EPA-650/2-74-046-0
June 1974
Environmental Protection Technology Series
I
LU
a
04 6
-------
EPA-650/2-74-046-Q
DEVELOPMENT OF A GAS LASER SYSTEM
TO MEASURE TRACE GASES
BY LONG PATH
ABSORPTION TECHNIQUES:
VOLUME I - GAS LASER SYSTEM MODIFICATIONS
FOR OZONE MONITORING
FINAL REPORT
by
S. E. Craig, D. R. Morgan,
D. L. Roberts, and L. R. Snowman
General Electric
Electronic Systems Division
100 Plastics Avenue
Pittsfield, Massachusetts 01201
Contract No. 68-02-0757
ROAP No. 26ACX
Program Element No. 1AA010
EPA Project Officer: W.A.McClenny
Chemistry and Physics Laboratory
National Environmental Research Center
Research Triangle Park, North Carolina 27711
Prepared for
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, D.C. 20460
June 1974
-------
This report has been reviewed by the Environmental Protection Agency
and approved for publication. Approval does not signify that the
contents necessarily reflect the views and policies of the Agency,
nor does mention of trade names or commercial products constitute
endorsement or recommendation for use.
11
-------
TABLE OF CONTENTS
Page No.
A. INTRODUCTION ,.... 1
B. SIGNAL PROCESSOR 2
1. Introduction 2
2. Mini Computer Processor 5
Interface Subsystem 6
Central Processor Subsystem 6
Program Input and Data Logging Subsystem 8
3. Logic Block Diagram 9
4. Software 13
Operational Modes 13
Collection Mode 14
Data Reduction and Display Mode 15
Program Structure 16
5. Interface Unit 21
C. SPECTRAL STUDIES 25
1. Introduction 25
2. Spectral Data , 28
Introduction and Summary , 28
Ozone (O ) 34
3
Carbon Dioxide (CO ) 36
£t
Water Vapor (HO) 39
tt
Ethylene (C H ) 42
ti 4
Ammonia (NH ) 46
u
3. Wavelength Selection 46
III
-------
TABLE OF CONTENTS (Cont'd)
Page No.
4. Linear Weight Computation (MFIL) 51
5. Factor Analysis of Drift 55
D. SPATIAL FILTER 58
E. SEALED E3OTOPIC LASER DESIGN 64
1. Introduction „ 64
2. Laser Plasma Tube Design Considerations 67
3. Stability Considerations 72
4. Intracavlty Windows .72
5. Spectral Tuner Design 74
6. Sealed Laser Design Considerations 79
7. Proposed Sealed Laser Design 84
F. REFERENCES 87
APPENDICES
A. INTERFERENCES A-l
B. OPTIMUM LINEAR WEIGHTS B-l
C. WAVELENGTH SELECTION C-l
D. LWSP - LASER WAVELENGTH SELECTION PROGRAM D-l
Iv
-------
LIST OF ILLUSTRATIONS
Page No.
Figure 1 - Signal Waveforms 2
Figure 2 - Signal and Gating Waveforms 4
Figure 3 - Data Collection and Reduction System 7
Figure 4 - Logic Block Diagram 9
Figure 5 - Program Modification 12
Figure 6 - Sequence of Events 23
Figure 7 - Composite Spectral Absorption In 9.4ji CO Band 31
2
Figure 8 - Composite Differential Spectral Absorption In P-Branch 33
of 9.4ji CO Band
Figure 9 - Ozone Absorption Coefficient Data Comparison in 9.4/i CO Band 35
Lt
Figure 10 - CO Coefficient Data Comparison 37
A
Figure 11 - Comparison of the Continuum Absorption Coefficient at Three 41
Temperatures
Figure 12 - Representative Spectrum of HO Between 800 and 1250 cm 44
Lt
Figure 13 - Spectrum of 383 Meters of Room Air at 750 Torr on a Rainy Day 45
Figure 14 - Resultant Line Selection From a 25 Iteration LWSP Run 48
Figure 15 - CMFIL Output Listing 49
Figure 16 - Signal to Noise Ratio 50
Figure 17 - Spatial Filter Experiment Layout 60
Figure 18 - Amplitude and Power Distribution of the Gaussian Fundamental Mode ... 61
Figure 19 - Contours of Equal Power Density ! 63
Figure 20 - CO Laslng Lines 65
2
Figure 21 - Calculated CO Isotope Band Centers 66
Figure 22 - Radial Gain Profile 69
Figure 23 - Unsaturated Gain of a CO Laser at a J-40 Transition 70
Lt
-------
LIST OF ILLUSTRATIONS (Cont'd)
Page No.
