration through the heavy

tree foliage. Fig. 4-25 shows top and side views of the point cloud colored by elevation.

In this case red points correspond to leaf returns close to 5 m, and orange points

correspond to leaf and roof returns close to 4 m. The system is unable to define much

of the roof structure due to almost total obscuration by the dense vegetation in this

case.














UFLSteady_RangeCal_24.SPR


Statistics
# of shots fired: 16000


total # of hits in file: 279197

# of channel error hits: 0

# of non-error hits: 1279197

total # of filtered hits: 922202 (72% of total)

average hits per shot for all hits: 79.9498

average hits per shot for filtered data: 57.6376

averaged range to target, all channels: 265.1215


i,~
e 'n
: "'6."


I ,]


au .LI


Slandaid Clenl.lon or 1 hr.nenl Mnls
ti --


S11I1 Mil I. Lurdl er P.r',el F IT, l, Edl i




LijIu


0 3


J I C i 310 icI -U u -U 11'
rqljei iijcgaQ Fcil L harl~ril F~rlitrc Dili


I:( Ii


1
..
1-' j
Oi I


2LCI





li --- -0 al.c E-- -0 T X-I'--
0' ilu Xi .10 1:1 s -a 0 "u 9l K *3 I l


1.r 1IU A J.r 1Lu bu I .'
Channels I.%


Figure 4-1. CATS static (no scan) returns analysis from a brick wall approximately 300
m from the instrument. A) Statistics from calibration test. B) Histograms
related to hit count per channel and average range per channel. C) Average
range per channel, zoom-in. D) Standard deviation of recorded ranges per
channel.


,-"L.. ..3A P jqj Pjii C .vi.re I F.1 ^*er-5-3C L'Sl











lCi l ri.TI r.. I ,..r hi: F i. har. rIl


3
2.5
2.
1.5,
1,


* 0
e
. Oep


% 0.
"~ ~ l 0
I-1


10" ToI flurr.nber ero I- r PeI rn' nn.l
**







*
*


* **** **
) ) )
) i
)1
SI:


Figure 4-2. Relative accumulated hit counts per channel for a CATS range calibration
test. A) Hits are associated with height and plotted to emphasize variation in
response levels. B) Colors associated with these heights are more easily
identified for spatial trends.


Figure 4-3. CATS static footprint test setup. A 450 turning mirror deflects four central
pixels onto reflecting surfaces below. The pixels immediately surrounding
these four are masked by cardboard to aid in channel discrimination, half of
one pixel -- channel 15 -- is distributed on the wood calibration surface, and
the other half is distributed 11 cm above on a wood sheet.











beamlet cross-section
/


Figure 4-4. Illustration of beamlet cross-section projected onto wood surfaces.


First / Last Hit Range Differentials for 3+ Return Shots
60
atmospheric returns

R BV 4- signal returns
20 -

0 200 400 600 800 1000 1200 1400 1600 1800 2000
Shot Number
2500-
2000 -
AD 1500 o
1000
500

0 5 10 15 20 25 30 35 40 45 50
Difference (# of Range Bins)
%rnn


2000
AD 1500
1000
500
n


10 20 30 40 50
Difference (cm), zoomed-in


60 70


Figure 4-5. Calculation of reported separation for 19 cm distance between multiple
scattering surfaces (trial D). A) Range bin value (RBV) difference between
first and last returns is plotted on the y-axis versus shot number on the x-axis.
B) Histogram of accumulated differentials (AD). C) Histogram of accumulated
differentials, zoom-in.


I I I I I


I I,










Table 4-1. Ranging trials for shots reporting 3+ returns
Actual Reported Error # of % of
Trial Separation Separation (cm) Multiple Total
(cm) (cm) Hit Shots Shots
A 0 n/a n/a 259 2.6
B 11.0 15.0 4.0 2094 29.4
C 14.0 15.0 1.0 2435 24.4
D 19.0 22.5 3.5 1815 18.2
E 24.0 30.0 6.0 831 8.3
F 31.0 37.5 6.5 2112 21.1
G 41.0 45.0 4.0 1471 14.7

Table 4-2. Effect of range bias offset on measurement error (in cm)
Trial IRS-ASI I(RS-5)-ASI I(RS-4)-ASI I(RS-3)-ASI I(RS-2)-ASI
A n/a n/a n/a n/a n/a
B 4.0 1.0 0 1.0 2.0
C 1.0 4.0 3.0 2.0 1.0
D 3.5 1.5 0.5 0.5 1.5
E 6.0 1.0 2.0 3.0 4.0
F 6.5 1.5 2.5 3.5 4.5
G 5.0 1.0 0 1.0 2.0
average 4.3 1.7 1.3 1.8 2.5
error


R1712U (4X4 MATRIX ANODE ARRAY)


No.5 No.6 No.7 No.8
80


60-
FWHM FWHM FWHM FWHM
2.9 2.9 2.9 2.9
40

20



0 3 6 9 12 15


LIGHT-[
SCAN



SUPPLY VOLTAGE: 2000V
WAVELENGTH :400nm
LIGHT SPOT DIA. :0.02mm


POSITION ON PHOTOCATHODE (mm)
Figure 4-6. Output profile of multi-anode MCP-PMT when a light spot of approximately
20 pm diameter is scanned across the photocathode active areas (1.4 x 1.4
mm squares) [50].


























Figure 4-7. Contour map of beam energy for a single pixel. The inner contour is
indicated by a dashed circle. Edges of neighboring pixel energy patterns are
also visible in the image corners.














IF



Figure 4-8. CATS static footprint test setup. A 450 turning mirror approximately 560 m
from the sensor deflects four central pixels onto a wood sheet used as a
calibration surface. The pixels immediately surrounding these four are
masked by cardboard to evaluate crosstalk of returns from the four pixels.









Table 4-3. Channel orientations for crosstalk calculations


a30 a18 822
a31 100% X
a32 X X
Channel 19

as X X
a7 100% X
a6 a32 a31
Channel 20


a18 a22 814
X 100% a15
X X a16
Channel 23

X X a15
X 100% a16
a7 a8 854
Channel 24


Figure 4-9. Geometric orientation of isolated pixels (shaded central four) in CATS
footprint.











B
A
C
Figure 4-10. Photos of CATS barrel testing. A) A 55 gallon barrel filled with water of
variable constituency is placed below a 450 turning mirror. B) Beam
propagation through 560 m. C) Four pixels are reflected down into the barrel
by the mirror.



















Figure 4-11. Histograms illustrating accumulated signal events for a heavily turbid
water-sand solution. Blue colored bins correspond to expected water surface
returns, green to middle of the column, and red to the bottom of the barrel. A)
Returns from four barrel channels combined. B) Returns from channel 22
only.

Atmosphenc Test Total Accumulated Events Per Channel (10k shots)
72300V
2400V
7000 2500V


4000
z 3000
2000
1000
0


10 15 20 25 30
Channel


Figure 4-12. Stacked histograms depicting accumulated atmospheric events per
channel at three different PMT supply voltages.


..ji


.m-illnna. mm mIn
























A B
Figure 4-13. Histograms depicting the consistency of accumulated events through the
duration of an 8 ps range gate for three different PMT supply voltages. Blue
colored bins correspond to events early on in the gate, and red to events near
the end. A) Returns from all channels combined. B) Returns from channel 14
only.

4 MN 07


Figure 4-14. Point cloud (green) resulting from CATS horizontal scan after timetag
structure modification. The sensor was situated on rooftop approximately 300
m from building.


100


ILnnmnmnnr






















*/







a_


Figure 4-15. Top view of point cloud generated from rotating line scan data at 2100 V.


10 20
20 2O0


320
310
300


-- 350
340
3M


20--


15-


10-

>-



0-4
6-


-40


F 4 s c ot cloud ge
Ir *,t *~ rK-
*~ '' ,,.^n^11^


Figure 4-16. Isometric view of point cloud generated from rotating line scan data at 2100
V.
















101


c
















350 -


340-


330-


320-


310-


300-


290-


Figure 4-17. Top view of point cloud generated from rotating line scan data at 2200 V.








20-


15




'0

o40


-30


10


X ml]


-" 350
34b
330
09
_,(320
30310
S290
20 280


Figure 4-18. Isometric view of point cloud generated from rotating line scan data at 2200
V.









102


- I.J





















I.~1

-L:x.
* c


*1


Figure 4-19. Front view of point cloud generated from rotating line scan data at 2200 V,
zoom-in of tree structure.


Photograph (front view) of a building with a large number of nearby trees.


103


'0


C, : t 1(
r h, ;"
i
i '~L ii


































Figure 4-21. Overhead view of large point cloud generated from the building with heavy
tree obstruction. Data was colored according to height (m) using QTModeler
software. Dark blue points correspond to the low slope of the grassy field
leading up to the building. Dark red points correspond to the tops of the trees
directly in front of the building.


Figure 4-22. Overhead or top view of 0-3 m data segment associated v
location.


104


























Figure 4-23. Isometric view of 0-3 m data segment to emphasize heights.


105












































Figure 4-24. Zoom-in on isometric view of 0-3 m data segment. A) Photo near building.
B) Isometric view with bushes, tree trunks and white bench chair with rounded
top in foreground and building walls in background.


106













































Figure 4-25.Top and side view of 0-5 m data segment. A) Top view expanded to 5 m
height (red points). B) Side view expanded to 5 m height.


107









CHAPTER 5
LSNR AIRBORNE LASER BATHYMETRY

ALB Background

The ability of light detection and ranging (LIDAR) hydrography to create detailed

maps via flexible data collection methods supports a variety of applications, including

nautical charting, hazard identification, erosion modeling, storm damage assessment,

and beach nourishment / design [59]. By the late 1980s, the idea of airborne laser

bathymetry (ALB) showed promise from systems such as the U. S. Airborne

Oceanographic LIDAR, the Canadian Larsen 500, the Australian WRELADS, and the

Swedish FLASH [60]. Operational systems matured during the early part of the next

decade, with successes by systems such as the Royal Australian Navy's Laser Airborne

Depth Sounder (LADS), the U.S. Army Corps of Engineer's Scanning Hydrographic

Operational Airborne Laser Survey (SHOALS), and the Swedish Maritime

Administration's HawkEye.

As volume of data and rate of acquisition became less restrictive concerns for

progressing computer technology, the processing workflow moved from a manually

intensive line based approach to an automatic area based methodology [59]. In the

early 2000s and through much of this decade, improvements have been made to

increase the pulse rate, spot spacing, and portability of ALB systems [61], but the nature

of the sensing representation of the air/water interface, water column, and bottom via

green channel waveforms remains largely the same.

Multiple Information Channels in Laser Ranging

The combination of green channel information with other types of observations has

proven useful for feature separation, especially in the bathymetric context. Data from


108









the SHOALS, for example, have been used to make high-density measurements of the

coastal zone for a number of years [62]. Analysis of 1064 nm (infrared) channel data in

conjunction with parallel green waveforms is beneficial for distinguishing the specular

interface reflection, or water surface, from the diffuse bottom reflection [62], [63], [64],

[65]. As a general rule of thumb, the SHOALS is sensitive to bottom returns out to 2-3

times the Secchi depth and out to a maximum of 60 m in the clearest coastal waters.

Shallow water detection (<5 m), however, has been shown to be difficult. Older

ALB data sets such as those collected prior to 2004 tend to have gaps in coverage

related to complexities in processing very shallow water data [61]. Traditional depth

extraction algorithms have problems dealing with merged waveforms even with manual

interventions. Recent enhancements to the traditional SHOALS depth extraction

algorithms from green APD waveform data have shown interesting developments for

dealing with mixed-mode signals in waters less than 1.5 m [61].

Changes in laser-induced water Raman backscatter can be used to detect surface

oil slicks [66], [67] and to measure chlorophyll concentrations to estimate phytoplankton

dynamics [68]. Research also indicates that novel analysis of red-channel waveform

data associated with Raman scattering can provide an approach for measuring very

shallow (<2 m) water depths [64].

Additional channels of information are a luxury that modern systems can exploit to

produce high-resolution area maps. The work pursued here, however, is aimed at

evaluating whether information from only the green channel can be used to provide

similar types of feature detection when a different type of detection methodology is

utilized.


109









Challenges Facing Single-Channel Detection

Ideally, a light detection and ranging (LIDAR) system will record a digitization of

the return waveform resulting from each laser shot. This allows for a post-processing

algorithm to identify time intervals associated with inflection points in the reconstructed

signal. Small footprint LIDARs then have the potential to provide detailed

characterization of vertical structure at high spatial resolution [14]. The idea of building

an "all-green" airborne bathymetry system that meets depth accuracy standards faces

several impediments. For full-waveform mapping, the interface component signal

amplitudes depend strongly on the beam nadir angle, wind speed, and specific

irradiated wave slopes, causing signals to vary with huge standard deviation [47].

Subject to these varying conditions, there may only be a small fraction of the outgoing

laser energy that is reflected back into the atmosphere to be collected by the ranging

receiver. The resulting surface uncertainty problem is a serious issue for system

designers attempting to decide which portions of the return waveform to interpret as sea

surface and bottom in a real-time, automatic manner.

At 532 nm in a typical coastal environment, less than 2% [69] of the laser light is

reflected back from the air-surface interface. A large portion of the green laser pulse is

refracted into the water column. The beam undergoes scattering from entrained

microscopic particulates that causes it to spread into a cone of continuously increasing

angle. A fraction of the transmitted energy in the water column, which decreases

exponentially in magnitude with greater depth, is incident on the bottom with a lateral

extent that depends on the scattering properties of the medium and optical depth. A

portion of this incident energy, typically 4 to 15% depending on bottom composition [64],

is reflected from the sea floor back into the water column. Absorption and scattering


110









again attenuate and stretch the pulse as it passes back to the surface, where most is

refracted back into the atmosphere.

A photodetector such as the avalanche photodiode or photomultiplier tube needs

to employ substantial amplification (>104) in order to magnify low amounts of this

backscattered energy (<1 fJ) into a measurable signal. Detection beyond 10 m depths

has typically required significant outgoing pulse energy at 532 nm (>50 pJ). The

generation of large amplitude pulses (>2 mJ) can simultaneously increase the outgoing

pulse width (>10 ns), which has reduced the capability of typical ALB systems to

accurately survey shallow (<5 m) coastal areas [61],[64]. Increasing output energy can

also cause safety issues at low altitudes and requires more sophisticated lasers.

Pulse Width and Output Energy Considerations

In past ALB systems, green beams were purposely expanded to a diameter of at

least several meters in order to achieve eye-safe operation with sufficient pulse

energies to provide reasonable signal-to-noise ratios at greater depths [70]. For

example, the Optech SHOALS-3000 system uses a 3 kHz, 6 ns, 3 mJ laser [Tuell,

personal communication, 2009] to provide bathymetric measurements up to 60 m depth

at 2x2 m spot spacing [62]. The NASA Experimental Advanced Airborne Research

LIDAR (EAARL) takes a different approach by using 532 nm, 10 kHz, 1.2 ns, 70 pJ

pulses (~1/43 the amount of energy) with a much narrower FOV; ~2 mrad [70]. This is

an interesting system because the combination of short pulse and small FOV is

expected to overcome the ambiguity traditionally encountered with surface location

detection in green-only ALB; an idea that must hold true if the CATS is to achieve any

success. Four 1 ns waveform digitizers connected to photodetectors operating at

varying sensitivity allow for cross-environment surveys that include abrupt surface


111









brightness boundaries, typically a cause of saturation in sensors optimized for viewing

dark water targets. This system has been shown to be capable of accurately surveying

12 m depths [71]. Optical rugosity analysis by the EAARL in Biscayne National Park

showed measurements consistent with field observations of coral features relative to

rubble-dominated bank reefs within the study area.

Given recent advances in low-signal level detector technology, a further question

is raised: how much can we lower the outgoing pulse energy and still achieve shallow

(0-5 m) bathymetric detection when paired with current commercially available

photomultiplier tubes? We have previously alluded that LSNR detection methodologies

work well for low-noise environments such as satellite laser ranging, but shallow-water

bathymetry presents a host of new variables with which to contend. The CATS sensor

is uniquely suited to answer this question. In relation to previous successful ALB

systems, the EAARL sensor is considered to be a low-power system. In comparison to

the EAARL, the CATS sensor outputs 35 nJ per channel; 1/2000 of the energy. This

represents quite a drastic shift from traditional ALB design philosophies.

Contrast to Waveform Digitization

In waveform digitization post-processing or embedded filtering (such as constant-

fraction discrimination), there is some level of decision making that occurs on an

automatic scale. For example, effects such as amplitude dependent range walk, slope

dependency of range and signal ringing can have negative effects without proper

compensation [72]. For LSNR shallow-water bathymetry, complete waveform

digitization may not be mandatory. It may be possible to move to a simple fixed-

threshold detection strategy and retain similar feature detection results. In Chapter 4,

static experiments showed that the estimated density of data points from the sea bottom


112









given 20-cm pixel spacing should be sufficient to establish near-shore depths (up to 5

m) at coarse spatial resolution (1 m scale) [56].

Experimental Setup

The CATS sensor was designed to be mounted alongside a commercial ALTM

unit such as the Optech Gemini, which is routinely operated by the University of Florida.

To minimize costs, it contains no inertial measurement unit (IMU). However, for the

tests reported here, the Gemini sensor was not used and the CATS sensor was

mounted in a Cessna Skymaster twin-engine aircraft, shown in Fig. 5-1. A special

mount was constructed out of aluminum plates and bolted directly to the rails that

normally support the co-pilot's seat. The system was therefore rigidly attached to the

aircraft frame, but no IMU data were collected with which to correct the observations for

changes in the orientation of the aircraft. Artifacts resulting from changes in the

orientation of the aircraft (roll, pitch, and yaw) include undulations of flat water surfaces

and lateral wandering of the surface profiles.

The mounted CATS sensor head fits down into an opening in the floor of the

aircraft with approximately 5 cm free space surrounding the main cylinder containing the

Risley prisms. A DC power supply provided 24 V at approximately 3 A. The DC supply

received 120 VAC from an inverter, connected to the aircraft's main power system. A

laptop PC was also connected to this inverter. The scanner control box and timetag

module were secured behind the pilot and at the feet of the operator.

Experiments were conducted to test the system's airborne profiling capability. The

two prisms were set 1800 out of phase so that the laser would be pointed straight along

the optical axis of the transmit / receive path, or at nadir. The aircraft was flown at

approximately 60 m/s and 500-600 m height above ground level for a series of ranging


113









trials. Because a large number of returns per channel were expected, only channels on

the second ranging board (17-32 or the central 4x4 square) were enabled for a majority

of the experiments. Initial collections with all 96 elements enabled in profiling mode at

the highest gain produced such a wealth of returns that the data throughput exceeded

system capacity.

Small Structure Identification

Several different types of buildings and other manmade objects were profiled at

nadir in a flight over the University of Florida campus in Gainesville, FL. Fig. 5-2(A)

shows an aerial photo of a region at the south end of campus and flight line. Fig. 5-2(B)

is the raw point data from channels 20 and 23. The image allows us to identify a small

outcropping in a building roof, a roof from a nearby parked car, angled roof geometry

from small office buildings, the outline of a stone bench, and a low-hanging awning

between two buildings.

Vertical Canopy Structure Identification

Fig. 5-3 shows an unfiltered point cloud generated from data over a forested

region in Mine Run, VA. Small leaf structure elements are definable here based on a

large number of multiple scattering events. The flight was conducted in autumn as the

deciduous forest was in the process of shedding leaves; this left small pockets of space

in the overhead canopy through which individual beamlets could propagate. Even

partial transmission of beam elements as a result of semi-occlusion by leaf surfaces

was enough to reconstruct intra-canopy elements and the ground surface below the tree

line. A zoom-in depicting a narrow tree trunk is pictured.


114









Optimizing System Performance

Analysis of data from test flights showed that overall signal strength at maximum

PMT gain (2500 V or 8.105) was high enough to warrant changing data collection

parameters. Responses from strongly reflective surfaces at this gain level produced a

stacking effect (hits in many consecutive range bins) near roof surfaces in the resulting

image. Since the microlaser output energy and detector type were selected to

maximize LSNR performance in turbid waters, this was an expected outcome. The

dynamic range between different types of features (especially when contrasting

topographic and bathymetric returns) may be very large. Responses from artificial

surfaces for on-nadir beam propagation at 2500 V were tens of range bins deep in

some cases.

Data collections at multi-photon sensitivity (2300 V or 2-105 gain) are capable of

producing responses 3 to 4 bins deep from surfaces such as building roofs, bare earth,

and tops of tree canopies. We could thus reduce the tube voltage for future surveys

and expect the top end of our dynamic range to be reduced. Generally speaking, if

interrogated far-field surfaces were of uniform reflectance and the gain of the tube is

matched to provide exactly enough amplification to barely produce a threshold crossing,

this might be an acceptable technique. Of course in practice, target reflectance

properties are not precisely known, and tailoring a detection strategy to one type of

feature negates the real benefit of this type of sensor-mixed mapping. As shown in

Chapter 4, shallow-water bathymetry is difficult at less than full amplification (2500 V).

For the LSNR detection paradigm to be worthwhile related to submerged surface

mapping, an amplification level should be maintained that produces returns from even

very weak scatterers.


115









Another solution may be viable: changing the type of airborne profiling from on-

nadir to off-nadir. By imposing a fixed incidence angle relative to ground for outgoing

laser light, the magnitude of the backscattered radiation is decreased due to less

significant contributions from specular reflections and increased slant range. This has

special relevance in the case of mirror-like water surfaces as large specular

components cause problems for shallow bathymetry on-nadir operation. A circular

pattern is one way to propagate this effect throughout a scan and retain contiguous

terrain coverage; in fact along-track resolution is effectively doubled.

Even with off-nadir operation, and the potential advantages of circular versus

raster scanning, CATS data shows that there is still a dynamic range issue because of

the large difference in signal strength between topography / sea surface elements and

distributed targets such as partially occluded forest canopies and submerged surfaces.

For large-scale uniform surfaces such as building roofs, it may be possible to

automatically collapse range tails into first returns and simultaneously provide some

measure of intensity about the response, dependant on the number of range bins that

the signal stretches over and the expected cross-section geometry [24]. Dynamic range

is one of the major challenges for adaptive algorithm development [73]; automatic

procedures are usually dependent on the local coherence of return information.

Data Artifacts

Accurate characterization of data artifacts can be one of the most challenging

obstacles for a developmental LIDAR system. This is especially relevant for LSNR

mapping as in some cases the noise floor can eclipse the true signal level. We expect,

however, that true signals will exhibit some form of spatial coherence even in lowly

populated areas due to the random nature of noise formation, allowing us to use


116









techniques such as Post-Detection Poisson Filtering (PDPF) and a Correlation Range

Receiver (CRR) [15] to beneficially exploit data trends. We will consider in detail

artifacts specific to the CATS to evaluate the applicability of this claim for mixed

topographic and bathymetric data processing.

A closer inspection of detected surfaces in Fig. 5-2 shows a potential problem; a

"ghost image" of roof surfaces appears at a constant offset beyond the first return.

Though we may be able to easily remove these types of artifacts for bare-earth and

building measurements via an automatic first-surface algorithm, we risk losing important

multiple-return information from other types of features. On the other hand, if no

algorithm is applied, we risk lumping the ghost returns in with partial reflections in close

proximity (<10 m) to the first. Our data sets may then also require significant manual

processing. In the worst-case scenario an automatic data classifier may erroneously

identify a ghost image from the water surface as a submerged element, yielding an

inaccurate depth estimation. Analyzing the scope of such artifacts is thus of prime

concern for shallow water bathymetry.

The influence of this type of noise is not always obvious because the magnitude of

the responses is suspected to be fairly low, meaning that certain types of data

visualizations can cause the effects to be classified as random thermal events or

volumetric scattering residuals rather than deterministic occurrences. It is important to

develop an automatic data processing technique that compensates for the noise effects

so that their influence does not lead to false feature characterizations. For example,

when first returns are plotted from channel 23 in Fig. 5-4(A), ghost surfaces are difficult

to identify. The image seems close to what we would expect; canopy returns are well-


117









distributed for partial leaf-on conditions, bare-earth surfaces (x<150 m) are singular, and

water returns (x>150 m) are generally bi-modal. The distribution of noise returns

relative to these features, however, should be analyzed in further detail.

Above the ground / water surface, we expect random contributions from dark /

solar events as well as atmospheric laser backscatter. Below ground and past ~5 m

water depth, we expect only contributions from dark / solar noise. Fig. 5-4(A) thus gives

us some indication that an aberrant effect may be in play; below ground noise events

are more numerous. The opposite should be true if our noise assumptions are valid.

When all returns from 16 channels of data are plotted in Fig. 5-4(B), artifacts become

much more evident; atmospheric noise increases slightly as expected, but the large

number of below-ground returns is now a stark contrast. This effect persists even

through overlapping first-return data; the theorized cause is a combination of noise and

crosstalk between PMT grid elements.

Though it is clear from Fig. 5-4(B) that artifacts below strongly reflective surfaces

are significant, further quantification may aid in noise identification and processing

efforts. For example, MATLAB can be used to compute estimates for local density in

selected regions within the cloud. A segment of flat ground data was manually

windowed to encompass expected signal returns. Ghost responses should exhibit well-

defined correlation to this window; the temporal offsets of expected highest correlation

are shown in Fig. 5-5 as "afterpulse 1/2." By then sliding the true signal window

vertically as in Fig. 5-6 and expressing the correlation coefficient as a ratio of true signal

density to the measured effect, the plot in Fig. 5-7(A) was generated. The plot

decreases exponentially starting at zero distance (correlation % of 1.0) down to the


118









noise floor, which is below a correlation % of 0.1. Here the correlation of ghost images

is not distinguishable. We then conclude that the spatial distributions of artifacts are

significantly less dense than the true signal. In order to analyze information near the

noise floor, the y-scale needs to be expressed logarithmically, as in Fig. 5-7(B). If we fit

Gaussian profiles to the output and measure the peaks, the largest correlations occur at

12.67 and 24.00 ns, with respective peaks of 0.0085 and 0.0053 (Fig. 5-7(C)).

We can posit several possible explanations for the noise. First, gas impurities in

the tube envelope can generate additional responses beyond the initial high-intensity

return; this is commonly referred to as "afterpulsing" in nuclear particle physics

experiments. Second, the cable connecting the PMT output to the high-speed data

interface could have an impedance mismatch. A mismatch could cause undesirable

reflections from the interface junctions to become significant at very high signal levels.

Third, the silicon germanium high speed gain block could be being overdriven in cases

of large signal output. This would cause the amplifier to operate beyond maximum

recommended ratings, thereby inducing anomalous effects.

PMT Afterpulsing

Exploration into the chemical and physical nature of PMT afterpulsing effects [74],

[75], [76] showed that diffusion of trace gases through the tube envelope contribute to

an increase in the probability of pulse formation. In particular, diffused helium was

found to be the main source of afterpulses in tubes more than two years old. In

contrast, the contribution from this impurity in new tubes was often small [76]. An

exponentially decaying low-level background signal has been shown to occur after the

abrupt termination of a sub microsecond light pulse [77]. Positive ions produced during

the ionization of residual gas in the PMT by photoelectrons produce secondary


119









electrons during their drift back onto the photocathode, which are then amplified [78],

[79]. Other possible explanations for both short (sub-microsecond) and long-term

(microsecond to millisecond) responses are electron-induced luminescence of the

dynodes or their support structures and microdischarges produced during high-gain

operation [77].

Because the CATS uses an MCP-PMT manufactured in 2004 that has undergone

significant use, it is possible that trace amounts of atmospheric gas diffused into the

envelope over time may have risen to significant levels. Quantifying the amounts of

impurities is beyond the scope of this work, but potential contributions from afterpulsing

should be mentioned due to the age of the tube.

Impedance Matching

For radio frequency (30 kHz to 300 GHz) circuitry, the impedance of the

transmission line itself becomes significant. In order to maintain maximum power

transfer, the characteristic impedance of the cable (Zo) should be closely matched to

the source (Z ) and load (Z,). If the total path length is greater than 10% of the signal

wavelength, mismatch effects can have greater impact. In the CATS system, 2-GHz

level signals mean that transmission lines longer than 4 cm must be well matched

(assuming velocity factor of 0.8). We should then consider the layout of associated

PCBs in the receiver design and estimate the total length between the PMT and data

capture elements. The minimum trace length is the sum of the following lengths:

* MCP-PMT pin to PMT adaptor board connector: 12.95 cm
* Samtec HFHM2 cable (1): 25.40 cm
* Samtec HFHM2 cable (2): 25.40 cm
* Frontplane board trace: 8.13 cm
* Connectors: 12.70 cm


120









The total trace is then 84.58 cm. Assuming an average propagation delay of

70.98 ps / cm, the total delay is 6 ns. The layout of potential cable reflections in relation

to a transmitted signal is shown in Fig. 5-8. In this model a signal pulse is initiated at

T=0, at the near end of the trace. The pulse reaches the far end of the trace at T=6 ns.

Here a portion of the signal is reflected back dependant on the match between Ro and

RL, which causes another reflection at T=12 ns based on the match between Rs and Ro.

This reflection reaches the far end of the trace at T=18 ns. This process continues

infinitely, and output pulses of diminishing magnitude will be received at T=6, 18, 30,

etc. intervals. The estimated spacings of the first two reflections are close to the

temporal spacings calculated from local density correlations (12.67 and 24 ns). The

next step is to calculate the expected magnitude of reflections based on manufacturer

specifications for impedances.

The line lengths of the Frontplane board are matched for a maximum channel-to-

channel delay difference of 18 ps. Signal routes are 50 0 single ended lines. Test data

using Time Domain Reflectometry (TDR) are given in Table 5-1. The measurements

show the average impedance for each layer; deviations are expected due to variations

in PCB manufacturing tolerances of the layer thickness and line width. If we let Z, =

49.9 0 and consider Layer 3 (L5), Board 2 as a worst-case scenario, we can use the

following analysis to calculate the magnitude of the first two reflections. When a signal

is impressed on the end of a transmission line, a fraction of the full source voltage

propagates down the line. This is referred to as the input acceptance function A, and is

expressed in terms of ZS and Zo [80],


121










A Zo (5-1)
Zs + Z

Next, the signal is attenuated by the propagation function H. For this analysis we

will assume an average value of 0.94. At the far end of the cable a fraction of the

attenuated signal amplitude emerges, which is determined by the output transmission

function,

T 2-ZL (5-2)
T-
ZL +Zo

When a fraction of the propagating signal emerges from the far end (T=6 ns in our

model), a reflected signal also travels back along the cable towards the source. The

magnitude of this reflection is determined by the far-end reflectance function,

SZ Zo (5-3)
R, =
ZL + Z

The reflected signal is again attenuated again by H as it travels towards the

source where it reflects a second time off the source impedance. The source end

reflection function is given by,

Zs Zo (5-4)
R, =-
Zs + Z

After the second reflection, the signal is attenuated by H as it travels back to the

load as it again emerges via the transmission function T. This process continues on

indefinitely. We can use these functions as coefficients to calculate the magnitude of

the first three signals:

Signal, = A H T, (5-5)

Signal2 =A-H R2 -H -R1 H-T


122









Signa/ = A.HR2 -H-R, HR2 -H-R, -H-T

For Z = Z = 49.9 0 and Zo = 52.84 0 (upper worst case), we find Signal, = 0.47,

Signal2 = 0.0004, and Signal= 2.46-10-7. For Zs=ZL = 49.9 0 and Zo= 47.52 0 (lower

worst case), we find Signal, = 0.47, Signal2= 0.0002, and Signal= 1.31 10-7. The first

reflection in both worst-case scenarios is then expected to be three orders of magnitude

smaller than the true signal, and the second reflection six orders or magnitude smaller.

Though potential reflections resulting from worst-case impedance mismatch may occur

at the same temporal spacing as the ghost images, the magnitudes appear to not be

significant.

Gain Block Overdrive

Output signals are amplified by a 4000 MHz 2.8V SiGe gain block manufactured

by Sirenza Microdevices (part number SGA-1263). The output of this device is fed

directly into the threshold comparator for each channel. Because the system needs to

operate on a single-photon sensitivity level, channel thresholds are set to a 4 mV

trigger. According to the manufacturer's data sheet, the gain block has a maximum RF

input power of -12 dBm, or 63 pW, which equates to 56 mV into a 50 0 load. An output

current over 1.1 mA will therefore begin to overdrive the gain block.

In Fig. 5-9, a large signal (110 mV) is injected into the gain block. This is

approximately 25 times larger than a signal expected to barely produce a threshold

crossing (4 mV). The resulting output is shown in Fig. 5-10. The desired response is

the first large negative pulse. Deviations from zero-level in the waveform after the first

negative pulse are anomalous effects resulting from the large signal input. Here we see

small negative deformations before and after a positive peak. If we inject a larger signal


123









(200 mV) (Fig. 5-11), the output shown in Fig. 5-12 is produced. A line is drawn here to

approximate the threshold crossing. The second crossing occurs at approximately 13

ns. If a still larger signal (300 mV), shown in Fig. 5-13, is injected into the block, we can

identify threshold crossings in Fig. 5-14 at approximately 13 and 20 ns.

The temporal spacings of these outputs are similar to that estimated from local

density correlation within the point cloud. Because the dynamic range between

submerged surfaces and highly reflective topography is so large, it is likely that the

overdrive of the gain block is the main cause of ghost images.

Point Cloud Filtering Based on Density Estimates

A data processing method should be developed to eliminate the noise artifacts

while preserving feature structure within the point cloud. Because the spatial densities

of data artifacts are fairly low (peaks less than 1% of true signal), we can perform point

cloud filtering based on local point information and expect not to lose a significant

amount of signal information about strongly reflective surfaces. To validate the potential

of the LSNR data, generated algorithms should be near-automatic; i.e. with the least

amount of manual input as possible.

First, a decision boundary between land and water surfaces was estimated using a

sliding horizontal window similar to Fig. 5-6. For each point in the data set used to

create Fig. 5-4(B), a 0.6 x 1.0 x 1.0 m box was fitted to produce density estimates in the

topographic data defined by the decision boundary. All other points were then classified

as potential bathymetric responses and a 0.3 x 1.5 x 1.0 m box was fitted to produce

density estimates. The results of this operation on data gathered from a single channel

of information are shown in Fig. 5-15. Next, the scene is split into coarse grid cells and

data within cells below the noise threshold are eliminated (Fig. 5-16(A)). The secondary


124









step helps to eliminate the few remaining artifacts, and the final processed image is

shown in Fig. 5-16(B). A dense clump of points 5 m above the water surface is visible

at x=180 in Fig. 5-16(B); this is the result of returns from an overhanging island tree

near the flight line. Because the density of returns suggests that this is not atmospheric

noise, the filter retains the associated points.

Fresh Water Bottom Mapping

The major goal of flight testing for the CATS instrument was to evaluate LSNR

water penetration capabilities. Application of the expectation maximization (EM)

Gaussian-mixture (GM) algorithm shows that surface and bottom can be reliably

separated in sea (total attenuation coefficient of 0.05 / m) and coastal (total attenuation

coefficient of 0.5 / m) water to a depth of 2 m [21]. At larger depths, simulation also

shows that horizontal resolution of the reconstructed sea bottom is greatly limited by

water clarity; meter-scale bathymetric mapping should only be possible at sea depths

less than three times the Secchi depth.

A zoom-in view of processed data from the shallow fresh water area (the Lake of

the Woods) in Mine Run, Virginia is shown in Fig. 5-17(B). The path of the aircraft near

the shoreline is indicated in an aerial photo (Fig. 5-17(A)). The water in this area

contained low amounts of colored dissolved organic material (CDOM); at the time of the

data collection the Secchi depth was measured out to 2 m. The lake bottom was a

mixture of dark sand and decaying organic material; bottom reflectivity was expected to

be very low. 16 channels of data were collected at 2500 V, or maximum PMT

amplification. Manual measurements of water depths in this area at 10 m intervals were

also performed using a measuring tape with an attached weight.


125









Point cloud values were interpolated to estimate the lake bottom surface, shown

as a dashed line in Fig. 5-17(B). The bottom was then corrected for the index of

refraction of water (1.33), shown as a solid line. Here we see a consistent response

from the water surface, shown as a flat line, and an exponential decay of water column

returns progressing to further depth. The lake bottom produces a strong signal at ~1.5

m depth near the shoreline at x=140 m. Water depths at 10 m intervals were extracted

from the LIDAR data in this region by differencing the interpolated water surface and

corrected lake bottom function. Lake bottom returns become negligible at

approximately 2 m water depth. There was a large amount of sediment buildup from

organic / inorganic particulate near a small island located at x=180 m. This could lead

to shortening of the estimated depth near the island. Table 5-2 compares LIDAR depth

estimates to measured values. The average difference between estimates and

measurements is 21 cm.

The lake bottom was slightly more reflective in other parts of the lake. Fig. 5-18(A)

shows a location to the northeast at the same lake where the bottom was composed of

a mix of lighter sand and organic material. The increase in reflectivity produced returns

from the lake bottom at greater depth. Fig. 5-18(B) shows an off-nadir profile for 16

channels at 2500 V depicting recovery of the lake bottom out to 2 m depth.

Salt Water Bottom Mapping

Further flight testing was conducted over the intracoastal waterway in St.

Augustine, Florida. Fig. 5-19(B) shows point cloud data starting from the shoreline

(aerial photo in Fig. 5-19(A)) and out to a boat dock 60 m away. In Fig. 5-19(B), the

water surface and bottom profile between boat docks and a shallow boat launch region

are depicted. Dense clusters of points 1 to 2 m above the water surface are a result of


126









narrow wooden walkways (x=175, 189, 203 m) leading out to the boathouses. The

sand bottom was reflective enough in this case for a single channel to provide bottom

reconstruction out to a depth of 2.5 m. Hand measurements of wood platform heights

above the water surface were performed at five locations, shown in the third row of

Table 5-3. LIDAR estimates of platform heights were calculated by differencing the first

above-water return associated with the wood surface and the interpolated bottom at the

five indicated locations. These estimates showed an average error of 17 cm.

Fig. 5-20(B) shows information from 3 channels over a sand bar. Bottom features

out to a depth of 1.5 m are recovered here in high detail; submerged pockets and dunes

are identifiable as currents have affected the sand geometry. An aerial photo of the

east coast of Florida near St. Augustine with the aircraft path is shown in Fig. 5-21(A).

Fig. 5-21(B) shows the associated point data from 3 channels. Of particular interest in

this image are the gradually increasing sea floor slope in the surf zone near the beach /

water interface (indicated in dark blue), and the sand bar located 40 m from shore.

Mapping Statistics and Discussion

Simulation [21] predicts that CATS bathymetric performance will largely be

dependent on water conditions; in turbid waters, the operational depth for mapping

should be limited to a few meters. Analysis of airborne profile data from the CATS over

a fresh water lake in Mine Run, Virginia and salt water areas near the eastern coast in

St. Augustine, Florida have largely confirmed these predictions for practical turbidities at

0 to 3 m depths. The central 16 channels 17% of the total footprint area were

sufficient in profile mode to reconstruct a fresh water lake bottom composed of dark

decaying organic / inorganic material (estimated to be highly non-reflective) out to 1.65

m. The Secchi depth in this area was measured to be 2 m. At a depth of 0.5 to 1.5 m,


127









on average 38 returns per meter were registered from the lake bottom. For 8 kHz laser

pulse rate and 60 m/s flight speed, this corresponds to an average of 1.77% of the

outgoing shots registering returns from the lake bottom. The central 16 channels were

also used to reconstruct a fresh water lake bottom composed of light sand and dark

decaying organic material (estimated to be slightly more reflective) out to 2 m. In Fig. 5-

18(B), less bottom returns near the shoreline were received as a result of overhanging

tree branches, but multiple returns still provided information about submerged surfaces.