Figure 24 - Layout of Laser Optical System 79
Figure 25 - Manifold for High Vacuum Fill Station 82
Figure 26 - Proposed Sealed Isotoplc Laser Design 85
APPENDICES
Figure A-l Detector Noise Versus Frequency A-"3
Figure A-2 Optical Scintillation Noise Versus Frequency A-3
Figure B-l Possible Transmission Patterns for Single Absorbers B-3
Figure B-2 Possible Transmission Patterns for Mixtures of Absorbers B-3
Figure B-3 Possible Patterns for the Natural Log of the Transmission B-5
Figure D-l Flow Chart of LWSP Program D-?
LIST OF TABLES
Table I Absorption Coefficients for 00°1 - 02°0 (9.4 Micron) CO9 Band 29
Table TI Absorption Coefficients for 00°1 - 10°0 (10.4 Micron) CO2 Band 30
Table HI Relative CO Absorption Coefficient Variation with Pressure 38
Li
and Temperature
Table IV Partial Tabulation of HO Vapor Line Data 43
It
Table V Absorption Coefficients and CL Variance of Atmospheric Species 52
Table VI Linear Weights and SNR's for Atmospheric Species 53
Table VH Cross Response of Linear Weights 54
Table VIII Eigenvector/Eigenvalue Analysis of Data Record 56
Table DC Comparison of Factors and Atmospheric Species 57
Table X Losses Versus Mirror Tilt for a Confocal Resonator 71
Table C-l CMFIL Input Data ^-7
vl
-------
A. INTRODUCTION
The Final Report of EPA Contract 68-02-0757, Development of a Gas Laser
System to Measure Trace Gases by Long Path Absorption Techniques consists of
two (2) volumes:
I Gas Laser System Modifications for Ozone Monitoring
n Field Evaluation of Gas Laser System for Ozone Monitoring
The work reported here stems from development activity begun in 1966 at
GE's Electronics Laboratory. Under this contract, a breadboard laser long path
monitor called ILAMS (Infrared Laser Atmospheric Monitoring System) was modified
to improve its sensitivity as indicated by previous field experience. System parameters
were selected to optimize system performance for ozone monitoring. A field
evaluation of the modified system was conducted.
In this volume of the Final Report, the work of ILAMS modification and the
selection of system parameters, lasing wavelengths and linear weights, is reported.
Significant system modifications included introduction of a spatial filter In the laser
output beam. In addition, a digital signal processor was Incorporated In the system,
replacing an analog device. The problems associated with Incorporating a sealed,
Isotoplc fill CO laser were studied and a laser design proposed. The selection of
£t
the four laser wavelengths were preceded by extensive spectral data collection to
determine the infrared absorption characteristics of target gases and expected Inter-
ferences. The selection process was facilitated by a previously developed computer
program. Similarly, the calculation of linear weights was done on an existing computer
program.
The change to digital processing was particularly important in the evolution of the
ILAMS design. It greatly enhanced the flexibility of the system, offering significant
advantages over the analog approach as discussed in the next section.
-------
B. SIGNAL PROCESSOR
1. INTRODUCTION
The spectrally scanning CO laser periodically steps through a number of
Lt
lasing modes. In each mode, the laser generates a quantity of energy at a par-
ticular wavelength. Figure la. Illustrates the energy output of the laser in time where
x (t) is the energy generated of wavelength,A and 7"is the period of one scan.