At a depth of 1 to 2 m, on average 74 returns per meter were registered from the lake

bottom. This corresponds to an average of 3.49% of the outgoing shots registering

returns from the lake bottom.

A single channel of information 1% of the total footprint area was used in

profile mode to reconstruct a salt water intracoastal floor (estimated to be highly

reflective) in water of high clarity out to 2.5 m. At a depth of 1 to 2 m, on average 45

returns per meter were registered from the submerged floor. This corresponds to an

average of 33.98% of the outgoing shots registering returns from the floor; these are

assumed to be near-ideal conditions. Three channels of information 3% of the total

footprint area were used in profile mode to reconstruct the sea floor (estimated to be

highly reflective) in the surf zone near a beach in water of optimum clarity out to 2 m. At

a depth of 1 to 2 m, on average 56 returns per meter were registered from the sea floor.

This corresponds to an average of 13.98% of the outgoing shots registering returns

from the floor.


128

























A B


Figure 5-1. Photos of the CATS mounted in aircraft. A) Cessna Skymaster aircraft. B)
Mounted CATS sensor head.


40 building roof outcropping .
S. angled toof geometry awning
E 30 t '* carroof I
0 stone bench **
1 0 :- r : .,. ', -- !...C ". .
1. .. .. '. i .a" .*.
o **. t I '
-10 --- -L "-- "J --- -- '. --- r -- | "-- -- '. --- I -- I --- I --- .1 ** t --- I --
940 960 980 1000 1020 1040 1060 1080 1100 1120 1140
Y(m)
B
Figure 5-2. Flight testing in Gainesville, FL over University of Florida campus. A) Aerial
photo with flight line. B) Channels 20 and 23, 2300 V, all returns, unfiltered
data showing vertical resolution capability in structure identification.


129

































Figure 5-3. Small point cloud from forested region in Mine Run, VA. Small leaf structure
elements are definable along with trunk and ground information.

50r


1< 3 :" O


S.'


.. *a *..l
I. -* *::.


Figure 5-4. Large point cloud from forested region in Mine Run, VA. A) Channel 23, 1st
returns plotted. B) Channels 17-32, all returns plotted.


130


' ~'~--,-.- -IllOnn~ARU~
'' ~' ~ ~'









































X(m)


Figure 5-5. Ghost images in raw data.


., 1.


~'5I


S *W .


I ** I *_* I __
S '... "* '. .. .




'. I,





'.
.
5, 5" 5 6'"0 65 ,


45 50 55 60 65


** : .".

S "raw data po* .


; 7 8
7.. : "* .0. .
*,
S*: I* IS" '*i* *4

1. .; '' ,,





raw data point .
windowed true surface.
i I, I
70 75 80


Figure 5-6. Vertical sliding window to evaluate magnitude of ghost responses.






131


e












Correlation of Windowed True Signal to Afterpulsing Signal

S0.8
.1 0.6 i








peaks

10' j



50 100 150 200 250
1 0 r ------------------------













050 100 150 200 250
0.01
0.008
E 0.006 -
o0o
c_ 0.004 -
0.002 -

00 50 100 150 200 250
Distance from true surface (0.1 m increment)

Figure 5-7. Results from sliding window. Correlation between responses in 0.1 m
increments, starting from true surface (correlation coefficient of 1.0). A)
Normal scale y-axis, peaks are indistinguishable. B) Log scale y-axis, ghost
peaks are visible. C) Magnitude of peaks are both less than 1%.


S----- T=6ns



S- T=18ns


3 ----- T=30ns
(3


Figure 5-8. Graphical model depicting reflections from detector cable interfaces.


132









Table 5-1. Time Delay Reflectometry test data (courtesy of Fibertek Inc.)
Board # L1 (0) L3 (0) L5 (0) L7 (0) L9 (0) L11 (0)
1 51.28 52.00 51.17 51.43 51.91 51.72
2 49.27 52.84 52.82 51.63 51.29 51.88
3 47.52 52.00 50.80 51.82 52.47 51.76


133






























Figure 5-9. Large signal (110 mV) injected into gain block (courtesy of Fibertek Inc.).


Figure 5-10. Output of gain block from 110 mV signal (courtesy of Fibertek Inc.).


134



























kine


Figure 5-11. Large signal (200 mV) injected into gain block (courtesy of Fibertek Inc.).

















Figure 5-12. Output of gain block from 200 mV signal (courtesy of Fibertek Inc.).
FMeam n200mVnts




Figure 5-12. Output of gain block from 200 mV signal (courtesy of Fibertek Inc.).


135























M0"i.nr4fl m mes I
Manats: scales


Figure 5-13. Large signal (300 mV) injected into gain block (courtesy of Fibertek Inc.).


ffP


Figure 5-14. Output of gain block from 300 mV signal (courtesy of Fibertek Inc.).


136

















40 a


=n- a retained points
f filtered points

0 50 100 150 200
X(m)


Figure 5-15. Result of first step of
airborne profile data.


50
40
30-
20-

I 10-


-10-
-20-


MATLAB local density filter for a single channel of raw


* retained points
o filtered points


CWO a


U Q a
0


-30
0 50 100 150 200
X (m)
50-
40-




10 -1. ?, '


-1 0 .
-20-
-30-


Figure 5-16. Result of second step of MATLAB local density filter for raw airborne profile
data. A) Retained and filtered points. B) Retained points only.







137


100
X (m)


200


a


o





























-6













98 *." :z. Z : .-:. "
-6.5"

-7








9* ....... ....lo


raw d po


1 5 0 1 6 0 1 7 0 1 8)0 1 9 0 b o t t o m


igure 5-17. Fligt es zoo 21o 220
Figure 5-. Flight testing in Mine Run, VA over the Lake of the Woo southwest. A)
Aerial Photo near boat ducks and shallow launch region ith flight line. )

Off-nadir Profile channels 17-32 2500 V, all returns Processed data
depicting fresh water Penetration out to 1.65 m.



138









Table 5-2. Comparison of LIDAR depth estimates and measurements
X(m) 150 160 170 180 190 200
LIDAR Depth 1.12 1.45 1.49 1.24 1.65 NA
Estimate (m)
Depth 1.27 1.52 1.65 1.52 2.03 2.44
Measurement (m)
Difference (m) 0.15 0.07 0.16 0.28 0.38 NA


139















































X(m)
B
Figure 5-18. Flight testing in Mine Run, VA over the Lake of the Woods. northeast. A)
Aerial photo near lake shoreline with flight line. B) Off-nadir profile, channels
17-32, 2500 V, all returns, processed data depicting fresh water penetration
out to 2 m.


140

















































X(m)
B
Figure 5-19. Flight testing in St. Augustine, FL over docks on the intracoastal waterway.
A) Aerial photo near boat docks and shallow launch region with flight line. B)
Off-nadir profile, single channel (18), 2500 V, all returns, processed data
depicting salt water penetration out to 2.5 m.


141









Table 5-3. Comparison of LIDAR platform distance estimates and measurements
X (m) 175 189 203 214 236
LIDAR Platform
1.68 2.13 2.37 2.53 2.90
Estimate (m)
Platform
1.82 2.20 2.59 2.74 3.12
Measurement (m)
Difference (m) 0.14 0.07 0.22 0.21 0.22


142


















































100 150 200 250 300 350
X(m)
B
Figure 5-20. Flight testing in St. Augustine, FL over a sandbar on the intracoastal
waterway. A) Aerial photo over sand bar with flight line. B) Off-nadir profile,
channels 21-23, 2500 V, all returns. Point cloud depicting sand bottom out to
1.5 m.


143







































water surface


sand bar
(-40m from shore)


surf zone
beach



gradual bottom slope
near shore


X (m)

B
Figure 5-21. Flight testing in St. Augustine, FL over the eastern coast. A) Aerial photo
of flight path relative to beach. B) Off-nadir profile, 2500 V, channels 21-23,
all returns, processed data related to beach / water interface region.


144


0

-2
N
-4









CHAPTER 6
CONCLUSIONS AND FUTURE WORK

Scope of Research

This dissertation has described the essential design elements necessary to

construct a low signal-to-noise ratio (LSNR) light detection and ranging (LIDAR) system

and assessed the capability of single wavelength, mixed topographic / bathymetric

mapping by analyzing the design and testing process of the University of Florida's

Coastal Area Tactical-mapping System (CATS).

Design and Testing Overview

Airborne laser swath mapping (ALSM) instruments have conventionally used high

energy / high-pulse rate designs. Exploration into alternate detection strategies with the

benefit of a highly sensitive detector suggests that detection efficiency can be improved

by moving to a LSNR LIDAR paradigm. For example, NASA's Satellite Laser Ranging

station (SLR2000) demonstrated the capability of employing passively Q-switched

microlasers with output energies of 135 pJ and rates below 5 kHz to achieve millimeter-

level ranging precision at satellite altitudes of 20,000 km [23]. Discrimination between

signal and noise was achieved by utilizing a range receiver able to identify and extract

laser echoes from solar background noise through Poisson statistical analysis of the

received photon distribution.

In a very low incident light state, a high gain (105) photodetector is able to produce

amplified responses due to weak backscattered laser radiation via the photoelectric

effect. Individual discrete pulses can be analyzed in a digital mode. If the time

resolution of a detector is sufficiently fast that the average time intervals between signal

pulses are wide in comparison, then individual incident photons can be "counted" and


145









the receiver is then capable of "photon counting" (i.e. single photon sensitivity). The

generation of a photoelectron in this configuration is a binary event given proper

discriminators, making the photon counting method superior to analog signal

measurement in terms of detection efficiency and stability.

The power level of incident radiation from signal events may therefore be on the

order of the noise, e.g. solar events, reflections off atmospheric particles, and detector

thermionic emissions (dark events). Noise events may occur randomly and with low

enough frequency to be separated from terrain signals; this has been practically

demonstrated by data collected and analyzed by Goddard Space Center in flights of a

photon counting laser altimeter, showing clear identification of trees, buildings, and even

potential shallow water penetration [23].

An operating wavelength between 500 to 600 nm allows for water penetration due

to minimum absorption in this spectral band. In order to estimate the output energy

required for single photon detection, signal strength can be estimated as a function of

transmitter, receiver, and environment characteristics. Using a modified form of the

LIDAR link equation, we can expect 63% chance of signal event registration at a one

photoelectron threshold can by transmitting greater than 13.5 nJ pulse energy given the

following parameters: a receiver optical efficiency of 0.4, detector quantum efficiency of

0.28, hologram optical efficiency of 0.8, 532 nm wavelength, 0.15 surface reflectivity

coefficient (representative of sand), 50 local surface incidence angle, and receiver

collecting area of 3.3.10-3 m2

The Coastal Area Tactical-mapping System (CATS), the University of Florida's

prototype LSNR mapping system, has provided data essential to exploring the viability


146









of a multi-element design with these parameters. Improved spatial resolution in single

beam scanning LIDAR systems has traditionally been achieved by increasing laser

pulse rate to decrease the amount of space between scanned ground footprints. This

provides an inherent limitation to avoid the condition of multiple pulses in flight at once;

the two-way travel time of electromagnetic radiation at an average flying height of 600 m

dictates a maximum pulse rate of 250 kHz. If, however, a single laser beam is split into

an array of elements and transmitted at once, spatial resolution improvements are then

dictated by the number of array elements and the capability of the receiver to process

returns over the array simultaneously. A moderate laser pulse rate can then

theoretically provide decimeter-scale resolution.

In the CATS, a passively Q-switched, frequency doubled neodymium-doped

yttrium aluminum garnet (Nd:YAG) microchip laser was selected to produce a stable

train of pulses. 3.48 pJ per pulse are output at 7.76 kHz. The beam diameter is 0.2 mm

with 2 mrad divergence. Laser light passes through a 10x magnification beam

expander, which also decreases the beam divergence to 0.2 mrad. After the beam

expander, each pulse is passed through a holographic diffraction element that

separates the beam into a 10x10 square array of beamlets rotated 450, each containing

~35 nJ of energy. Outgoing laser beamlets then pass through shared afocal lenses with

a total magnification of 4x. The laser light is deflected through two Risley prisms (V-

coated optical wedges with 99.5% light transmission). BK7 material is selected with a

refractive index of 1.52, wedge angle of 13.580, diameter of 88 mm, and central

thickness of 22.5 mm. The refractive properties of these elements combined with the

wedge surface angle deflect the beam through a predicted path. This allows for the


147









array of laser beamlets to be arranged in a variety of repeatable patterns. Different

types of patterns can provide enhanced coverage for specific applications.

Temporal filtering via a range gate eliminates atmospheric backscatter

immediately outside the aircraft and ensures that strongly reflective targets near the

output window do not produce a response. Incident radiation passes back through a

shared afocal lens. Spatial (2.8 mm field stop) and spectral (0.3 nm bandwidth about

532.1 nm) filters reduce unwanted interference, and the filtered beamlets are rotated

450 by a Dove prism and refocused via a telephoto lens before reaching the MCP-PMT

and the associated timing electronics.

Because a very low amount of backscattered radiation is expected from each laser

beamlet, a detector must be employed with high enough gain to amplify single photon

events. In order to achieve 105 gains with 200 ps rise time, 25 ps transit time spread

(TTS), and 200 ps dead time specifications, a micro-channel plate photomultiplier tube

(MCP-PMT) was selected. This device is capable of simultaneous detection of

independent events from a 10x10 pixel array, allowing for two-dimensional resolution.

The MCP-PMT provides exceptional gain while maintaining compact size (less than 10

cm3), immunity to anode saturation, and stability with respect to variations in supply

voltage. Varying threshold settings with regards to pulse height distribution plots can

provide application-specific versatility.

Ground-based field testing results have yielded 10 cm range accuracy at one

standard deviation from the mean and small-scale feature identification at variable scan

settings. Building edges, tree structure, bushes and narrow light poles are easily

discernable from background noise. GPS time tag information reconciles the position of


148









the scanner at any given time in relation to the temporal sequence of laser shot in a

data set and Risley prism wedge index marks. Analyzing this data allows for study of

potential data flow problems, e.g. if the laser fire slaving system happens to miss

synchronization pulses, or if the scanner rotation is malfunctioning. An accurate time

history with regards to scanning operations aids in point cloud rectification and

diagnosis of mechanical errors.

The CATS sensor is capable of recording multiple returns in each channel of the

PMT from each laser pulse. Analysis of distributed returns allows for identification of

ground reflections under dense foliage and penetration past the water surface. Static

experiments were conducted to test LSNR LIDAR performance metrics such as

minimum separation distance between multiple scattering objects, water penetration

depth for submerged surfaces, and multi- vs. single-photon sensitivity. Discriminating

between accumulated histogram waveforms of two separate objects required a

separation of 15 cm. An upper limit or "blind zone" for multiple intra-shot returns was

thus established.

Varying the PMT sensitivity changed the contrast between noise production and

detected signal strength. Atmospheric noise increases significantly on a consistent

basis as gain increases toward a single photoelectron threshold (due to solar noise

events generally being single photon events). Depending on the reflective properties of

targets in the scan area, employing lower gain can decrease the amount of noise

throughout the range gate.

The CATS offers the capability to change the PMT supply voltage between

experimental collections. Operating at multi-photon sensitivity may have value for


149









situations in which high ambient noise precludes single-photon sensitivity, as ground-

based testing has indicated that man-made objects such as buildings features are easily

distinguishable even at lower gain. Airborne data collections at multi-photon sensitivity

(2300 V or 2.105 gain) are capable of producing responses 3 to 4 bins deep from

surfaces such as building roofs, bare earth, and tops of tree canopies. Reducing the

tube voltage could help to reduce dynamic range variance. Generally speaking, if

interrogated far-field surfaces are of uniform reflectance and the gain of the tube is

matched to provide exactly enough amplification to barely produce a threshold crossing,

this might be an acceptable technique. It is difficult, however, to estimate target

reflectance properties ahead of time, and tailoring a detection strategy to one type of

feature negates the real benefit of this type of sensor-mixed mapping. Static testing

shows that shallow-water bathymetry is difficult at less than full amplification (2500 V or

8-105). For the LSNR detection paradigm to be worthwhile related to submerged

surface mapping, an amplification level should be maintained that produces returns

from even very weak scatterers.

Another solution has shown significant promise for reducing dynamic range

without significantly altering bathymetric responses: changing the type of airborne

profiling from on-nadir to off-nadir. By imposing a fixed incidence angle relative to

ground for outgoing laser light, the magnitude of the backscattered radiation is

decreased due to less significant contributions from specular reflections and increased

slant range. This has special relevance in the case of mirror-like water surfaces as

large specular components cause problems for shallow bathymetry on-nadir operation.


150









A circular pattern is one way to propagate this effect throughout a scan and retain

contiguous terrain coverage; in fact along-track resolution is effectively doubled.

Simulation has suggested that surface and bottom can be reliably separated in

sea water to a depth of 2 m [21]. At larger depths, simulation also suggests that the

horizontal resolution of the reconstructed sea bottom is greatly limited by water clarity.

Static tests of the CATS sensor show that at a depth of 50 cm in coastal water, on

average 12% of shots registered at least one signal event. This was estimated to be a

sufficient signal level to produce meter-scale shallow (2 to 5 m depth) bathymetry data

at coarse spatial resolution (1 m).

Before reporting on bathymetric statistics from airborne data, contributions from

several different noise sources were considered. Two "ghost" images commonly

appear after responses from highly reflective surfaces. The first is the largest in

amplitude and consistently occurs approximately 12.67 ns beyond the first return of the

true surface. The second is approximately half the amplitude of the second, and occurs

24.00 ns later. Three possible sources have been suggested for the ghost images:

afterpulsing from gas impurities in the tube envelope, impedance mismatching in PMT

cables, and gain block overdrive. The latter was shown to be the most likely cause of

the spatially coherent noise due to the dynamic range relation. Reflections from

impedance mismatches in transmission line components will be distributed at similar

temporal offsets, but their magnitudes are expected to not be significant. Increased

levels of random noise after strong input signals suggest contributions from PMT

afterpulsing; factory testing should be performed to identify diffused impurity levels.


151









Though more significant than dark / atmospheric noise, the magnitude of the

artifacts was evaluated with regards to local point cloud densities. Because the

calculated densities were fairly low (peaks less than 1% of the true signal), we can

perform point cloud filtering and expect not to lose a significant amount of signal

information about highly reflective surfaces. The result is an affirmation of the temporal

coherence idea for LSNR LIDAR expressed by Degnan in 2001 [15]. Even with

potentially varied sources of noise producing corruptive data, we can still identify

important features via automatic cleaning algorithms.

After post-processing, data from the CATS has largely confirmed bathymetric

predictions via airborne profile scans over a fresh water lake in Mine Run, Virginia. The

central 16 channels, or 17% of the total footprint area, were sufficient to reconstruct a

fresh water lake bottom composed of dark decaying organic / inorganic material

(estimated to be highly non-reflective) out to 2 m. Approximately 3.49% of the outgoing

laser shots registered a return from the lake bottom. With partial occlusion from an

overhanging tree near the lake shore, multiple returns still provided enough information

about the submerged surface to fit a gradually decreasing trend line consistent with

more highly dense points farther away from shore. An accuracy survey at five near-

shore locations yielded 21 cm average error.

In a salt water intracoastal area near St. Augustine, FL, a single channel of

information (1% of the total footprint area) was used in profile mode to reconstruct the

submerged floor out to 2.5 m depth. The light sand here was estimated to be highly

reflective and turbidity near ideal for these types of coastal waters. On average 33.98%

of the outgoing shots registered returns from the bottom. An accuracy survey at five


152









dock platforms yielded 17 cm average error. In the surf zone near the beach, three

channels of information (3% of the total footprint area) were able to reconstruct the sea

floor in water of optimum clarity out to 2 m depth. Sea surface points in this part of the

cloud are sparse, meaning that the local density filter produces voids near the surface in

expected white water areas. The sea floor profile underneath these areas is

uninterrupted, however, with limited discontinuity near the beach / water interface. On

average 13.98% of the outgoing shots register returns from the sea floor.

Future Work

Automatic processing of topographic and bathymetric data can be difficult with

high water turbidities and low-reflectivity bottom surfaces. Even with off-nadir operation,

and the potential advantages of circular versus raster scanning, CATS data shows that

there is still a dynamic range issue because of the large difference in signal strength

between topography / sea surface elements and distributed targets such as partially

occluded forest canopies and submerged surfaces. For large-scale uniform surfaces

such as building roofs, it may be possible to automatically collapse range tails into first

returns and simultaneously provide some measure of intensity about the response,

dependant on the number of range bins that the signal stretches over and the expected

cross-section geometry [24]. Dynamic range is one of the major challenges for adaptive

algorithm development [73]; automatic procedures are usually dependent on the local

coherence of return information.

It has been shown that gain block overdrive is the most likely cause of ghost

images in CATS data. Within the current receiver structure, we may be able add a

limiting circuit to ensure that input to the block does not exceed maximum amplifier

tolerances. Clipping large signals would result in a higher noise figure, but single-


153









photon level responses should occur below the -12 dBm level. We should then be able

to limit responses to this level and eliminate the ghost effects.

Another proposed method for combating the dynamic range problem is to

implement a "sensing" stage to modify signal amplification on a dynamic level. For

example, the high voltage in the CATS system is already stepped on a per-shot basis in

order to reduce power consumption and extend PMT life. If a separate photodetector

were placed in front of the primary MCP-PMT (whether physically or by temporal delay

sequencing) such that it were possible to determine the intensity of the backscattered

radiation before stepping up the tube high voltage, the level to which the voltage was

stepped could be dynamically altered. It would be possible, then, to obtain different

amplification for sea surface returns as opposed to sea bottom returns with fairly

minimal modification to the current system structure.

The MCP-PMT is already operating at maximum amplification, but if a sensing

stage negated the possibility of full amplification during periods of high intensity laser

backscatter, it would be possible to fly lower than 500 m and potentially increase the

maximum water penetration depth. The 10x10 array output is eye safe down to 100 m

altitudes, so the only restricting factor even for populated areas are FAA regulations.

Flying lower would also decrease the ground point spacing, providing a denser cloud

and discrimination of finer features.

The total power draw for the CATS system is 372 W during profiling and 396 W

during scanning. In comparison, the Optech Gemini system power draw is average 560

W and peak 980 W. If a dedicated ASIC or FPGA could be developed to encapsulate

system control on a scripted level, the need for a control laptop could be eliminated from


154









the CATS design. The laptop currently requires 200 W, or approximately half of the

total power draw of the entire system. Compensating for this part of the system with a

low-power processing unit would move the unit closer to the realm of unmanned air

vehicle (UAV) payloads.

The potential for small-feature identification in an airborne scanning capability has

been shown in a second-generation system developed by Sigma Space Corporation. A

footbridge, gazebo, and street lamps are visible in Fig. 6-1 from flight testing over the

U.S. Naval Academy Campus in Annapolis, MD. Design parameters for this system are

similar to the CATS design, with the exception of higher (1.5x) output power and faster

pulse rate (22 kHz) [81]. A profile section in the Chesapeake Bay is shown in Fig. 6-2

depicting bottom reconstruction out to a depth of 4 m. The maximum depth tolerance in

this turbid media (estimated water clarity 14/100) is larger than CATS results in water of

higher clarity, meaning that increasing laser pulse energy has significant effect. An

interesting result is shown in Fig. 6-3 for a profile over a melt pond in Greenland; here

we have some of the best water clarity and bottom reflectivity possible. We can see the

frozen lake bottom out to 15 m depth-this suggests potential for LSNR LIDAR in

submerged glacial mapping.

Conclusions

We can break down the essential qualities required for a successful ALB system

into three general categories: accuracy, capability, and cost effectiveness [70]. For

CATS data, comparisons to submerged ground truth in two different environments have

shown average error within IHO Order 1 standards (50 cm at 95% confidence interval)

for shallow (<5 m) waters. Seamless mapping at the land/water interface and within the

surf zone has been demonstrated, and reconstructions of submerged bottom profiles


155









were shown for a variety of cases. The potential for larger maximum depth tolerance in

more turbid media has been indicated via reference to a second-generation system; this

system used slightly higher output energy (50 nJ) and the same detection methodology.

The successes with drastically lower output energies (five orders of magnitude less than

the traditional mJ-level) suggest that a compact, low-power microlaser and small-form

factor MCP-PMT can serve as a basis for potential UAV miniaturization. Elimination of

full-waveform digitization, which is commonly found in most modern ALB systems,

further supports this approach and also allows for significant cost reduction.

We have arrived at these conclusions by lowering the detection threshold down to

the single photoelectron level, allowing the system to record all effective observations

without having to resort to an interpretation strategy such as a classic CFD

implementation. This has the benefit of avoiding the requisite optical power integration

problem normally encountered when working with a large amount of backscattered

photons. The natural caveat is increased probability of false positives, but we have

shown that spurious noise can be kept to a minimum even at high amplification with

appropriate selection of spatial, spectral, and temporal filters.

Using a very short laser pulse width (500 ps) allows for a sufficiently fast timing

receiver to differentiate surface and bottom to a small margin (approximately 15 cm

accounting for returns in separate range bins). This is not quite zero depth, but for

bathymetric mapping beyond 0.5 m it is an acceptable margin. Combining this

approach with expansion of the primary beam into an array of sub-beamlets produces

ultra-dense point clouds. As a result, even if uncertainty exists in representation of

volume backscatter and bottom reflection, geometric patterns can be exploited to aid


156









classification. This idea was illustrated by eliminating data artifacts in LSNR data

associated with large dynamic range based on spatial density coherence. The

possibility of exploiting the related detection paradigm has previously been evaluated in

only a few developmental systems.


157

























Figure 6-1. Flight testing of second generation LSNR LIDAR by Sigma Space
Corporation over US Naval Academy campus in Annapolis, MD [81].


Figure 6-2. Flight testing of second generation LSNR LIDAR by Sigma Space
Corporation in Chesapeake Bay off US Naval Academy Campus. Point cloud
depicts bay bottom out to approximately 4 m [81].


158





















Figure 6-3. Flight testing of second generation LSNR LIDAR by Sigma Space
Corporation in Greenland. Point cloud depicts ice bottom penetration out to
15 m [courtesy of Sigma Space Corporation].


159









APPENDIX A
DATA PROCESSING ELECTRONICS

Ranging Receiver

Gated timing events from 100 "channels" in a 10x10 array are recorded by a multi-

stop ranging receiver in the CATS. Signals from the detector elements are processed

at several different stages, as dictated by the system controller. Fig. A-1 shows a.

receiver block diagram for n channels. The four stages are:

* Pre-amplification
* Threshold Detection
* Serial to Parallel Conversion
* Data Capture / Control / Communications Logic

The receiver system detects fixed threshold level crossings from multi-element

detector pixels. A narrow pulse received by any one of the detector elements is

amplified and low pass filtered. The low pass filter is used to stretch the pulse due to

the fact that the input signal pulse width could be of the same order as the sample

period of the threshold comparator. Given test data, the bandwidth of the low pass filter

has been calculated to achieve greater than 90% counting efficiency for single

photoelectron events. The detector has high quantum efficiency and low noise with a

bandwidth greater than 1 GHz. The preamplifier is a wide band (DC to 1 GHz) low

noise RF gain block that amplifies the detected signal to a level suitable for the

processing electronics to operate.

Fig. A-2 shows a data capture board block diagram. After the signal is filtered, it

passes through an automatic gain control (AGC) amplifier. The AGC amplifier is used

to compensate for the non-uniform response between singular elements across the

detector array. AGC amplifier settings are voltage controlled via a digital to analog


160









converter (DAC). A data capture field programmable gate array (FPGA) stores

calibrated data values that are sent to the AGC DAC.

Next, the signal goes to a threshold comparator. DAC channels, associated with

the threshold comparators, provide the threshold voltage reference. This voltage

reference corresponds to the expected level for a single photoelectron event. The

FPGA control logic provides reference point data to the DAC. The output of the

threshold comparator is low voltage positive emitter coupled logic (LVPECL).

Serial to parallel conversion changes sequential pulse outputs from the

comparator to an 8-bit parallel output at 1/8th of the sample clock speed. The operation

of this block de-multiplexes sequential photon detection from the threshold comparator

and spreads the data out into eight data lines. This provides a relatively low-speed data

input for the data-capture FPGA to synchronize to its internal clock and latch the data.

Each device receives a separately driven differential clock based on the system clock.

A divide by eight clock based on the input clock is an 8-bit word framer to synchronize

the parallel data outputs. This clock output is used to latch the 8-bit word into the data-

capture FPGA. A reset pin synchronizes multiple devices together. Because each

serial to parallel converter requires its own separate clock, the high-speed system clock

is buffered with fan-out drivers. In order to sample all the signals at the same time,

each of the clock edges must be as close as possible. Programmable delay chips are

used for this purpose in order to match propagation delays.

Data management is dictated by a capture FPGA, which performs three main

functions: (1) to collect and store data from two serial to parallel converters, (2) send

commands to the corresponding DACs associated with each channel, and (3)


161









communicate with an FPGA dedicated to control logic. Each four-channel section

requires the use of two capture FPGAs. A 16-channel board will thus have eight

capture FPGAs. These units carry out actions based on instructions from control logic.

High level control functions are built into a separate layer to support communication with

the system controller. This forms the structure allowing the download of captured data

via specific commands.

Event Timer Circuitry

The practical design for implementing the aforementioned theoretical structure can

be separated into five sections:

* Back Plane
* Range/control board
* Detector Assembly
* Data Capture boards
* Front Plane

In the CATS, all boards except for the front plane and detector interface board are

plugged vertically into the back plane. The back plane accommodates a range/control

board, a detector high voltage assembly, and three 32-channel data capture modules.

The front plane board lies across and mates perpendicular to the top of the 32-channel

data capture modules. The four outer pixels of the 10x10 array are currently not

implemented because a 96 channel design was more easily created.

The back plane supports a JTAG data bus shared by the three data capture

modules. The JTAG data bus is used for command/control and reading/writing

individual channel parameters. A 4-bit differential data bus is used for downloading

captured data to the system controller via the range/control board. Each of the data

capture assemblies has four sets of 4-bit data bus outputs. For three data capture


162









modules, 48 data bus lines to the range/control board are therefore supported. In

addition, the back plane provides power and clock distribution.

The range/control board has four main functions: (1) to generate a stable high

frequency system clock, (2) provide control for the high voltage section of the detector

assembly, (3) perform range gate sequences, and (4) support central communications

between the data capture assemblies and the system controller. A JTAG link from the

system controller interface is used to send command/control information to the

range/control board. The command/control section interprets this information.

The TO pulse circuit is driven by an optical pickup monitoring bleed light from the

Nd:YaG laser. The leading edge of each pulse in the train of laser output triggers a

sequence of events related to data capture, similar to how the incremental rotation of

each optical wedge is triggered. A range gate circuit first delays applying power to the

MCP-PMT according to a user set value; this ensures that the outgoing laser pulse is far

enough away from the sensor to avoid noise associated with power-up and returns from

objects outside the intended return window. Once the gate delay operation is complete,

initializations are sent to the data capture boards and detector assembly.

The command / control circuit shown in Fig. A-3 monitors the detector assembly

data lines, and issues commands to the data capture boards via the JTAG link when the

data lines are ready for transfer. A multiplexor circuit allows for all three boards to be

controlled via one control block, and transports return information from the 96 detector

elements.

High voltage power supplies are used for detector bias. In order to control the

high voltage supplies, the detector assembly has both a DAC and an ADC. A low


163









voltage reference input to the DAC is used to program the high voltage supplies. The

system controller provides the voltage level settings to the range/control board. A

corresponding digital value is sent to the DAC located at the detector assembly. The

range/control board also reads the detector assembly ADC and uses the data for a

voltage level feedback. To maintain detector gain stability, a command circuit

incorporates closed loop control using the DAC data as the reference and the ADC data

as the feedback value.

For maximum sensitivity (and to prolong detector life) the photon-counting detector

is normally off and then gated for data collection. The high voltage is also switched to a

lower level between range measurements in order to limit the average current

consumption and build-up of heat in the associated electronics. In order for the photon-

counting detector's gain to settle, the gate pulse must be sent out prior to the start

capture signal. The detector gate pulse is derived from the start capture circuit's coarse

counter. A register within the FPGA circuit contains the terminal count that initiates the

detector gate pulse. Terminal count values for both the start capture pulse and the

detector gate pulse are received from the system controller via a circuit on the

range/control board.

Fig. A-4 shows the organization of the receiver components in the processing

electronics block assembly. Each data capture board has 16 channels. Two data

capture FPGAs and one control FPGA are used for every four-channel section. A 32-

channel module has two data capture boards placed back-to-back. An aluminum plate

is sandwiched in between these two boards for thermal transfer.


164









The front plane provides equal path routes for the high-speed detector signals

from the detector assembly. Signals from the photon-counting detector are routed via

the front plane to each of the 100 data capture channels. The signal routes are 50 0

matched transmission lines. The individual data capture channels must simultaneously

collect all signal outputs from each detector element. The front plane is used to

equalize these signal paths among all channels.

FPGA programming is performed over the backplane bus. The data capture and

data control FPGAs receive programming information from the master range/control

FPGA located on the range/control board. The same source code is uploaded to all of

the FPGAs. Individual board variations can be built into the source code and selected

using the board select address information. Additional board settings and parameters

can be sent over the system JTAG bus after the program has been uploaded. This

provides a number of advantages: only one program needs to be maintained, the

program resides within the system controller (can be easily changed), and receiver

channel boards can be interchangeable without reconfiguration.

These five parts back plane, range/control board, detector assembly, data

capture boards, and front plane are contained within a cube below the cylindrical

motor head shown in Fig. A-5. This cube housing the timing electronics can be

decoupled from the optical elements for maintenance and modification.


165














































Figure A-1. Receiver block diagram for n channels.


Each of the four Data Capture
FPGAs requires two start signals


Start pulse
Drivers



Start Pulse from
Range Gate circuit


Figure A-2. Data capture board block diagram.


166


ControUer
































SAssembly 1 Data
Assembly 2 Data
Assembly 3 Data


Figure A-3. Range/control board block diagram.


3 37


e A. 50R r
Figure A-4. Receiver components.


S181 Front Plane


U Iu uI


Photon
Counting
Detector






Detector Adapter
Board
HV Board
Control Board
- Three 32-Channel
Assemblies

Back Plane


167


- -


























Figure A-5. Assembled CATS system photo with timing electronics "box" outlined in red.


168









APPENDIX B
SYSTEM TROUBLESHOOTING

Range Offset Error

In early versions of CATS processing software, the RGS value was multiplied by

0.5 ns. This was found to be in error and current versions of code have incorporated

the correct value of 1.0 ns. Carrying out the old calculation for a nominal value of

RGS=719 would give,

t2 way,error =719 0.5 10 9 + RBN 0.5.10-9 (s) (B-1)
719 0.5 10-9 +RBN 0.5 9 10
1-way,error .3.108

719-0.5-10-9 0+ RBN-0.5.10-9
= 3.108 .3.108
2 2
RBN 0.5.10-9
=53.925+ .3.108 (m).
2

With the correct code, however,

t2-waycorr =719.1.0 10-9 +RBN 0.5.10-9 (s) (B-2)
719-1.0 10-9 +RBN 0.5 -10-9
way,corr 2 3.10

719-1.0-10-9 RBN-0.5-10-9
= .3.108 -+ .3.108
2 2
RBN -0.5.10-9
=107.850+ .3.108 (m).
2

These calculations show that ranges previously used to generate three-

dimensional point clouds would be off by a constant value of 107.85 53.925 = 53.925

meters. Small errors in coding techniques can produce large-scale errors in range

calculation. This leads into the ramifications of local oscillator variation and resulting

range bin stability.


169









Oscillator Variation

The accuracy of range event timing is dictated by the stability of a 2.00 GHz

system clock, which is directly related to the RBN recorded in data files. Any significant

variation about the central frequency requires modification of the aforementioned

equations, and correspondingly modification of data processing software.

This significance of this idea was illustrated during system construction. Concern

about the reliability of this main clock was raised after the manufacture of a new control

board, which consolidated the function of the external GPS box and eliminated the need

for its use. A PLL Control chip (controlled by an FPGA) together with a 1.95-2.15GHz

VCO (Z-Comm V610ME04) generates the system clock source of 2.00 GHz.

In the process of finishing the new control board, receiver assembly, and system

verification, the clock frequency was discovered to be about 1.90 GHz. The problem

source was traced back to the FPGA design, which did not provide an appropriate

startup reset. A new control board fixed the problem. It was speculated that the system

at some point in the past had the correct clock frequency, but there is no way of

knowing at exactly what point in time the skew began. Since the problem was traced

back to original design files, however, it is assumed that the problem presides over most

of the files collected with the old system.

The old control board at this point in time has not been re-integrated to get a

measure of the presumed incorrect frequency. Estimating that the clock reached a

minimum value of 1.90 GHz, the Range Bin length would be lengthened to 1 / 1.90

[GHz] = 0.5263 [ns]. Ranges would then be calculated as,


170









(RGS)-(. 1246-10-9)+(RBN)- 0.5263 -9) (B-3)
-way 3-10' (m).

Processing for previous data collections were altered to accommodate this value.

The PLL chip used has since been replaced and verified at 2.00 GHz. It is instructive to

note the importance of this central component; any drift over time is difficult to identify in

point cloud data without calibration ranging for known distances.

Scan Synchronization

Image Generation: Early Steps

The first attempts to synchronize laser shot time tags to scanner index mark time

tags from test data sets exposed some potential errors in both laser and scanner

timetag outputs. For example, the consistency of the observed scan periods using the

recorded absolute GPS time was checked. To do this, scanner time tags from

T000024.STD (a horizontal building scan at 2100 V) were imported into MATLAB.

In this particular file there were 596 entries of scanner index times. This scan was

run for 30 seconds, resulting in 30/596 or 0.05034 s per revolution; roughly 19.86 Hz.

Observing the rotational period recorded from each scanner wedge, i.e., IndexTime2-

IndexTimel, IndexTime3-IndexTime2, etc. produces the plots in Fig. B-1.

Notable in the bottom image are large deviations in what should be a relatively

constant period of approximately 20 Hz. The calculated differential times occasionally

spike to negative values. Looking at these spikes more closely in the actual data

revealed a possible explanation: the microsecond counter not rolling over correctly. For

example, consider the raw data from the scanner index times of file T000024.STD at

entries 256 to 262, shown in Fig. B-2.


171









Entry 257 should read 18:38:24.010854, instead of 18:38:23.1010854. There was

a problem with the carry flag in the rollover calculation; the GPS seconds eventually

were updated correctly from xx:xx:23 xx:xx:24, but not until entry 261. This counter

error explains the large discontinuities in the observed scan period.

Assuming that these spikes were due to this error in time tag production and not

an error in scanner motion, the time outputs were filtered to what the maximum range of

the scan period should have been for this small interval. During testing it was observed

that the laser pulse rate deviated from anywhere between 7800 and 8100 shots/second.

Given that the scanner was correctly syncing 400 laser shots to one cycle of revolution,

this yields a potential data window of 19.5 to 20.25 Hz,

19.50 0.05128 (B-4)
-- (Hz) -> (s), per period .
20.25 0.04938

The resulting filtered data contains 578 of the 597 total time tag outputs (97% of

data retained), so it appears to be a reasonable window of data to look at given that

there are around 20 errors in tag generation on rollovers. This filtered data is plotted in

Fig. B-3. The images indicate that the scanner working as expected given that the laser

fire rate was slightly wandering around 8 kHz.

These observations about scanner index time tags were corroborated by analyzing

several more datasets; the motion controller appeared to be working per the design

specification. Unfortunately because of problems with shot time tag generation, it was

not possible to observe the position of the scanner in relation to successive periods of

laser fire unless the laser was assumed to be perfectly slaved to the scanner via the

motion control system. Other data collection runs were made without the benefit of time


172









tag information, however, so several previously developed techniques were employed

to try and view the test ranges from that day regardless of accurate timing records.

Point Cloud Results from a Building: Problems

The procedure for creating a point cloud from early CATS test data began by

simply plotting the recorded ranges about a 400 shot period. In order to compensate for

deviant non-uniform behavior in the receiver channels (i.e. multiple returns in certain

channels and zero returns in others) only the per-shot range average was plotted, rather

than plotting individual returns from all 96 independent channels. These averages were

then staggered across 400 angular intervals a projected relative to a fixed origin. Fig.

B-4 (left) shows a top view of a point cloud created from a single horizontal scan line.

The right side of Fig. B-4 shows an image resulting from overlay of 400 shot

intervals in a 10,000 shot dataset. The image of the bank building wall is expected to

look similar to the image on the left with some deviation about the general structure.

The right image, however, shows a cloud that cannot be attributed to excessive noise

effects. Possible explanations for the garbled image included dropped synchronization

pulses, incorrect prism starting locations, or data overflow issues. The importance of

having a diagnostic tool to pinpoint the problem namely a correctly functioning timetag

structure is evident here.

Shot-By-Shot Analysis of Early Data

Methods for point cloud production resulting in the distorted image assumed that

the data was distributed evenly according to a 400-shot period (i.e. shots N to N+399

comprise one scan period). If, however, there happened to be an error in laser sync

pulse detection, there would be no correction to shift the period. The resulting data

would then have "missing shots" where no shot time tag was recorded. Initial plots of


173









accumulated 400-shot data indicated that this was a plausible theory given that were

problems with the repeatability of the period.

Manual analysis of 400-shot segments identified a repeated set of features, shown

in the Fig. B-5(A). This image is a top view of a point cloud generated using average

ranges from a single 400-shot window distributed across angular intervals from a fixed

origin point. Fig. B-5(B) shows a zoom-in of a pattern contained within a single 400-

shot set. When average ranges from three sets of 400-shot increments (blue, red, and

green points in the bottom image) are plotted on top of each other as a function of

footprint number, they should depict the same geometric features-being that they

correspond to multiple repeated scans of the same building face. Fig. B-5(C) illustrates

how the geometric patterns for three 400-shot increments did not align, however. From

this image, it is likely that the scanner rotation rate was changing significantly across

these three data windows.

It was later discovered that the electronics used to detect the synchronization

pulses from the laser pulse train and the scanner index marks was not sampling at a

high enough rate. As a result, slight deviations in pulse width were causing "dropouts"

in pulse detection, which led to such errors as the perceived wandering scan period.

These types of hardware malfunctions could have been easily identified with a correctly

operating timetag structure.

Data processing operations such as these, however, solidify the importance of the

fifth data field data time tagging and motivate its further development. Further data

collection was put on hold until a new timetag structure was implemented.


174









Data Time Tagging


Timetag Theory Re-Evaluation

Analysis of scan data had shown that data processing was not able to reliably

predict the shot number corresponding to the start of a scan period. It was necessary to

re-evaluate the timetag structure. Each individual data collection resulted in the output

of four timing files. They were as follows:

* A.STD: the time at which the A-scan wedge index mark was detected. This is
recorded as a coarse time (seconds) and a fine time microsecondss).

* B.STD: the time at which the B-scan wedge index mark was detected. This is
recorded as a coarse time (seconds) and a fine time microsecondss).

* Ranges.SPR: all relevant ranging information (channel #, range bin, etc.). This file
also contains the output value of a 1 MHz microsecond counter which is reset
every 100 Hz. This is the only time information individually recorded for every
shot.

* I. STD: an internal system clock that is updated when a 100 Hz reset signal is fired.
This file also contains information about coarse time read from a GPS 1 Hz signal.

It is possible to calculate the timetag of each shot number through combination of

a local 1 MHz counter value in the range file and the current time value in the I file. This

involved matching the microsecond counter value in the Ranges.SPR file to the

appropriate time in the I.STD file. The resulting shot time tag would be:

t.r = I.STDoarse (B-5)
tfe = I.STDfne +Ranges.SPRfne.

The position in the I.STD file is incremented when a 100 Hz reset in the

Ranges.SPR file is detected. Shot timetags are matched up with either the A.STD or

B.STD times to get the shot corresponding to the start of a scan period (hereon be

called shot0. The shot time tag occurring closest to the first A-index time is labeled as

shot.


175









Problems with Previous Timetag Implementation

A direct comparison to the timetag of the scanner index wedge times should then

yield the shot number that corresponds to the start of a scan period. This did not turn

out to be true. The point clouds derived from these predicted shot had the forescans

and backscans mis-aligned, shown in Fig. B-6. Only when shot was manually shifted

(by anywhere from 20 to 350 shots) was the proper point cloud generated. This

procedure is illustrated in Fig. B-7.

A number of issues identified in the individual timing files could have contributed to

the difficulties of predicting shot. A series of data showed that the 100 Hz micro-

second counter reset was often missed, causing gaps in the I.STD data as well as

microsecond counter values in the range data file that exceeded 10000. The

microsecond counter must be reset before rolling over to 10000 or an error results.

There were also aberrations in the microsecond counter value where the increment

from one shot to the next was 2-3 times the normal value (implying dropped shots).

Additionally, the number of entries in the I.STD data file did not match the number of

resets detected from the 1 MHz counter in the range data file. Because of the issues

plaguing the I-file and detection of the corresponding 100-Hz signal, this part of the

timetag structure was eliminated. Shot timetag operations were moved onboard the

FPGA, as described in the next section.

Restructuring of Timetag Generation

The logic for timetag operations was stored onboard a Xilinx FPGA. Index mark

timetags were produced as a result of four input pins tied to the following signals: (1) 1

Hz GPS signal, (2) a free running counter based on a local 1 MHz oscillator, (3) A index

mark signal, and (4) B index mark signal. The leading edge of a pulse detected on pins


176









3 or 4 generated an interrupt which latched in data from the stored GPS time (updated

every second via a signal on the first pin) plus the current value of the free-running

counter. The files A.STD and B.STD were generated by this process. The I/O is

expressed graphically in Fig. B-8(A).

Shot timetags were created by essentially the same method, but using a different

fine precision counter. This method used a different 1 MHz local oscillator from inside

the main electronics "cube" rather than index mark oscillator, which was found on the

controller board itself. The microsecond counter was reset by a 100 Hz internal signal,

found in I.STD. The value of the counter was matched to the appropriate time in the

I.STD file to create the shot timetag as previously described.

In reference to the problem of determining shot, questions were raised about the

synchronization between the two different local oscillators. It was theorized that the

accuracy of the sync might be related to the throughput capacity of the USB link versus

the Ethernet link due to the control loops routing certain signals. High load conditions

could produce an offset that would create a disparity producing a calculated shot

offset. To solve this potential problem, laser fire pulse train was directly input into the

FPGA used to tag the index mark signals, shown in Fig. B-8(B). A new output file in the

same format as the index mark files was then created to record the shot timetags. This

required modification of the FPGA programming and signal routing.

Testing of Timetag Restructuring

The new structure was tested in Fairfax, VA. Data collections with the sensor

pointed at a building edge in the distance provided point cloud data with distinct

geometries, allowing for characterization of scanner behavior for consecutive 400 shot

periods. Timetag files were analyzed to determine shot for each data set. Out of 11


177









data sets, 9 registered the correct value of shot, allowing for the scan pattern to be

correctly rectified in the point cloud.

During system testing to evaluate the new timetag structure, the scanner exhibited

some movement errors visible to the eye. For example, during night operation of a

static horizontal line scan, the observed output pattern began to slowly rotate such that

the pattern eventually became vertical. These rotations did not occur throughout all

conducted trials, so a second round of tests was conducted. The same large-scale

scanner error was observed, and further post-processing of timing data collected during

aberrant scanner behavior showed the B wedge rotational period varying slowly from

trial to trial.

It was theorized that a physical malfunction of the B wedge was causing the

system to produce inconsistent scan patterns. The system was brought into a

laboratory for observations of diagnostic signals on an oscilloscope. The width of the A

index pulse was consistent at about 1.5 ps, but the width of the B index pulse fluctuated

between 1 and 3 ps. This pointed to a malfunction in the optical encoder associated

with the B wedge. Upon opening the sensor head for visual inspection of the encoders,

a large crack was found on the B encoder plate. Without timetag processing, this issue

may have not been explored for an unacceptable amount of time.

Correctly operating timetags are a significant part of the CATS instrument design.

Without accurate knowledge of when the start of a scan occurs, it would be difficult to

align recorded return events to scanner behavior without distinct point cloud geometry.

Since intended targets for airborne operation are expected to not be ideally symmetric,

it is essential that a correct shot is calculated. Timetag comparison also allows for the


178









diagnosis of potential problems with individual system components. For example, laser

double pulsing can be identified and corrected via post-processing of event time

histories.


179

















Time Difference Between Successive Index Scans (observed from GPS Timetags)


1-

0.5-
SA
0-

-0.5-

-1
0 100 200 300 400 500 600
Number of Index Pulses-1
Scanner Rotation Rate Between Successive Scans (observed from GPS Timetags
30



0
0-


-10

-20 -
0 100 200 300 400 500 600
Number of Index Pulses-1



Figure B-1. Differential measurements of CATS scanner periods. A) Time difference
between successive index scans. B) Scanner rotation rate between
successive scans. Possible errors in scanner timetags are indicated by large
discontinuities.



File Edit Format View Help
00247. 05/26/2006 18:38:23.505770 A
00248. 05/26/2006 18:38:23.556276
00249. 05/26/2006 18:38:23.606789
00250. 05/26/2006 18:38:23.657289
00251. 05/26/2006 18:38:23.707786
00252. 05/26/2006 18:38:23.758314
00253. 05/26/2006 18:38:23.808706
00254. 05/26/2006 18:38:23.859326
00255. 05/26/2006 1 Q Q7nR 4
00256. 05/26/2006 18:38:23.960344
00257. 05/26/2006 18:38:23.1010854
00258. 05/26/2006 18:38:23.012790
00259. 05/26/2006 18:38:23.063306
00260. 05/26/2006 18:38:23.113805
00261. 05/26/2006 18:38:24.164312
00262. 05/26/2006 18:38:24.214818
00263. 05/26/20061 ": :M4. '2bS3
00264. 05/26/2006 18:38:24.315841
00265. 05/26/2006 18:38:24.366347
00266. 05/26/2006 18:38:24.416860
00267. 05/26/2006 18:38:24.467368
00268. 05/26/2006 18:38:24.517885
00269. 05/26/2006 1: 38:24.56i874
Figure B-2. Time data file associated with possible scanner timetag errors. A segment
of the data is windowed to demonstrate rollover errors in timetag generation.


180












Filtered Time Difference Between Successive Scans (no errors)


0.0508

S 0.0506-|- A


0.0502
0.0502 --------------------------------
0 100 200 300 400 500 600
Number of Index Pulses -1
Filtered Scanner Rotation Rate Between Successive Scans (no errors)
19.9-,-,-----,

19.85

19.8 B



19.7'
0 100 200 300 400 500 600
Number of Index Pulses-1



Figure B-3. Filtered differential measurements of CATS scanner periods. A) Time
difference between successive scans. B) Scanner rotation rate between
successive scans. The large-scale discontinuities are notably not present,
and high-frequency characteristics are consistent with laser repetition rate
fluctuation.



3,30
285
320-
310- 280 l. o U





270
265
260 *

250 260
-60 -50 .40 .30 -20 .10 0 -80 -60 -40 -20 0 20 40 60 80


A B
Figure B-4. Point clouds created from preliminary CATS sensor data. A) Top view of a
point cloud created from a single horizontal scan line of a bank building face
approximately 300 m away. B) Top view of a point cloud created from a
horizontal scan of bank building face, resulting from overlay of 400 shot
intervals in a 10,000 shot dataset.


181














500
450
400
350 .
300 -
200 +
200 *
150
100
-100 80 -60 40 -20 0 20 40 60 80 100

A


310




,* ...,."I



270

260

30 20 -10 0 10 20

B


Raw Range Data for Various Scans

Footprints 51240-51639
Footprints 51640-52039
Footprints 63240-63640


*


.


%
A. V4X A
f
.b *<

/


* -
I-i



**
* "

+


- .
'~ +


* *


50 100 150 200 250
Footprint # (0-400)


Figure B-5. Illustration of scan period wandering. A) Top view of a point cloud
generated using average ranges from a single 400-shot window distributed
across angular intervals from a fixed origin point. Ranges from one scan
period corresponding to 400 laser shots are plotted with constant angular
intervals. B) Zoom-in of geometric pattern contained within a 400-shot
window. C) Average ranges for three 400-shot windows are plotted on top of
each other as a function of footprint number to illustrate wandering scan
period.



















182


i 140
a,

- 120


100


80


350 400


"'


.*
r *













Point Cloud Generated Using ShotO = 121
.1


250

240 -

230 -

220 -


190

180

170


areas scanned on forescan/backscan.
Should be overlapping!


1501 1 i'- I I I I I
-60 -40 -20 0 20 40 60
X (m]


Figure B-6. Point cloud for horizontal profile scan of bank building. Figure was
generated using shotO calculated from GPS timetags. The two highlighted
areas represent the same building structure and should overlap, but do not
because of the error in shot calculation.


Pot- C aud 6ww d Ui Sh fm 121


Figure B-7. Effect of shoO variation in generated image. As shot is varied, the
forescan (the image produced from the first 200 shots) and the backscan (the
image produced from the last 200 shots) begin to align and merge. The point
clouds are generated using predicted shot (A), shot = 150 (B), and shot =
145 (C). 145 is the value of shot that results in the best point cloud.
















183


P... C(d C-.,.s VIII Sh C 11s)


PMM Cld GtfwMB Lon Sho 0 1GS










i I a M0 Qo w
1


O
~e,~o~ooloaa,
rlni


4 x3 0 20 n 2 Q W0
Moo os















Input pins

GPS signal (1 Hz)

free-running counter (1 Mhz)

A pulse (-20 Hz)

B pulse (-20 Hz)


old design


output pin




- ASTD
i B.STD


I
GPS signal (1 Hz) -
2
free-running counter (1 Mhz)
3
Apulse (~20 Hz) -
4
B pulse (20 Hz)
laser fire pulse (-8 kHz) -


Figure B-8. FPGA block diagram showing redesign of CATS timetag structure to
improve calculation of start of scan periods. A) Original structure. B) New
structure.


184


A.STD

B.STD

ShoTag.STD


new design









LIST OF REFERENCES


[1] J. B. Campbell, Introduction to Remote Sensing, 4th ed.: Taylor and Francis Ltd,
2007.

[2] "The Imaging and Geospatial Information Society," Bethesda, MD: American Society
for Photogrammetry and Remote Sensing, 2009.

[3] D. Satale and M. Kulkarni, "LiDAR in Mapping," 2003.

[4] H. Lee, K. Kampa, and K. C. Slatton, "Segmentation of ALSM point data and the
prediction of subcanopy sunlight distribution," in Geoscience and Remote Sensing
Symposium, 2005, IGARSS '05 Proceedings, 2005 IEEE International, p.4, 2005.

[5] B. J. Luzum, K. C. Slatton, and R. L. Shrestha, "Identification and analysis of
airborne laser swath mapping data in a novel feature space," Geoscience and
Remote Sensing Letters, IEEE, vol. 1, pp. 268-271, 2004.

[6] B. J. Luzum, K. C. Slatton, and R. L. Shrestha, "Analysis of spatial and temporal
stability of airborne laser swath mapping data in feature space," Geoscience and
Remote Sensing, IEEE Transactions on, vol. 43, pp. 1403-1420, 2005.

[7] N. Haala and C. Brenner, "Extraction of buildings and trees in urban environments"
ISPRS Journal of Photogrammetry and Remote Sensing, vol. 54, pp. 130-137,
1999.

[8] Yu, Yang, and Chen, "Automatic feature extraction and stereo image processing with
genetic algorithms for LiDAR data," in International Conference on Computer
Graphics, Imaging and Vision: New Trends, 2005.

[9] A. Alharthy and J. Bethel, "Heuristic Filtering and 3D Feature Extraction From LIDAR
Data," in ISPRS Technical Commission III, Graz, Austria, 2002.

[10] S. C. Popescu, R. H. Wynne, and J. A. Scrivani, "Fusion of Small-Footprint Lidar
and Multispectral Data to Estimate Plot-Level Volume and Biomass in Deciduous
and Pine Forest in Virginia, USA," Forest Science, vol. 50, 2004.

[11] M. Lee and G. Tuell, "A Technique for Generating Bottom Reflectance Images from
SHOALS Data," in U.S. Hydro 2003 Conference, Biloxi, MI, 2003.

[12] K. C. Slatton and W. E. Carter, "A Primer for Airborne LIDAR," 2008.

[13] W. Wagner, A. Ullrich., T. Melzer, C. Briese, and K. Kraus, "From single-pulse to
full waveform airborne laser scanners: Potential and practical challenges," in
International Society for Photogrammetry and Remote Sensing XXth Congress,
Commission 3 Istanbul, Turkey, vol. XXXV, part B3: pp. 6. July 2004.


185









[14] R. Gutierrez, A. Neuenschwander, and M.M. Crawford, "Development of Laser
Waveform Digitization for Airborne LIDAR Topographic Mapping Instrumentation,"
in Geoscience and Remote Sensing Symposium, IGARSS, 2005.

[15] J. J. Degnan, "Unified Approach to Photon-Counting Microlaser Rangers,
Transponders, and Altimeters," Surveys in Geophysics, vol. 22, pp. 431-447,
2001.

[16] J. Degnan, "Photon-Counting Multikilohertz Microlaser Altimeters for Airborne and
Spaceborne Topographic Measurements," Journal of Geodynamics (Special Issue
on Laser Altimetry), 2002.

[17] U. Stilla and B. Jutzi, "Waveform analysis for small-footprint pulsed laser systems,"
in: Shan J, Toth CK (Eds), Topographic Laser Ranging and Scanning: Principles
and Processing, CRC Press, Boca Raton, pp. 215-234, 2008.

[18] K. Albright, R. C. Smith, C. Ho, S. K. Wilson, J. Bradley, A. Bird, D. E. Casperson,
M. Hindman, R. Whitaker, J. Theiler, R. Scarlett, and W. C. Priedhorsky, "RULLI: A
Photon Counting Imager," in International Advanced Studies Institute (IASI)
Symposium on Detection and Analysis of Sub-surface Objects and Phenomena,
Monterey, CA, 1998.

[19] B. F. Aull, A. H. Loomis, D. J. Young, R. M. Heinrichs, B. J. Felton, P. J. Daniels,
and D. J. Landers, "Geiger-Mode Avalanche Photodiodes for Three-dimensional
Imaging," Lincoln Laboratory Journal, vol. 12, pp. 383-396, 2000.

[20] J. J. Degnan, R. Machan, E. Leventhal, D. Lawrence, G. Jodor, and C. Field,
"Inflight performance of a second generation, photon counting, 3D imaging lidar,"
Laser Radar Technology and Applications XIII, Proceedings of SPIE, vol. 6950,
2008.

[21] T. Cossio, "Prediction of Topographic and Bathymetric Measurement Performance
of Airborne Low-SNR Lidar Systems" (Ph.D. dissertation, University of Florida,
2009).

[22] M. Flood, "LIDAR Activities and Research Priorities in the Commercial Sector," in
Land Surface Mapping and Characterization Using Laser Altimetry, Annapolis,
MD, 2001.

[23] J. Degnan, McGarry, Zagwodzki, Dabney, Geiger, Chabot, Steggerda, Marzouk,
and Chu, "Design and Performance of an Airborne MultiKilohertz Photon-
Counting, Microlaser Altimeter," in Land Surface Mapping and Characterization
Using Laser Altimetry Annapolis, Maryland, 2001.

[24] P. Cho, H. Anderson, R. Hatch, and P. Ramaswami, "Real-Time 3D Ladar
Imaging," Lincoln Laboratory Journal, vol. 16, 2006.

[25] 0. Svelto, Principles of Lasers, 4th ed. (trans. David Hanna), Springer, 1998.


186









[26] A. Einstein, "On the quantum theory of radiation," Physikalische Zeitschrift, vol. 18,
issue 121, 1917.

[27] A. E. Seigman, Lasers, University Science Books, 1986.

[28] T. H. Maiman, "Stimulated Optical Radiation in Ruby," Nature, vol. 187, issue 4736,
pp. 493-494, 1960.

[29] G. R. Gould, "The LASER, Light Amplification by Stimulated Emission of
Radiation," The Ann Arbor Conference on Optical Pumping, June 1959.

[30] W. Koechner, Solid-state laser engineering, 2nd ed. Springer Verlag, 1988.

[31] R. Paschotta, "Q Factor," Encyclopedia of Laser Physics and Technology, RP
Photonics, 2003.

[32] F. J. McClung and R. W. Hellwarth, "Giant optical pulsations from ruby," Journal of
Applied Physics, vol. 33, issue 828, pp. 103-105, 1962.

[33] J. J. Degnan, "Optimization of Passively Q-Switched Lasers," IEEE Journal of
Quantum Electronics, vol. 31, no. 11, pp. 1890-1901, 1995.

[34] R. Daly and S. D. Sims, "An Improved Method of Mechanical Q-Switching Using
Total Internal Reflection," Applied Optics, vol. 3, issue 9, pp. 1063-1066, 1964.

[35] A. Szabo and R. A. Stein, "Theory of Giant Pulsing by a Saturable Absorber,"
Journal of Applied Physics, vol. 36, no. 5, May 1965.

[36] J. J. Degnan, "Theory of the optimally coupled Q-switched laser," IEEE Journal of
Quantum Electronics, vol. 25, pp. 214-220, 1989.

[37] J. J. Zayhowski and P. L. Kelley, "Optimization of Q-switched lasers," IEEE Journal
of Quantum Electronics, vol. 27, pp. 2220-2225, 1991.

[38] P. Peterson, A. Gavrielides, M. P. Sharma, and T. Erneux, "Dynamics of Passively
Q-Switched Microchip Lasers," IEEE Journal of Quantum Electronics, vol. 35, no.
8, pp. 1247-1256, August 1999.

[39] J. J. Zayhowski and A. L. Wilson, "Pump-Induced Bleaching of the Saturable
Absorber in Short-Pulse Nd:YAG / Cr4:YAG Passively Q-Switched Microchip
Lasers," IEEE Journal of Quantum Electronics, vol. 39, no. 12, pp. 1588-1593,
December 2003.

[40] ISO Standard 11146, "Lasers and laser-related equipment Test methods for laser
beam widths, divergence angles and beam propagation ratios," 2005.

[41] Segelstein, Hale, Querry, Wielickza, and Weng, "Absorption Spectra of Water
Compendium," 1973, 1981, 1989.


187









[42] P. B. Russel, T. J. Swissler, and M. P. McCormick, "Methodology for error analysis
and simulation of lidar aerosol measurements," Applied Optics, vol. 18, pp. 3783-
3797, 1979.

[43] Z. Liu, W. Hunt, M. Vaughan, C. Hostetler, M. McGill, K. Powell, D. Winkler, and Y.
Hu, "Estimating random errors due to shot noise in backscatter lidar observations,"
Applied Optics, vol. 45, pp. 3783-3797, 2006.

[44] T. Cossio, C. Slatton, W. Carter, K. Shrestha, and D. Harding, "Predicting
topographic and bathymetric performance for LSNR airborne lidar," IEEE
Transactions on Geosciences and Remote Sensing, Vol. 47, No. 7, Part II, pp.
2298-2315, 2009.

[45] Hamamatsu-Photonics, "Characteristics and use of Si APD (Avalanche
Photodiode)," 2004.

[46] Perkin-Elmer, "Avalanche Photodiode: A User Guide. Understanding Avalanche
Photodiodes for improving system performance," 2006.

[47] G. C. Guenther, A. G. Cunningham, P. E. LaRocque, and D. J. Reid, "Meeting the
Accuracy Challenge in Airborne LIDAR Bathymetry," in EARSeL-SIG Workshop on
LIDAR, 2000.

[48] R. Nairn, "Royal Australian Navy Laser Airborne Depth Sounder: The first year of
operations," in International Hydro Rev., Monaco, 1994.

[49] B. Billard and P. J. Wilsen, "Sea surface and depth detection in the WRELADS
airborne depth sounder," Applied Optics, pp. 2059-2066, 1986.

[50] Hamamatsu-Photonics, "Photon Counting Using Photomultiplier Tubes," 2001.

[51] Hamamatsu-Photonics, "Photomultiplier Tubes: Basics and Applications." vol. 2,
1999.

[52] H. R. Hertz, "Ueber sehr schnelle electrische Schwingungen," Annalen der Physik,
vol. 267, 1887.

[53] I. Moreno, "Jones Matrix for Image-Rotation Prisms," Applied Optics, vol. 43, pp.
3373-3381, 2004.

[54] R. M. Marino and W. R. Davis, "Jigsaw: A Foliage-Penetrating 3D Imaging Laser
Radar System," Lincoln Laboratory Journal, Vol. 15, No. 1, 2005.

[55] B. Jutzi and U. Stilla, "Range determination with waveform recording laser systems
using a Wiener Filter," ISPRS Journal of Photogrammetry & Remote Sensing 61,
pp. 95-107, 2006.


188









[56] K. Y. Shrestha, K. C. Slatton, W. E. Carter, T. K. Cossio, "Performance Metrics for
Single-Photon Laser Ranging," IEEE Geoscience and Remote Sensing Letters,
vol. 7, issue 2, pp. 338-342, 2010.

[57] W. Carter, R. Shrestha., C. Slatton, and W. Dietrich, "Research Quality Airborne
Laser Swath Mapping: The Defining Factors," in A White paper issued by NCALM,
2007.

[58] Y. Shao, R. W. Silverman, S. R. Cherry, "Evaluation of Hamamatsu R5900 series
PMTs for readout of high-resolution scintillation arrays," Nuclear Instruments and
Methods in Physics Research, A 454, pp. 379-388, 2000.

[59] K. Francis, P. LaRocque, L. Gee, and M. Paton, "Hydrographic Lidar Processing
Moves into the Next Dimension," Proc. of U.S. Hydro, 2003.

[60] R. W. Pope, P. Johnson, U. Lejdebrink, and J. W. Lilycrop, "Airborne Lidar
Hydrography: Vision for Tomorrow," Proc., 2001.

[61] P. E. LaRocque and E. Yang, "Enhancements to the New SHOALS 3000 System,"
International Lidar Mapping Forum, Denver CO, March 3-5, 2010.

[62] J. M. Wozencraft, "Complete coastal mapping with airborne lidar," Oceans 2002
MTS/IEEE, vol. 2, pp. 1194-1198, 2002.

[63] G. C. Guenther, P. E. LaRocque, and J. W. Lillycrop, "Multiple surface channels in
SHOALS airborne lidar," Proc. SPIE, vol. 2258, pp. 422-430, 1994.

[64] S. Pe'eri and W. Philpot, "Increasing the Existence of Very Shallow-Water LIDAR
Measurements Using the Red-Channel Waveforms," IEEE Transactions on
Geoscience and Remote Sensing, vol. 45, no. 5, May, 2007.

[65] A. Collin, P. Archambault, and B. Long, "Mapping the Shallow Water Seabed
Habitat With the SHOALS," IEEE Transactions on Geoscience and Remote
Sensing, vol. 46, no. 10, October 2008.

[66] R. A. O'Neil, L. Buja-Bijunas, and D.M. Rayner, "Field performance of a laser
fluorosensor for the detection of oil spills," Appl. Opt., vol. 19, no. 6, pp. 863-870,
Mar., 1980.

[67] M. R. Bristow, D. Nielsen, D. Bundy, and R. Furtek, "Use of water Raman emission
to correct airborne laser fluorosensor data for effects of water optical attenuation,"
Appl. Opt., vol. 20, no. 17, pp. 2889-2906, Sep., 1981.

[68] L. W. Harding Jr., E. C. Itsweire, and W. E. Esaias, "Determination of phytoplankton
chlorophyll concentrations in the Chesapeake Bay with aircraft remote sensing,"
Remote Sensing of Environment, vol. 40, issue 2, May, 1992.


189









[69] G. C. Guenther, "Airborne Laser Hydrography: System Design and Performance
Factors," NOAA Prof Paper Ser., Rockville, MD: NOS, NOAA, March, 1985.

[70] G.C. Guenther, Digital Elevation Model Technologies and Applications: The DEM
Users Manual, 2nd Edition, D. Maune, ed., American Society for Photogrammetry
and Remote Sensing, Chapter 8: Airborne Lidar Bathymetry, pp. 253-320, 2007.

[71] J. C. Brock, C. W. Wright, T. D. Clayton, and A. Nayegandhi, "LIDAR optical
rugosity of coral reefs in Biscayne National Park, Florida," Coral Reefs, vol. 23, no.
1, April, 2004.

[72] C. Hug, A. Ullrich, and A. Grimm, "Litemapper-5600- A Waveform-Digitizing LIDAR
Terrain and Vegetation Mapping System," International Archives of
Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol.
XXXVVI, part 8/W2, pp. 24-29, 2005.

[73] T. Cossio, C. Slatton, W. Carter, and K. Shrestha, "Predicting small target detection
performance of low-SNR airborne lidar," Journal of Selected Topics in Applied
Earth Observations and Remote Sensing (in press), 2010.

[74] G.A. Morton, H.M. Smith, R. Wasserman, "Afterpulses in Photomultipliers," IEEE
Trans. Nucl. Sci., NS-14, no. 1, pp. 443-448, 1967.

[75] R. Staubert, E. Bohm, K. Hein, K. Sauerland, and J. Trumper, "Possible effects of
photomultiplier-afterpulses on scintillation counter measurements," Nuclear
Instruments and Methods, vol. 84, issue 2, April, 1970.

[76] P.B. Coates, "The origins of afterpulses in photomultipliers," J. Phys. D: App. Phys.,
vol. 6, 1973.

[77] M. P. Bristow, "Suppression of afterpulsing in photomultipliers by gating the
photocathode," Applied Optics, vol. 41, no. 24, August, 2002.

[78] Y. Kanaya and H. Akimoto, "Gating a channel photomultiplier with a fast high-
voltage switch: reduction of afterpulse rates in a laser-induced fluorescence
instrument for measurement of atmospheric OH radical concentrations," Applied
Optics, vol. 45, no. 6, February, 2006.

[79] N. Akchurin and H. Kim, "A study on ion initiated photomultiplier afterpulses,"
Nuclear Instruments and Methods in Physics Research, pp. 121-126, 2007.

[80] H. Johnson and M. Graham, High-Speed Digital Design- A Handbook of Black
Magic, pp. 160-165, 1993.

[81] J. J. Degnan, "Photon-Counting Lidars for Airborne and Spaceborne Topographic
Mapping", OSA 2010 Optics and Photonics Congress, Applications of Lasers for
Sensing and Free Space Communications (LS&C) Topical Meeting, Paper
LSWA4, San Diego, CA, January 31 February 4, 2010.


190









BIOGRAPHICAL SKETCH

Kristofer Y. Shrestha was born in Madison, Wisconsin and raised in Gainesville,

Florida. He graduated from Eastside High School in 2000 with International

Baccalaureate diploma and went on to study electrical engineering at the University of

Florida. Kristofer received the Bachelor of Science degree in December 2004, and is

currently a Ph.D. student in the department of Civil and Coastal Engineering. He was

awarded the Alumni Fellowship in 2005, and now works as a research assistant as a

member of the Adaptive Signal Processing Lab under Dr. K. Clint Slatton. He continues

to search for improvements in applied science by exploring new technologies.


191





PAGE 1

1 MIXED TOPOGRAPHIC AND SHALLOW BATHYMETRIC MAPPING VIA MULTI STOP SINGLE PHOTOELECTRON LASER RANGING By KRISTOFER Y. SHRESTHA A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLM ENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2010

PAGE 2

2 2010 Kristofer Y. Shrestha

PAGE 3

3 To my mother for instilling in me a determination bordering on stubbornness

PAGE 4

4 ACKNOWLEDGMENTS This research was made po ssible only with the valued advice and support of my family, friends fellow students and professors I would like to thank Dr. Clint Slatton for directing me into this line of work and for exhibiting continual patience in the face of very difficult obsta cles. Dr. Slatton continues to serve as my inspiration to work even harder in times of tough circumstance. I would also like to thank Dr. Bill Carter for pushing me to find the alternative solution in times where I genu inely believed success to be an imp ossibility. His expertise and guidance in the field of low signal laser ranging have been critical to this work. Next, I would like to express my appreciation for the support of my fellow Geosensing students. This small piece of text is probably all that will ever document the selfless field work of these students on my behalf so I need to take this opportunity to thank Tristan Cossio, Juan Fernandez, Pang Wei Liu, Kitiipat Kampa, Pravesh Kumari, Tolee Matamas, Mike Starek, Abhinav Singhania, and Bidiya Yadav. I will not forget the freezing cold nights in Citra, the struggles at Lake Wauburg, or the countless other time s that they aided me in CATS testing. I would also like to thank the staff at the CE Structures Lab / Plant Science Research for use of the their facilities, William Gavert / Jacob Hwang for their support in diagnosing electronics issues, and Sidney Schofield for his invaluable help in preparing the system for airborne operation. Finally, I would like to express my deep gratitude to my fat her Ramesh and my mother Chew Ying for providing me with opportunities that few children get to have, and to my sister Veena for all her love. Finally, I would like to thank Cynthia Gillikin for her patience and support through this long process.