K K y
a. Laser Energy
Output
« *"
X](t)
X2(t)
X3(t)
Xi(t+r)
x2l
_T
b. Signal
Return
Energy
c. Reference
Preamp
Output
d. Return K T X, (t)
Signal s\ KB T2 X2 (t)
KBT3X3(t)
Signal
Preamp
Output
e. Normalized
Return ^
Signal
-d/c
KT2
T2 X 2 (t +T)
(K= KB/KA)
TIME
Figure 1. SIGNAL WAVEFORMS
-------
This energy is transmitted over a path in which a spectral attenuation takes
place that is dependent upon the nature of the medium. Figure Ib. illustrates the
energy after passing through a medium where T is the transmission of the medium
at wavelength,
'k.
The desired information consists of the transmissions T T T
1, 2, 3,
which characterize the medium. As shown in Figure la., the laser output at each
wavelength is generally not constant from wavelength to wavelength, or even from
period to period. This problem necessitates the use of a reference that directly
senses the output of Figure la. Conversion of the optical energy to an electrical
signal is accomplished with a suitable IR detector followed by a preamplifier. If
the return signal, Figure Id. is divided by the reference signal, Figure Ic., the
desired information is obtained, as in Figure le.
Information is extracted at each wavelength by synchrounous demodulation
which is performed by filtering or integrating gated portions of the signal. Typical
signal waveforms and gating signals are shown in Figure 2. A null gate is also
provided in order to derive a zero energy reference for dc restoration since the
signal is ac coupled through the preamp.
The normalization division process may take place before or after synchronous
demodulation as long as the reference signal-to-noise-ratio is large enough.
Division before synchronous demodulation requires a fast, stable divider and
necessitates state-of-the-art hardware for precision measurements. On the other
hand, division after synchronous demodulation requires a relatively slow separate
divider for each wavelength and twice as many synchronous demodulators.
-------
a.) Analytic Signal
b.) Reference Signal
c.) Null Gate
d.) Signal Gate
e.) Signal Gate #2 °
f.) Signal Gate #3 0
g.) Signal Gate ^4 0
JLJLJLJl
, Figure 2. SIGNAL AND GATING WAVEFORMS
-------
For large absorptions, logarithmic processing is desirable in order to
linearize the exponential transmission-concentration characteristic exhibited
by gases. Again, there is some choice as to where this transformation is applied.
If the signal-to-noise-ratio (SNR) is very high, then log processing can be accom-
plished before synchronous detection without any degradation in performance.
However, in general, it is preferable to do as much filtering as possible before
the log conversion in order to maintain reliable performance.
Linear weighted sums of the filtered log transmission measurements are then
derived in order to estimate the concentrations of particular atmospheric gases.
This technique is discussed in detail in Appendix B.
2. MINI-COMPUTER PROCESSOR
The mini-computer signal processor includes a general purpose (stored
program) mini-computer and appropriate interface electronics. The collection
and reduction of data is entirely under computer, i.e., program control; results
are displayed on simple displays incorporated in the equipment, and on an optional
teletype, which need not be used (or even be connected) during field or test range
exercise of the system.
The use of the stored program control and data reduction means:
• changes in system design, or variations in data reduction algorithms,
may be accommodated without alteration of the data collection or re-
duction hardware; only changes in the control program will be required.
• modification of signal processor parameters such as number of wave-
lengths (up to 8), gate locations, system response time, weighting
factors, etc., do not even require software changes, these parameters
are expediently entered by the teletype input.
• the precision of data processing may be made as accurate as desired;
similarly the impact of imprecise calculations may be assessed by direct
simulation for purposes of evaluating future low cost special purpose
instruments.
-------
• additional data, e.g., environmental conditions, time, date, signal
variability, laser parameters, etc., may be measured and recorded
without modification of or addition to the existing system hardware.
• the performance of one or more data processing and display systems
can be directly analyzed, e.g., data from several ozone monitors
could be crosscorretated and recorded.