PAGE 5

5 TA BLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 9 LIST OF FIGURES ................................ ................................ ................................ ........ 10 ABSTRACT ................................ ................................ ................................ ................... 15 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 17 Introduction to Remote Sensing ................................ ................................ .............. 17 Photogrammetry and LIDAR ................................ ................................ ............ 17 Advantages of LIDAR ................................ ................................ ....................... 18 Applications of LIDAR ................................ ................................ ...................... 18 Improving Image Fidelity ................................ ................................ ......................... 19 LIDAR Point Clouds ................................ ................................ ......................... 19 Existing Met hods for Improving Feature Detection ................................ ........... 19 Existing Methods for Improving Data Quality ................................ .................... 20 The LSNR Mapping Approach ................................ ................................ ................ 21 Problems with High Energy / High Pulse Rate Designs ................................ ... 21 Motivations for a LSNR LIDAR System ................................ ............................ 22 Research Aims ................................ ................................ ................................ 24 2 LSNR COMPONENTS ................................ ................................ ............................ 25 Commercial Sector Activities and Scientific Impact ................................ ................ 25 COTS Market and Costs ................................ ................................ .................. 25 LSNR Concept Systems ................................ ................................ ................... 25 Overview of Lasers ................................ ................................ ................................ 27 Quantum Dynamics Fundamentals ................................ ................................ .. 27 Coherence ................................ ................................ ................................ ........ 29 Three and Four Level Laser M odels ................................ ............................... 30 Optical Resonators ................................ ................................ ........................... 31 Q Switching ................................ ................................ ................................ ...... 32 Microchip Laser Ad vancements ................................ ................................ ....... 34 Beam Quality Measures ................................ ................................ ................... 36 Effects of Laser Parameters ................................ ................................ ................... 36 Operating Wavelength ................................ ................................ ...................... 37 Laser Pulse Width ................................ ................................ ............................ 37 Pulse Energy ................................ ................................ ................................ .... 38 Selecting a LSNR Photodetector ................................ ................................ ............ 40 Avalanche Photodiode ................................ ................................ ..................... 40

PAGE 6

6 Photomultiplier Tube Advantages ................................ ................................ ..... 41 Combinations ................................ ................................ ................................ ... 42 Photodetector Improvements ................................ ................................ .................. 43 Photon Counting ................................ ................................ ............................... 43 Semiconductor Principles ................................ ................................ ................. 44 Micro channel Plate Photomultiplier Tube Concept ................................ .......... 46 MCP PMT Gain ................................ ................................ ................................ 46 Detection Efficiency and Pulse Height Distribution ................................ ........... 48 Variation in PMT Gain and Output Instability ................................ .................... 50 3 LSNR PROTOTYPE DESIGN ................................ ................................ ................ 60 Photon Counting System Prototype: CATS ................................ ............................ 60 General Scanning Pri nciples ................................ ................................ ............ 60 System Proposal ................................ ................................ .............................. 60 Optics ................................ ................................ ................................ ...................... 61 Microchip Laser ................................ ................................ ................................ 61 Beam Expander ................................ ................................ ................................ 62 Risley Prisms ................................ ................................ ................................ .... 63 Range Gate ................................ ................................ ................................ ...... 64 Spatial and Spectral Filters ................................ ................................ ............... 64 Dove Prism ................................ ................................ ................................ ....... 65 Telephoto Lens and Mechanical Shutter ................................ .......................... 66 MCP PMT ................................ ................................ ................................ ......... 66 Multi stop Event Timer ................................ ................................ ............................ 67 Assembly and Footprint Di mensions ................................ ................................ ...... 68 4 LSNR PERFORMANCE METRICS ................................ ................................ ........ 75 Laser Ranging Concepts and Implementation ................................ ........................ 75 Range Calculation Fundamentals ................................ ................................ .... 75 Range Discretization ................................ ................................ ........................ 76 Data Processing Fields ................................ ................................ ..................... 77 Ground based Static Testing ................................ ................................ .................. 78 Experimental Setup ................................ ................................ .......................... 78 Range Accuracy ................................ ................................ ............................... 79 Minimum Separation Distance ................................ ................................ .......... 79 Crosstalk Evaluation ................................ ................................ ......................... 82 Water Penetration Depth ................................ ................................ .................. 86 Multi vs. Single Photon Sensitivity ................................ ................................ .. 88 Ground based Scan Testing ................................ ................................ ................... 90 Horizontal Scan Building Features ................................ ................................ 90 Rotating Line Scan Building Features ................................ ........................... 90 Rotating Line Scan Building Features (Partial Obscuration) .......................... 91 5 LSNR AIRBORNE LASER BATHYMETRY ................................ .......................... 108

PAGE 7

7 ALB Background ................................ ................................ ................................ ... 108 Multiple Information Channels in Laser Ranging ................................ ............ 108 Challenges Facing Single Channel Detection ................................ ................ 110 Pulse Width and Output Energy Considerations ................................ ............ 111 Contrast to Waveform Digitization ................................ ................................ .. 112 Experimental Setup ................................ ................................ .............................. 113 Small Structure Identification ................................ ................................ ................ 114 Vertical Canopy Structure Identification ................................ ................................ 114 Optimizing Sys tem Performance ................................ ................................ .......... 115 Data Artifacts ................................ ................................ ................................ ........ 116 PMT Afterpulsing ................................ ................................ ............................ 119 Impe dance Matching ................................ ................................ ...................... 120 Gain Block Overdrive ................................ ................................ ..................... 123 Point Cloud Filtering Based on Density Estimates ................................ ................ 124 Fresh Water Bottom Mapping ................................ ................................ ............... 125 Salt Water Bottom Mapping ................................ ................................ .................. 126 Mapping Statistics and Discussion ................................ ................................ ....... 127 6 CONCLUSIONS AND FUTURE WORK ................................ ............................... 145 Scope of Research ................................ ................................ ............................... 145 Desi gn and Testing Overview ................................ ................................ ............... 145 Future Work ................................ ................................ ................................ .......... 153 Conclusions ................................ ................................ ................................ .......... 155 APPENDIX A DATA PROCESSING ELECTRONICS ................................ ................................ 160 Ranging Receiver ................................ ................................ ................................ 160 Event Timer Circuitry ................................ ................................ ............................ 162 B SYSTEM TROUBLESHOOTING ................................ ................................ .......... 169 Range Offset Error ................................ ................................ ................................ 169 Oscillator Variation ................................ ................................ ................................ 170 Scan Synchronization ................................ ................................ ........................... 171 Image Generation: Early Steps ................................ ................................ ...... 171 Point Cloud Results from a Buildin g: Problems ................................ .............. 173 Shot By Shot Analysis of Early Data ................................ .............................. 173 Data Time Tagging ................................ ................................ ............................... 175 Timetag Theory Re Evaluation ................................ ................................ ....... 175 Problems with Previous Timetag Implementation ................................ ........... 176 Restructuring of Timetag Genera tion ................................ ............................. 176 Testing of Timetag Restructuring ................................ ................................ ... 177

PAGE 8

8 LIST OF REFERENCES ................................ ................................ ............................. 185 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 191

PAGE 9

9 LIST OF TABLES Table page 2 1 Sample system parameters for use in LIDAR link equation. ............................... 55 4 1 Ran ging trials for shots reporting 3+ r eturns ................................ ....................... 96 4 2 Effec t of range bias offset on measurement e rror (in cm) ................................ ... 96 4 3 Channel orientations for crosstalk c alculations ................................ ................... 98 5 1 Time Delay Reflectometry test d ata (courtesy of Fibertek Inc.) ........................ 133 5 2 Comparison of LIDAR depth estimates and m easurements ............................. 139 5 3 Comparison of LIDAR p latf orm distance estimates and m easurements ........... 142

PAGE 10

10 L IST OF FIGURES Figure page 2 1 Point clouds resulting from dat a collected by the DARPA Jigsaw ..................... 51 2 2 Point clouds res ulting from Jigsaw sensor data after filtering algorithms are applied. ................................ ................................ ................................ ............... 51 2 3 Diagram of energy level transitions in four level laser. ................................ ....... 52 2 4 Maximum pulse repetition rate versus pulsewidth in SPMCL PQS microchip laser [39]. ................................ ................................ ................................ ............ 52 2 5 Typical absorption coefficients of water as a function of wavelength [41]. .......... 53 2 6 Gaussian shaped function generated in MATLAB to simula outgoing laser pulse. ................................ ................................ .......................... 53 2 7 Simulation of ideal water surface and o pulse width. ................................ ................................ ................................ ......... 54 2 8 Gaussian outgoing laser pulse width. ................................ ................................ ................. 54 2 9 outgoing pulse width. ................................ ................................ .......................... 55 2 10 Data from NASA Goddard Space Cente nstrument [23]. ........ 56 2 11 A typical return waveform from the Optech SHOALS bathymetric mapping system, measured by a PMT detector [11]. ................................ ........................ 56 2 12 Output pulses from a PMT at different light levels. ................................ ............. 57 2 13 PMT signal under varying light intensities. ................................ ......................... 57 2 14 Sc hematic structure of a MCP PMT and the principle of multiplication [50]. ...... 57 2 15 Typical pulse height distribution plot (PHD) [50]. ................................ ................ 58 2 16 PHD plots depicting an incre ase in incident light level and supply voltage ........ 58 2 17 Typical PMT plateau characteristics. ................................ ................................ .. 59 2 18 Typical PMT output variation versus supply voltage. ................................ .......... 59 3 1 Block diagram of optical path layout. ................................ ................................ .. 70

PAGE 11

11 3 2 Detailed op tical path with turning mirrors, mechanical shutter, receiver electronics, and fiber optic detector. ................................ ................................ ... 70 3 3 Photo of JDS Uniphase NanoPulse lasers. ................................ ........................ 71 3 4 CATS beam polarization test ................................ ................................ ............. 71 3 5 Isometric and cutout views of scanner head. ................................ ...................... 72 3 6 Lab trials of scanner. ................................ ................................ .......................... 72 3 7 Photo of Hamamatsu Electronics Inc. model R4110U 74 M100D MCP PMT and quantum efficiency curve. ................................ ................................ ............ 72 3 8 Photos of the CATS and related components. ................................ ................... 73 3 9 Photo of several building structures in Lanham MD. ................................ .......... 74 3 10 Footprint projected on a bri ck wall at a distance of 303.7570 m. ........................ 74 4 1 CATS static (no scan) returns analysis from a brick wall approximately 300 m from the instrument ................................ ................................ ............................ 93 4 2 Relative accumulated hit counts per channel for a CATS range calibration test ................................ ................................ ................................ ...................... 94 4 3 CATS static footprint test setup.. ................................ ................................ ........ 94 4 4 Illustration of beamlet cross section projected onto wood surfaces. ................... 95 4 5 Calculation of reported separation for 19 cm distance between multiple scattering surfaces (trial D) ................................ ................................ ................ 95 4 6 Output profile of multi anode MCP PMT when a light spot of approximately 20 m diameter is scanned acros s the photocathode active areas [50]. ............ 96 4 7 Contour map of beam energy for a single pixel ................................ ................. 97 4 8 CAT S static footprint test setup. ................................ ................................ ........ 97 4 9 Geometric orientation of isolated pixels (shaded central four). ........................... 98 4 10 Photos of CATS barrel testing ................................ ................................ ........... 98 4 11 Histograms illus trating accumulated signal events for a heavily turbid water sand solution. ................................ ................................ ................................ ..... 99 4 12 Stacked histograms depicting accumulated atmospheric events per channel at three different PMT supply voltag es. ................................ .............................. 99

PAGE 12

12 4 13 Histograms depicting the consistency of accumulated events through the e different PMT supply voltages .......... 100 4 14 Point cloud (green) resulting from CATS horizontal scan after timetag structure modification ................................ ................................ ....................... 100 4 15 Top view of point cloud generated from rotating line s can data at 2100 V. ...... 101 4 16 Isometric view of point cloud generated from rotating line scan data at 2100 V. ................................ ................................ ................................ ...................... 101 4 17 Top view o f point cloud generated from rotating line scan data at 2200 V. ...... 102 4 18 Isometric view of point cloud generated from rotating line scan data at 2200 V. ................................ ................................ ................................ ...................... 102 4 19 Front view of point cloud generated from rotating line scan data at 2200 V, zoom in of tree structure. ................................ ................................ .................. 103 4 20 Photograph (front view) of a building with a l arge number of nearby trees. ...... 103 4 21 Overhead view of large point cloud generated from the building with heavy tree obstruction. ................................ ................................ ................................ 104 4 22 Overhead or top view of 0 3 m data segment associated with building location. ................................ ................................ ................................ ............ 104 4 23 Isometric view of 0 3 m data segment to emphasize heights. .......................... 105 4 24 Zoom in on isometric view of 0 3 m data segment. ................................ ......... 106 4 2 5 Top and side view of 0 5 m data segment ................................ ....................... 107 5 1 Photo s of the CATS mounted in aircraft ................................ .......................... 129 5 2 Flight testing in Gainesville, FL over University of Florida campus.. ................. 129 5 3 Small point cloud from forested region in Mine Run, VA ................................ 130 5 4 Large point cloud from forested region in Mine Run, VA ................................ 130 5 5 Ghost images in raw data. ................................ ................................ ................ 131 5 6 Vertical sliding window to evaluate magnitude of ghost responses. ................. 131 5 7 Resu lts from sliding window. ................................ ................................ ............ 132 5 8 Graphical model depicting reflections from detector cable interfaces. .............. 132

PAGE 13

13 5 9 L arge signal (110 mV) injected into gain block. ................................ ................ 134 5 10 Output of gain block from 110 mV signal ................................ ......................... 134 5 11 L arge signal (200 mV) injected into ga in block. ................................ ................ 135 5 12 Output of gain block from 200 mV signal. ................................ ......................... 135 5 13 L arge signal (300 mV) injected into gain block. ................................ ................ 136 5 14 Output of gain block from 300 mV signal ................................ ......................... 136 5 15 Result of first step of MATLAB local density filter for a single channel of raw airborne profile data. ................................ ................................ ......................... 137 5 16 Result of second step of MATLAB local density filte r for raw airborne profile data ................................ ................................ ................................ ................. 137 5 17 Fli ght testing in Mine Run, VA over the Lake of the Woods southwest ............ 138 5 18 Flight testing in Mine Run VA over the Lake of the Woods northeast ............. 140 5 19 Flight testing in St. Augustine, FL over docks on the intracoastal waterway .... 141 5 20 Flight testing in St. Augustine, FL over a sandbar on the intracoastal waterway. ................................ ................................ ................................ ......... 143 5 21 Flight testing in St. Augustine, FL over the eastern coast. ................................ 144 6 1 Flight testing of second generation LSNR LIDAR by Si gma Space Corporation over US Naval Academy campus in Annapolis, MD [81]. ............. 158 6 2 Flight testing of second generation LSNR LIDAR by Sigma Space Corporation in Chesapeake Bay off US Naval Acade my Campus [81]. ............ 158 6 3 Flight testing of second generation LSNR LIDAR by Sigma Space Corporation in Greenland ................................ ................................ ................ 159 A 1 Receiver block diagram for n channels. ................................ ............................ 166 A 2 Data capture board block diagram. ................................ ................................ ... 166 A 3 Range/control board block diagram. ................................ ................................ 167 A 4 Receiver components. ................................ ................................ ...................... 167 A 5 ........... 168 B 1 Differential measu rements of CATS scanner periods ................................ ....... 180

PAGE 14

14 B 2 Time data file associated with possible scanner timetag errors ........................ 180 B 3 Filtered differential measurements of CATS scanner periods .......................... 181 B 4 Point clouds created from preliminary CATS sensor data. ............................... 181 B 5 Illustration of scan period wandering. ................................ ............................... 182 B 6 Point cloud for horizontal profile scan of bank building ................................ ..... 183 B 7 Effect of shot0 variation in generated image ................................ .................... 183 B 8 FPGA block diagram showing redesign of CATS timetag structure to improve calculation of start of scan periods ................................ ................................ ... 184

PAGE 15

15 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy MIXED TOPOGRAPHIC AND SHALLOW BAT HYMETRIC MAPPING VIA MULTI STOP SINGLE PHOTOELECTRON LASER RANGING By Kristofer Y. Shrestha August 2010 Chair: Kirk Hatfield Major: Civil Engineering Airborne laser swath mapping (ALSM) systems collect thousands of range measurements per second, enabli ng efficient high resolution surface mapping. Low signal to noise ratio (LSNR) detection techniques allow for implementation of light detection and ranging (LIDAR) instrumentation on platforms with prohibitive power, size, and weight restrictions. The vi ability of single wavelength photon counting techniques applied to mixed target detection and shallow water bathymetry has not been previously investigated in detail The system design process for constructing a L SNR LIDAR system is described with emphasi s on component selection as related to theoretical performance. Cost effective microchip lasers are now commercially available that produce sub nanosecond pulse widths and stable peak energies (>1 J at 10 kHz) In order to detect backscattered laser rad iation at low power levels, detector elements must be capable of single photon sensitivity; a high gain (10 5 ), fast r esponse (sub 250 ps rise time) detector is therefore central to photon counting operation. Micro channel plate photomultiplier tubes (MCP PMTs) are capable of attaining these specifications with simultaneous two dimensional detection over an array.

PAGE 16

16 s prototype LSNR LIDAR sensor the Coastal Area Tactical mapping System (CATS) was constructed with these major comp onents in mind. The instrument is designed to operate in a fixed wing aircraft flying 600 m above by scanning a 10x10 array of laser beamlets. Ground based field te sting results yield 10 cm range accuracy and small scale feature identification at variable scan settings. Experimental data is presented establishing the ability to record multiple returns from the same laser pulse. Evaluation of system dead time shows t hat the minimum separation distance required to discriminate multi ple scattering objects is 15 cm, indicating the possibility of seamless topographic / bathymetric mapping with no gap at the waterline. Data collected during tests display vertical structur e in shallow water areas based on fixed threshold crossings at very low signal levels. A major concern for the binary detection strategy is dark / atmospheric / crosstalk noise producing false positive measurements. Potential causes of image ghosting are described, and data collection / processing solutions based on local density estimation are explored LIDAR depth estimates from airborne profiles are compared to on site measurements, and near shore submerged feature identification is presented. Analysi s of CATS data confirms simulation estimates of decimeter scale spatial resolution for hybrid topographic and shallow bathymetric mapping. Dark noise production and atmospheric scattering in daylight operation at maximum signal amplification does not sign ificantly impede fea t ure identification procedures. In order to combat the high variability in dynamic range, modifications to the detector structure may be necess ary to reduce anomalous effects.

PAGE 17

17 CHAPTER 1 INTRODUCTION Introduction to Remote Sensing Our s urrounding environment is a diverse collection of refractive and reflective media. The interaction of light with matter is of special interest for remote sensing, a branch of science concerned with the acquisition of information about an object or phenomen on without direct physical contact. Information about positions, sizes, and interrelationships between objects can be concisely conveyed through imagery. Because humans possess a high level of proficiency in deriving information from such images, we expe rience little difficulty in interpreting even those scenes that are visually complex. In fact, it is only when we attempt to replicate these capabilities using computer programs that we realize how powerful our abilities are to derive this kind of intr ica te information [1 ]. Photogrammetry and LIDAR Photogrammetry is one method of obtaining reliable information about the environment through processes of recording, measuring, and interpreting images and patterns of electromagnetic radia nt energy and other ph enomena [ 2 ]. Recent advances in technology have given rise to parallel methods of remote sensing, one example being light detection and ranging (LIDAR). This acronym is commonly used to describe the intersection of an inertial navigation system (INS), a light amplification through stimulated emission of radiation (laser) device, and a global positioning system (GPS ) [ 3 ]. Precise timing of transmitted and received light is the focus of LIDAR instruments. Given that light travels at a known velocity, time of flight observations allow for

PAGE 18

18 measurement of propagation distances. Detailed geo spatial information is obtained by the combination of these observations and enhanced navigation tools. Advantages of LIDAR LIDAR offers advantages over photogrammetry wh ich has caused airborne LIDAR to emerge as an attractive alternative to traditional methods of large scale geospatial data capture [ 3 ]. Photogrammetric methods for digital terrain model (DTM) generation can be labor intensive and time consuming. LIDAR al lows for fast surveying of remote areas because each data coordinate is individually geo referenced; aerial triangulation or orthorectification of data is not required. Because of a dependence on solar illumination, aerial photography systems are limited to daytime operation and are affected by weather. LIDAR systems can operate at night and through a variety of weather conditions. In addition, LIDAR offers a distinct improvement over photogrammetry in the collection of elevation data in heavily forested areas. Dense canopy cover generally prevents retrieval of accurate terrain informati on by photogrammetric sensors [ 3 ]. Airborne laser mapping sensors, on the other hand, can be used to acquire 3D discrete point measurements of canopy structure and the u nderlying surface topogra phy with sub meter resolution [ 4 ]. We will extend this concept t o airborne bathymetry in later chapter s Applications of LIDAR A typical airborne LIDAR sensor provides tens to hundreds of thousands of range measurements per second which can be used to construct a high resolution map of an area. Airborne laser swath mapping (ALSM) is a proven method for obtaining high quality t hree dimensional mapping data [ 5], [ 6 ]. The impact of this technology is easily seen from a militaristic viewpoint; improved target detection capabilities create

PAGE 19

19 advantages for surveillance and reconnaissance of enemy operations. The contributions are not necessarily limited to warfare, however airborne mapping can provide valuable information about rural o r inaccessible areas, record observations of delicate or highly sensitive terrain, and aid in hazardous environment mapping. Improving Image Fidelity LIDAR Point C louds The basic visualization of LIDAR data is a projection of three dimensional points into scales. For example, large scale topographic undulations can be mapped, while at the same time small scale road edges, roof outlines and tree locations can be pinpointed. Analysis of post processed data has yielded automatic methods for feature segmentat ion and target identification [ 7 ] [8], [ 9 ]. Point density within the cloud determines the spatial resolution at which slowly varying continuous features are monitored and sharply varying discrete features are identified. The ability to plot more points per square meter results in higher image fidelity. Existing Methods for Improving Feature D etection It is possible to merge LIDAR points with data from another type of senso r to improve feature detection capabilities. For example, it has been shown that the fusion of small footprint LIDAR and multi spectral data can help to estimate plot level volume and biomass in decidu ous forests [10 ]. Gaps in point cloud data over water can be filled with information from sonar but usually at coarse intervals. It has also been shown that the addition of high resolution hyperspectral imagery aids in classification of benthic LIDAR waveforms [ 11 ]. Data fusion and weighting of feature de tection algorithms continues to be a growing area of study in the field of remote sensing.

PAGE 20

20 Preferably, however, a single sensor would be able to provide all the necessary information and eliminate the need for data fusion considerations. For LIDAR alone, increasing the laser pulse repetition rate decreases the time between illuminations of ground footprints; thus subject to a static scan pattern (not dependent on laser rate) and a consistent platform velocity, spacing between ground points is reduced and h orizontal resolution is improved This approach eventually results in having more than one laser pulse in flight at the same time, which can cause difficulties when the range to the surface changes rapidly and unpredictably, as is often the case in mounta inous terrain or urban settings. The two way travel time of the laser pulses set a maximum pulse rate of 250 kHz at an average flying height of 600 m. If, however, a single laser beam is split into an array of elements and transmitted at once, the spatia l resolution is limited by the number of array elements and the capability of the receiver to process returns over the array simultaneously. A moderate laser pulse rate (< 10 kHz) can then theoretically provide contiguous coverage of the surface of the ter rain at decimeter scale resolution on a single pas s of an airborne LIDAR system [9 ]. Existing Methods for Improving Data Q uality Low level signal classification is often difficult in LIDAR systems given practical considerations for variations in surface re flectivity ; diffuse reflections of laser light are not always easily detectable C haracterization of system noise is important fo r discrimination of true events, meaning that n oise components should be kept to a minimum where possible. Superior contrast between signal and noise may generally be achieved by transmitting higher laser pulse energy. P roviding large outgoing pulse energy increase s the probability that weak light matter interactions will produce a detectable response. Lasers operating at high energy can be paired with a variety of

PAGE 21

21 backscattered radiation or low amplification. An assumption for this case is that the paired photodetector can operate at an acce ptable dynamic range; i.e., be able to detect the faintest and strongest measured signals with equal accuracy. Another assumption is that noise returns do not increase at the same rate as signal returns given a more energetic pulse; otherwise false events would obscure the desired response. The LSNR Mapping Approach Problems w ith High Energy / High Pulse Rate D esigns Meter scale spatial resolution in airborne scanning laser altimeter systems can be achieved by employing high energy / high pulse rate design s. Currently deployed ALSM implementations are designed to produce a high signal to noise ratio (SNR); typical specifications at standard altitudes of 500 to 1500 m are transmitted energy of [ 12 ]. These systems are capable of generating thousands of return photons incident on the receiver for each outgoing laser pulse. High SNR ensures that a system will be able to separate true return signals from sources of spurious noise events such as a tmospheric scattering and detector dar k noise, but may still contain erroneous points associated with multi path (sequential reflections from different surfaces prior to returning to the sensor). High energy, high pulse rate systems are not always practica lly desirable because of the size, complexity and cost of hardware. For example, attempting to integrate a high energy LIDAR design into a small unmanned aerial vehicle (UAV) platform is problematic. This type of platform operates on a fixed power budget so introducing a heavy payload impedes flight operations. Utilizing a high energy laser can shorten

PAGE 22

22 mission duration by draining battery life; high power lasers require complex electronics and sophisticated cooling systems. Bulky equipment is unsuitabl e for scalability with low and the loss of an expensive laser mapping unit can be prohibitive. Operation in populated areas also restricts maximum laser power and sensor altitude due to eye safety requirements. Designs based on a high energy requirement per illuminated area limit the size of the footprint, resulting in less than contiguous coverage of the terrain. Motivations for a LSNR LIDAR S ystem Practical requirements such as those mentioned above have motivated alternate methods to detect weak light matter interactions. Current commercial terrain mapping systems provide time of flight measurements for multiple reflections from each laser shot. The event timer used t o determine the time of flight of the recorded returns can use various discriminator designs to analyze the photodetector voltage response, including threshold, center of gravity, maximum, zero crossing of the second derivative, and/or constant fraction [ 1 3 ]. Classically a constant fraction discriminator has been used to obtain first / last responses related to separate reflecting objects in complex waveforms [ 14 ]. This approach may limit the minimum distance between two distinguishable objects in the sur veyed scene because lasers of relatively high energy (greater than 100 per pulse) generally have pulse widths of a few nanoseconds or longer. It has been suggested that a conventional single element detector approach operating under high SNR does not ma ke the most efficient use of available laser photons [ 15 ]. This idea has shown merit in low SNR (LSNR) LIDAR experiments on the Multi kiloHertz Micro Laser Altimeter (MMLA), developed at the NASA Goddard Space

PAGE 23

23 Flight Center under the Instrument Incubator Program [ 16 ]. It is estimated that the surface sampling rate can be increased by over three orders of magnitude for a given laser power/telescope aperture product by emitting the available photons in a high frequency train of low energy pulses, as opposed to a low frequency train of high energy pulses, and by employing photon counting detector arrays with multi channel timing receivers [ 16 ]. In theory, surface mapping efficiency can be significantly improved by using a LSNR LIDAR paradigm based on lower en ergy and shorter pulse lasers [ 15 ]. Small footprint LIDAR are intended to illuminate only a few surfaces within the footprint, yielding distinct return pulses corresponding to separate reflecting objects [ 17 ]. LSNR concept systems such as the Remote Ultr a Low Light Imaging system developed at the Los Alamos National Laboratory [ 18 ], the Jigsaw sensor created under the DARPA program at the Massachusetts Institute of Technology Lincoln Laboratory [ 19 ], and a second generation 3D imaging and polarimetric LID AR developed by Sigma Space Corporation [ 20 ] have established LSNR LIDAR as an effective approach to terrain mapping. Though these ventures have shown the promise of the LSNR technique, they still represent a limited number in comparison to the many HSNR p arallels. In the context of modern technology advancements, this has l eft significant gray areas in a comprehensive understanding of LSNR signal analysis. Simulation can be a powerful tool in this case, as a modeling capability can analyze performance as a function of system specifications and mission environment. Current mathematical models support

PAGE 24

24 the idea that a single wavelength LSNR mapping instrument can provide hybrid topographic and bathymetric mapping in shallow coastal areas [ 21 ]. Research A ims The goal of this work is to move beyond simulation and explore the viability of topographic and shallow bathymetric mapping via a single wavelength (532 nm), low altitude (500 m) low energy (3 J), short pulse (5 00 ps) laser ranging system. A prototype LSNR system has been developed by the University of Florida for this purpose, and detailed analysis of sensor data is shown here. In Chapter 2 of this work, the fundamentals of selecting an appropriate illumination source and photodetector are presented. In Chapter 3, the specifics related to the University of Florida's LSNR prototype design are listed. In Chapter 4, analyses of static experiments yield performance metrics associated with LSNR feature detection In Chapter 5 analyses of airborne experi ments explore the sensor's ability to provide shallow water bathymetry in a variety of water types. Chapter 6 presents conclusions about the viability of the technique and suggestions for improvement.

PAGE 25

25 CHAPTER 2 LSNR COMPONENTS Commercial Sector Activities and Scientific Impact COTS Market and C osts Prior to 1995 laser altimetry was generally conducted using custom designed sensors operated by research groups or built by commercial survey firms to exploit niche markets. The majority of these efforts were based on single prototype designs, limiting their ability to create and service a broad, sustainable, sector wide demand for the technology. Since 1995, however, a commercial off the shelf (COTS) instrument market has developed which has removed many of t he constraints for incorporating laser altimetry into operations [ 22 ]. These conditions have allowed for new types of lasers and photodetectors to become available without prohibitive cost to a variety of research organizations. LSNR Concept S ystems The S atellite Laser Ranging S tation (SLR2000) [ 15 ] was one of the earliest experimental platforms to demonstrate LSNR potential. The SLR2000 employed passively Q switched microlasers with output energies of 13 to achieve mm level ranging precision at satellite altitudes of 20,000 km. Using a s ub nanosecond pulse width, a laser beam divergence of 10 arcseconds, and a receive aperture of 76 cm result ed in a signal strength of less than one photoelectron per laser shot Target discrimination was made possible by utilizing a range receiver able to identify and extract single photon laser echoes from solar background noise through Poisson statistical analysis of the received photon distrib ution. The Post Detection

PAGE 26

26 Poisson Filtering (PDPF) technique has been extended to the MMLA sensor for signal extraction in an airborne scanning altimeter capable of daylight operations [ 23 ]. The Remote Ultra Low Light Imaging ( RULLI ) system also responds to individual photons using a low power pulsed laser and Pulse Absolute Timing (PAT) channels for time discrimination in the front end electronics [ 18 ]. Three dimensional imaging was accomplished with a vertical resolution of 5 cm. The detector is made u p of a micro channel plate (MCP) image intensifier front end with high gain for photon counting and a wire would cross delay line (CDL) anode. A stack of three MCPs allow for electronic gain up to 10 7 Each PAT channel is designed around a number of inte rnal circuits in a hybrid electronics module called a time interval meter (TIM). This includes a constant fraction discriminator (CFD) front end, a dual interpolator stage, and a ten bit time to digital converter. This allows for detection of signals wit h amplitude variation of 40 dB while maintaining a time walk of less than 100 ps. The Jigsaw system uses arrays of single photon sensitivity avalanche photodiodes (APD) for integration onto various surveillance platforms [ 19 ]. The sensor was also tested i n an indoor environment for use in surface reconnaissance experiments. An example of raw data from the Jig saw system is shown in Fig. 2 1 these images illustrates the challenges associated with LSNR LIDAR detection, as most of the points originate from avalanche photodiode dark counts. These challenges can be met through post processing techniques, as researchers have created a suite of novel algorithms to perform noise reduction, ground plane identification, detector response deco nvolution, and illumination pattern renormalization [ 24 ]. Results of thes e efforts are shown in Fig. 2 2

PAGE 27

27 Selection of a well paired illumination source and photodetector can improve SNR so that data processing presents less of an obstacle. The following sections describe how to select components with appropriate characteristics to maximize photon counting performance. A brief history of lasers is presented to introduce microchip laser design parameters. We will then transition into classic detector pai rings and the photon counting concept Overview of Lasers Quantum Dynamics F undamentals Lasing is made possible by the creation of population inversion in a group of atoms or molecules. This phenomenon is best introduced with quantum mechanics. In the si mplest form, a group of N atoms can exist in two basic states or energy levels: the ground state, and the excited state, with The number of atoms in the ground and excited states a re given by and respectively. The total number of atoms in the system is N, where The energy difference between the two states is given by This difference determines the characteristic frequency of light which will interact with the atoms, following the expression [ 25 ] (2 1) where h is Planck's constant (6.626110 J s). Thermodynamics shows us that for a system in thermal equilibrium, the ratio of a number of atoms in each state is give n by the Boltzmann distribution, (2 2)

PAGE 28

28 where T is the thermodynamic temperature of the group of atoms and k is Boltzmann's constant (1.3807 10 J K 1 ). The ratio of the atom populations at room temperature ( T ~ 300 K) for an energy difference delta E that corresponds to visible light ( f ~ 500 kT = 0.026 eV. Since this ratio kT it follows that the ratio is very small; i.e., there are very few atoms in the excited state for a system in thermal equilibrium. The Bolt zmann distribution dictates that in this case, even at infinite temperature, the population will not exceed Achieving population inversion thus requires pushing the system into a non equilibrated state. Met hods of achieving population inversion are based on precise manipulation of optical properties. Light can interact with a system of atoms in the following three ways: absorption, spontaneous emission, and stimulated emission. In the first, photons of fre quency may be absorbed by ground state atoms, causing them to transition to the higher energy level. The probability of occurrence is proportional to incident light intensity and In the second, atoms in th e excited state may also relax back to the ground state, and in the process release photons of frequency ; this is called spontaneous emission. The final interaction, stimulated emission, is the most important for the optical amplifi cation process which makes lasing possible. An atom in the excited state may be perturbed by the passage of a photon with frequency (exactly corresponding to the ssion of a second photon at the same frequency. Albert Einstein showed that the base probability of a

PAGE 29

29 photon causing stimulated emission in a single atom is equal to the probability of a photon being absorbed by an atom in the ground state [ 26 ]. Conseque ntly, when the number of atoms in the excited and ground states are equal, the rate of absorption is equal to the rate of stimulated emission for a given radiation density. Coherence The stimulated transition process is the sum of two separate processes. The lower energy level contains atoms that are stimulated by the applied signal to make upward transitions, while the upper energy level contains atoms that are stimulated by the applied signal to make downward transitions. The net flow of atoms is always from the higher population level to the lower population level. If the lower level is more populated, the signal is attenuated. If the higher level is more populated, the signal is amplified. The stimulated transition process results from a resonant re sponse of the atomic wave function, or of the atomic charge cloud in each individual atom, to the applied signal. The internal induced oscillation or dipole response that is produced in each atom is stimulated by the applied signal [ 27 ]. Atomic oscillatio ns follow the driving optical signal in full coherence, meaning that they follow the same amplitude and phase. An amplifying laser medium produces a linear rep roduction of the input signal and correspondingly any amplitude or phase mod ulation present in the signal except that a) the output signal is increased in magnitude; b) the signal modulation may be decreased in bandwidth because of the finite bandwidth of the atomic response; and c) the signal generally has a small amount of spontaneous emission noise added to it [ 27 ].

PAGE 30

30 Three and Four Level Laser M odels It has been shown that the ideal two level system is not enough to achieve population inversion. A more realistic approach considers three levels : and with Initially the majority of the atoms are in the ground state These atoms transition to by absorbing radiation at a frequency This is often referred to as the "pump transition." For a medium suitable for laser operation, it is required that these excited atoms quickly decay to Spontaneous emission may occur as a result, but in practice the to transition is radiationless; energy is transferred to vibrational motion of the host material surrounding the atoms without the generation of a photon. Correspondingly, an atom may decay from to and release a photon of frequency This is the intended laser transition, and it is achieved by ensuring that the lifetime of the transition is much longer than the lifetime of the transition ; often called a favorable lifetime ratio. Over time, the population in this case will become very small, and a population of excited atoms will accumulate at Optical a mplification then can be achieved with the population inversion The three level framework was the basis for many early lasers, including the first ruby laser as implemented by Theodore Maiman in 1960 [ 28 ]. In the context of modern lasers the three level system is inefficient because the medium must be very strongly pumped in order to achieve inversion. Most lasers now operate via a four level process.

PAGE 31

31 A four level laser energy level diagram is shown in Fig. 2 3 The fast, radiati onless transition from the pump band to the upper laser level results in the population being rapidly depleted. The same is true for atoms at the lower laser level ; they will make a quick transition to the ground state via The laser medium is selected such that the transition of atoms from to is the intended la ser transition. A four level system is much more efficient than a three level, however, because any significant population accumulating at produces a population inversion with respect to the negligible population Optical amplification is thus possible for the frequency Optical R esonators Now that laser operation has been introduced from the quantum perspective, the device characteristics behind resonance or amplification can be desc ribed. While at Columbia University in 1959, graduate student Gordon Gould introduced the idea of the open resonator [ 29 ] and its potential applications. An optical cavity or resonator is an arrangement of mirrors that forms a standing light wave. The c avity surrounds the gain medium and provides feedback for laser light. Multiple reflections of light from mirrors at either end of the resonance chamber produce standing waves for certain resonance frequencies. Certain patterns of standing waves are susta ined via constructive interference, called the modes of the laser, while others are suppressed due to destructive interference. There are two types of modes. Longitudinal modes refer to the wavelengths of the wave reinforced by constructive interference after many reflections from the cavity mirrors, and transverse modes refer to the electromagnetic field pattern

PAGE 32

32 measured in a plane perpendicular to the beam propagation direction. The most basic transverse mode is the Gaussian energy distribution. The si mplest optical cavity is created by two facing plane (flat) mirrors, commonly called the Fabry Prot arrangement. The difficulty in using flat mirrors comes in the alignment; even a few arc seconds of error can cause the intracavity beam to leak out the s ides of the cavity. This problem becomes much less significant when dealing with short mirror separation distances ( L < 1 cm) [ 30] however, allowing for their use in microchip and semiconductor lasers. Q S witching Pulsed lasers are particularly important for laser ranging applications, as the pulse duration and energy are of prime relevance to a system's timing and detection capabilities. "Q switching" is a technique for increasing the energy of short pulses by modulating the intracavity losses and the "Q factor" of the laser resonator. Q factor, or quality factor, is a term borrowed from electronic circuits (for example, describing inductive capacitive elements in the context of a continuously operating oscillator). The term was adapted for optical reso nators to measure the strength of the oscillation damping in the optical cavity [ 31 ]. Manipulation of the Q factor allows for generation of "giant" pulses [ 32 ], referring to a large amount of energy output in a short amount of time. Q switching is achieve d by putting a variable attenuator inside the laser's optical resonator. When the attenuator is functioning, light which leaves the gain medium does not return, and lasing cannot begin. The attenuation in the cavity corresponds to a decrease in the Q fac tor of the optical resonator. A high Q factor corresponds to low

PAGE 33

33 resonator losses per round trip, and vice versa. The variable attenuator is commonly called a "Q switch" when used for this purpose. Initially the laser medium is pumped while the Q switch is set to prevent feedback of light into the gain medium (producing an optical resonator with low Q). This produces a population inversion, but laser operation cannot yet occur since there is no feedback from the resonator. Since the rate of stimulated e mission is dependent on the amount of light entering the medium, the amount of energy stored in the gain medium increases as the medium is pumped. Due to losses from spontaneous emission and other processes, after a certain time the stored energy will rea ch some maximum level; the medium is said to be "gain saturated." At this point, the Q switch device is quickly changed from low to high Q, allowing feedback and the process of optical amplification by stimulated emission to begin. Because of the large a mount of energy already stored in the gain medium, the intensity of light in the laser resonator builds up very quickly; this also causes the energy stored in the medium to be depleted almost as quickly. The net result is a short pulse of light output fro m the laser, known as a giant pulse, which may have very high peak intensity. The Q factor can be reduced via both active and passive methods; the output of an energy pulse in this manner is called Q switching. For active Q switching, losses in the cavity are typically modulated via acousto or electro optics. A pulse is formed after an electrical trigger signal is received. With a sufficiently fast trigger, pulse widths on the order of the cavity round trip transit time can be achieved [ 33 ]. These are often referred to as "cavity dumped lasers." Mechanical switches such as a Lummer Gehrcke plate in conjunction with a rotating prism also been used [ 34 ], but are less

PAGE 34

34 common. The idea of using a saturable absorber to achieve passive Q switching (PQS) was presented in 1965 by Szabo and Stein [ 35 ], along with derivation of rate equations and numerical analysis. Microchip Laser A dvancements To obtain closed form solutions for key laser parameters such as output energy and pulsewidth in modern lasers, one can solve the three coupled differential equations which describe PQS. Optimization of the amplification system depends on the combination of characteristics related to the laser gain medium, absorber medium, and pump intensity. The Lagrange multiplier tech nique introduced by Degnan [ 36 ] can determine the optimum mirror reflectivites and unsaturated absorber transmissions as a function of two dimensionless variables: z the ratio of the logarithmic round trip small signal gain to the dissipative optical loss amplifying medium divided by the saturation energy of the absorber. These theoretical calculations were shown to be in agreement with experimental results, and independent optimization theories for rapid PQS l asers were subsequently formed [ 37 ]. The early part of the 21st century gave rise to significant advancements in continuous wave (CW) pumped PQS microchip lasers, with emphasis on large amplitude repetitive behavior of PQS in solid state devices [ 38 ]. Num erical and analytic solutions were presented to describe the gain medium as an ideal four level atom and the saturable absorber as a two level atomic system. The region of interest in this case is where the saturation intensity for the saturable absorber is larger than that for the gain, and when both decay rates are fast. Numerical bifurcation studies and asymptotic analysis show [ 38 ] that PQS is possible if these restrictions are met, and microchip laser design equations can be derived that are valid wh enever the gain is depleted and the

PAGE 35

35 saturable absorber is bleached. When applied to puls ing in a microchip laser with a neodymium doped yttrium aluminum garnet (Nd:YAG) gain medium and chromium doped yttrium aluminum garnet (Cr 4+ :YAG) saturable absorber, theoretical results were shown to be within 10% of the experimental values [ 38 ]. Studies such as these formed the theoretical basis for microchip laser development. For the PQS Nd:YAG / Cr 4+ :YAG laser specifically, it has been shown that the maximum repet ition rate is correlated with the pulse width; specifically, it decreases as the pulse width gets shorter [ 39 ]. If the absorption cross section of the saturable absorber is large in comparison to the gain cross section of the gain medium, relatively few s timulated photons are necessary to rapidly bleach the absorber. Absorption in Cr 4+ :YAG occurs at 808 nm the wavelength used t o pump the Nd:YAG gain medium which can cause partial bleaching of absorption. This results in a smaller absorption coefficie nt at the oscillating wavelength and a smaller effective Cr 4+ :YAG fill factor (the normalized fraction of the filled cavity) for the laser. The fraction of the laser cavity filled with Nd:YAG decreases as the Cr 4+ fill factor increases, meaning that the Nd :YAG must be more highly inverted to reach threshold. More heat is thus deposited per unit length of active material. The gain cross section decreases and a greater inversion is required as a result of the increase in thermal energy. Positive feedback o ccurs, leading to a well defined threshold for thermal runaway [ 39 ]. The maximum pulse repetition rate is correlated with the Cr 4+ fill factor, meaning that the maximum repetition rate and pulse width are also correlated. Experimental results for several SPMCL microchip lasers used in the Zayho wski study are shown in Fig. 2 4

PAGE 36

36 Beam Quality M easures A common measure of beam quality is the factor. This is defined as the product of the beam radius (measured at the beam waist) and the beam divergence half angle (measured in the far field) divided by accor ding to ISO Standard 11146 [ 40 ], (2 3) A laser beam is often said to be diffraction limited by times, as a di ffraction limited Gaussian beam has a factor of 1. For a given beam divergence angle, which is limited by the numerical aperture of the focusing lens, the factor limits the degree to which the beam can be fo cused. Together with the optical power, the beam quality factor determines the radiance of a laser beam. The degree of linear polarization is normally quantified by the polarization extinction ratio (PER). This is defined as the ratio of the optical pow ers in the two polarization directions. Effects of Laser Parameters Ideally, a LIDAR system would be built with a custom laser through which pulse outputs are directly controlled. Pulse repetition frequency (PRF), energy, and width could then be changed t o suit the particular application. While this kind of direct control provides inherent operating advantages related to quality monitoring, designs driven by free running lasers allow the use of cheaper COTS components. The tradeoff is that system operati ons may then be affected by small per pulse variations (if not directly monitored). The capability to swap in cheaply available lasers operating at different PRF, however, allows for such systems to approximate the versatility of a more

PAGE 37

37 expensive directly controlled laser. Additional diagnostic operations may prove critical for ensuring accuracy; without direct control of the laser, generating a time history of events related to laser fire is essential to limiting errors in point cloud generation. For ex similarly tagging certain scanner characteristics, mechanical components can be simultaneously monitored for stability. Operating Wavelength An imaging LIDAR is an active remote sensing instrument, meaning that it illuminates the investigated area with its own energy source (via a collimated beam of light) rather than relying on passive solar illumination. The selection of the specific wavelength and amount of output energy per laser pulse has direct relation to the sensing capabilities of the system. The propagation characteristics of electromagnetic radiation through a given medium are dependent on the incident wavelength. For example, visible green light (532 nm) exhibits lo w attenuation through water [ 41 ] making it the preferred wavelength for bath ymetric applications. Fig. 2 5 shows the spectral dependence of typical water absorption coefficients. Laser Pulse W idth The width of the outgoing laser pulse has particular impo rtance for a system operating at low output energies Bathymetry is one area where this can be demonstrated, because discrimination between sea surface and sea bottom returns is of prime concern. For this application, it is necessary that light matter in teractions produce sufficiently short return pulses to differentiate the two responses. It is possible to produce feature classification given that these requirements are met, as is illustrated in the following simulations.

PAGE 38

38 MATLAB was used to convolve the Gaussian function shown in Fig. 2 6 with two i.e., no contributions due to roughness or slope of a laser pulse from a sea surface and bottom. The height of the two discrete elements was set to 0.3 and 0.7 to produce more contrast in convolution operation result. The separation of the two elements was varied from 1 to 6 m to simulate different water depths. The resulting responses for convolution operations of the 10 ns FWHM (3 m in air) pu lse in Fig. 2 7 shows how differentiation between surface and bottom is only possible at the larger depths, i.e. 4 m and larger. An outgoing laser pulse width of 480 ps (Fig. 2 8 ) 1/20 of the previous value requires much less depth to produce s ignificant differentiation between surface and bottom signals. Simulating a 1 m depth, which produced no temporal separation of signal events in the previous simulation, now shows a clear division. 480 ps corresponds to a distance of 0.144 m in air. The two signals are resolvable at tw ice this value, or 0.3 m (Fig. 2 9 ). Pulse Energy If we consider outgoing laser pulse energy as an accumulation of highly coherent individual photons, the ideal photodetector pairing to sense backscattered radiation would b e sensitive to even a single returning photon. As will be described in the following section, detector improvements have brought this ideal closer to the realm of possibility. In order to estimate the output energy required for single photon detection, t he signal strength is estimated as a function of transmitter, receiver, and environment characteristics. The uncertainty of an observation can be measured as a function of the various errors sources [ 42 ]. Even for a radiation field of constant intensity, the number of photons arriving at the photodetector for a given time interval is inherently uncertain

PAGE 39

39 due to the discrete nature of the incident light, charge carriers, and the interaction of light with the photodetector [ 43 ]. The relationship between th e number of photons incident on the detector and the number of emitted photoelectrons (p.e.) can be expressed via a Poisson distribution. The probability of detected p.e. is therefore (2 4) where is the expected number of p.e. recorded by the receiver. The probability of registering at least one signal event can then be calculated as a function of [ 44 ]. For example, it is expected that for a signal level of one p.e., there is a 63.21% chance that there will be a signal event registered. The LIDAR link equation relates basic elements to estimate backscattered signal strength. A modified form of this equation yields the following expression for transmitted energy per laser pulse [ 15 ], (2 5) where is the hologram (in the case of an altered beam) efficiency, is the detector quantum efficiency, is the receiver optical efficiency, is t he photon frequency in hertz, is the wavelength dependent surface coefficient, is the local incidence angle on the surface, is the collecting area of the receiver aperture, is the at mospheric extinction coefficient per meter, and is the range to surface. Table 2 1 shows sample system parameters for moderate laser output power, modest telescope aperture, and average efficiencies for commercially available materi als. The sample values indicate that we expect 20% transmission loss from the

PAGE 40

40 hologram creating the beamlet pattern; 28% quantum efficiency in the detector; 60% loss due to spatial and spectral filters; 15% reflectivity for typical ground soil at 532 nm; a 6.5 cm diameter telescope; and 600 m flying height. To have a 63.21% chance of detecting one signal event, the transmitted energy should be larger than 13.5 n J Goddard Space Center test flights of a scanning laser altimeter abo ard a NASA P3B aircraft has shown potential for instruments operating at low puls e power and PRF [ 16 ]. Fig. 2 10 describes the test parameters and ranging results with respect to buildings, tree canopy heights, and shallow water bathymetry. Selecting a LS NR Photodetector Operating at low pulse energy and thereby LSNR requires a detector with high gain capabilities. There are several types of light sensors commonly used for LIDAR mapping. Avalanche P hotodiode One type of sensor classically used to convert weak light backscatter into analog electrical signals is the avalanche photodiode (APD). APDs are high speed (up to 250 ps rise time) photodiodes using an internal gain mechanism that functions by applying a reverse voltage. When light enters a photodiod e, electron hole pairs are generated if the incident energy is higher than the band gap energy. The drift speed of these electron hole pairs or carriers depends on the electric field strength. As these carriers collide with the crystal lattice, ionizatio n in which new electron hole pairs are generated takes place. This starts a chain reaction referred to as the avalanche multip lication of photocurrent [ 45 ].

PAGE 41

41 The internal gain effect of an APD is normally between 100 and 500 [ Perkin Elmer 2006 ]. It is pos sible to achieve very high gain 10 4 to 10 6 by operating an APD in will normally conduct a large current when biased above breakdown, but if this current is limited to l ess than the APD latch current, there is a strong statistical probability that the current will fluctuate to zero in the multiplication region. The APD will then remain in a nonconductive state until an avalanche pulse is triggered by either a bulk (dark c urrent) or photo generated carrier. By keeping the number of bulk pulses low, the APD is then used to count individual current pulses from incident photons. Minimization of the dark current is a significant factor for this type of APD operation, and is r educed by cooling. APDs are commercially available that span the spectral range from 300 to 1700 nm [ 46 ]. Silicon (Si) APDs can be used from 300 to 1100 nm, germanium (Ge) between 800 and 1600 nm, and indium gallium arsenide (InGaAs) from 900 to 1700 nm. InGasAs APDs ar e typically available with low noise current extended spectral response to 1700 nm, and high frequency bandwidth for a given active area. Ge APDs are useful for applications under high electro magnetic interference, where amplifier noise is significantly higher than APD noise [ 45 ]. APDs with high gain and fast rise times however, often exhibit slow relative decays (10 100 ns), however, causing range tails as shown in the Jigsaw sensor [ 24 ]. Photomultiplier T ube Advantages The photomultip lier tube (PMT) is another type of light sensor that provides very high internal gains similar to those of a Geiger mode APD. Generally PMTs are able to achieve higher internal gains up to 10 7 than APDs with superior recovery times. They also, ho weve r, require voltages from 1 to 3 kV, are more fragile, and more

PAGE 42

42 sensitive to stray magnetic fields. For system designs intending to amplify single photoelectron events, PMTs can be attractive as long as they remain robust under normal system operation. Com binations A LIDAR system can use a combination of an APD and a PMT in order to compensate for varying environmental response. In the bathymetric case, for example, deep water causes rapid decay in the power of the received signal. Depending on the water depth and clarity, there can be more than six orders of magnitude dynamic range between strong water interface returns and weak bottom returns [ 47 ]. This difference must be handled by the detector without anomalous effects and compressed into the useful input range of the detector, which is typically only two or three orders of magnitude. The use of a PMT or high gain APD alone for bathymetric mapping can cause several problems related to the strong interface reflections that occur frequently near nadir. Even with the tube gated off, high incident light levels can cause a build up of space charge around the photocathode [ 47 ]. This can affect output signal levels, linearity, and bandwidth. The result is a decrease in performance and may even be damaging time after the impulse as a result of overly strong input signals [ 48 ]. This idea will be explored in greater depth in a subsequent chapter Unexpectedly large surface return s may also create detector saturation, and corresponding distortion in the output waveform [ 49 ]. The U.S. Army C orps of Engineers' Scanning Hydrographic Operational Laser Survey (SHOALS) system uses an APD to measure the bottom response in depths

PAGE 43

43 shallower than 14 m, and a PMT for depths deeper than 8 m. A measure of the returned power is recorded via waveform digitization. The received power and resulting waveform is a function of the transmitter and receiver characteristics, target laser cross section, and atmospheric effects, and tar get reflectance [ 11 ]. Fig. 2 11 shows how this information can be analyzed to produce classification of the surface return, volume backscatter, and bottom return. Analyzing responses from two types of detectors can thus im prove characterization of benthic information. Ideally, a single highly sensitive detector would be able to handle difficult classification problems such as the aforementioned issues associated with LIDAR bathymetry without negative consequences If this were possible, t he need for multiple sensors or waveform digitization could be eliminated, allowing for a more simplistic syste m structure with low overhead. Photodetector Improvements We now present an overview of recent photodetector improvements in orde r to estimate the capabilities of a highly sensitive detector when paired with a low energy laser. Photon C ounting Large pulse energy lasers produce enough light backscatter from illuminated targets to generate multiple photoelectric events at the detector For highly sensitive detectors, this is referred to the multiple photoelect ron state. As shown in Fig. 2 12 [ 50 ], repeated photoelectron emission in this state creates output pulse intervals so narrow that they overlap each other. The combination of t hese overlapped pulses results in a signal of consistent amplitude with high f requency components, or Fig. 2 13 (a) [ 51 ]. The output signal is effectively the mean value of this waveform containing the AC

PAGE 44

44 component. When the incident light level drops to a fraction of i ts previous level as in Fig. 2 13 (b), the AC component becomes more significant. In a very low inc ident light state as in Fig. 2 13 (c), the individual discrete pulses can be analyzed in a digital mode. Recent advancements in PMT design have led to improved precision, sensitivity, and robustness in commercially available units. Commercial and developmental single photon sensitivity detectors now exist which are capable of providing centimeter level ranging resolution [ 15 ]. Typical photocath ode / dynode / anode configurations had previously been singular; new units are available in multi channel arrays. These configurations allow for the receiver to angularly resolve the source of a single photon event within the receiver field of view [ 15 ]. If the time resolution of the PMT is sufficiently fast that the average time intervals between signal pulses are wide in ity. It can be argued that because the generation of a photoelectron in this configuration is essentially a binary event given proper discriminators, the photon counting method is superior to analog signal measurement in terms of detection efficiency and stability. Semiconductor P rinciples Semiconductor PMTs operate on the principle of the external photoelectric effect, by which photoelectrons are emitted into a vacuum from a material called the photocathode [ 52]. The wavelength of maximum response and lo ng wavelength cutoff are determined by the combination of alkali metals used for the photocathode and its fabrication process. In a semiconductor band model, there exists a forbidden band gap or energy gap (EG) that cannot be occupied by electrons. The e lectron affinity (EA) is an interval between the conduction band and the vacuum level barrier (vacuum level).

PAGE 45

45 The work function ( ) is the energy difference from the Fermi level and the vacuum level. When photons strike a photocatho de, electrons in the valence band absorb energy and become excited, diffusing toward the photocathode surface. If the diffused electrons have enough energy to overcome the vacuum level barrier, they are emitted into the vacuum as photoelectrons. The quan tum efficiency is the ratio of output electrons to incident photons, given by (2 6) : reflection coefficient : full absorption coefficient of photons : probability that light absorption may excite electrons to a level greater than vacuum level : mean escape length of excited electrons : probability that electrons reaching the photocat hode sur face may be released into the vacuum : frequency of light In the above equation, if an appropriate material is chosen to determine parameters and th e factors that dominate the quantum efficiency will be (mean escape length of excited electrons) and (probability that electrons may be released into the vacuum). L becomes larger by use of a better crystal and greatly depends on electron affinity (EA). If a surface layer of electropositive material such as Cs 2 O is applied to a photocathode, a depletion layer is formed, causing the band structure to be bent downward. This bending can make the electron affinity negative. This state is called

PAGE 46

46 NEA (negative electron affinity). The NEA effect increases the probability ( ) that the electrons reaching the photocathode surface may be emitted into the vacuum. In partic ular, it enhances the quantum efficiency at long wavelengths with lower excitation energy. In addition, it lengthens the mean escape distance ( ) of excited electrons due to the depletion layer [ 50 ]. Micro channel Plate Photomultipli er Tube C oncept PMTs normally operate via a chain of dynode stages. After a photoelectric event, each dynode stage increases electron production by a certain degree; the product of these combined amplifications results in the overall gain. MCP PMTs are d ifferent in structure and operation from conventional discrete dynodes. An MCP consists of millions of very thin, conductive glass capillaries (4 to 25 microns in diameter) fused together and sliced into a thin plate. The two surfaces of the plate are co vered with thin metal electrodes, and a voltage is applied to create an electric field in the direction of the channel axes. Electrons emitted by the photocathode of the PMT enter the capillaries. After a wall electron collision, secondary electrons are accelerated by the electric field and travel along parabolic trajectories determined by their initial ejection velocity. As a result of repeated secondary electron emissions, the electron current increases exponentially as the electrons travel through the micro channel plate [ 51 ]. The schematic structure of the micro ch annel plate is shown in Fig. 2 14 along with the principle of multiplication. MCP PMT G ain The gain of a MCP PMT is determined by the length to diameter ratio of a c hannel, and the secondary emission characteristics gain factor The gain factor is a

PAGE 47

47 product of the inherent characteristics of the channel wall material. The overall gain is approximated as, (2 7) In gene ral, it is possible to obtain a higher gain by increasing which also moves the gain rising point to the higher voltage side. This is not without upper bound, however, because at some point the noise begins t o increase significantly due to ion feedback events. An upper threshold for =40 is usually selected so that a single MCP provides a gain of 10 4 at 1 kV supply voltage [ 51 ]. A two stage MCP PMT can then provide high enough gain to e nable photon counting measurements, i.e. from 10 5 to 10 6 Employing thousands of independent amplifying micro channels allows for a MCP PMT to avoid geometric saturation effects that can hinder more conventional high gain, segmented anode, metal dynode cha in photomultipliers. In the latter, any noise photons that hit a particular portion of the photocathode can uniformly deplete the gain in the corresponding anode. In an MCP PMT, however, a noise photon can drain a particular micro channel (or set of micr o channels) without affecting the others in a given anode. When a single photon enters a MCP PMT, the photocathode converts it into an electron which travels to the anode while being multiplied. The transit time of the resulting electron bunch differs dep ending on the input photon. The distribution of this transit time is referred to as the transit time spread (TTS). In most photomultiplier tubes, the electron transit time is dependent on the position of photocathode illumination. When the entire photoc athode is uniformly illuminated, the difference in transit time with respect to position is referred to as the cathode transit time difference

PAGE 48

48 (CTTD). The TTS is usually affected by the CTTD, but in the case of proximity focused MCP PMTs, it has little ef fect [ 51 ]. During the multiplication process from cathode to anode in a metal dynode chain PMT, the signal pulse can be broadened due to the emission angle distribution and initial velocity distribution of photoelectrons and secondary electrons, as well as the effect of the focusing lens. MCP PMTs can largely ignore this effect because a strong electric field is applied nearly parallel from the photocathode to MCPin and MCPout to anode. Since the plate takes the place of a conventional dynode structure, t he electron transit time in the secondary electron multiplication process is also very short, allowing for TTS up to 25 ps in Hamamatsu MCP PMTs [ 51 ]. MCP PMTs are available in small packages, with 100 anode configurations less than 6 x 6 x 3.5 cm. In mag netic fields parallel to the tube axis, the tube can be stably operated for fields up to 2.0 T. For fields perpendicular to the axis, the output drops drastically if fields exceed 0.07 T. Current micro channel plate implementations are able to offer comp act size, fast time response, and two dimensional detection with high spatial resolution, and stable operation in high magnetic fields [ 51 ]. Detection Efficiency and Pulse Height D istribution The effective capability of a PMT to collect photoelectrons emit ted from the photocathode is called the collection efficiency ( ). This value, combined with the photocathode quantum efficiency ( ), directly affects the measured signal. The ratio of the number of output puls es ( ) to the number of incident photons ( ) is called the detection efficiency [ 51 ],

PAGE 49

49 (2 8) The output pulse height can vary depending on fluctuations in the secondary electron emission ratio. In order to gauge the time varying response of a PMT, it is useful to consider a pulse height distribution (PHD) plot. This plot is a histogram of accumulated pulse heights given a series of test data, shown in Fig. 2 15 ( A ) [ 51 ]. A typical PHD s ignal to noise comparison is shown in Fig. 2 15 (B ). Note that the noise line exists even with no illumination of the photocathode; these events result from thermionic emissions (thermal or dark noise events). When the incident light level is increased, th e amplitude of the signal + noise increases as shown in Fig. 2 16 ( A ). Increasing the supply voltage causes the PHD curve to be stretched and flattened as shown in Fig. 2 16 ( B ). These images illustrate how detection efficiency is affected by threshold lev el. For example, the first valley in the PHD waveform is usually selected as the lower level discriminator (LLD), and a point midway down the falling edge of the waveform is selected for the upper level discriminator (ULD) [ 50 ]. Since a large portion of the noise often occurs at lower pulse heights, an appropriate setting can allow for filtering of thermal (dark noise) events. If, for instance, the LLD is selected far to the right of the waveform peak, the majority of signal events may not be passed once the output pulses are brought into a binary signal. Experimenting with LLD settings other than the initial valley is not without merit, however, depending on the application. If a large amount of noise is causing signal classification problems in data po st processing, one solution may be to increase the LLD to a point where only multi photon events have a high probability of being detected.

PAGE 50

50 Re collecting the data with in a multi photon state rather than a single photon state can have advantages for cleani ng up very noisy signal as long as the signal provides a strong response. Decreasing the PMT supply voltage can have the same effect; both values have significant impact on the overall detection efficiency. Variation in PM T Gain and Output I nstability The effect of changing the supply voltage can be seen by counting the number of pulses with the LLD fixed. Although the analog gain increases exponentially with supply voltage, the number of counted pulses remains fairly constant for a certain range. This r ange is referred to as the plateau characteristic, and analyzing the associated plot indicates an appropriate region for setting the supply voltage. An example of a plateau characteristic is shown in Fig. 2 17 which also illustrates why the photon counti ng method is relatively insensitive to small scale supply voltage fluctuations. The signal to noise ratio remains flat throughout the plateau range. Even when the gain changes significantly as a result of supply voltage instability (Fig. 2 18 ), the photo n counting mode remains stable with respect to count rate [ 50 ].

PAGE 51

51 A B Figure 2 1 Point clouds resulting from data collected by the DARPA Jigsaw program sensor at MIT Lincoln Laboratory. A) Raw Jigsaw data colored according to r ange relative to the senso r. B) V iew from inside the raw point cloud of genuine signal and background noise [ 24 ]. Figure 2 2 Point clouds resulting from Jigsaw sensor data after filtering algorithms are applied. Following preliminary cleaning of the 3D image, human subjects an d test range objects are much more apparent than in the raw input [ 24 ].

PAGE 52

52 Figure 2 3 Diagram of energy level transitions in four level laser. Figure 2 4 Maximum pulse repetition rate versus pulsewidth in SPMCL PQS microchip laser [ 39 ].

PAGE 53

53 A B C Figure 2 5 Typical absorption coefficients of water as a function of wavelength [ 41 ]. A) Segelstein 1981 ; B) Hale and Querry 1973 ; and C) Wieliczka, Weng and Querry 1989 Figure 2 6 Gaussian shaped function ge 10 ns at full width half maximum (FWHM), or 3 m (in air).

PAGE 54

54 A B C D Figure 2 7 reflectors, and result ing response. Depth is vari ed for four plots. A) D epth of 1 m. B) D epth of 3 m. C) D epth of 4 m. D) D epth of 6 m. Figure 2 8 Gaussian outgoing laser pulse width. This plot corresponds to 48 0 ps at FWHM, or 0.1441 m (in air).

PAGE 55

55 A B C D Figure 2 9 outgoing pulse width (480 ps). Three subplots are shown for the outgoing pulse, reflectors, and response. Dept h is varied for four plots. A) D epth of 1 m. B) D epth of 0.144 m. C) D epth of 0.2 m. D) D epth of 0.3 m. Table 2 1 Sample system parameters for use in LIDAR link equation. 0.8 0.28 0.4 HZ 0.15 5 M 2 M 1 600 M

PAGE 56

56 A B C D Figure 2 10 A) Engineering flight parameters. S ample profiling da t a is shown from B ) small buildings and trees in the town of Chincot ea gue, VA, from 6.7 km altitude, C ) tree lines and underlying terrain from 3.4 km altitude; and D ) shallow water bathymetry over the Chesapeake Bay from 3.4 km altitude [ 23 ]. Fig ure 2 11 A typical return waveform from the Optech SHOALS bathymetric mapping system, measured by a PMT detector [ 11 ].

PAGE 57

57 Fig ure 2 12 Output pulses from a PMT at different light levels. Figure 2 13 PMT signal under varying light intensities. Figure 2 14 Sch ematic structure of a MCP PMT and the principle of multiplication [ 50 ].

PAGE 58

58 A B Figure 2 15 Typical pulse height distribution plot (PHD) [ 50 ]. A) The histogram variation is shown mapped to pulse heights from a train of incoming pulses. B) Noise and s ignal are compared in a PHD plot. A B Figure 2 16 PHD plots depicting an increas e in incident light level (A) and supply voltage (B ).

PAGE 59

59 Figure 2 17 Typical PMT plateau characteristics. Figure 2 18 Typical PMT output variation versus supply voltage.

PAGE 60

60 CHAPTER 3 LSNR PROTOTYPE DESIG N Photon Counting System Prototype: CATS Building upon the major concepts of Chapter 2, we now present a design for the optical framework and receiver package to enable single photon laser ranging. General Scanning P rincipl es Laser mapping systems typically employ a scanning mechanism to direct the path of outgoing laser pulses. The repetition rates (5 to 10 kHz) of commercially available micro lasers combined with the flight speeds of commercial aircraft (60 m/s) make it n coverage of the terrain on a single pass. Expanding the size of the ground footprint can be achieved by using a multi element design. Instead of transmitting a singl e high energy beam, the outgoing laser pulse can be separated into many low energy beamlets via holographic expansion to sustain coverage requirements. The result is nearly overlapping footprints with small (decimeter) inter footprint pixel spacing. Syste m P roposal The availability of a cost effective microchip laser and high gain MCP PMT served mapping System (CATS ). The CATS was built through a partnership with Fibert ek, Inc. and Sigma Space Corporation. This LSNR prototype instrument operates at low power and low laser pulse rate while attempting to achieve contiguous ground coverage at a spatial resolution of a few decimeters. In order to meet these goals, a receiv er processes returns at the single photoelectron level resulting from an array of outgoing light pixels. While the main application is the mapping of coastal areas, including the

PAGE 61

61 shallow depth shore, single photoelectron laser altimetry has demonstrated t hat CATS could have advantages in developing high resolution digital topographic databases, monitoring biomass and vegetation heights, sea and lake levels, ice sheet thickness, etc. [ 23 ]. Optics Fig. 3 1 shows a block diagram of the CATS optical path eleme nts as they relate to the transmit and receive paths. Fig. 3 2 shows a more detailed image with the turning mirrors, mechanical shutter, receiver electronics, and fiber optic detector. Several tip tilt mirrors deflect the beam at right angles in order to reduce the overall extent of the optical path. The chosen mirror coatings have 99.5% reflectivity at a wavelength of 532 nm. Microchip L aser In the CATS, a NanoPulse class IIIB microchip laser manufactured by JDS Uniphase produces high repetition rates, high peak power, and short pulses. The laser cavity consists of a Nd:YAG gain medium bonded to a YAG saturable absorber. Pulsing is achieved through PQS. The cavity mirrors are vapor deposited directly on each side of the crystal to form a monolithic o scillator that is end pumped by a CW diode laser. The NanoPulse laser product line shown in Fig. 3 3 generates 1064 nm output and can be optimized for either 50 mW of average power or 8 J pulse energy. In either case, the output is a pulse train of high peak power, sub 1 ns, near diffraction limited pulses. The beam diameter is 0.2 mm with 2 mrad divergence. These lasers are capable operating at temperatures from 0 to 50 C at the base plate, and can be stored at temperatures from 20 to 60 C. Th e power stability over 6 hours was

PAGE 62

62 measured to be 3%, the ellipticity ratio is 1.1:1, and the warm up time is approximately 10 minutes. A variation of the NanoPulse laser, called the NanoGreen, utilizes a potassium titanyl phosphate (KTP) frequency doubl er at the same output as the crystal that is used for the NanoPulse to produce 532 nm pulses. An IR blocking filter is added to remove residual 1064 nm light. The PER is greater than 100:1 and the M 2 mode approximately 1.2 for a Gaussian beam profile. F ield tests for beam polarization are shown in Fig. 3 4. The NanoGreen can be specified for up to 15 mW of average energy. The entire laser assembly is contained within a compact 15 x 4 x 4 cm package. Each manufactured laser has slightly different output characteristics. For the operational CATS laser, a thermal head sensor has measured 27 mW at the direct laser output (7.76 kHz), which Beam E xpander Pulsed light passes through a beam expander, decreasing the beam spot size at large distances. There are two main types of beam expanders: Keplerian and Galilean. In the latter, the combination of a positive and negative focal length lens provide magnification of the beam diameter and then reduction of the divergence. Keplerian types achieve the same effect of producing a larger but highly collimated beam size, but with two positive focal length lenses. Contrasting the two, a Galilean design contains no internal focus point and the overall length of the Galilean as compared to a similar power Keplerian is smaller. In the CATS, a plano concave and achromatic lens p roduce 10x total magnification; this reduces the beam divergence to 0.2 mrad. BK 7 lens material with magnesium fluoride (MgF 2 ) coating is chosen for its compatibility with visible light applications, along with an anti reflection coating. After

PAGE 63

63 the beam expander, each pulse is passed through a holographic diffraction element that separates the beam into a 10x10 square array of beamlets rotated 45. The efficiency of the hologram is approximately 80%. A lens in the transmit path is shimmed to increase th e amount of defocus and resulting angular divergence of laser spots in the far field. The spot sizes then have 77 % linear fill factor at the beam contour, or 46 % area fill factor for 95% of the total energy. Outgoing laser beamlet s pass through a shared afocal element resulting in a total magnification of 4x. Risley P risms Laser light is deflected out of the shared afocal element through two Risley prisms, shown as cross hatched elements in Fig. 3 5. Each Risley prism is simply an optical wedge V coated for 99.5% light transmission. The refractive properties of this material combined with the wedge surface angle deflect the beam through a predicted path. BK7 material is selected with a refractive index of 1.52, wedge angle of 13. 58, diameter of 88 mm, and central thickness of 22.5 mm. Each incremental wedge rotation is slaved directly to the leading edge of a synchronization pulse associated with the fire of the laser. The wedges are set to rotate a constant angular amount betw een synchronization pulses, and can be independently controlled to move at varying speeds in either direction. Software routines created prior to system operation determine the frequency of the two rotations. This allows for the array of laser beamlets to be deflected in a variety of repeatable patterns. The line scan in Fig. 3 6( A ) closely approximates the familiar raster or sawtooth scan, with platform motion responsible for creating the along track displacement. Raster scans are estimated to produce a maximum 225 m swath width

PAGE 64

64 at an altitude of 600 m. An example of a different scan pattern available with the Risley prism design is shown in Fig. 3 6( B ). By slightly offsetting the rotation speed of one of the prisms, the line scan can be made to rotate over time. This type of scan results in much denser coverage of points near nadir. Different types of patterns can provide enhanced coverage for specific applications. A test of time deviation of the nearest laser synchronization pulse from the wedge r eference pulse indicates that the pulse position on the ground deviates by approximately 0.07 pixels from scan to scan; an acceptable margin. Range G ate The system is operated with a range gate of 100 to 150 m. The supply voltage on the tube provides full amplification only after a set amount of time has passed (related to the range gate setting); meaning that the power is temporally stepped in relation to the laser pulse train. This allows for the outgoing laser pulse to be sufficiently far away from the output window as to minimize electronic noise variables associated with laser fire. It also eliminates atmospheric backscatter immediately outside the aircraft and ensures that strongly reflective targets near the output window (atmospheric reflections, birds) do not produce a response. This helps to prolong the life of the photomultiplier tube. Spatial and Spectral F ilters Resulting backscattered radiation passes back through the shared afocal element. Dust particles residing on optical surfaces may pr oduce scattered laser light resulting in interference patterns. These patterns can produce contrasting background light that interferes with desired signal information. A simple way to deal with this interference is through a tight spatial filter. The i dea here is to reduce the receiver field of view to the minimum value which encompasses the transmit beam divergence and potential

PAGE 65

65 pointing variations and/or co alignment errors with the receiver [ 15 ]. Scattered light can be blocked by centering a small a perture at the focal spot of the direct beam, while allowing the direct beam to pass through. The resulting cone of light has a smooth irradiance distribution and can be refocused to form a collimated beam that is almost equally smooth. The laser return s are also passed through a spectral filter to further improve return signal quality before reaching the timing electronics. For systems operating at low signal to noise ratios, it is necessary that filtering operations be precise. In the CATS, an interf erence filter accepting wavelengths centered at 532.1 (+/ 0.05) nm with FWHM 0.25 (+/ 0.05) nm is used. The filter has a temperature coefficient of 0.002 nm / C. To further reduce scattered light reaching the active area of the PMT, a central mask is placed on the spectral filter to block out on axis backscatter. Dove P rism The pattern of filtered beamlets is rotated 45 by a Dove prism. This optical element is made from truncated right angle prisms such that a beam of light entering one of the sloped prism faces undergoes total internal reflection (TIR) from inside of the bottom face and emerges from the opposite sloped face in flipped image orientation [ 53 ]. By rotating the prism about its longitudinal axis, the transmitted image rotates at twice th e rate of the prisms; meaning that they can rotate a beam of light by an arbitrary angle ( 45 in this case ) The losses occurring as a result of the entrance and exit faces are reduced to a small fraction of a percent by using a dielectric V coating for b oth surfaces.

PAGE 66

66 Telephoto Lens and M ech anical S hutter The resulting pixels are refocused by a telephoto lens before being imaged onto separate cathodes of a 10x10 segmented anode photomultiplier tube. Detected signals are sent to a 100 channel multi stop ra nging receiver. A mechanical shutter before the telephoto lens ensures that the photosensitive material is not subject to stray light during system transport and setup. The scanner window is also tilted relative to the optical axis to minimize reflection of photon back into the receiver. The receiver field of view (FOV) is a 3.5 m square. This is a larger aperture than the diameter of the transmitted bundle of beamlets. MCP PMT Because a very low amount of reflected radiation is expected from each laser beamlet, a detector must be employed with high enough gain to amplify single photon events. In order to achieve 10 5 gains with 200 ps rise time, 25 ps transit time spread (TTS), and 200 ps dead time specifications, a micro channel plate photomultiplier t ube (MCP PMT) was selected. This device is capable of simultaneously detecting independent events from a 10x10 pixel array, allowing for two dimensional resolution. The MCP PMT provides exceptional gain while maintaining compact size. In the CATS, a 10x1 0 segmented anode array built by Hamamatsu Electronics Inc (model number R4110U 74 M100D), shown in Fig. 3 7( A ), is capable of sensing single photon events The aluminum housing of the MCP PMT measures 5.330 x 5.33 0 x 3.175 cm Three pins on the narrow e nd of the housing fit into a PCB that controls the high voltage applied to the tube. Detector signals are routed through a PCB attached to the back of the housing. Taking into account the aluminum mounts for these components, the combined volume of the d etector module is less than 1000 cm 3 A

PAGE 67

67 GaAsP (gallium arsenide phosphide) crystal activated with cesium is used as a transmission type photocathode, meaning that photoelectrons are emitted in the same direction as the incident light. This type of photoc athode has high quantum efficiency (QE) in the visible region [ 51 ], with a plateau curve from 300 to 850 nm. The QE plot for the R4110U is shown in Fig. 3 7( B ). The nominal photocathode uniformity, MCP gain uniformity, and crosstalk values are 76%, 70%, and 3.8% respectively. The dark noise 5 gain. Multi stop Event Timer Accurately detecting scattered light returns using single photon events can be complex. The negative impact of reduced signal levels relative to conventional altimeters is largely offset by the corresponding reduction in the detection threshold to about one photoelectron. Instrument sensitivity in this case is maintained or even enhanced [ 15 ]. The caveat to higher sensitivity i s the increa sed presence of noise effects. Reduction of the detection threshold to approximately one photoelectron allows for robust detection at lower light levels. However, e ven with a range gate, narrow spectral filter, and tight spatial filter, typic ally there are sufficient noise events from atmospheric backscattering (especially in areas with atmospheric haze or fog) to warrant the use o f a multiple stop event timer. N oise events can be separated from terrain or vegetation returns because they are e xpected to be randomly distributed in time, or "temporally incoherent," in comparison to "temporally coherent" signals which rise above the noise floor [ 15 ]. This permits single photon detection in daylight with wide range gates as well as multiple return s per pixel per laser fire from volumetric scatterers such as tree can opies or turbid water columns.

PAGE 68

68 More information on the processing electronics associated with multiple stop event timing is found in the Appendices Multi stop LIDAR sensors are capable of providing information about biomass in forest understories related to sunlight distribution [ 11 ], surveying the sea surface / floor in navigation channels, ports and harbors, coral reefs, beaches, and shorelines [ 11], [ 47 ], and classification between s patially mixed topographic features in urban and suburban areas [ 5 ]. Man made objects that would be obscured to pilots in mission fly overs such as tanks concealed in heavy brush or camouflage netting are able to be identified via observations of spatial structure in LIDAR point clouds [ 54 ]. Assembly and Footprint Dimensions Fig. 3 8 (A ) shows the CATS sensor mounted on a tripod with laptop and power supply connections supported by an equipment rack. A close up of the optics, enclosed scanner, and electron ics cube contained within the black colored metal assembly is shown in Fig. 3 8(B ) Fig. 3 8(C ) shows a zoom in on the communications ports for laser and scanner operations. Fig. 3 8(D ) shows communications ports for laser, timetag outputs, and range dat a throughput. Fig. 3 8(E) / (F ) show the front / back of the scanner interface module. Flying at nominal altitude (600 m), the system is expected to produce a horizontal spatial resolution of 20 cm at nadir. Fig. 3 9 shows a photo of several buildings in Lanham, MD. These structures were used to verify footprint dimensions. The diamond shaped footprint is projected onto a brick wall in Fig. 3 10. Manual measurements of the footprint projected onto the brick surface at night produced a side length of 0.9 9 to 1.01 m, an average of 1.00 m. The center to center spot spacing was 9.52 to 10.16 cm, an average of 9.84 cm. The spot size diameter was

PAGE 69

69 4.45 to 5.08 cm, an average of 4.77 cm. In addition to a CATS distance measurement, the range to the wall was me asured using an Optech ILRIS laser mapping sensor, and by Google Earth satellite photography. The average distance to the brick wall by these three independent measures was 303.76 m. Extrapolating these results to 600 m, the average side length would be 1.97 m, the center to center spot spacing would be 19.44 cm, and the spot size diameter would be 9.42 cm.

PAGE 70

70 Figure 3 1. Block diagram of optical path layout. Figure 3 2 Detailed optical path with turning mirrors, mechanical shutter, receiver electronic s, and fiber optic detector.

PAGE 71

71 Figure 3 3. Photo of JDS Uniphase NanoPulse lasers. A B C Figure 3 4. CATS beam polarization test. A) Effect of r otating linear polarizing film square is shown without attenuation B) perpendicular to polarization axis an d C) paral lel to polarization axis

PAGE 72

72 Figure 3 5. Isometric and cutout views of scanner head. The scanner window is tilted relative to the optical axis to minimize reflection of photon back into the receiver. A B Figure 3 6. Lab trials of scanner, showi ng A) line and B) rotating line scans A B Figure 3 7. A) Photo of Hamamatsu Electronics Inc model R4110U 74 M100D MCP PMT and B) quantum efficiency curve.

PAGE 73

73 A B C D E F Figure 3 8. Photos of the CATS and related components. A) View of s ensor mounted on tripod with laptop and powe r supply components. B) Output window, enclosed scanner, electronics cube, enclosed optical components. C) Communications ports for laser and scanner operations. D) Communications ports for laser, timetag ou tputs, range data throughput. E) Front and F) back of scanner interface module.

PAGE 74

74 Figure 3 9. Photo of several building structures in Lanham MD where tests were conducted to observe the footprint of the CATS sensor. A brick wall and bank overhang are in dicated. A B C Figure 3 10. Footprint projected on a brick wall at a distance of 303.7570 m. A) Side view. B) Front view. C) Close up of four pixels.

PAGE 75

75 CHAPTER 4 LSNR PERFORMANCE METRICS Experimental results from the CATS pro totype have yielded proof of concept LSNR operation and techniques for improving data processing related to system effects. Because the CATS is able to process multiple returns and has the ability to change sensitivity level between datasets, its penetrat ion and noise reduction capabilities were assessed in a variety of different test setups There are a number of situations that can create difficulties for classic first / last return differentiation techniques. For example, Jutzi and Stilla describe slop ed surfaces within the footprint of the beam leading to shifted mean distances, different elevated areas occurring at building walls leading to dilation and erosion, slightly different elevations leading to range superimposition, and small objects distribu ted randomly about a mean elevation value leading to a widened, low intensity pulse that is problematic for a fixed threshold analog system [ 55 ]. The ability of post processing methods to overcome these difficulties in the LSNR case is dependent on measur able system properties. A number of experimental configurations were designed to investigate performance metrics for the CATS related to minimum separation distance, water penetration depth, and atmospheric backscatter [ 56 ]. Laser Ranging Concepts and Imp lementation Range Calculation F undamentals In order to determine the range to a target, the two way propagation delay of a laser pulse is measured. The ability of a ranging receiver to record multiple unique events, i.e. multiple canopy returns, is depend ent on the length of the transmitted laser

PAGE 76

76 a region after a registered return in which no subsequent returns can be detected [ 57 ]. The blind zone extent is determin ed largely by the light propagation speed and the pulse length In terms of estimated terrain resolution, this creates a distance below the initially reflecting object in which no other objects can be detect ed. For a 10 ns pulse length, the blind zone distance is (4 1) The CATS sensor utilizes a short pulse length in comparison; approximately 480 picoseconds FWHM, which results in (4 2) Considering only pu lse width restrictions, the instrument should be able to discriminate a distinct second object if it is 14.4 cm away from the first. The overall resolution capability of a LIDAR system is not solely a function of transmitted pulse length, but rather the pr oduct of simultaneously occurring factors. Photodetector recovery time is another important element to consider. If the detector is not capable of a recovery faster than the amount of time related to the blind zone distance, the former will constrain the system specifications. The CATS photodetector, for example, has a recovery time of approximately 210 ps, which ensures that it is not the limiting factor. Range D iscretization In the CATS, the laser pulse propagation delay is measured in discrete time in tervals called range bins. A registered return is assigned a range bin number (RBN) based on the measured time since the transmission of the laser pulse. One of the

PAGE 77

77 factors in the final range uncertainty is the length of the range bins used. This bin le ngth is related to the effective range resolution. To complement the sub nanosecond pulse length of outgoing laser pulses, the internal system clock is set to 2 GHz. The range bin length is 1 / 2 GHz, or 0.5 ns. Accounting for the two way travel time of the laser pulse, this makes the range resolution for consecutive single photon events (4 3) In order to eliminate noise returns associated with backscatter as the laser pulse is just exiting the output window and to protect the P MT from being overexposed to returns from close proximity targets, a range gate is used to temporally filter responses. The range gate setting (RGS) is an offset to two way travel time beyond the range gate value, in nanoseconds. The total two way travel time and the related range are (4 4) Data Processing F ields CATS range information is bundled into a binary file format with the .SPR extension. The raw files contain the following information about rec orded returns: shot number channel number hit number range information time tag information The first four fields are self explanatory for the stream of data being recorded in a multi element, multi stop laser ranging system. The goal of the fifth field global

PAGE 78

78 positioning system (GPS) time tag information is to reconcile the position of the scanner at any given time in relation to the temporal sequence of laser shots in a data set and Risley prism wedge index marks. Incorporating this data field allo ws for analysis of potential data flow problems, e.g. if the laser fire slaving system happens to miss synchronization pulses, or if scanner rotation is malfunctioning. Once correctly implemented, time tag information greatly aids in post processing an d p otential troubleshooting. This idea is further developed in Appendix B. Ground based Static Testing Experimental S etup Practical testing of the instrument was conducted on the rooftop of a building in Lanham, MD approximately 15 m tall. The sensor was mo unted on a tripod with associated power and communications components (Fig. 3 8). The output window was pointed at buildings, vegetation, and other targets 200 to 800 m in the far field. Initial data collections focused mainly on static pointing of the fo otprint, to measure standard deviation of measured ranges. Several attempts were made to range to nearby buildings. To establish static straightforward orientation of the laser footprint, the two prisms were set 180 out of phase so that the laser would be pointed straight along the optical axis of the transmit / receive path. C++ range processing code and data runs made on the rooftop in hopes that testing modific ations could be made without incurring the delay of post processing steps. PMT control software also provided a simple graphical output showing accumulated channel returns in a 10x10 intensity grid every second.