The data collection and reduction system is sketched in Figure 3. A Digital
Equipment Corporation PDF 11/05 is used for the central processor. The data
collection and reduction equipment in Figure 3 consists of three major subsystems:
Interface Subsystem
The Interface Subsystem is composed of an 8 input analog signal multiplexer,
which is followed by a sample-and-hold amplifier and an analog-to-dlgital converter
at 10-bit precision. (The analysis path detector preamplifier output is connected to
one multiplexer input, the reference path to a second multiplexer input, the re-
maining 6 are available for sensing other voltage levels of interest). Additional
subsystem elements include an AGC attenuator, a wheel position counter and
demultiplexer/storage capability for analog data displays like the meters shown
in Figure 3.
Central Processor Subsystem
The central processor and its own control panel form this subsystem. Power
supplies for this equipment are contained within the CPU cabinet proper. The
central processor control panel ordinarily is disabled during operation.
-------
Analysis Path
Amplifier Output
Reference Path
Amplifier Output
Chopper Wheel
Encoder Timing ••
Signals
Interface
Multiplexer,
Sample and
Hold, A/D
AGC
Attenuators,
D/A Converter
Storage Registers
I
Teletype and Paper Tape Reader
(Teletype Not Required for Equipment
Operation; May Be Removed After
Control Program Has Been Loaded
Meters
Central Processor
POP 11/05
CPU
Program Input and
Data Logging
Data Display
Fiaure 3. DATA COLLECTION AND REDUCTION SYSTEM
-------
Program Input and Data Logging Subsystem
A Teletype Corporation ASR-33 teletype with appropriate interface circuits
constitutes this subsystem. As indicated, it plays two roles. First, it permits
entry (ordinarily via paper tape) of the control program. Second, it permits detailed
reporting of directly measured quantities, or derviced (computed) quantities.
The central processor is designed so that programs stored in its core memory
may be caused to remain intact during periods of no primary power. This option
is exercised, so that once a control program has been entered in the CPU, it
need not be reentered until there is a need to change it, regardless of whether the
CPU remains energized or not. The control program is designed so that it will
run properly regardless of whether the teletype is connected or not. Thus the
teletype unit is an optional data display device, not an essential component of the
system once the control program has been entered.
Details of the mini-computer signal processor are described in the following
sections.
3. LOGIC BLOCK DIAGRAM
A logic block diagram of the complete system is shown in Figure 4. The
analysis path or analytic detector is followed by a preamplifier which generates a
10 mV - 10 V signal level depending upon path attenuation. In this diagram a wide-
band analytic detector is shown. However for the modified ILAMS system, a thermistor
bolometer detector with equalization will be alternatively used.
-------
METE*
OUTPUTS
iUi
•ll
*. 3
*
i
J
CD
Figure 4. LOG.'C 8LOCK CTiaG
-------
A digitally-controlled amplifier, using a multiplying D/A converter (MDAC),
automatically controls this signal level to about 10 V in order to limit the dynamic
range requirements of the A/D converter. The mini-computer generates a feedback
signal to control the gain of this circuit and also performs the required filtering
necessary to realize a given loop response time which is usually set to about 1 second.
A single pole low-pass filter preceeds the A/D converter in order to limit the
bandwidth and thereby establish the required sampling rate. The cutoff frequency
is chosen large enough in order to limit the interpulse interference to an acceptable
level. An analysis of bandwidth, sampling time, and interpulse interference as
well as other effects appears in Reference 1.
For the ILAMS system, the reference detector is followed by a similar pre-
amplifier, equalizer, and filter. However, no gain control is necessary for this
channel.
An A/D converter and multiplexer alternately samples the analytic and reference
channels at a rate compatible with the signal bandwidth.
A digital shaft encoder which is attached to the laser wavelength selector
wheel is used with a counter in order to provide the synchronous gating information
shown in Figure 2. This arrangement provides an 8-bit number which determines
the wheel position at any instant in time. The lowest order bit is also used to control
the A/D converter and multiplexer so that the samples are synchronized with the
waveforms.