PAGE 79

79 Range A ccuracy Ranging to a brick wall prov ided a calibration test for the consistency of returns per channel. Coding techniques for ranging are further explained in Appendix B, and the significance of a slight calculation error and oscillator variation is presented. The range gate value was inco rrectly entered into software routines for the initial processing, so 40 m was later added to results to determine absolute range values. Relative measurements are still applicable, however. Hits were filtered for returns from 264 to 266 m to discard any noise returns not associated with scattering off the brick. The associated statistics and images are shown in Fig. 4 1. At one standard deviation from the mean, 99% of the recorded range data falls within the 10 to 20 cm level. The outlier at channel 33 only registered 6 hits, a very small sample in relation to other channels. The distribution of channel hits, on the other hand, showed significant channel to channel deviation. The images in Fig. 4 2 show the orientation of the channels in the footprint and the relative hit counts related to footprint geometry. In Fig. 4 2(A) hit counts are associated with height are plotted to emphasize variation in response levels. In Fig. 4 2(B) colors associated with these heights are more easily identified for spatial trends. Of note is that the bottom left row and far right row of channels are where most of the low hit count problems occur; this was theorized to be related to a problem aligning the reflected beamlets to the PMT anodes. Re alignment procedures improved later ground based data collections. Minimum Separation D istance In the case of a partially occluded laser beam, a multiple stop receiver allows for recognition of multiple objects within the same laser shot. Given the 250 ps MCP PMT recovery ti me and 0.5 ns (1 / 2 GHz) timing bins, it is expected that the "dead time"

PAGE 80

80 time before a subsequent mu ltiple return can be registered will be dominated by the 480 ps laser pulse width. More specifically, the system dead time should theoretically be li mited primarily by the pulse width (14.4 cm in the atmosphere). Experiments were conducted to verify this value; the minimum separation distance (MSD) required to discriminate multiple scattering objects. Measurement of effective dead time establishes a boundary for sensor penetration A controlled experimental setup was developed for analysis of MSD in the situation where multiple hits would be necessary for target discrimination. The system was mounted on a tripod and pointed at the structures shown in Fig. 4 3. A 2 cm thick piece of sanded plywood was placed 1 m below a mirror approximately 560 m from the CATS to act as a calibration surface. At this distance, the 10x10 array of laser beamlets produced a rotated square footprint measuring approximate ly 2 m on one side. A central section of the large diamond shaped array (shown reflecting off the chain link fence in the background of Fig. 4 3) was intercepted by a cardboard mask. A square hole in the mask allowed transmission of only four beamlets to a 45 turning mirror located behind the cardboard. The mirror redirected the beamlets from the horizontal propagation direction to the vertical, and onto reflecting surfaces below. The cardboard mask assisted in channel discrimination. Returns from the mask were filtered out (temporally), allowing for post processing analysis to more easily identify desired signal responses. Half of a beamlet cross section was projected onto the wood calibration surface, and the other half projected at a varying distanc e above on a similarly reflective wood sheet. Fig. 4 4 illustrates the intercepted surfaces.

PAGE 81

81 Seven ranging trials (A G) were conducted at various separation distances (Act ual Separation column in Table 4 1 ). Data was recorded for 10,000 laser shots (1.25 sec) in each trial. Returns were filtered for a coarse 5 m window around the central response to include contributions from atmospheric returns near the primary reflecting surface. The return signal at the calibration surface (0 cm separation) was app ro ximately two range bins wide as expected for 0.5 ns timing bins, 480 ps pulse width, and a flat optical cross section. Data was thus retained for laser shots reporting three or more total intra shot returns, as hits beyond the second return were potential ly secondary surface returns. For each laser shot, the differential range bin value (RBV) between the first and last hit was calculated to identify the secondary surface distance, shown in Fig. 4 5 plots. The analysis process for trial D is descr ibed her e. In Fig. 4 5(A) large (>10 RBV) data values occur sparsely, meaning that they represent atmospheric returns. These peaks resulting from 3+ return shots account for less than 5% of the total data in trial D. Conversely, small (<5 RBV) data values occu r frequently and at a consistent level, meaning that they represent secondary surface returns. In Fig. 4 5 (B) the differentials were accumulated into a total count per range bin distance (AD) and displayed in a histogram. The third column (Reported Separ ation) in Table 4 1 was generated from the peak histogram value in each trial. The fourth column in Table 4 1 shows the error in secondary surface measurement for each trial. Results from trial A (zero separation) are included to show that a single targe t produces a signal response less than three returns wide. The 2.6% of shots reporting multiple hits with 3 or more returns in this trial are attributed to atmospheric scattering. Increasing the separation of the two surfaces beyond 10 cm, as in trial B, begins to

PAGE 82

82 produce significant (non atmospheric) multiple returns. Because the resolution limit of the timing electronics is 2 GHz, the smallest surface that can be detected is the nearest range bin edge, or 15 cm resolution. Though the timing oscillator is expected to be stable about 2 GHz, it is possible that the central frequency drifts over time with extended system usage. The significance of potential drift is further explored in Appendix B. In order to evaluate this empirically, the error values in Table 4 1 were calculated for known target separations. T o determine a systemic range bias, the effect of imposing a constant offset on range estimation was tested. Table 4 2 shows analysis of four values of range bias in each trial 5, 4, 3, and 2 cm. The values in each column are calculated by taking the absolute value of the Reported Separation (RS) and Actual Separation (AS) difference in each trial. The average error for each offset is then calculated. Analysis of distributed returns allows for ide ntification of ground reflections under dense foliage and penetration past the water surface. Given CATS receiver specifications, dead time is theoretically limited primarily by laser pulse width. Assuming that the PMT is not saturated by the return puls e and that the intercepted cross section is not vertically elongated, experiments show that a secondary surface can be detected at a distance of 15 cm from the first. Range errors in secondary surface identification for six separate trials were within 7.5 cm, or within the timing resolution limit of the system. Systemic bias testing showed that an offset of 4 cm minimized the average error in secondary surface measurement (down to 1.3 cm). Crosstalk E valuation In the CATS, backscattered radiation from a 1 0x10 square of beamlets is imaged onto an equally spaced grid of photocathode material. Photoelectric events amplified

PAGE 83

83 from each grid section are collected by separ ate anodes. Because all of the sub beamlets are transmitted at once the MCP PMT m ust be c apable of simultaneous sensing over each element in its grid. It is expected that each anode response will provide unique XY position information about intra footprint responses. The spatial resolution of the system with regards to a single photon event is dependent on its ability to accurately register independent responses from individual anodes. A concern in adopting this type of detector structure is the influence of crosstalk on recorded observations. Multi anode PMT designs are commercially availab le with the following geometric plate layout: squares of photosensitive material associated with each anode are separated by narrow borders of non conductive mat erial in a single plate. Fig. 4 6 shows the output profile of a multi anode MCP PMT when a lig ht spot of approximately Light incident on a singular portion of the photocathode material may produce a response at an adjacent anode crosstalk between anodes. The overlapping re gions of the waveforms in Fig. 4 6 are crosstalk areas. The amount of crosstalk between adjacent anodes of the photomultiplier tube contributes to the uncertainty of resolving a single return at the specified spatial resolution. This can result from broadening of electron flow when lig ht is converted into electrons, or incident light spread within the faceplate. Testing of two types of Hamamatsu R5900 series photomultiplier tubes showed fairly low (~5%) nearest neighbor crosstalk and even lower (1%) next nearest neighbor measurements [ 58 ]. These experiments indicated that crosstalk is mainly due to the spreading of optical photons in the entrance window; the M64 PMT used had a

PAGE 84

84 1.3 mm thick glass entrance window. Reduction in window thickness to 0.8 mm further reduced inter channel cro sstalk. In comparison, the R4110U model PMT used in the CATS system has a 4.77 mm thick window; the observed crosstalk for this device may thus be higher than 5%. The contributions from overlapping waveforms can be mitigated by restricting the area of foc used energy. In the CATS, a lens in the transmit path is shimmed to increase the amount of defocus and resulting angular divergence of laser spots in the far field. The spot sizes then have 77% linear fill factor, or 46% area fill factor. A typical CATS pixel energy distribution is shown in Fig. 4 7 The teal contour line (and superimposed dotted circle) corresponds to 50% of the peak amplitude (FWHM). The blue line is the to a smaller area on the active photocathode has the potential to reduce crosstalk, but can also reduce the effective lifetime of the tube. In order to practically evaluate crosstalk co ntributions in the CATS the setup pictured in Fig. 4 8 was created. Four pixels of the footprint were masked to create temporal spacing in receiver returns between pixels outside the central square. Pixels 19, 20, 23, and 24 were isolated for this exper iment. For each laser pulse, returns were filtered temporally for a 5 ns window centered on the wood signal response. This separates any possible stray returns from the mask. Of the 8 surrounding pixels, 5 channels are considered to get a measure of pix el to pixel crosstalk. For example, channels 18, 22, 30, 31, and 32 are considered in the channel 19 crosstalk analyses. The contributions from these channels are labeled as a 18 a 32 in Table 4 3 to express geometric relation within the CATS footprint (Fi g. 4 9 ).

PAGE 85

85 An experiment was first performed for multi photon sensitivity (2300 V supply), or 5 gain. The dark noise count at this gain is expected to be 119 counts per second (cps). Over 10,000 shots or 1.3 seconds of data, on average 25,000 retu rns were registered for each signal pixel (2 3 hits per shot). Each surrounding non signal pixel registered on average 150 returns. This is close to the dark noise specification at this gain. Hits from channels a 18 a 32 were accumulated and divided by the number of total non signal pixels (5). This value was 0.17% of the accumulated hits from channel 19 (100% level). Average crosstalk at the other three channels was calculated as the following: channel 20 0.62%, channel 23 0.22%, and channel 24 0.6 5%. For multi photon sensitivity, crosstalk between receiver channels is small relative to MCP PMT design specifications (3.80%). The same analysis was performed for single photon sensitivity (2500 V supply), or 5 gain. Over 10,000 shot s or 1.3 seconds of data, on average 45,000 returns were registered for each signal pixel (4 5 hits per shot). Each surrounding non signal pixel registered on average 5,000 returns. This is an order of magnitude larger than the expected dark noise level at this gain, so channel crosstalk is assumed to have significant effect here. The accumulated hits from a 18 a 32 in this case were on average 12.22% of the accumulated hits from channel 19 (100% level). Crosstalk at the other three channels was calcula ted as the following: channel 14 7.24%; channel 22 15.39%; and channel 23 9.96%. For single photon sensitivity, the measured crosstalk is in some cases nearly four times as large as PMT specifications (3.80%).

PAGE 86

86 Because of the large disparity between val ues for example, channel 22 exhibited more than two times the amount of crosstalk than channel 14 it was expected that sub pixel backscatter was not properly aligned to the PMT grid. Even a very slight (0.1 mm) offset in the tube alignment can cause o ne channel in the grid to receive significantly more light than a neighboring channel. Re alignment procedures were carried out to ensure that the backscatter was being uniformly projected on the grid before conducting airborne testing. Water Penetration D epth The potential for bathymetric mapping of near shore waters was evaluated by recording ranges through water columns of variable depth. To gauge the signal response in a wide breadth of bathymetric scenarios, the optical clarity of tested waters was a lso altered A 55 gallon, 90 cm tall barrel was placed below a turning mirror so that four pixels of the CATS footprint were distributed down into th e barrel (Fig. 4 10(A) ). With the barrel filled with 90 cm of distilled water, the barrel bottom (painted white for optimal reflectivity) was clearly visible to the human eye. A 180 backscatter turbidimeter (infrared) reported an average value of 1.5 nephelometric turbidity units (NTU) for this sample. The barrel was also filled with natural water transport ed from the Gulf Coast. This salt water was skimmed off the top of a near shore area. The turbidity of the sample was measured at 18.6 NTU. The white barrel bottom was no longer visible to the eye at a depth of 43 cm. Measurements were performed at nigh t to aid in positioning of the 10x10 footprint on the cardboard mask / large turning mirror shown in Fig. 4 8 During the day, the laser reflections were sufficiently weak at 600 m that they were not visible to the naked

PAGE 87

87 eye; meaning that precise position ing of a single channel down through a water column, for example, was not possible. Operating at night did have the benefit of reducing potential solar events in these trials, but previous experiments have shown that daytime operation has minimal effect o n SNR for wood surfaces at 550 m. Because the location of the barrel bottom was precisely known 60 cm water column depth for this experiment, a narrow range window corresponding to 3 bins around the expected response was chosen. This made it possible to accumulate a history of returns related exclusively to the barrel bottom response, rather than including sporadic returns from the water column and potential noise events (by using a wide range window). For each laser shot in the accumulated return hi story, data from a selected channel corresponding to a beamlet projected down into the barrel was considered only within the narrow range window. For example, if on average 70% of the transmitted laser shots registered one or more signal events from the b arrel bottom, this means that 300 out of a total 1000 accumulated laser shots did not register at least one hit within the selected window. For 60 cm of this clear water, on average 97% of the transmitted laser shots registered one or more signal events fr om the barrel bottom. Estimating the beam attenuation coefficient from this data and extrapolating the signal intensity to further depths, it is expected that approximately 90% of the incoming beamlets would register a signal event at 5 m. Likewise, at a depth of 50 cm in coastal water, on average 12% of shots registered at least one signal event. Extrapolation of experimental results to further depths leads to the suggestion that this is a sufficient signal level to produce meter scale shallow (2 to 5 m depth) bathymetry data [ 56 ].

PAGE 88

88 A 22.7 kg bag of sand was poured into distilled water. Data was collected while silica particulate was settling down into the water column. This was a heavily turbid suspension; visibility to the eye was approximately 7.5 cm Histograms from a data window of 10,000 l aser shots are shown in Fig. 4 11 for three different supply voltages. The blue and red signals are spaced approximately 15 range bins peak to peak; close to the height of the water column after accounting for t he change in the index of refraction. The blue signal is thus associated with the water surface return, the green from the middle of the column, and the red from the bottom of the barrel. A plot of all four channels merged together (left image) and a sin gle channel (right image) are shown to illustrate response consistency. Although detection of small low reflectivity targets submerged in highly turbid waters may be difficult, the density of data points from the sea bottom given 20 cm pixel spacing should be sufficient to establish near shore depths (up to 5 m) at coarser spatial resolution (1 m scale). For example, if the signal strength at 5 m degrades such that 4% of the pixels within a footprint produce ocean bottom returns, a 600 m flying height, 8 k Hz laser rep rate, and 60 m/s platform velocity is enough to reconstruct the digital terrain model at 1 m grid spacing. Partially submerged targets with intercepted cross sections smaller than 0.25 m 2 however, may not be detected. Additionally, for the heavily turbid case, the multi photon state (2300 V or 2.5 5 ) was not sufficient for feature identification. Extraction of the bottom signal in a heavily turbid suspension necessitated maximum MCP 5 ). Multi vs. Single Photon S ensitivity Varying levels of photomultiplier tube sensitivity were assessed to weigh noise production versus variations in supply voltage. For example, Fig. 4 12 displays stacked

PAGE 89

89 histograms of accumulated atmospheric events per channel for a 1.25 second data capture. Changing the supply voltage from a multi photon state (blue) to a single photon state (red) drastically affected the recorded number of atmospheric scattering events across most channels. Atmospheric backscatter as a result of laser light propagating through an 8 s range gate and sky irradiance was me asured. The sensor output window was pointed towards the sky on a clear day and returns recorded for the duration of the range gate. This procedure was repeated for three different PMT voltages, or gain settings of 5 5 5 Fig 4 13( A) displays results from all channels combined and Fig. 4 13(B) displays results from one channel. Blue colored bins correspond to events early on in the gate, and red to events near the end. There were significantly more events recorded early in the ra nge gate at the single photon level; the theorized cause was a switching problem related to the high voltage supply. The majority of the gate showed a fairly constant level of returns, however. The contributions from laser light / diffuse solar irradiance scattering by atmospheric particulate and direct solar irradiance were significantly increased for the central 32 channels as gain increased toward a single photoelectron threshold. This was an expected result as atmospheric events are generally single p hoton events, and reaching a single photon threshold required full gain ( 5 ) On average, the quantity of recorded returns increased by a factor of about 70 by increasing the gain from 5 5 The CATS offers the capability to change the PMT supply voltage between experimental collections. Operating at multi phot on sensitivity may have value for

PAGE 90

90 situations in which high ambient noise precludes single photon sensitivity, as man made objects such as buildings features may be easily distinguishable even at lower gain 5 ). This idea is explored in the following sections describing scan results. It is 5 ) will produce the best fine scale resolution and foliage/water penetration, but large amounts of noise can obscure some features. Decreasing PMT gain may then provide hi gher contrast between signal and noise for highly reflective features. Ground based Scan Testing Horizontal Scan Building F eatures Early attempts at creating a three dimensional image from range and timetag information are chronicled in Appendix B. Diag nostic signal analysis allowed for identification of mechanical problems such as a cracked encoder plate. After rectifying these issues, point cloud generation aided by diagnostic tools based on time histories has produced significantly cleaner point clou ds. With scan motion set to project a horizontal line pattern, the CATS was run for 30 seconds. Post processing of returns yielded the point cloud sh own in the top left of Fig. 4 14 The footprint shape reflected off the main scattering surface is promi nently shown across the solid brick building face. Metal light poles located approximately 10 m in front of the building are also discernable from the building front. Noise events occurring throughout the scan are easily separated from the bulk of signal returns. Rot ating Line Scan Building F eatures Data from a rotating line scan (shown in Fig. 3 6) was collected to produce a larger three dimensional view of the same building. A top view of the re sulting point cloud in Fig. 4 15 is shown for 9 seconds of scan data (approximately 72000 laser shots). An

PAGE 91

91 isometric view of the same scene is shown in Fig. 4 1 6 This data was collected with a 2100 V PMT supply voltage. Raising the voltage level on the PMT to 2200 V produced the expected result of a larger v ariety of distinguishable fine features at the cost of more noise returns. This is shown in Fig s 4 17 a nd 4 18 Tree struct ure is also visible in Fig. 4 19 During this data collection, there were semi occlusions as a result of leaves; points on the bu ilding face were discernable through the spaces between tree trunks / branches. Rotating Line Scan Building Features (Partial O bscuration) Test scans of a building with heavy vegetation obstructions o ccluding the building (Fig. 4 20 ) yielded the larg e po int cloud shown in Fig. 4 21 A top view of the cloud is plotted here using the QTModeler software package and colored according to elevation. To get a better idea of the sensor response from building faces and objects directly in front of the building, points were filtered for a 3 m elevation window in this region. The result s shown in Fig s 4 22 and 4 23 depict features more clearly. Here we see range associated with strong responses from highly reflective surfaces. While these can pose a s erious problem for automated imagery generation and display, such effects may be beneficially exploited. Isolated dark noise points do not exhibit the same type of geometrically organized response. The deconvolution algorithm employed in the MIT Jigsaw s ystem, for example, eliminates these responses while simultaneously enhancing SNR [ 54 ]. Fig. 4 24(A) shows a photo near the building and further zoom in on the isometric cloud view. Here we see small bushes, narrow tree trunks, and a white stone bench. I n Fig. 4 24(B), t he white stone surface is the most highly reflective and produces the most returns; the geometry of the small top arch is discernable. The dark tree bark is also

PAGE 92

92 highly reflective. There is a significant difference between the number of the returns from the trunks and the dense small bushes. The system demonstrates in this case the potential to separate the trunk and foliage features. Expanding the points filter to 5 m elevation shows penetration through th e heavy tree foliage. Fig. 4 2 5 shows top and side views of the point cloud colored by elevation. In this case red points correspond to leaf returns close to 5 m, and orange points correspond to leaf and roof returns close to 4 m. The system is unable to define much of the roof struc ture due to almost total obscuration by the dense vegetation in this case.

PAGE 93

93 UFL_Steady_RangeCal_24.SPR Statistics # of shots fired: 16000 total # of hits in file: 279197 # of channel error hits: 0 # of non error hits: 1279197 total # of filtered hit s: 922202 (72% of total) average hits per shot for all hits: 79.9498 average hits per shot for filtered data: 57.6376 averaged range to target, all channels: 265.1215 A B C D Figure 4 1. CATS static (no scan) returns analysis from a brick wal l approximately 300 m from the instrument. A) Statistics from calibration test B ) Histograms related to hit count per channel and average range per channel. C ) Average range per channel, zoom in. D ) Standard deviation of recorded ranges per channel.

PAGE 94

94 A B Figure 4 2. Relative accumulated hit counts per channel for a CATS range calibration test. A) H its are associated with height and plotted to emphasize variation in response levels. B) C olors associated with these heights are more easily identifie d for spatial trends. Figure 4 3. CATS static footprint test setup. A 45 turning mirror deflects four central pixels onto reflecting surfaces below. The pixels immediately surrounding these four are masked by cardboard to ai d in channel discrimination. half of one pixel -channel 15 -is distributed on the wood calibration surface, and the other half is distributed 11 cm above on a wood sheet.

PAGE 95

95 Figure 4 4. Illustration of beamlet cross section projected onto wood surfaces. A B C Figure 4 5. Calculation of reported separation for 19 cm distance between multiple scattering surfaces (trial D). A) Range bin value (RBV) difference between first and last returns is plotted on the y axis versus shot number on the x axis. B ) Histogram of accumulated differentials (AD). C) Histogram of accumulated differentials, zoom in.

PAGE 96

96 Table 4 1 Ranging trials for shots reporting 3+ r eturns Trial Actual Separation (cm) Reported Separation (cm) Error (cm) # of Multiple Hit Shots % of Total Shots A 0 n/a n/a 259 2.6 B 11.0 15.0 4.0 2094 29.4 C 14.0 15.0 1.0 2435 24.4 D 19.0 22.5 3.5 1815 18.2 E 24.0 30.0 6.0 831 8.3 F 31.0 37.5 6.5 2112 21.1 G 41.0 45.0 4.0 1471 14.7 Table 4 2 Effect of range bias offset on measurement e rror (in cm ) Trial |RS AS| |(RS 5) AS| |(RS 4) AS| |(RS 3) AS| |(RS 2) AS| A n/a n/a n/a n/a n/a B 4.0 1.0 0 1.0 2.0 C 1.0 4.0 3.0 2.0 1.0 D 3.5 1.5 0.5 0.5 1.5 E 6.0 1.0 2.0 3.0 4.0 F 6.5 1.5 2.5 3.5 4.5 G 5.0 1.0 0 1.0 2.0 average error 4.3 1.7 1.3 1.8 2.5 Figure 4 6 Output profile of multi anode MCP PMT when a light spot of approximately 20 m diameter is scanned across the photocathode active areas (1.4 x 1.4 mm squares) [ 50 ].

PAGE 97

97 Figure 4 7 Contour map of beam energy for a single pixel. The inner co ntour is indicated by a dashed circle. Edges of neighboring pixel energy patterns are also visible in the image corners. Figure 4 8 CATS static footprint test setup. A 45 turning mirror approximately 560 m from the sensor deflects four central pixels onto a wood sheet used as a calibration surface. The pixels immediately surrounding these four are masked by cardboard to evaluate crosstalk of returns from the four pixels.

PAGE 98

98 T able 4 3 Channel orientations for crosstalk c alculations a 30 a 18 a 22 a 31 10 0% X a 32 X X Channel 19 a 18 a 22 a 14 X 100% a 15 X X a 16 Channel 23 a 8 X X a 7 100% X a 6 a 32 a 31 Channel 20 X X a 15 X 100% a 16 a 7 a 8 a 54 Channel 24 Figure 4 9. Geometric orientation of isolated pixels (shaded central four) in CATS footprint. A B C Figure 4 10. Photos of CATS barrel testing. A) A 55 gallon barrel filled with water of variable constituency is placed below a 45 turning mirror. B) Beam propagation through 560 m. C) Four pixels are reflected down into the barrel by the mirror.

PAGE 99

99 A B Figure 4 11 Histograms illustrating accumulated signal events for a heavily turbid water sand solution. Blue colored bins correspond to expected water surface returns, green to middle of the column, and red to the bottom of the barrel. A) Returns from four b arrel channels combined. B) Returns from channel 22 only. Figure 4 12 Stacked histograms depicting accumulated atmospheric events per channel at three different PMT supply voltages.

PAGE 100

100 A B Figure 4 13 Histograms depicting the consistency of accumulated events through the colored bins correspond to events early on in the gate, and red to events near the end. A) Returns from all channels combined. B) Returns from channel 14 only. Figure 4 14 Point cloud (green) resulting from CATS horizontal scan after timetag structure modification. The sensor was situated on rooftop approximately 300 m from building.

PAGE 101

101 Figure 4 15 Top view of point cloud generat ed from r otating line scan data at 2100 V. Figure 4 16 Isometric view of point cloud generated from rotating line scan data at 2100 V.

PAGE 102

102 Figure 4 17 Top view of point cloud generated from r otating line scan data at 2200 V. Figure 4 18 Isometric view of point cloud generated from rotating line scan data at 2200 V.

PAGE 103

103 Figure 4 19 Front view of point cloud generated from rotating line scan data at 2200 V, zoom in of tree structure. Figure 4 20 Photograph (front view) of a building with a large number of nearby trees.

PAGE 104

104 Figure 4 21 Overhead view of large point cloud generated from the building with heavy tree obstruction. Data was colored according to height (m) using QTModeler software. Dark blue points correspond to the low slope of the grassy fie ld leading up to the building. Dark red points correspond to the tops of the trees directly in front of the building. Figure 4 22 Overhead or top view of 0 3 m data se gment associated with building location.

PAGE 105

105 Figure 4 23 Isometric view of 0 3 m data segment to emphasize heights.

PAGE 106

106 A B Figure 4 24. Zoom in on isometric view of 0 3 m data segment. A) Photo near building. B) Isometric view with bushes, tree trunks and white bench chair with rounded top in foreground and building walls in background

PAGE 107

107 A B Figure 4 25 Top and side view of 0 5 m data segment. A) Top view expanded to 5 m heig ht (red points). B) Side view expande d to 5 m height

PAGE 108

108 CHAPTER 5 LSNR AIRBORNE LASER BATHYMETRY ALB Background The ability of light detection and ranging (LIDA R) hydrography to create detailed maps via flexible data collection methods supports a variety of applications, including nautical charting, hazard identification, erosion modeling, storm damage assessment, a nd beach nourishment / design [ 59 ]. By the late 1980s, the idea of airborne laser bathymetry (ALB) showed promise from systems such as the U. S. Airborne Oceanographic LIDAR, the Canadian Larsen 500, the Australian W RELADS, and the Swedish FLASH [ 60 ]. Operational systems matured during the early part of the next decade, with successes by systems such as the Royal Australian Navy's Laser Airborne Depth Sounder (LADS), the U.S. Army Corps of Engineer's Scanning Hydrographic Operational Airborne Laser Survey (SHOALS), and the Swedish Mari time Administrati on's HawkEye. As volume of data and rate of acquisition became less restrictive concerns for progressing computer technology, the processing workflow moved from a manually intensive line based approach to an automatic area based methodology [ 59 ]. In the e arly 2000s and through much of this decade, improvements have been made to increase the pulse rate, spot spacing, and portability of ALB systems [ 61 ], but the nature of the sensing representation of the air/water interface, water column, and botto m via g reen channel waveforms remains largely the same. Multiple Information Channels in Laser Ranging The combination of green channel information with other types of observations has proven useful for feature separation, especially in the bathymetric context. Data from

PAGE 109

109 the SHOALS, for example, have been used to make high density measurements of the coast al zone for a number of years [ 62 ]. Analysis of 1064 nm (infrared) channel data in conjunction with parallel green waveforms is beneficial for distinguishing the specular interface reflection or water surface, from the diffuse bottom reflection [ 62], [ 63 ] [ 64], [ 65 ]. As a general rule of thumb, t he SHOALS is sensitive to bottom returns out to 2 3 times the Secchi depth and out to a maximum of 60 m in the cl earest coastal waters Shallow water detection (<5 m), however, has been shown to be difficult. Older ALB data sets such as those collected prior to 2004 tend to have gaps in coverage related to complexities in proce ssing very shallow water data [ 61 ]. Tr aditional depth extraction algorithms have problems dealing with merged waveforms even with manual interventions. Recent enhancements to the traditional SHOALS depth extraction algorithms from green APD waveform data have shown interesting developments fo r dealing with mixed mode sign als in waters less than 1.5 m [ 61 ]. Changes in laser induced water Raman backscatter can be used to detect surface oil slicks [ 66 ], [ 67 ] and to measure chlorophyll concentrations to estimate phytoplankton dynamics [ 68 ]. Resea rch also indicates that novel analysis of red channel waveform data associated with Raman scattering can provide an approach for measuring ver y shallow (<2 m) water depths [ 64 ]. Additional channels of information are a luxury that modern systems can exploi t to produce high resolution area maps. The work pursued here, however, is aimed at evaluating whether information from only the green channel can be used to provide similar types of feature detection when a different type of dete ction methodology is util ized.

PAGE 110

110 Challenges Facing Single Channel Detection Ideally, a light detection and ranging (LIDAR) system will record a digitization of the return waveform resulting from each laser shot. This allows for a post processing algorithm to identify time intervals associated with inflection points in the reconstructed signal. Small footprint LIDARs then have the potential to provide detailed characterization of vertical structure at high spatial resolution [ 14 ]. The idea of building an "all green" airborne bathym etry system that meets depth accuracy standards faces several impediments. For full waveform mapping, the interface component signal amplitudes depend strongly on the beam nadir angle, wind speed, and specific irradiated wave slopes, causing signals to va ry with huge standard deviation [ 47 ]. Subject to these varying conditions, there may only be a small fraction of the outgoing laser energy that is reflected back into the atmosphere to be collected by the ranging receiver. The resulting surface uncertain ty problem is a serious issue for system designers attempting to decide which portions of the return waveform to interpret as sea surface and bottom in a real time, automatic manner. At 532 nm in a typical coastal environment, less than 2% [ 69 ] of the lase r light is reflected back from the air surface interface. A large portion of the green laser pulse is refracted into the water column. The beam undergoes scattering from entrained microscopic particulates that causes it to spread into a cone of continuou sly increasing angle. A fraction of the transmitted energy in the water column, which decreases exponentially in magnitude with greater depth, is incident on the bottom with a lateral extent that depends on the scattering properties of the medium and opti cal depth A portion of this incident energy, typically 4 to 15% de pending on bottom composition [ 64 ], is reflected from the sea floor back into the water column. Absorption and scattering

PAGE 111

111 again attenuate and stretch the pulse as it passes back to the su rface, where most is refracted back into the atmosphere. A photodetector such as the avalanche photodiode or photomultiplier tube needs to employ substantial amplification (>10 4 ) in order to magnify low amounts of this backscattered energy (<1 fJ) into a m easur able signal. Detection beyond 10 m depths has typically required significant outgoing pulse energy at 532 nm (> 50 J). The generation of large amplitude pulses (>2 mJ) can simultaneously increase the outgoing pulse width (>10 ns), which has reduced the capability of typical ALB systems to accurately survey shallow (<5 m) coastal areas [ 61], [ 64 ]. Increasing output energy can also cause safety issues at low altitudes and require s more sophisticated lasers. Pulse Width and Output Energy Considerations In past ALB systems, green beams were purposely expanded to a diameter of at least several meters in order to achieve eye safe operation with sufficient pulse energies to provide reasonable signal to noise ratios at greater depths [ 70 ]. For example, the O ptech SHOALS 3000 system uses a 3 kHz, 6 ns, 3 mJ laser [Tuell, personal communication, 2009] to provide bathymetric measurements up to 60 m depth at 2x2 m spot spacing [ 62 ]. The NASA Experimental Advanced Airborne Research LIDAR (EAARL) takes a different approach by using 532 nm, 10 kHz, 1.2 ns, 70 J pulses ( ~ 1/43 the amount of energy) with a much narrower FOV; ~2 mrad [ 70 ]. This is an interesting system because the combination of short pulse and small FOV is expected to overcome the ambiguity tradition ally encountered with surface location detection in green only ALB; an idea that must hold true if the CATS is to achieve any success. Four 1 ns waveform digitizers connected to photodetectors operating at varying sensitivity allow for cross environment s urveys that include abrupt surface

PAGE 112

112 brightness boundaries, typically a cause of saturation in sensors optimized for viewing dark water targets. This system has been shown to be capable of accurately surveying 12 m depths [ 71 ] Optical rugosity analysis by the EAARL in Biscayne National Park showed measurements consistent with field observations of coral features relative to rubble dominated ba nk reefs within the study area. Given recent advances in low signal level detector technology a further question i s raised: how much can we lower the outgoing pulse energy and still achieve shallow (0 5 m) bathymetric detection when paired with current commercially available photomultiplier tubes? We have previously alluded that LSNR detection methodologies work well for low noise environments such as satellite laser ranging, but shallow water bathymetry presents a host of new variables with which to contend. The CATS sensor is uniquely suited to answer this question. In relation to previous successful ALB systems, the EAARL sensor is considered to be a low power system. In comparison to the EAARL, the CATS sensor outputs 35 nJ per channel; 1/2000 of the energy. This represents quite a drastic shift from traditional ALB design philosophies. Contrast to Waveform Dig itization In waveform digitization post processing or embedded filtering (such as constant fraction discrimination), there is some level of decision making that occurs on an automatic scale. For example, effects such as amplitude dependent range walk, slo pe dependency of range and signal ringing can have negative effects without proper compensation [ 72 ]. For LSNR shallow water bathymetry, complete waveform digitization may not be mandatory. It may be possible to move to a simple fixed threshold detection strategy and retain similar feature detection results. In Chapter 4, static experiments showed that the estimated density of data points from the sea bottom

PAGE 113

113 given 20 cm pixel spacing should be sufficient to establish near shore depths (up to 5 m) at coar se sp atial resolution (1 m scale) [ 56 ]. Experimental Setup The CATS sensor was designed to be mounted alongside a commercial ALTM unit such as the Optech Gemini, which is routinely operated by the University of Florida. To minimize costs, it contains no i nertial measurement unit (IMU). However, for the tests reported here, the Gemini sensor was not used and the CATS sensor was mounted in a Cessna Skymaster twin engine aircraft, shown in Fig. 5 1 A special mount was constructed out of aluminum plates and bolted directly to the rails that normally support the co aircraft frame, but no IMU data were collected with which to correct the observations for changes in the orientation of the aircraft. Artifacts resulting from changes in the orientation of the aircraft (roll, pitch, and yaw) include undulations of flat water surfaces and lateral wandering of the surface profiles. The mounted CATS sensor head fits down into an opening in the floor of th e aircraft with approximately 5 cm free space surrounding the main cylinder containing the Risley prisms. A DC power supply provided 24 V at approximately 3 A. The DC supply A laptop PC was also connected to this inverter. The scanner control box and timetag module were secured behind the pilot and at the feet of the operator. Experiments were conducted to test the system's airborne profiling capability. The two prisms we re set 180 out of phase so that the laser would be pointed straight along the optical axis of the transmit / receive path, or at nadir. The aircraft was flown at approximately 60 m/s and 500 600 m height above ground level for a series of ranging

PAGE 114

114 trials. Because a large number of returns per channel were expected, only channels on the second ranging board (17 32 or the central 4x4 square) were enabled for a majority of the experiments. Initial collections with all 96 elements enabled in profiling mode a t the highest gain produced such a wealth of returns that the data throughput exceeded system capacity. Small Structure Identification Several different types of buildings and other manmade objects were profiled at nadir in a flight over the University of Florida campus in Gainesville, FL. Fig. 5 2(A ) shows an aerial photo of a region at the south end of campus and flight line. Fig. 5 2(B ) is the raw point data from channels 20 and 23. The image allows us to identify a small outcropping in a building roo f, a roof from a nearby parked car, angled roof geometry from small office buildings, the outline of a stone bench, and a low hanging awning between two buildings. Vertical Canopy Structure Identification Fig. 5 3 shows an unfiltered point cloud generated from data over a forested region in Mine Run, VA. Small leaf structure elements are definable here based on a large number of multiple scattering events. The flight was conducted in autumn as the deciduous forest was in the process of shedding leaves; th is left small pockets of space in the overhead canopy through which individual beamlets could propagate. Even partial transmission of beam elements as a result of semi occlusion by leaf surfaces was enough to reconstruct intra canopy elements and the grou nd surface below the tree line. A zoom in depicting a narrow tree trunk is pictured.

PAGE 115

115 Optimizing System Performance Analysis of data from test flights showed that overall signal strength at maximum PMT 5 ) was high enough to warrant chan ging data collection parameters. Responses from strongly reflective surfaces at this gain level produce d a stacking effect (hits in many consecutive range bins) near roof surfaces in the resulting image. Since the micro laser output energy and detector ty pe were selected to maximize LSNR performance in turbid waters, t his was an expected outcome T he dynamic range between different types of features (especially when contrasting topographic and bathymetric returns) may be very large. Responses from artifi cial surfaces for on nadir beam propagation at 2500 V were tens of range bins deep in some cases. Data collections at multi photon sensitivity ( 5 gain) are capable of producing responses 3 to 4 bins deep from surfaces such as building roofs, bare earth, and tops of tree canopies We could thus reduce the tube voltage for future surveys and expect the top end of our dynamic range to be reduced. Generally speaking, i f interrogated far field surfaces were of uniform reflectance and the gain of the tube is matched to provide exactly enough amplification to barely produce a threshold crossing, this might be an acceptable technique. Of course in practice target reflectance properties are not precisely known and tailoring a detection strategy to one type of feature negates the real b enefit of this type of sensor mixed mapping. As shown in Chapter 4, shallow water bathymetry is difficult at less than full amplification ( 2500 V ) For the LSNR detection paradigm to be worthwhile related to submerge d surface mapping an amplification level should be maintained that produces returns from even very weak scatterers.

PAGE 116

116 Another solution may be viable: changing the type of airborne profiling from on nadir to off nadir. By imposing a fixed incidence angle re lative to ground for outgoing laser light, the magnitude of the backscattered radiation is decreased due to less significant contributions from specular reflections and increased slant range. This has special relevance in the case of mirror like water sur faces as large specular components cause problems for shallow bathymetry on nadir operation. A circular pattern is one way to propagate this effect throughout a scan and retain contiguous terrain coverage; in fact along track resol ution is effectively dou bled. Even with off nadir operation, and the potential advantages of circular versus raster scanning, CATS data shows that there is still a dynamic range issue because of the large difference in signal strength between topography / sea surface elements and distributed targets such as partially occluded forest canopies and submerged surfaces. For large scale uniform surfaces such as buildi ng roofs, it may be possible to automatically collapse range tails into first returns and simultaneously provide some me asure of intensity about the response, dependant on the number of range bins that the signal stretches over and the expected cross section geometry [ 24 ]. Dynamic range is one of the major challenges for adaptive algorithm development [ 73 ]; automatic proce dures are usually dependent on the local co herence of return information. Data Artifacts Accurate characterization of data artifacts can be one of the most challenging obstacles for a developmental LIDAR system. This is especially relevant for LSNR mappin g as in some cases the noise floor can eclipse the true signal level. We expect, however, that true signals will exhibit some form of spatial coherence even in lowly populated areas due to the random nature of noise formation, allowing us to use

PAGE 117

117 technique s such as Post Detection Poisson Filtering (PDPF) and a Correlation Range Receiver (CRR) [ 15 ] to beneficially exploit data trends. We will consider in detail artifacts specific to the CATS to evaluate the applicability of this claim for mixed topographic and bathymetric data processing. A closer inspection of detected surfaces in Fig. 5 2 shows a potential problem; a "ghost image" of roof surfaces appears at a constant offset beyond the first return Though we may be able to easily remove these types of a rtifacts for bare earth and building measurements via a n automatic first surface algorithm we risk losing important multiple return information from other types of features On the other hand, if no algorithm is applied, we risk lumping the ghost returns in with partial reflections in close proximity (<10 m) to the first O ur data sets may then also require significant manual processing. In the worst case scenario an automatic data classifier may erroneously identify a ghost image from the water surface as a submerged element, yielding an inaccurate depth estimation. Analyzing the scope of such artifacts is thus of prime concer n for shallow water bathymetry. The influence of this type of noise is not always obvious because the magnitude of the responses is suspected to be fairly low, meaning that certain types of data visualizations can cause the effects to be classified as random thermal events or volumetric scattering residuals rather than deterministic occurrences. It is important to develop an autom atic data processing technique that compensates for the noise effects so that their influence does not lead to false feature characterizations. For example, when first returns are p lotted from channel 23 in Fig. 5 4(A ), ghost surfaces are difficult to ide ntify The image seems close to what we would expect; canopy returns are well

PAGE 118

118 distributed for partial leaf on conditions, bare earth surfaces (x<150 m) are singular, and water returns (x>150 m) are generally bi modal. The distribution of noise returns re lative to these features, however, should be analyzed in further detail. Above the ground / water surface, we expect random contributions from dark / solar events as well as atmospheric laser backscatter. Below ground and past ~5 m water depth, we expect only contributions from dark / solar noise. Fig. 5 4(A) thus gives us some indication that an aberrant effect may be in play; below ground noise events are more numerous. The opposite should be true if our noise assumptions are valid. W hen all returns f rom 16 channels of data are plotted in Fig. 5 4(B) artifacts become much more evident; atmospheric noise increases slightly as expected, but the large number of below ground returns is now a stark contrast. This effect persists even through overlapping f irst return data; the theorized cause is a combination of noise and crosstalk between PMT grid elemen ts. Though it is clear from Fig. 5 4(B) that artifacts below strongly reflective surfaces are significant, further quantification may aid in noise identifi cation and processing efforts. For example, MATLAB can be used to compute estimates for local density in selected regions within the cloud. A segment of flat ground data was manually windowed to encompass expected signal returns. Ghost responses should exhibit well defined correlation to this window; the temporal offsets of expected highest correlation are show n in Fig. 5 5 as "afterpulse 1/2." By then sliding the true signa l window vertically as in Fig. 5 6 and expressing the correlation coefficient as a ratio of true signal density to the measured effect, the plot in Fig. 5 7 (A) was generated. The plot decreases exponentially starting at zero distance (correlation % of 1.0) down to the

PAGE 119

119 noise floor, which is below a correlation % of 0.1. Here the corre lation of ghost images is not distinguishable. We then conclude that the spatial distributions of artifacts are significantly less dense than the true signal. In order to analyze information near the noise floor, the y scale needs to be expressed logarit hmically as in Fig. 5 7(B). If we fit Gaussian profile s to the output and measure the peaks, the largest correlation s occur at 12.67 and 24.00 ns, with respective peaks of 0.0085 and 0.005 3 (Fig. 5 7(C)). W e can posit several possible explanations for th e noise. First gas impurities in the tube envelope can generate additional responses beyond the initial high intensity return ; this is commonly referred to as "afterpulsing" in nuclear particle physics experiments Second the cable connecting the PMT o utput to the high speed data interface could have an impedance mismatch. A mismatch could cause undesirable reflection s from the interface junction s to become significant at very high signal levels. Third the silicon germanium high speed gain block coul d be being overdriven in cases of large signal output. This would cause the amplifier to operate beyond maximum recommended ratings, thereby inducing anomalous effects. PMT Afterpulsing Exploration into the chemical and physical nature of PMT afterpulsing effects [ 74 ] [ 75 ] [ 76 ] showed that diffusion of trace gases through the tube envelope contribute to an increase in the probability of pulse formation. In particular, diffused helium was found to be the main source of afterpulses in tubes more than two years old. In contrast, the contribution from this impurity in new tubes was often small [ 76 ]. An exponentially decaying low level background signal has been shown to occur after the abrupt termination of a sub microsecond light pulse [ 77 ]. Positive ion s produced during the ionization of residual gas in the PMT by photoelectrons produce secondary

PAGE 120

120 electrons during their drift back onto the photocathode, which are then amplified [ 78], [ 79 ]. Other possible explanations for both short (sub microsecond) and long term (microsecond to millisecond) responses are electron induced luminescence of the dyno des or their support structures and microdischarges produced during high gain operation [ 77 ]. Because the CATS uses an MCP PMT manufactured in 2004 that has under gone significant use, it is possible that trace amounts of atmospheric gas diffused into the envelope over time may have risen to significant levels Quantifying the amounts of impurities is beyond the scope of this work, but potential contributions from afterpulsing should be mentioned due to the age of the tube. Impedance Matching For radio frequency (30 kHz to 300 GHz) circuitry, the impedance of the transmission line itself becomes significant. In order to maintain maximum power transfer, the characte ristic impedance of the cable ( ) should be closely matched to the source ( ) and load ( ) If the total path length is greater than 10% of the signal wavelength, mismatch effects can have greater impact In the CATS system, 2 GHz level signals mean th at transmission lines longer than 4 cm must be well matched (assuming velocity factor of 0.8). We should then consider the layout of associated PCBs in the receiver design and estimate the to tal length between the PMT and data capture elements. The minimum trace length is the sum of the following lengths: MCP PMT pin to PMT adaptor board connector: 12.95 cm Samtec HFHM2 cable (1): 25.4 0 cm Samtec HFHM2 cable (2): 25.4 0 cm Frontplane board tra ce: 8.13 cm Connectors: 12.70 cm

PAGE 121

121 The total trace is then 84.58 cm Assuming an average propagation delay of 70.98 ps / cm the total delay is 6 ns. The layout of potential cable reflections in relation to a transmitted signal is shown in Fig. 5 8 In t his model a signal pulse is initiated at T=0 at the near end of the trace T he pulse reaches the far end of the trace at T=6 ns. Here a portion of the signal is reflected back dependant on the match between R o and R L which causes another reflection at T= 12 ns based on the match between R s and R o This reflection reaches the far end of the trace at T= 18 ns. This process continues infinitely, and output pulses of diminishing magnitude will be received at T= 6, 18, 30, etc. intervals. The estimated spaci ngs of the first two reflections are close to the temporal spacings calculated from local density correlations (12.67 and 24 ns). The next step is to calculate the expected magnitude of reflections based on manufacturer specifications for impedances. The line lengths of the Frontplane board are matched for a maximum channel to channel delay difference of 18 ps. Signal routes are 50 single ended lines. Test data using Time Domain Reflectometry (TDR) are given in Table 5 1. The measurements show the ave rage impedance for each layer; deviations are expected due to variations in PCB manufacturing tolerances of the layer thickness and line width. If we let = 49.9 consider Layer 3 (L5), Board 2 as a worst case scenario, we can use the following analysis to calculate the magnitude of the first two reflections. When a signal is impressed on the end of a transmission line, a fraction of the full source voltage propagates down the line. This is referred to as the input acceptance function and is expressed in terms of and [ 80 ]

PAGE 122

122 (5 1) Next, the signal is attenuated by the propagation function For this analysis we will assume an average value of 0.94. At the far end of the cable a fraction of the attenuated signal amplitude emerges, which is determined by t he output transmission function, (5 2) When a fraction of the propagating signal emerges from the far end (T=6 ns in our model), a reflected signal also travels back along the cable towards the source. The magnitude of this reflection is determined by t he far end reflectance function, (5 3) The reflected sign al is again attenuated again by as it travels towards the source where it reflects a second time off the source impedance. The source end reflection function is given by, (5 4) After the second reflection, the signal is attenuated by as it travels back to the load as it again emerges via the transmission function This process continues on indefinitely. We can use these functions as coefficients to calculate the magnitude of the first three signals: (5 5)

PAGE 123

123 For = = 49.9 and = 52.84 upper worst case), we find = 0.47, = 0.0004, and = 2.46 10 7 For = = 49.9 and = 47.52 worst case), we find = 0.47, = 0.0002, and = 1.31 10 7 The first reflection in both worst case scenario s is then expected to be three orders of magnitude smaller than the true signal, and the second reflection s ix orders or magnitude smalle r. Though potential reflections resulting from worst case impedance mismatch may occur at the same temporal spacing as the ghost images, the magnitudes appear to not be significant. Gain Block Overdrive Output signals are amplified by a 4000 MHz 2.8V SiGe gain block manufactured by Sirenza Microdevices (part number SGA 1263). The output of this device is fed directly into the threshold comparator for each channel Because the system needs to operate on a single photon sensitivity level, channel threshold s are set to a 4 mV trigger. According to the manufacturer's data sheet, t he gain block has a maximum RF input power of 12 dBm or 63 W, which equates to 56 mV into a 50 load. An output current over 1.1 mA will therefore begi n to overdrive the gain b lock. In Fig. 5 9 a large signal (1 1 0 mV) is injected into the gain block. This is approximately 25 times larger than a signal expected to barely produce a threshold crossing (4 mV). The resulting output is shown in Fig. 5 10 The desired response is t he first large negative pulse. Deviation s from zero level in the waveform after the first negative pulse are anomalous effects resulting from the large signal input. Here we see small negative deformations before and after a positive peak. If we inject a larger signal

PAGE 124

124 (2 0 0 mV) (Fig. 5 11 ), the output shown in Fig. 5 12 is produced. A line is drawn here to appr oximate the threshold crossing. T he second crossing occurs at approximately 13 ns. If a still larger signal (300 mV), shown in Fig. 5 13, is inj ected in to the block, we can identify threshold crossings in Fig. 5 14 at approximately 13 and 20 ns The temporal spacing s of these outputs are similar to that estimated from local density correlation within the point cloud. Because the dynamic range bet ween submerged surfaces and highly reflective topography is so large, it is likely that the overdrive of the gain block is the main cause of ghost images. Point Cloud Filtering Based on Density Estimates A data processing method should be developed to elim inate the noise artifacts while preserving feature structure within the point cloud. Because the spatial densities of data artifacts are fairly low (peaks less than 1% of true signal), we can perform point cloud filtering based on local point information and expect not to lose a significant amount of signal information about strongly reflective surfaces. To validate the potential of the LSNR data, generated algorithms should be near automatic; i.e. with the least amoun t of manual input as possible. First, a decision boundary between land and water surfaces was estimated using a sliding hor izontal window similar to Fig. 5 6 For each point in th e data set used to create Fig. 5 4(B) a 0.6 x 1.0 x 1.0 m box was fitted to produce density estimates in the top ographic data defined by the decision boundary. All other points were then classified as potential bathymetric responses and a 0.3 x 1.5 x 1.0 m box was fitted to produce density estimates. The results of this operation on data gathered from a single cha nnel of information are shown in Fig. 5 15 Next, the scene is split into coarse grid cells and data within cells below the noise threshold are eliminated (Fig. 5 16 (A) ). The secondary

PAGE 125

125 step helps to eliminate the few remaining artifacts and the final pr ocessed image is shown in Fig. 5 16 (B) A dense clump of points 5 m above the water surface is visible at x=180 in Fig. 5 16(B) ; this is the result of returns from an overhanging island tree near the flight line. Because the density of returns suggests t hat this is not atmospheric noise, the filter retains the associated points. Fresh Water Bottom Mapping The major goal of flight testing for the CATS instrument was to evaluate LSNR water penetration capabilities. A pplication of the expectation maximizati on (EM) Gaussian mixture (GM) algorithm shows that surface and bottom can be reliably separated in sea (total attenuation coefficient of 0.05 / m) and coastal (total attenuation coefficient of 0.5 / m) water to a depth of 2 m [ 21 ]. At larger depths, simul ation also shows that horizontal resolution of the reconstructed sea bottom is greatly limited by water clarity; meter scale bathymetric mapping should only be possible at sea depths less than three times the Secchi depth. A zoom in view of processed data from the shallow fresh water area (the Lake of the Woods) in Mine Run, Virginia is shown in Fig. 5 17 (B ). The path of the aircraft near the shoreline is indic ated in an aerial phot o (Fig. 5 17 (A )). The water in this area contained low amounts of colored dissolved organic material (CDOM) ; at the time of the data collection the Secchi depth was measured out to 2 m. The lake bottom was a mixture of dark sand and decaying organic material; bottom reflectivity was expected to be very low. 16 channels of data were collected at 2500 V, or maximum PMT amplification. Manual measurements of water depths in this area at 10 m intervals were also performed using a measuring tape with an attached weight.

PAGE 126

126 Point cloud values were interpolated to estimate the lake botto m surface, shown as a dashed line in Fig. 5 17 (B ) The bottom was then corrected for the index of refraction of water (1.33), shown as a solid line. Here we see a consist ent response from the water surface shown as a flat line, and an exponential decay of water column returns progressing to further depth. The lake bottom produces a strong signal at ~1.5 m depth near the shoreline at x=140 m. Water depths at 10 m intervals were extracted from the LIDAR data in this region by differencing the interpolate d water surface and corrected lake bottom function Lake bottom returns become negligible at approximately 2 m water depth. There was a large amount of sediment buildup from organic / inorganic particulate near a small island located at x=180 m. This co uld lead to shortening of the estimated depth near the island. Table 5 2 compares LIDAR depth estimates to measured values. The average difference between estimates and measurements is 21 cm. The lake bottom was slightly more reflective in other parts of the lake. Fig. 5 18 (A ) shows a location to the north east at the same lake where the bottom was composed of a mix of lighter sand and organic material. The increase in reflectivity produced returns from the lake b ot tom at greater depth. Fig. 5 18 (B ) sho ws an off nadir profile for 16 channels at 2500 V depicting reco very of the lake bottom out to 2 m depth. Salt Water Bottom Mapping Further flight testing was conducted over the intracoastal waterway in St. Augustine, Florida. Fig. 5 19 (B) shows point clo ud data starting from the shoreline ( aerial photo in Fig. 5 19(A)) and out to a boat dock 60 m away In Fig. 5 19(B ), the water surface and bottom profile between boat docks and a shallow boat launch region are depicted. Dense clusters of points 1 to 2 m above the water surface are a result of

PAGE 127

127 na rrow wooden walkways (x=175, 189, 203 m) leading out to the boathouses. The sand bottom was reflective enough in this case for a single channel to provide bottom reconstruction o ut to a depth of 2 .5 m. Hand meas urements of wood platform heights above the water surface w ere performed at five locations, shown in the third row of Table 5 3. LIDAR estimates of platform heights were calculated by differencing the first above water return associated with the wood surf ace and the interpolated bottom at the five indicated locations. These estimates showed an average error of 17 cm. Fig. 5 20 (B) shows information from 3 channels over a sand bar. Bot tom features out to a depth of 1 .5 m are recovered here in high detail; submerged pockets and dunes are identifiable as currents have affected the sand geometry. An aerial photo of the east coast of Florida near St. Augustine with the aircraft path is shown in Fig. 5 21 (A ). Fig. 5 21 (B ) shows the associated point data from 3 channels. Of particular interest in this image are the gradually increasing sea floor slope in the surf zone near the beach / water interface (indicated in dark blue) and the sand bar located 40 m from shore. Mapping Statistics and Discussion Simulation [ 21 ] predicts that CATS bathymetric performance will largely be dependent on water conditions; in turbid waters, the operational depth for mapping should be limited to a few meters. A nalysis of airborne profile data from the CATS over a fresh water lake in Mine Run, Virginia and salt water areas near the eastern coast in St. Augustine, Florida have largely confirmed these predict i ons for practical turbidities at 0 to 3 m depths. The central 16 channels 17 % of the total footprint area were sufficient in profile mode to reconstruct a fresh water lake bottom composed of dark decaying organic / inorganic material (estimated to be highly non reflective) out to 1.65 m. The Secchi depth in this area was meas ured to be 2 m. At a depth of 0.5 to 1.5 m,

PAGE 128

128 on av erage 38 returns per meter were registered from the lake bottom. For 8 kHz laser pulse rate and 60 m/s flight speed, this corresponds to an average of 1.77% of the outgoing shots registering returns from the lake bottom. The central 16 channels were also used to reconstruct a fresh water lake bottom composed of light sand and dark decaying organic material (estimated to be sl ightly more reflective) out to 2 m. In Fig. 5 18(B ), less bottom returns near the shoreline were received as a result of overhangin g tree branches, but multiple returns still provided information about submerged surfaces. At a depth of 1 to 2 m, on average 74 returns per meter were registered from the lake bottom. This corresponds to an average of 3.49% of the outgoing shots registe ring returns from the lake bottom. A single channel of information 1 % of the total footprint area was used in profile mode to reconstruct a salt water intracoastal floor (estimated to be highly reflective) i n water of high clarity out to 2 .5 m. At a d epth of 1 to 2 m, on average 45 returns per meter were registered from the submerged floor. This corresponds to an average of 33.98% of the outgoing shots registering returns from the floor; these are assumed to be near ideal conditions. Three channels o f information 3 % of the total footprint area were used in profile mode to reconstruct the sea floor (estimated to be highly reflective) in the surf zone near a beach in water of optimum clarity out to 2 m. At a depth of 1 to 2 m, on average 56 returns per meter were registered from the sea floor. This corresponds to an average of 13.98% of the outgoing shots registering returns from the floor.

PAGE 129

129 A B Figure 5 1 Photo s of the CATS mounted in aircraft. A) Cessna Skymaster aircraft. B) Mounted CAT S sensor head. A B Figure 5 2 Flight testing in Gainesville, FL over University of Florida campus. A ) A erial photo with flight line. B ) Channels 20 and 23, 2300 V, all returns, unfiltered data showing vertical resolution capability in structure iden tification.

PAGE 130

130 Figure 5 3 Small point cloud from forested region in Mine Run, VA. Small leaf structure elements are definable along with trunk and ground information. A B Figure 5 4 Large point cloud from forested region in Mine Run, VA A) Channe l 23, 1st returns plotted. B) Channels 17 32, all returns plotted.

PAGE 131

131 Figure 5 5 Ghost images in raw data. Figure 5 6 Vertical sliding window to evaluate magnitude of ghost responses.

PAGE 132

132 A B C Figure 5 7 Results from sliding window. Correlation between responses in 0.1 m increments, starting from true surface (correlation coefficient of 1.0). A) Normal scale y axis, peaks are indistinguishable. B) Log scale y axis, ghost peaks are visible. C) Magnitude of peaks are both less than 1%. Figure 5 8 Graphical model depicting reflections from detector cable interfaces.

PAGE 133

133 Table 5 1. Time Delay Reflectometry test d ata (courtesy of Fibertek Inc.) Board # L1 L3 L5 L7 L9 L11 1 51.28 52.00 51.17 51.43 51.91 51.72 2 49.27 52.84 52.82 51.63 51.29 51.88 3 47.52 52.00 50.80 51.82 52.47 51.76

PAGE 134

134 Figure 5 9 L arge signal (11 0 mV) injected into gain block (courtesy of Fibertek Inc ). Fig ure 5 10 Output of gain block from 110 mV signal (courtesy of Fibertek Inc.)

PAGE 135

135 Figure 5 11 L arge signal (20 0 mV) injected into gain block (courtesy of Fibertek Inc.) Figure 5 12 Outpu t of gain block from 200 mV signal (courtesy of Fibertek Inc.)

PAGE 136

136 Figure 5 13 L arge signal (30 0 mV) injected into gain block (courtesy of Fibertek Inc.) Figure 5 14 Output of gain block from 300 mV signal (courtesy of Fibertek Inc.)

PAGE 137

137 Figure 5 15 Result of first step of MATLAB local density filter for a single cha nnel of raw airborne profile data. A B Figure 5 16 Result of second step of MATLAB local density filter for raw airborne profile data. A) Retain ed and filtered points. B) Retained points only.

PAGE 138

138 A B Figure 5 17 Flight testing in Mine Run, VA ove r the Lake of the Woods southwest A ) Aerial photo near boat docks and shallow la unch region with flight line. B ) Off nadir profile, channels 17 32, 2500 V, all returns, processed data depicting fresh water penetration out to 1.65 m.

PAGE 139

139 Table 5 2 Comparis on of LIDAR depth estimates and m easurements X (m) 150 160 170 180 190 200 LIDAR Depth Estimate (m) 1.12 1.45 1. 4 9 1. 24 1. 65 NA Depth Measurement (m) 1.27 1.52 1.65 1.52 2.03 2.44 Difference (m) 0. 1 5 0. 07 0. 16 0. 28 0. 38 NA

PAGE 140

140 A B Figure 5 18 Flight testing in Mine Run, VA over the Lake of the Woods northeast A ) Aerial photo near lak e shoreline with flight line. B ) Off nadir profile, channels 17 32, 2500 V, all returns, processed data depicting fresh water pe netration out to 2 m.

PAGE 141

141 A B Figure 5 19 Flight testing in St. Augustine, FL over docks on the int racoastal waterway. A ) Aerial photo near boat docks and shallow la unch region with flight line. B ) Off nadir profile, single channel (18), 2500 V, all returns, processed data depicting salt w ater penetration out to 2 .5 m.

PAGE 142

142 Table 5 3. Comparison of LIDAR p latform distance e stimates and m easurements X (m) 175 189 20 3 214 2 36 LIDAR Platform Estimate (m) 1.68 2.13 2.37 2.53 2.90 Platform Measurement (m) 1.82 2.20 2.59 2.74 3.12 Difference (m) 0 .14 0.07 0.22 0.21 0.22

PAGE 143

143 A B Figure 5 20 Flight testing in St. Augustine, FL ov er a sandbar on the intracoastal waterway. A ) Aerial photo ov er sand bar with flight line. B ) Off nadir profile, channels 21 23, 2500 V, all returns. Point cloud depict ing sand bottom out to 1.5 m.

PAGE 144

144 A B Figure 5 21 Flight testing in St. Augustine, FL over the eastern coast. A ) Aerial photo of f light path relative to beach. B ) Off nadir profile, 2500 V, channels 21 23, all returns, processed data related to beach / water interface region.

PAGE 145

145 CHAPTER 6 CONCLUSIONS AND FUTU RE WORK Scope of Research This dissertation has described the essential design elements necessary to construct a low signal to noise ratio (LSNR) light detection and ranging (LIDAR) system and assesse d the capability of single wavelength, mixed topographic / bathymetric mapping by analyzing the design and testing process of the University of Florida's Coastal Area Tactical mapping System (CATS). Design and Testing Overview Airborne laser swath mapping (ALSM) instruments have conventionally used high energy / high pulse rate designs. Exploration into alternate detection strategies with the benefit of a highly sensitive detector suggests that detection efficiency can be improved by moving to a LSNR LIDAR station (SLR2000) demonstrated the capability of employing passively Q switched microlasers with output energies of level ranging precision at satellite altitudes of 20,000 km [ 23 ]. Discrimination between signal and noise was achieved by utilizing a range receiver able to identify and extract laser echoes from solar ba ckground noise through Poisson statistical analysis of the received photon distribution. In a very low incident light state, a high gain (10 5 ) photodetector is able to produce amplified responses due to weak backscattered laser radiation via the photoelect ric effect. Individual discrete pulses can be analyzed in a digital mode. If the time resolution of a detector is sufficiently fast that the average time intervals between signal

PAGE 146

146 generation of a photoelectron in this configuration is a binary event given proper discriminators, making the photon counting method superior to analog signa l measurement in terms of detection efficiency and stability. The power level of incident radiation from signal events may therefore be on the order of the noise, e.g. solar events, reflections off atmospheric particles, and detector thermionic emissions ( dark events). Noise events may occur randomly and with low enough frequency to be separated from terrain signals; this has been practically demonstrated by data collected and analyzed by Goddard Space Center in flights of a photon counting laser altimeter showing clear identification of trees, buildings, and even potential shallow water penetration [ 23 ]. An operating wavelength between 500 to 600 nm allows for water penetration due to minimum absorption in this spectral band. In order to estimate the out put energy required for single photon detection, signal strength can be estimated as a function of transmitter, receiver, and environment characteristics. Using a modified form of the LIDAR link equation, we can expect 63% chance of signal event registrat ion at a one photoelectron th reshold can by transmitting greater than 1 3 5 n J pulse energy given the following parameters: a receiver optical efficiency of 0.4, detector quantum efficiency of 0.28, hologram optical efficiency of 0.8, 532 nm wavelength, 0.1 5 surface reflectivity coefficient (representative of sand), 5 local surface incidence angle, and receiver 3 m 2 The Coastal Area Tactical prototype LSNR mapping system, has prov ided data essential to exploring the viability

PAGE 147

147 of a multi element design with these parameters. Improved spatial resolution in single beam scanning LIDAR systems has traditionally been achieved by increasing laser pulse rate to decrease the amount of spac e between scanned ground footprints. This provides an inherent limitation to avoid the condition of multiple pulses in flight at once; the two way travel time of electromagnetic radiation at an average flying height of 600 m dictates a maximum pulse rate of 250 kHz. If, however, a single laser beam is split into an array of elements and transmitted at once, spatial resolution improvements are then dictated by the number of array elements and the capability of the receiver to process returns over the array simultaneously. A moderate laser pulse rate can then theoretically provide decimeter scale resolution. In the CATS, a passively Q switched, frequency doubled neodymium doped yttrium aluminum garnet (Nd:YAG) microchip laser was selected to produce a stabl e train of pulses. with 2 mrad divergence. Laser light passes through a 10x magnification beam expander, which also decreases the beam divergence to 0.2 mrad. After the beam expande r, each pulse is passed through a holographic diffraction element that separates the beam into a 10x10 square array of beamlets rotated 45 each containing ~35 nJ of energy Outgoing laser beamlets then pass through shared afocal lenses with a total magn ification of 4x. The laser light is deflected through two Risley prisms (V coated optical wedges with 99.5% light transmission). BK7 material is selected with a refractive index of 1.52, wedge angle of 13.58, diameter of 88 mm, and central thickness of 22.5 mm. The refractive properties of these elements combined with the wedge surface angle deflect the beam through a predicted path. This allows for the

PAGE 148

148 array of laser beamlets to be arranged in a variety of repeatable patterns. Different types of patte rns can provide enhanced coverage for specific applications. Temporal filtering via a range gate eliminates atmospheric backscatter immediately outside the aircraft and ensures that strongly reflective targets near the output window do not produce a respon se. Incident radiation passes back through a shared afocal lens. Spatial (2.8 mm field stop) and spectral (0.3 nm ba ndwidth about 532.1 nm) filters reduce unwanted interference, and the filtered beamlets are rotated 45 by a Dove prism and refocused via a telephoto lens before reaching the MCP PMT and the associated timing electronics. Because a very low amount of backscattered radiation is expected from each laser beamlet, a detector must be employed with high enough gain to amplify single photon events. In order to achieve 10 5 gains with 200 ps rise time, 25 ps transit time spread (TTS), and 200 ps dead time specifications, a micro channel plate photomultiplier tube (MCP PMT) was selected. This device is capable of simultaneous detection of independent events from a 10x10 pixel array, allowing for two dimensional resolution. The MCP PMT provides exceptional gain while maintaining compact size (less than 10 cm 3 ), immunity to anode saturation, and stability with respect to variations in supply voltage. Varying threshold settings with regards to pulse height distribution plots can provide application specific versatility. Ground based field testing results have yielded 10 cm range accuracy at one standard deviation from the mean and small scale feature id entification at variable scan settings. Building edges, tree structure, bushes and narrow light poles are easily discernable from background noise. GPS time tag information reconciles the position of

PAGE 149

149 the scanner at any given time in relation to the tempo ral sequence of laser shot in a data set and Risley prism wedge index marks. Analyzing this data allows for study of potential data flow problems, e.g. if the laser fire slaving system happens to miss synchronization pulses, or if the scanner rotation is malfunctioning. An accurate time history with regards to scanning operations aids in point cloud rectification and diagnosis of mechanical errors. The CATS sensor is capable of recording multiple returns in each channel of the PMT from each laser pulse Analysis of distributed returns allows for identification of ground reflections under dense foliage and penetration past the water surface. Static experiments were conducted to test LSNR LIDAR performance metrics such as minimum separation distance betwee n multiple scattering objects, water penetration depth for submerged surfaces, and multi vs. single photon sensitivity. Discriminating between accumulated histogram waveforms of two separate objects required a shot returns was thus established. Varying the PMT sensitivity changed the contrast between noise production and detected signal strength. Atmospheric noise increases significantly on a consistent basis as gain increases toward a single photoelectron threshold (due to solar noise events generally being single photon events). Depending on the reflective properties of targets in the scan area, employing lower gain can decrease the amount of noise throughout the range gate. The CA TS offers the capability to change the PMT supply voltage between experimental collections. Operating at multi photon sensitivity may have value for

PAGE 150

150 situations in which high ambient noise precludes single photon sensitivity, as ground based testing has in dicated that man made objects such as buildings features are easily distingu ishable even at lower gain. Airborne d ata collections at multi photon sensitivity ( 5 gain) are capable of producing responses 3 to 4 bins deep from surfaces such as building roofs, bare earth, and tops of tree canopies Reducing the tube voltage could help to reduce dynamic range variance. Generally speaking, i f interroga ted far field surfaces are of uniform reflectance and the gain of the tube is matched to provide exactly enough amplification to barely produce a threshold crossing, this might be an acceptable technique. It is difficult however, to estimate target refle ctance properties ahead of time, and tailoring a detection strategy to one type of feature negates the real benefit of this type of sensor mixed mapping. Static testing shows that s hallow water bathymetry is difficult at less than full amplification (2500 V or 8 10 5 ). For the LSNR detection paradigm to be worthwhile related to submerged surface mapping an amplification level should be maintained that produces returns from even very weak scatterers. Another solution has shown significant promise for reduc ing dynamic range without significantly altering bathymetric responses : changing the type of airborne profiling from on nadir to off nadir. By imposing a fixed incidence angle relative to ground for outgoing laser light, the magnitude of the backscattered radiation is decreased due to less significant contributions from specular reflections and increased slant range. This has special relevance in the case of mirror like water surfaces as large specular components cause problems for shallow bathymetry on n adir operation.

PAGE 151

151 A circular pattern is one way to propagate this effect throughout a scan and retain contiguous terrain coverage; in fact along track resolution is effectively doubled. Simulation has suggested that surface and bottom can be reliably separa ted in sea water to a depth of 2 m [ 21 ]. At larger depths, simulation also suggests that the horizontal resolution of the reconstructed sea bottom is greatly limited by water clarity. Static tests of the CATS sensor show that at a depth of 50 cm in coast al water, on average 12% of shots registered at least one signal event. This was estimated to be a sufficient signal level to produce meter scale shallow (2 to 5 m depth) bathymetry data at coarse spatial resolution (1 m) Before reporting on bathymetric statistics from airborne data, contributions from several different noise sources were considered. Two "ghost" images commonly appear after responses from highly reflective surfaces The first is the largest in amplitude and consistently occur s approxima tely 12.67 ns beyond the first return of the true surface The second is approximately half the amplit ude of the second, and occurs 24.00 ns later. Three possible sources have been suggested for the ghost images : afterpulsing from gas impurities in the t ube envelope impedance mismatching in PMT cables, and gain block overdrive T he latter was shown to be the most likely cause of the spatially coherent noise due to the dynamic range relation Reflections from impedance mismatches in transmission line co mponents will be distributed at similar temporal offsets, but their magnitudes are expected to not be significant. Increased levels of random noise after strong input signals suggest contri butions from PMT afterpulsing; factory testing should be performed to identify diffused impurity levels.

PAGE 152

152 T hough more significant than dark / atmospheric noise, the magnitude of the artifacts was evaluated with regards to local point cloud densities. Because the calculated densities were fairly low (peaks less than 1% of the true signal), we can perform point cloud filtering and expect not to lose a significant amount of signal information about highly reflective surfaces. The result is an affirmation of the temporal coherence idea for LSNR LIDAR expressed by Degnan in 2 001 [ 15 ]. Even with potentially varied sources of noise producing corruptive data, we can still identify important features via automatic cleaning algorithms. After post processing, data from the CATS has largely confirmed bathymetric predictions via airb orne profile scans over a fresh water lake in Mine Run, Virginia The central 16 channels, or 17% of the total footprint area, were sufficient to reconstruct a fresh water lake bottom composed of dark decaying organic / inorganic material (estimated to be highly non reflective) out to 2 m. Approximately 3.49% of the outgoing laser shots registered a return from the lake bottom. With partial occlusion from an overhanging tree near the lake shore, multiple returns still provided enough information about th e submerged surface to fit a gradually decreasing trend line consistent with more highly dense points farther away from shore. An accuracy survey at five near shore locations yielded 21 cm average error. In a salt water intracoastal area near St. Augustin e, FL, a single channel of information (1% of the total footprint area) was used in profile mode to reconstruct the submerged floor out to 2 .5 m depth The light sand here was estimated to be highly reflective and turbidity near ideal for these types of c oastal waters. On average 33.98% of the outgoing shots registered returns from the bottom. An accuracy survey at five

PAGE 153

153 dock platforms yielded 17 cm average error. In the surf zone near the beach, three channels of information (3% of the total footprint a rea) were able to reconstruct the sea floor in water of optimum clarity out to 2 m depth. Sea surface points in this part of the cloud are sparse, meaning that the local density filter produces voids near the surface in expected white water areas. The se a floor profile underneath these areas is uninterrupted, however, with limited discontinuity near the beach / water interface. On average 13.98% of the outgoing shots register returns from the sea floor. Future Work Automatic processing of topographic and bathymetric data can be difficult with high water turbidities and low reflectivity bottom surfaces. Even with off nadir operation, and the potential advantages of circular versus raster scanning, CATS data shows that there is still a dynamic range issue because of the large difference in signal strength between topography / sea surface elements and distributed targets such as partially occluded forest canopies and submerged surfaces. For large scale uniform surfaces such as buildi ng roofs, it may be poss ible to automatically collapse range tails into first returns and simultaneously provide some measure of intensity about the response, dependant on the number of range bins that the signal stretches over and the expected cross section geometry [ 24 ]. Dynam ic range is one of the major challenges for adaptive algorithm development [ 73 ]; automatic procedures are usually dependent on the local co herence of return information. It has been shown that gain block overdrive is the most likely cause of ghost images i n CATS data. Within the current receiver structure, we may be able add a limiting circuit to ensure that input to the block does not exceed maximum amplifier tolerances. Clipping large signals would result in a higher noise figure, b ut single

PAGE 154

154 photon leve l responses should occur below the 12 dBm level. We should then be able to limit responses to this level and eliminate the ghost effects. Another proposed method for combat ing the dynamic range problem is to implement a "sensing" stage to modify signal a mplification on a dynamic level. For example, the high voltage in the CATS system is already stepped on a per shot basis in order to reduce power consumption and extend PMT life. If a separate photodetector were placed in front of the primary MCP PMT (wh ether physically or by temporal delay sequencing) such that it were possible to determine the intensity of the backscattered radiation before stepping up the tube high voltage, the level to which the voltage was stepped could be dynamically altered. It wo uld be possible, then, to obtain different amplification for sea surface returns as opposed to sea bottom returns with fairly minimal modification to the current system structure. The MCP PMT is already operating at maximum amplification, but if a sensing stage negated the possibility of full amplification during periods of high intensity laser backscatter, it would be possible to fly lower than 500 m and potentially increase the maximum water penetration depth. The 10x10 array output is eye safe down to 1 00 m altitudes, so the only restricting factor even for populated areas are FAA regulations. Flying lower would also decrease the ground point spacing, providing a denser cloud and discrimination of finer features. The total power draw for the CATS system is 372 W during profiling and 396 W during scanning. In comparison, the Optech Gemini system power draw is average 560 W and peak 980 W. If a dedicated ASIC or FPGA could be developed to encapsulate system control on a scripted level, the need for a con trol laptop could be eliminated from

PAGE 155

155 the CATS design. The laptop currently requires 200 W, or approximately half of the total power draw of the entire system. Compensating for this part of the system with a low power processing unit would move the unit c loser to the realm of unmanned air vehicle (UAV) payloads. The potential for small feature identification in an airborne scanning capability has been shown in a second generation system developed by Sigma Space Corporation. A footbridge, gazebo, and stree t lamps are visible in Fig. 6 1 from flight testing over the U.S. Naval Academy Campus in Annapolis, MD. Design parameters for this system are similar to the CATS design, with the exception of higher (1.5x) output power and faster pulse rate (22 kHz) [ 81 ] A profile section in the Chesapeake Bay is shown in Fig. 6 2 depicting bottom reconstruction out to a depth of 4 m. The maximum depth tolerance in this turbid media (estimated water clarity 14/100) is larger than CATS results in water of higher clarity meaning that increasing laser pulse energy has significant effect An interesting result is shown in Fig. 6 3 for a profile over a melt pond in Greenland; here we have some of the best water clarity and bottom reflectivity possible. We can see the froz en lake bottom out to 15 m depth this suggests potential for LSNR LIDAR in submerged glacial mapping. Conclusions We can break down the essential qualities required for a successful ALB system into three general categories: accuracy, capa bility, and cost e ffectiveness [ 70 ] For CATS data, c omparisons to submerged ground truth in two different environments have shown average error within IHO Order 1 standards ( 50 cm at 95% confidence interval) for shallow (<5 m) waters Seamless mapping at the land/water interface and within the surf zone has been demonstrated, and reconstruction s of submerged bottom profiles

PAGE 156

156 were shown for a variety of cases. The potential for larger maximum depth tolerance in more turbid media has been indicated via reference t o a secon d generation system ; this system used slightly higher output energy (50 nJ) and the same detection methodology. The successes with drastically lower output energies (five orders of magnitude less than the traditional mJ level) suggest that a compact, low power microlaser and small form factor MCP PMT can serve as a basis for potential UAV miniaturization. Elimination of full waveform digitization, which is commonly found in most modern ALB systems, further supports this approach and also allows for signif icant cost reduction We have arrived at these conclusions by lowering the detection threshold down to the single photoelectron level allowing the system to record all effective observations without having to resort to an interpretation strategy such as a classic CFD implementation. This has the benefit of avoiding the requisite optical power integration problem normally encountered when working with a large amount of backscattered photons. The natural caveat is increased probability of false positives, but we have shown that spurious noise can be kept to a minimum even at high amplification with appropriate selection of spatial, spectral, and temporal filters. Using a very short laser pulse width (500 ps) allows for a sufficiently fast timing receiver to differentiate surface and bottom to a small margin (approximately 15 cm accounting for returns in separate range bins). This is n ot quite zero depth, but for bathymetric mapping beyond 0.5 m it is an acceptable margin. Combining this approach with expan sion of the primary beam into an array of sub beamlets produces ultra dense point clouds. As a result, even if uncertainty exists in representation of volume backscatter and bottom reflection, geometric patterns can be exp loited to aid

PAGE 157

157 classification T h is idea was illustrated by eliminating data artifacts in LSNR data associated with large dynamic range based on spatial density coherence. The possibility of exploiting the related detection paradig m has previously been evaluated in only a few development al systems.