10
-------
Each data collection cycle consists of serially reading a signal and wheel
position sample into the DR11-A interface unit. The samples for each analytic
and reference wavelength and nulls are accumulated in separate registers according
to wheel position. This performs the synchronous demodulation of the signals.
Digital filtering of the data is done in two steps. In the first step, samples
are accumulated for l/4s at each analytic and reference wavelength. A null is
also accumulated for analytic and reference. Both accumulated nulls are
appropriately weighted and subtracted from the corresponding accumulated wave-
length samples. The null weight for each wavelength is determined so as to
account for the different number of samples accumulated in the wavelength and null
accumulators over one scan; i.e.
nbr of samples in wavelength accumulator in one scan
nbr of null samples in one scan
The second step consists of low pass filtering the l/4s sequences to obtain the
desired system response time. This approach allows a quasi-continuous readout of
all monitoring signals. In addition, the l/4s signal sequences can also be filtered
for AGC feedback with arbitrary response time.
The factor E shown in Figure 5 controls the system response time
T - (1 - In E)/4 seconds
where E is a positive number less than one.
11
-------
1 *
2 ^
u3 p
U4 »
5 *
U6 *
v2 *
^
^3 >
i. ^
v4 *
V5 ^
V6 P
DIVIDER
X Ui
xi
f
X]
y _
A2
X0
3
X4
X5
X4
XA
M -
1
NAA;
i
t
< f,.
IA
2")
LOW
PASS
FILTER
Yp(l-E)*Xi
+ E Y|
t
1
Yl
— T~+
Y2 .
3 M
Y4
YS k
Yrt k
LOG
i log Y;
FigureS. PROGRAM MODIFICATION
-------
A "geometric" integrator is used for AGC feedback filtering. The error signal
is derived in such a manner as to maintain each analytic level at less than or equal
to half of its respective reference level, on the average. For this arrangement,
the response time of the AGC loop is dependent upon the reference level. However,
an improved design which eliminates this dependency as well as simplifies the
computations is shown in Figure 5. This version will probably be adopted in any
future generation software.
After the final low-pass filtering, the log ratios of the 1/4 second analytic and
reference samples are computed. A balance register is provided at this point in
order to initially calibrate the instrument for a zero output in the absence (hopefully)
of any absorbing gas.
The normalized log ratios are then weighted, scaled and outputed to the
teletype unit as the final concentration estimates. A later modification also
provided meter readouts. A threshold feature Is also included In order to signal
exceptionally large concentration levels.
An existing analog signal simulator (Reference 2) is being used to generate arbitrary
absorption patterns and thereby checkout the entire signal processor in the laboratory.
4. SOFTWARE
Operational Modes
The system normally operates in either one of two modes: collection, and data
reduction and display.
13
-------
Collection Mode
In this mode repeated measurements of the analysis path signal and
reference path signals are made and recorded. Additional measurements
on the behavior of other system elements also may be made before, during or
immediately after the repeated transmission measurements.
The output of the analysis path detector amplifier is connected to one
of the multiplexer inputs indicated in Figure 3; the output of the reference
path amplifier is connected to a second multiplexer input.
The chopping wheel, which both modulates the laser beam output and
controls the laser wavelength, is driven by a motor which is monitored by
a digital shaft encoder. The net effect of this apparatus is to generate a binary
number which reports the position of the wheel and thus laser wavelength and
modulation. The wheel position is reported to 1 part in 256, i.e., to about 1.4
precision. One can think of a 256 tooth gear or index on the chopper wheel. The
wheel rotates about 3000 rpm, or 50 revolutions/second, so that the wheel
advances one "tooth" every 78 microseconds. (In reality, the binary counter
advances 1 count each 78 microseconds.) At each new count, the CPU is in-
terrupted and the present reading of the counter and one A/D converter reading
are transmitted to the computer. The counter reading is interpreted to deter-
mine the nature and proper disposition of the A/D converter reading; that dis-
position is made and the processor then waits for another interrupt. (Typical
data dispositions are the addition of a current reading to the previously
14
-------
accumulated sum for that wavelength and that path, or ignoring the reading
because it is the transition interval between "on" and "off".)