PAGE 158

158 Figure 6 1. Flight testing of second generation LSNR LIDAR by Sigma Space Corporation over US Naval Academy campus in Annapolis, MD [ 81 ] Figure 6 2 Flight testing of second generation LSNR LIDAR by Sigma Space Corporation in Chesapeake B ay off US Naval Academy Campus. Point cloud depicts bay bottom out to approximately 4 m [ 81 ]

PAGE 159

159 Figure 6 3 Flight testing of second generation LSNR LIDAR by Sigma Space Corporation in Greenland Point cloud depicts ice bottom penetration out to 15 m [ co urtesy of Sigma Space Corporation ]

PAGE 160

160 APPENDIX A DATA PROCESSING ELEC TRONICS Ranging Receiver Gated timing events from 100 "channels" in a 10x10 array are recorded by a multi stop ranging receiver in the CATS. Signals from the detector elements are processe d at several different stages, as dictated by the system controller. Fig. A 1 shows a. receiver block diagram for n channels. The four stages are: Pre amplification Threshold Detection Serial to Parallel Conversion Data Capture / Control / Communications Logic The receiver system detects fixed threshold level crossings from multi element detector pixels. A narrow pulse received by any one of the detector elements is amplified and low pass filtered. The low pass filter is used to stretch the pulse due t o the fact that the input signal pulse width could be of the same order as the sample period of the threshold comparator. Given test data, the bandwidth of the low pass filter has been calculated to achieve greater than 90% counting efficiency for single photoelectron events. The detector has high quantum efficiency and low noise with a bandwidth greater than 1 GHz. The preamplifier is a wide band (DC to 1 GHz) low noise RF gain block that amplifies the detected signal to a level suitable for the process ing electronics to operate. Fig. A 2 shows a data capture board block diagram. After the signal is filtered, it passes through an automatic gain control (AGC) amplifier. The AGC amplifier is used to compensate for the non uniform response between singular elements across the detector array. AGC amplifier settings are voltage controlled via a digital to analog

PAGE 161

161 converter (DAC). A data capture field programmable gate array (FPGA) stores calibrated data values that are sent to the AGC DAC. Next, the signal g oes to a threshold comparator. DAC channels, associated with the threshold comparators, provide the threshold voltage reference. This voltage reference corresponds to the expected level for a single photoelectron event. The FPGA control logic provides r eference point data to the DAC. The output of the threshold comparator is low voltage positive emitter coupled logic (LVPECL). Serial to parallel conversion changes sequential pulse outputs from the comparator to an 8 bit parallel output at 1/8 th of the s ample clock speed. The operation of this block de multiplexes sequential photon detection from the threshold comparator and spreads the data out into eight data lines. This provides a relatively low speed data input for the data capture FPGA to synchroni ze to its internal clock and latch the data. Each device receives a separately driven differential clock based on the system clock. A divide by eight clock based on the input clock is an 8 bit word framer to synchronize the parallel data outputs. This c lock output is used to latch the 8 bit word into the data capture FPGA. A reset pin synchronizes multiple devices together. Because each serial to parallel converter requires its own separate clock, the high speed system clock is buffered with fan out dr ivers. In order to sample all the signals at the same time, each of the clock edges must be as close as possible. Programmable delay chips are used for this purpose in order to match propagation delays. Data management is dictated by a capture FPGA, whic h performs three main functions: (1) to collect and store data from two serial to parallel converters, (2) send commands to the corresponding DACs associated with each channel, and (3)

PAGE 162

162 communicate with an FPGA dedicated to control logic. Each four channel section requires the use of two capture FPGAs. A 16 channel board will thus have eight capture FPGAs. These units carry out actions based on instructions from control logic. High level control functions are built into a separate layer to support commun ication with the system controller. This forms the structure allowing the download of captured data via specific commands. Event Timer Circuitry The practical design for implementing the aforementioned theoretical structure can be separated into five sect ions: Back Plane Range/control board Detector Assembly Data Capture boards Front Plane In the CATS, all boards except for the front plane and detector interface board are plugged vertically into the back plane. The back plane accommodates a range/control board, a detector high voltage assembly, and three 32 channel data capture modules. The front plane board lies across and mates perpendicular to the top of the 32 channel data capture modules. The four outer pixels of the 10x10 array are currently not i mplemented because a 96 channel design was more easily created. The back plane supports a JTAG data bus shared by the three data capture modules. The JTAG data bus is used for command/control and reading/writing individual channel parameters. A 4 bit dif ferential data bus is used for downloading captured data to the system controller via the range/control board. Each of the data capture assemblies has four sets of 4 bit data bus outputs. For three data capture

PAGE 163

163 modules, 48 data bus lines to the range/con trol board are therefore supported. In addition, the back plane provides power and clock distribution. The range/control board has four main functions: (1) to generate a stable high frequency system clock, (2) provide control for the high voltage section of the detector assembly, (3) perform range gate sequences, and (4) support central communications between the data capture assemblies and the system controller. A JTAG link from the system controller interface is used to send command/control information to the range/control board. The command/control section interprets this information. The T0 pulse circuit is driven by an optical pickup monitoring bleed light from the Nd:YaG laser. The leading edge of each pulse in the train of laser output triggers a sequence of events related to data capture, similar to how the incremental rotation of each optical wedge is triggered. A range gate circuit first delays applying power to the MCP PMT according to a user set value; this ensures that the outgoing laser p ulse is far enough away from the sensor to avoid noise associated with power up and returns from objects outside the intended return window. Once the gate delay operation is complete, initializations are sent to the data capture boards and detector assemb ly. The command / control circuit shown in Fig. A 3 monitors the detector assembly data lines, and issues commands to the data capture boards via the JTAG link when the data lines are ready for transfer. A multiplexor circuit allows for all three boards t o be controlled via one control block, and transports return information from the 96 detector elements. High voltage power supplies are used for detector bias. In order to control the high voltage supplies, the detector assembly has both a DAC and an ADC. A low

PAGE 164

164 voltage reference input to the DAC is used to program the high voltage supplies. The system controller provides the voltage level settings to the range/control board. A corresponding digital value is sent to the DAC located at the detector assemb ly. The range/control board also reads the detector assembly ADC and uses the data for a voltage level feedback. To maintain detector gain stability, a command circuit incorporates closed loop control using the DAC data as the reference and the ADC data as the feedback value. For maximum sensitivity (and to prolong detector life) the photon counting detector is normally off and then gated for data collection. The high voltage is also switched to a lower level between range measurements in order to limit the average current consumption and build up of heat in the associated electronics. In order for the photon capture signal. The detector gate pulse is derived from the counter. A register within the FPGA circuit contains the terminal count that initiates the detector gate pulse. Terminal count values for both the start capture pulse and the detector gate pulse are received from the syste m controller via a circuit on the range/control board. Fig. A 4 shows the organization of the receiver components in the processing electronics block assembly. Each data capture board has 16 channels. Two data capture FPGAs and one control FPGA are used for every four channel sec tion. A 32 channel module has two data capture boards placed back to back. An aluminum plate is sandwiched in between these two boards for thermal transfer.

PAGE 165

165 The front plane provides equal path routes for the high speed detector signals from the detector assembly. Signals from the photon counting detector are routed via the front plane to each of the 100 data captur matched transmission lines. The individual data capture channels must simultaneously collect all signal outputs from each detector element. The front plane is used to equalize these signal paths among all channels. FPGA programming is performed over the backplane bus. The data capture and data control FPGAs receive programming information from the master range/control FPGA located on the range/control board. The same source code is uploaded to all of the FPGAs. I ndividual board variations can be built into the source code and selected using the board select address information. Additional board settings and parameters can be sent over the system JTAG bus after the program has been uploaded. This provides a numbe r of advantages: only one program needs to be maintained, the program resides within the system controller (can be easily changed), and receiver channel boards can be interchangeable without reconfiguration. These five parts back plane, range/control boa rd, detector assembly, data capture boards, and front plane are contained within a cube below the cylindrical motor head shown in Fig. A 5. This cube housing the timing electronics can be decoupled from the optical elements for maintenance and modificat ion.

PAGE 166

166 Figure A 1. Receiver block diagram for n channels. Figure A 2 Data capture board block diagram.

PAGE 167

167 Figure A 3. Range/control board block diagram. Figure A 4. Receiver components.

PAGE 168

168 Figure A 5. Assembled CATS system photo with timing electroni

PAGE 169

169 APPENDIX B SYSTEM TROUBLESHOOTI NG Range Offset Error In early versions of CATS processing software, the RGS value was multiplied by 0.5 ns. This was found to be in error and current versions of code have incorporated the correct value of 1.0 ns. Carrying out the old calculation for a nominal value of RGS =719 would give, (B 1) With the correct code, however, (B 2) These calculations show that ranges previously used to generate th ree dimensional point clouds would be off by a constant value of 107.85 53.925 = 53.925 meters. Small errors in coding techniques can produce large scale errors in range calculation. This leads into the ramifications of local oscillator variation and r esulting range bin stability.

PAGE 170

170 Oscillator Variation The accuracy of range event timing is dictated by the stability of a 2.00 GHz system clock, which is directly related to the RBN recorded in data files. Any significant variation about the central frequen cy requires modification of the aforementioned equations, and correspondingly modification of data processing software. This significance of this idea was illustrated during system construction. Concern about the reliability of this main clock was raised after the manufacture of a new control board, which consolidated the function of the external GPS box and eliminated the need for its use. A PLL Control chip (controlled by an FPGA) together with a 1.95 2.15GHz VCO (Z Comm V610ME04) generates the system c lock source of 2.00 GHz. In the process of finishing the new control board, receiver assembly, and system verification, the clock frequency was discovered to be about 1.90 GHz. The problem source was traced back to the FPGA design, which did not provide a n appropriate startup reset. A new control board fixed the problem. It was speculated that the system at some point in the past had the correct clock frequency, but there is no way of knowing at exactly what point in time the skew began. Since the proble m was traced back to original design files, however, it is assumed that the problem presides over most of the files collected with the old system. The old control board at this point in time has not been re integrated to get a measure of the presumed incor rect frequency. Estimating that the clock reached a minimum value of 1.90 GHz, the Range Bin length would be lengthened to 1 / 1.90 [GHz] = 0.5263 [ns]. Ran ges would then be calculated as,

PAGE 171

171 (B 3) Processing for previous data coll ections were altered to accommodate this value. The PLL chip used has since been replaced and verified at 2.00 GHz. It is instructive to note the importance of this central component; any drift over time is difficult to identify in point cloud data witho ut calibration ranging for known distances. Scan Synchronization Image Generation: Early S teps The first attempts to synchronize laser shot time tags to scanner index mark time tags from test data sets exposed some potential errors in both laser and scanne r timetag outputs. For example, the consistency of the observed scan periods using the recorded absolute GPS time was checked. To do this, scanner time tags from T000024.STD (a horizontal building scan at 2100 V) were imported into MATLAB. In this partic ular file there were 596 entries of scanner index times. This scan was run for 30 seconds, resulting in 30/596 or 0.05034 s per revolution; roughly 19.86 Hz. Observing the rotational period recorded from each scanner wedge, i.e., IndexTime2 IndexTime1, I ndexTime3 IndexTime2, etc. produces the plots in Fig. B 1. Notable in the bottom image are large deviations in what should be a relatively constant period of approximately 20 Hz. The calculated differential times occasionally spike to negative values. Lo oking at these spikes more closely in the actual data revealed a possible explanation: the microsecond counter not rolling over correctly. For example, consider the raw data from the scanner index times of file T000024.STD at entries 256 to 262, shown in Fig. B 2.

PAGE 172

172 Entry 257 should read 18:38:24.010854, instead of 18:38:23.1010854. There was a problem with the carry flag in the rollover calculation; the GPS seconds eventually were updated correctly from xx:xx:23 xx:xx:24, but not until entry 261. This c ounter error explains the large discontinuities in the observed scan period. Assuming that these spikes were due to this error in time tag production and not an error in scanner motion, the time outputs were filtered to what the maximum range of the scan p eriod should have been for this small interval. During testing it was observed that the laser pulse rate deviated from anywhere between 7800 and 8100 shots/second. Given that the scanner was correctly syncing 400 laser shots to one cycle of revolution, t his yields a potential data wind ow of 19.5 to 20.25 Hz, (B 4) The resulting filtered data contains 578 of the 597 total time tag outputs (97% of data retained), so it appears to be a reasonable window of data to look at given that there are around 20 errors in tag generation on rollovers. This filtered data is plotted in Fig. B 3. The images indicate that the scanner working as expected given that the laser fire rate was slightly wandering around 8 kHz. These observations about s canner index time tags were corroborated by analyzing several more datasets; the motion controller appeared to be working per the design specification. Unfortunately because of problems with shot time tag generation, it was not possible to observe the pos ition of the scanner in relation to successive periods of laser fire unless the laser was assumed to be perfectly slaved to the scanner via the motion control system. Other data collection runs were made without the benefit of time

PAGE 173

173 tag information, howeve r, so several previously developed techniques were employed to try and view the test ranges from that day regardless of accurate timing records. Point Cloud Results from a Building: P roblems The procedure for creating a point cloud from early CATS test dat a began by simply plotting the recorded ranges about a 400 shot period. In order to compensate for deviant non uniform behavior in the receiver channels (i.e. multiple returns in certain channels and zero returns in others) only the per shot range average was plotted, rather than plotting individual returns from all 96 independent channels. These averages were then staggered across 400 angular intervals a projected relative to a fixed origin. Fig. B 4 (left) shows a top view of a point cloud created from a single horizontal scan line. The right side of Fig. B 4 shows an image resulting from overlay of 400 shot intervals in a 10,000 shot dataset. The image of the bank building wall is expected to look similar to the image on the left with some deviation a bout the general structure. The right image, however, shows a cloud that cannot be attributed to excessive noise effects. Possible explanations for the garbled image included dropped synchronization pulses, incorrect prism starting locations, or data ove rflow issues. The importance of having a diagnostic tool to pinpoint the problem namely a correctly functioning timetag structure is evident here. Shot By Shot Analysis of Early D ata Methods for point cloud production resulting in the distorted image assumed that the data was distributed evenly according to a 400 shot period (i.e. shots N to N+399 comprise one scan period). If, however, there happened to be an error in laser sync pulse detection, there would be no correction to shift the period. The resulting data

PAGE 174

174 accumulated 400 shot data indicated that this was a plausible theory given that were problems with the repeatability of the period. Manual analysis of 4 00 shot segments identified a repeated set of features, shown in the Fig. B 5 ( A ). This image is a top view of a point cloud generated using average ranges from a single 400 shot window distributed across angular intervals from a fixed origin point. Fig. B 5(B) shows a zoom in of a pattern contained within a single 400 shot set. When average ranges from three sets of 400 shot increments (blue, red, and green points in the bottom image) are plotted on top of each other as a function of footprint number, th ey should depict the same geometric features being that they correspond to multiple repeated scans of the same building face. Fig. B 5 (C) illustrates how the geometric patterns for three 400 shot increments did not align, however. From this image, it is likely that the scanner rotation rate was changing significantly across these three data windows. It was later discovered that the electronics used to detect the synchronization pulses from the laser pulse train and the scanner index marks was not sampling at a in pulse detection, which led to such errors as the perceived wandering scan period. These types of hardware malfunctions could have been easily identified with a correctly operating timetag structure. Data processing operations such as these, however, solidify the importance of the fifth data field data time tagging and motivate its further development. Further data collection was put on hold until a new ti metag structure was implemented.

PAGE 175

175 Data Time Tagging Timetag Theory Re E valuation Analysis of scan data had shown that data processing was not able to reliably predict the shot number corresponding to the start of a scan period. It was necessary to re evalu ate the timetag structure. Each individual data collection resulted in the output of four timing files. They were as follows: A.STD : the time at which the A scan wedge index mark was detected. This is recorded as a coarse time (seconds) and a fine time (microseconds). B.STD : the time at which the B scan wedge index mark was detected. This is recorded as a coarse time (seconds) and a fine time (microseconds). Ranges.SPR : all relevant ranging information (channel #, range bin, etc.). This file also conta ins the output value of a 1 MHz microsecond counter which is reset every 100 Hz. This is the only time information individually recorded for every shot. I.STD : an internal system clock that is updated when a 100 Hz reset signal is fired. This file also c ontains information about coarse time read from a GPS 1 Hz signal. It is possible to calculate the timetag of each shot number through combination of a local 1 MHz counter value in the range file and the current time value in the I file. This involved mat ching the microsecond counter value in the Ranges.SPR file to the appropriate time in the I.STD file. The resulting shot time tag would be: (B 5) The position in the I.STD file is incremented when a 100 Hz reset in the Ranges.SPR file is detected. Shot timetags are matched up with either the A.STD or B.STD times to get the shot corresponding to the start of a scan period (hereon be called shot0). The shot time tag occurring closest to the first A index time is labeled as shot0.

PAGE 176

176 P roblems with Previous Timetag I mplementation A direct comparison to the timetag of the scanner index wedge times should then yield the shot number that corresponds to the start of a scan period. This did not turn out to be true. The point clouds derived from these predicted shot0 had the forescans and backscans mis aligned, shown in Fig. B 6. Only when shot0 was manually shifted (by anywhere from 20 to 350 shots) was the proper point cloud generated. This procedure is illustrated in Fig. B 7. A number o f issues identified in the individual timing files could have contributed to the difficulties of predicting shot0. A series of data showed that the 100 Hz micro second counter reset was often missed, causing gaps in the I.STD data as well as microsecond c ounter values in the range data file that exceeded 10000. The microsecond counter must be reset before rolling over to 10000 or an error results. There were also aberrations in the microsecond counter value where the increment from one shot to the next w as 2 3 times the normal value (implying dropped shots). Additionally, the number of entries in the I.STD data file did not match the number of resets detected from the 1 MHz counter in the range data file. Because of the issues plaguing the I file and de tection of the corresponding 100 Hz signal, this part of the timetag structure was eliminated. Shot timetag operations were moved onboard the FPGA, as described in the next section. Restructuring of Timetag G eneration The logic for timetag operations was stored onboard a Xilinx FPGA. Index mark timetags were produced as a result of four input pins tied to the following signals: (1) 1 Hz GPS signal, (2) a free running counter based on a local 1 MHz oscillator, (3) A index mark signal, and (4) B index mark signal. The leading edge of a pulse detected on pins

PAGE 177

177 3 or 4 generated an interrupt which latched in data from the stored GPS time (updated every second via a signal on the first pin) plus the current value of the free running counter. The files A.STD and B.STD were generated by this process. The I/O is expressed graphically in Fig. B 8 (A) Shot timetags were created by essentially the same method, but using a different fine precision counter. This method used a different 1 MHz local oscillator from insid e controller board itself. The microsecond counter was reset by a 100 Hz internal signal, found in I.STD The value of the counter was matched to the appropriate time i n the I.STD file to create the shot timetag as previously described. In reference to the problem of determining shot0, questions were raised about the synchronization between the two different local oscillators. It was theorized that the accuracy of the s ync might be related to the throughput capacity of the USB link versus the Ethernet link due to the control loops routing certain signals. High load conditions could produce an offset that would create a disparity producing a calculated shot0 offset. To solve this potential problem, laser fire pulse train was directly input into the FPGA used to tag the index mark signals, shown in Fig. B 8 (B) A new output file in the same format as the index mark files was then created to record the shot timetags. Thi s required modification of the FPGA programming and signal routing. Testing of Timetag R estructuring The new structure was tested in Fairfax, VA. Data collections with the sensor pointed at a building edge in the distance provided point cloud data with di stinct geometries, allowing for characterization of scanner behavior for consecutive 400 shot periods. Timetag files were analyzed to determine shot0 for each data set. Out of 11

PAGE 178

178 data sets, 9 registered the correct value of shot0, allowing for the scan p attern to be correctly rectified in the point cloud. During system testing to evaluate the new timetag structure, the scanner exhibited some movement errors visible to the eye. For example, during night operation of a static horizontal line scan, the obse rved output pattern began to slowly rotate such that the pattern eventually became vertical. These rotations did not occur throughout all conducted trials, so a second round of tests was conducted. The same large scale scanner error was observed, and fur ther post processing of timing data collected during aberrant scanner behavior showed the B wedge rotational period varying slowly from trial to trial. It was theorized that a physical malfunction of the B wedge was causing the system to produce inconsiste nt scan patterns. The system was b rought into a laboratory for observations of diagnostic signals on an oscilloscope. The width of the A cal encoder associated with the B wedge. Upon opening the sensor head for visual inspection of the encoders, a large crack was found on the B encoder plate. Without timetag processing, this issue may have not been explored for an unacceptable amount of t ime. Correctly operating timetags are a significant part of the CATS instrument design. Without accurate knowledge of when the start of a scan occurs, it would be difficult to align recorded return events to scanner behavior without distinct point cloud g eometry. Since intended targets for airborne operation are expected to not be ideally symmetric, it is essential that a correct shot0 is calculated. Timetag comparison also allows for the

PAGE 179

179 diagnosis of potential problems with individual system components. For example, laser double pulsing can be identified and corrected via post processing of event time histories.

PAGE 180

180 A B Figure B 1. Differential measurements of CATS scanner periods. A) Time difference between successive index scans. B) Scanner ro tation rate between successive scans. Possible errors in scanner timetags are indicated by large discontinuities. F igure B 2. Time data file associated with possible scanner timetag errors. A segment of the data is windowed to demonstrate rollover erro rs in timetag generation.

PAGE 181

181 A B Figure B 3. Filtered differential measurements of CATS scanner periods. A) Time difference between successive scans. B) Scanner rotation rate between successive scans. The large scale discontinuities are notably no t present, and high frequency characteristics are consistent with laser repetition rate fluctuation. A B Figure B 4. Point clouds created from preliminary CATS sensor data. A) Top view of a point cloud created from a single horizontal scan line of a ba nk building face approximately 300 m away. B) Top view of a point cloud created from a horizontal scan of bank building face, resulting from overlay of 400 shot intervals in a 10,000 shot dataset.

PAGE 182

182 A B C Figure B 5. Illustration of scan period wand ering. A) Top view of a point cloud generated using average ranges from a single 400 shot window distributed across angular intervals from a fixed origin point. Ranges from one scan period corresponding to 400 laser shots are plotted with constant angula r intervals. B) Zoom in of geometric pattern contained within a 400 shot window. C) Average ranges for three 400 shot windows are plotted on top of each other as a function of footprint number to illustrate wandering scan period.

PAGE 183

183 Figure B 6. Point clou d for horizontal profile scan of bank building. Fig ure was generated using shot0 calculated from GPS timetags. The two highlighted areas represent the same building structure and should overlap, but do not because of the error in shot0 calculation. A B C Figure B 7. Effect of sho0 variation in generated image. As shot0 is varied, the forescan (the image produced from the first 200 sho ts) and the backscan (the image produced from the las t 200 shots) begin to align and merge. The point clouds are gene rated using predicted shot0 ( A ), shot0 = 150 ( B ), and shot0 = 145 ( C ). 145 is the value of shot0 that results in the best point cloud.

PAGE 184

184 A B Figure B 8. FPGA block diagram showing redesign of CATS timetag structure to improve calculation of start of sc an periods. A) Original structure. B) New structure

PAGE 185

185 LIST OF REFERENCES [1] J. B. Campbell, Introduction to Remote Sensing 4th ed.: Taylor and Francis Ltd, 2007. [2] "The Imaging and Geospatial Information Society," Bethesda, MD: American Society for Ph otogrammetry and Remote Sensing, 2009. [3] D. Satale and M. Kulkarni, "LiDAR in Mapping," 2003. [4] H. Lee, K. Kampa, and K. C. Slatton, "Segmentation of ALSM point data and the prediction of subcanopy sunlight distribution," in Geoscience and Remote Sensi ng Symposium, 2005, IGARSS '05 Proceedings, 2005 IEEE International p.4, 2005. [5] B. J. Luzum, K. C. Slatton, and R. L. Shrestha, "Identification and analysis of airborne laser swath mapping data in a novel feature space," Geoscience and Remote Sensing L etters, IEEE, vol. 1, pp. 268 271, 2004. [6] B. J. Luzum, K. C. Slatton, and R. L. Shrestha, "Analysis of spatial and temporal stability of airborne laser swath mapping data in feature space," Geoscience and Remote Sensing, IEEE Transactions on, vol. 43, p p. 1403 1420, 2005. [7] N. Haala and C. Brenner, "Extraction of buildings and trees in urban environments ISPRS Journal of Photogrammetry and Remote Sensing, vol. 54, pp. 130 137, 1999. [8] Yu, Yang, and Chen, "Automatic feature extraction and stereo ima ge processing with genetic algorithms for LiDAR data," in International Conference on Computer Graphics, Imaging and Vision: New Trends 2005. [9] A. Alharthy and J. Bethel, "Heuristic Filtering and 3D Feature Extraction From LIDAR Data," in ISPRS Technica l Commission III Graz, Austria, 2002. [10] S. C. Popescu, R. H. Wynne, and J. A. Scrivani, "Fusion of Small Footprint Lidar and Multispectral Data to Estimate Plot Level Volume and Biomass in Deciduous and Pine Forest in Virginia, USA," Forest Science, vol. 50, 2004. [11] M. Lee and G. Tuell, "A Technique for Generating Bottom Reflectance Images from SHOALS Data," in U.S. Hydro 2003 Conference Biloxi, MI, 2003. [12] K. C. Slatton and W. E. Carter, "A Primer for Airborne LIDAR," 2008. [13] W. Wagner, A. Ullrich., T. Melzer, C. Briese, and K. Kraus, "From single pulse to full waveform airborne laser scanners: Potential and practical challenges," in International Society for Photogrammetry and Remote Sensing XXth Congress, Commission 3 Istanbul, Turkey, vo l. XXXV, part B3: pp. 6. July 2004.

PAGE 186

186 [14] R. Gutierrez, A. Neuenschwander, and M.M. Crawford, Development of Laser Waveform Digitization for Airborne LIDAR Topographic Mapping Instrumentation ," in Geoscience and Remote Sensing Symposium IGARSS, 2005. [1 5] J. J. Degnan, "Unified Approach to Photon Counting Microlaser Rangers, Transponders, and Altimeters," Surveys in Geophysics, vol. 22, pp. 431 447, 2001. [16] J. Degnan, "Photon Counting Multikilohertz Microlaser Altimeters for Airborne and Spaceborne To pographic Measurements," Journal of Geodynamics (Special Issue on Laser Altimetry), 2002. [17] U. Stilla and B. Jutzi, "Waveform analysis for small footprint pulsed laser systems," in: Shan J, Toth CK (Eds), Topographic Laser Ranging and Scanning: Principl es and Processing CRC Press, Boca Raton, pp. 215 234, 2008. [18] K. Albright, R. C. Smith, C. Ho, S. K. WIlson, J. Bradley, A. Bird, D. E. Casperson, M. Hindman, R. Whitaker, J. Theiler, R. Scarlett, and W. C. Priedhorsky, "RULLI: A Photon Counting Imager ," in International Advanced Studies Institute (IASI) Symposium on Detection and Analysis of Sub surface Objects and Phenomena Monterey, CA, 1998. [19] B. F. Aull, A. H. Loomis, D. J. Young, R. M. Heinrichs, B. J. Felton, P. J. Daniels, and D. J. Landers, "Geiger Mode Avalanche Photodiodes for Three dimensional Imaging," Lincoln Laboratory Journal, vol. 12, pp. 383 396, 2000. [20] J. J. Degnan, R. Machan, E. Leventhal, D. Lawrence, G. Jodor, and C. Field, "Inflight performance of a second generation, phot on counting, 3D imaging lidar," Laser Radar Technology and Applications XIII, Proceedings of SPIE vol. 6950, 2008. [21] T. Cossio, "Prediction of Topographic and Bathymetric Measurement Performance of Airborne Low SNR Lidar Systems" (Ph.D. dissertation, U niversity of Florida, 2009). [ 22 ] M. Flood, "LIDAR Activities and Research Priorities in the Commercial Sector," in Land Surface Mapping and Characterization Using Laser Altimetry Annapolis, MD, 2001. [ 23 ] J. Degnan, McGarry, Zagwodzki, Dabney, Geiger, Ch abot, Steggerda, Marzouk, and Chu, "Design and Performance of an Airborne MultiKilohertz Photon Counting, Microlaser Altimeter," in Land Surface Mapping and Characterization Using Laser Altimetry Annapolis, Maryland, 2001. [ 24 ] P. Cho, H. Anderson, R. Hatc h, and P. Ramaswami, "Real Time 3D Ladar Imaging," Lincoln Laboratory Journal, vol. 16, 2006. [ 25 ] O. Svelto, Principles of Lasers 4th ed. (trans. David Hanna), Springer, 1998.

PAGE 187

187 [ 26 ] A. Einstein, "On the quantum theory of radiation," Physikalische Zeitschr ift vol. 18, issue 121, 1917. [ 27 ] A. E. Seigman, Lasers University Science Books, 1986. [ 28 ] T. H. Maiman, "Stimulated Optical Radiation in Ruby," Nature vol. 187, issue 4736, pp. 493 494, 1960. [ 29 ] G. R. Gould, "The LASER, Light Amplification by Sti m ulated Emission of Radiation," The Ann Arbor Conference on Optical Pumping June 1959. [ 30 ] W. Koechner, Solid state laser engineering 2nd ed. Springer Verlag, 1988. [ 31 ] R. Paschotta, Q Factor," Encyclopedia of Laser Physics and Technology RP Photonics 2003. [ 32 ] F. J. McClung and R. W. Hellwarth, "Giant optical pulsations from ruby," Journal of Applied Physics vol. 33, issue 828, pp. 103 105, 1962. [ 33 Switched Lasers," IEEE Journal of Quantum Electronics vol. 31, no. 11, pp. 1890 1901, 1995. [ 34 ] R. Daly and S. D. Sims, "An Improved Method of Mechanical Q Switching Using Total Internal Reflection," Applied Optics vol. 3, issue 9, pp. 1063 1066, 1964. [ 35 ] A. Szabo and R. A. Stein, "Theory of Giant Pulsing by a Saturable Absorber," Journal of Applied Physics vol. 36, no. 5, May 1965. [ 36 ] J. J. Degnan, "Theory of the optimally coupled Q switched laser," IEEE Journal of Quantum Electronics vol. 25, pp. 214 220, 1989. [ 37 ] J. J. Zayhowski and P. L. Kelley, "Optimization of Q switched lasers," IEEE Journal of Quantum Electronics vol. 27, pp. 2220 2225, 1991. [ 38 ] P. Peterson, A. Gavrielides, M. P. Sharma, and T. Erneux, "Dynamics of Passively Q Switched Microchip Lasers," IEEE Journal of Quantum Electronics vol. 35, no. 8, pp. 1247 1256, August 1999. [ 39 ] J. J. Zayhowski and A. L. Wilson, "Pump Induced Bleaching of the Saturable Absorber in Short Pulse Nd:YAG / Cr 4+ :YAG Passively Q Switched Microchip Lasers," I EEE Journal of Quantum Electronics vol. 39, no. 12, pp. 1588 1593, December 2003. [ 40 related equipment Test methods for laser [ 41 ] Segelstein, Hale, Querry, Wielickza, and Weng, "Absorption Spec tra of Water Compendium," 1973, 1981, 1989.

PAGE 188

188 [ 42 ] P. B. Russel, T. J. Swissler, and M. P. McCormick, "Methodology for error analysis and simulation of lidar aerosol measurements," Applied Optics, vol. 18, pp. 3783 3797, 1979. [ 43 ] Z. Liu, W. Hunt, M. Vaugha n, C. Hostetler, M. McGill, K. Powell, D. Winkler, and Y. Hu, "Estimating random errors due to shot noise in backscatter lidar observations," Applied Optics, vol. 45, pp. 3783 3797, 2006. [ 44 ] T. Cossio, C. Slatton, W. Carter, K. Shrestha, and D. Harding, IEEE Transactions on Geosciences and Remote Sensing Vol. 47, No. 7, Part II, pp. 2298 2315, 2009. [ 45 ] Hamamatsu Photonics, "Characteristics and use of Si APD (Avalanche Photodi ode)," 2004. [ 46 ] Perkin Elmer, "Avalanche Photodiode: A User Guide. Understanding Avalanche Photodiodes for improving system performance," 2006. [ 47 ] G. C. Guenther, A. G. Cunningham, P. E. LaRocque, and D. J. Reid, "Meeting the Accuracy Challenge in Airb orne LIDAR Bathymetry," in EARSeL SIG Workshop on LIDAR 2000. [ 48 ] R. Nairn, "Royal Australian Navy Laser Airborne Depth Sounder: The first year of operations," in International Hydro Rev. Monaco, 1994. [ 49 ] B. Billard and P. J. Wilsen, "Sea surface and depth detection in the WRELADS airborne depth sounder," Applied Optics, pp. 2059 2066, 1986. [ 50 ] Hamamatsu Photonics, "Photon Counting Using Photomultiplier Tubes," 2001. [ 51 ] Hamamatsu Photonics, "Photomultiplier Tubes: Basics and Applications." vol. 2, 1999. [ 52 ] H. R. Hertz, "Ueber sehr schnelle electrische Schwingungen," Annalen der Physik, vol. 267, 1887. [53 ] I. Moreno, "Jones Matrix for Image Rotation Prisms," Applied Optics, vol. 43, pp. 3373 3381, 2004. [54 ] R. M. Marino and W. R. Davis, "Jigsaw: A Foliage Penetrating 3D Imaging Laser Radar System," Lincoln Laboratory Journal Vol. 15, No. 1, 2005. [55 ] B. Jutzi and U. Stilla, "Range determination with waveform recording laser systems using a Wiener Filter," ISPRS Journal of Photogrammetry & Remote Sensing 61 pp. 95 107, 2006.

PAGE 189

189 [56 ] K. Y. Shrestha, K. C. Slatton, W. E. Carter, T. K. Cossio, Performance Metrics for Single Photon Laser Ranging," IEEE Geoscience and Remote Sensing Letters vol. 7 issue 2 pp. 338 342 2010. [57 ] W. Carter, R. Shrest ha., C. Slatton, and W. Dietrich, "Research Quality Airborne Laser Swath Mapping: The Defining Factors," in A White paper issued by NCALM 2007. [58] Y. Shao, R. W. Silverman, S. R. Cherry, "Evaluation of Hamamatsu R5900 series PMTs for readout of high res olution scintillation arrays," Nuclear Instruments and Methods in Physics Research A 454 pp. 379 388 2000. [59 ] K. Francis, P. LaRocque, L. Gee, and M. Paton, "Hydrographic Lidar Processing Moves into the Next Dimension," Proc. of U.S. Hydro 2003. [60 ] R. W. Pope, P. Johnson, U. Lejdebrink, and J. W. Lilycrop, "Airborne Lidar Hydrography: Vision for Tomorrow," Proc. 2001. [61 International Lidar Mapping Forum Denver CO, March 3 5, 2010. [6 2 Oceans 2002 MTS / IEEE vol. 2 pp. 1194 1198 2002. [63 ] G. C. Guenther, P. E. LaRocque, and J. W. Lillycrop, "Multiple surface channels in SHOALS airborne lidar," Proc. SPIE vol. 2258, pp. 422 430 1994. [64] S. Pe'eri and W. Philpot, "Increasing the Existence of Very Shallow Water LIDAR Measurements Using the Red Channel Waveforms," IEEE Transactions on Geoscience and Remote Sensing vol. 45 no. 5 May 2007. [65] A. Colli n, P. Archambault, and B. Long, "Mapping the Shallow Water Seabed Habitat With the SHOALS," IEEE Transactions on Geoscience and Remote Sensing vol. 46 no. 10 October 2008. [66] r Appl. Opt. vol. 19 no. 6 pp. 863 870, Mar. 1980. [67] to correct airborne laser fluorosensor data for effects of water opt Appl. Opt. vol. 20 no. 17 pp. 2889 2906 Sep. 1981. [68] Remote Sens ing of Environment vol. 40 issue 2 May 1992.

PAGE 190

190 [69] G. C. Guenther, "Airborne Laser Hydrography: System Design and Performance Factors," NOAA Prof. Paper Ser. Rockville, MD: NOS, NOAA, March 1985. [70] G.C. Guenther, Digital Elevation Model Technologie s and Applications: The DEM Users Manual 2nd Edition, D. Maune, ed., American Society for Photogrammetry and Remote Sensing, Chapter 8: Airborne Lidar B athymetry, pp. 253 320 2007. [71] J. C. Brock, C. W. Wright, T. D. Clayton, and A. Nayegandhi, "LIDAR optical rugosity of coral reefs in Biscayne National Park, Florida," Coral Reefs vol. 23 no. 1 April 2004. [72] C. Hug, A. Ullrich, and A. Grimm, "Litemapper 5600 A Waveform Digitizing LIDAR Terrain and Vegetation Mapping System," International Archiv es of Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol. XXXVVI, part 8/W2 pp. 24 29 2005. [73] T. Cossio, C. Slatton, W. Carter, and K. Shrestha, "Predicting small target detection performance of low SNR airborne lidar," Journal of Se lected Topics in Applied Earth Observations and Remote Sensing (in press), 2010. [ 74 IEEE Trans. Nucl. Sci. NS 14, no. 1 pp. 443 448 1967. [75] R. Staubert, E. Bohm, K. Hein, K. photomultiplier Nuclear Instruments and Methods vol. 84 issue 2 April 1970. [76] J. Phys. D: App. Phys ., vol. 6 1973. [77] Applied Optics vol. 41 no. 24 August 2002. [78] voltage switch: reduction of afterpulse rates in a laser induced fluorescence Applied Optics vol. 45 no. 6 February 2006. [79] Nuclear Instruments and Methods in Physics Research pp. 121 126 2007. [80] H. Johnson and M. Graham, High Speed Digital Design A Handbook of Black Magic pp. 160 165 1993. [81] J. J. Degnan, "Photon Counting Lidars for Airborn e and Spaceborne Topographic Mapping", OSA 2010 Optics and Photonics Congress Applications of Lasers for Sensing and Free Space Communications (LS&C) Topical Meeting Paper LSWA4, San Diego, CA, January 31 February 4, 2010.

PAGE 191

191 BIOGRAPHICAL SKETCH Kristofer Y. Shrestha was born in Madison, Wisconsin and raised in Gainesville, Florida. He graduated from Eastside High School in 2000 with International Baccalaureate diploma and went on to study electrical engineering at the University of Florida Kristofer re ceived the Bachelor of Science degree in December 2004, and is currently a Ph.D. student in the department of Civil and Coastal Engineering. He was awarded the Alumni Fellowship in 2005, and now works as a research assistant as a member of the Adaptive Si gnal Processing Lab under Dr. K. Clint Slatton. He continues to search for improvements in applied science by exploring new technologies.