At preset intervals, e.g., once every N cycles of transmissivity
data collection, the normal data collection sequence may be altered to
measure any other parameter of the system which is connected to one of the
multiplexer ports. Each of these measurements may be recorded separately
if the total number is small (e.g., 100), or averaged. Each measurement
would require 50 to 100 ps to complete. Upon completion of the measure-
ment, data collection would continue as before, picking up again at the next
full revolution of the chopper wheel.
At the conclusion of a preset number (e.g., M) of transmission
observations, operation in the data collection mode is terminated.
The parameters M and N are stored in the processor at some
previous time, either by teletype, or by use of the numeric input controls
of the operator control subsystem.
Data Reduction and Display Mode
The second mode of operation is data reduction and display.
At the conclusion of a data collection sequence, signal measurements
have been accumulated on each of typically 4 wavelengths, at both the reference
and analysis paths. The transmission measurements consist of both "peak"
15
-------
measurements at each wavelength and "valley" measurements made during
the off period between open shutter intervals. In all, for 4 wavelengths, 10
quantities will be available, (4 reference paths, 4 analysis paths, reference
amplifier null and analysis amplifier null).
The data reduction program now operates on this data, reporting,
for example, relative transmission at each wavelength or estimated con-
centration of each of one or more gases.
The first two results might best be reported via the teletype; the
latter results might be reported both via the teletype and via the simpler
displays included in the operator control subsystem, (As indicated earlier,
the teletype system and software are configured so that the absence of
the instrument does not prevent successful computation and presentation of
the results intended for display on the operator control subsystem.)
It may be noted that the only hardware that would have to be changed
to accommodate the use of any reasonable number (say up to 10) different
wavelengths of laser radiation would be the chopper wheel. The "data dis-
position" table stored in the processor would have to be changed, but this is
merely a matter of reading a paper tape. A few additional memory locations
would be needed for the accumulation of transmission measurements, but the
22 accumulators required for 10 wavelengths are readily available.
Program Structure
Software of the ILAMS system is composed of one real time analysis program
and two non-real time support programs that modify tables used in the analysis
16
-------
program. The functions of each program are:
• Real-time Analysis Program - a) collect real time laser out-
puts and compute a series of measurements for printing on the
teletype, b) monitor and feedback of AGC.
• Parameter Entry Program - supplies and modifies, via the tele-
type keyboard, various weighting factors, integration constants,
time constants, scaling factor, etc.
• Channel Assignment Program - builds tables of values which are
used to assign individual data samples to specific input channels.
Instructions for Operating Real-time Analysis Program
These instructions apply when the analysis program and required
support programs are already loaded into memory. For instructions on the
loading of programs, see Chapters 2 and 6 of the "PDP-11 Paper Tape Soft-
ware Programming Handbook". Steps required to start program are:
1. Turn console power key to ON.
2. Turn teletype console switch to LINE position.
3. Depress ENABLE/HALT switch.
4. Enter start location 12400 in the Switch register.
8
5. Depress LOAD ADRS switch.
6. Lift ENABLE/HALT switch.
7. Depress START switch.
The above starting point will cause the AGC value to be initialized
to the minimum gain setting. An alternate restart location of 12414g will
leave the AGC at its last computed value.
17
-------
The program may be stopped at any time by depressing the
ENABLE/HALT switch.
Program Output
a) A set of eight measures are printed on the teletype for
approximately every 16th set of data gathered.
b) A new AGC value is computed and transmitted for every
set of data gathered.
c) Eight meter outputs are transmitted for every set of data gathered.
Program Input - Keyboard
The depression of selected keys on the teletype allows the user to
alter the flow of the analysis program.
Key Reaction
"B" Program performs the balancing function and then
continues with normal processing.
"U" Program performs an unbalancing function and then
continues with normal processing.
"T" Control is transferred to the channel assignment
program which prints a "K" when ready to receive
data. Restart at !2414 is necessary to return to
analysis program.
"P" Control is transferred to the parameter entry program
which prints a "$" when ready to receive data. Restart
at 124148 is necessary to return to analysis program.
Parameter Entry Program
Various parameters used by the program are stored in an 12 x 8 array
which may be altered by the parameter entry program. The array structure and
contents are as follows:
18
-------
Array Location Parameter
1.1
2,1
3,1
4,1
1,4
•2,4
•3,4
• 4,4
5,1 — 5,4
6,1 -—6,4
7,1 — 7,4
8,1 -—8,4
9,1
9,2
10,1
10,2
11,1—11,8
12,1 —12,8
Measure 1 weight
Measure 2 weights
Measure 3 weights
Measure 4 weights
Measure 5 weights
Measure 6 weights
Measure 7 weights
Measure 8 weights
E
1-E
F
AGC value
Scaling constants
(Multiply results by
2n where n = 1—15)
Meter offsets
Range
-1.0—1.0
-1.0—1.0
-1.0—1.0
-1.0 —1.0
-1.0 -*1.0
-1.0—1.0
-1.0 —1.0
-1.0—1.0
0 — < 1
0-^1
0 ^( 1
0—7777
8
1-15
Maximum
Input Units
± 250
± 250
± 250
± 250
± 250
± 250
± 250
fc 250
32767
32767
32767
4095
15
Input
Units
± 1/250
± 1/250
± 1/250
± 1/250
-t 1/250
± 1/250
:t 1/250
± 1/250
1/32768
1/32768
1/32768
1/4095
1
63 64
1/64
The parameter entry program may be started by depressing the "P" key
while the analysis program is running - or - starting at 14000 . The program
o
will print a "$" when ready to accept data.
Input format:
x, xbnnnnn(CR)
(CR) is a return
where
Examples:
b - blank
x, x - signifies the array coordinates
nnnnn - any value up to 5 digits - may be preceded by a
minus sign to signify a negative value
1, lb!25(CR) - Value of 1/2 entered for set 1 weight 1
3, 7b-10(CR) - Value of -1/2 5 entered for set 3 weight 3
5,lb32767(CR) - Value of l-2~15 entered for E
19
-------
Channel Assignment Program
This program is used to assign data samples (as designated by
wheel position count) to specific input channels. Channels are numbered:
01
02
03
04
-Analytic
09
10
11
12
> Reference
07 Analytic nulls 15 Reference nulls
The channel assignment program may be started by depressing the "T" key
while the analysis program is running - or - starting at 7000 . The program
will print a "K" when ready to accept data.
Input format:
xxbnn,nnn n(CR)
Examples
03bl7,19,123,7(CR)
04bl24(CR)
etc.
where
b - blank
xx - is a two digit channel number
n or nn or nnn - is the wheel position count
Notes: 1. 00 - as a channel entry will clear the channel
assignment table.
2. Wheel position counts are separated by commas
with the last one followed by a return.
8
3. Any positive octal value placed in location 7446
will be added to the wheel positions as they
are entered. Truncation at 256 is performed by
the program (256 = 0, 257= 1, etc.).
20
-------
5. INTERFACE UNIT
The Interface Unit conditions the analog signals from the reference and
analytic detectors and the signal from the shaft position encoder, converts them
to digital form, and makes them available in the proper sequence as inputs to the
computer. The capability for automatic gain control (AGC) is provided for the
analytic signal channel. In addition the Interface Unit Includes storage elements
and digital-to-analog converters for output indicators such as meters or chart
recorders. Up to eight values can be indicated or recorded simultaneously, and
the eight quantities to be displayed can be selected by the computer program.
The Interface Unit is assembled in an aluminum enclosure measuring
approximately 28 X 18.7 X 18.7 cm. It consists of a socket panel which holds
most of the analog and digital circuitry, a Data Acquisition System, and power
supplies. The power supplies allow the unit to operate directly from the 117 V
AC line. They also furnish power for the shaft position encoder and the preamplifier
for the analytic detector.
The Data Acquisition System is a modular unit made by Datel Systems, Inc.
(Model DAS-16-L10B). It has a 10-bit analog-to-digital converter, a sample-and-
hold circuit, and an analog multiplexer, connected to allow either random or
sequential digitizing of up to 8 analog voltages. The Interface Unit uses only two
of the eight inputs - one for the analytic signal and one for the reference signal.
These two inputs are sampled and digitized alternately.
21
-------
The functional elements of the Interface Unit are shown at the left side of
the signal processor block diagram, Figure 4. The detectors and their pre-
amplifiers, and the shaft position encoder, are not physically a part of the Inter-
face Unit.
The AGC circuit makes use of a multiplying D/A converter (MDAC) in the
feedback path of an operational amplifier. The closed-loop gain of this amplifier
is inversely proportional to the value of a 12-bit binary number that is applied to
the MDAC, except that the gain cannot exceed 39. Since the AGC control comes
from the mini-computer, the algorithm used and its response time are the choice
of the programmer.
The shaft position encoder is of the incremental type with a zero index. It
produces 1024 pulses per revolution on one output, and one pulse per revolution
on the other. The 1024 pulse output clocks a 10-bit binary counter, which under-
goes a complete cycle once per revolution. The index resets the counter to insure
that it always starts from exactly the same shaft position. In this way, the 10-bit
number furnished by the counter indicates the position of the chopper wheel with an
accuracy of about 0.1%.
Figure 6 shows the sequence of events and their relative timing during a con-
version cycle. A time equivalent to 1/256 of a wheel revolution, or 78 p.sec. at
3000 rpm, is allowed for this cycle. The first four lines in Figure 6 show the logic
states of the four least significant bits of the wheel position counter. The 128-count
line is used to switch the analog multiplexer in the Data Acquisition System, so that
22
-------
1024
r~L
512
256
l_
128
1_
CONV CMD
to
co
CONV BSY
REQ B
DATA TRANS
POSAAL
-78//SEC
Figure 6. SEQUENCE OF EVENTS
-------
the analytic channel and the reference channel are sampled and digitized in
alternate cycles. After this multiplexer switches, a period of about 9.8 ^scc.
is allowed for settling before an A/D conversion is initiated by the negative
transition of the "Convert Command11 line. The A/D converter's "busy" line
stays high while the conversion is taking place, and then goes low. At that point
the "Request B" line goes high to indicate to the computer that data is ready for
input. This data passes through a digital multiplexer which selects between the
wheel position counter and the A/D converter. Initially, this multiplexer,
controlled by the "Position/Value" line, is switched toward the counter.
When the computer has read in the wheel position, it sends a "Data
Transmitted" pulse to the Interface Unit. This pulse terminates the "Request
B" and changes the digital multiplexer to the A/D converter. The computer then
reads the digitized input value and returns another "Data Transmitted" pulse.
(The Interface Unit does not use this second pulse.)
Since the A/D conversion requires only about IS^tsec. , the time available
for the computer to read both the shaft position and signal values is about 65 ^scc.
Only the eight most significant bits of the wheel position counter are furnished to
the computer, since that is sufficient to give each conversion cycle a unique wheel
position value.
24
-------
C. SPECTRAL STUDIES
1. INTRODUCTION
ILAMS measures directly the transmission (T1§ T2> Tg, T4>
of the sample region between the transmitter and the retroreflector at four wave-
lengths. Assuming that the line width of the resonance absorption is broad
compared to the transmitted laser line width, and assuming also that there is no
saturation in the absorbing media, then, for a uniform concentration of the absorber
over the path, the transmission at each discrete wavelength is of the form, T^ =
exp (-A C L): where A is the absorption coefficient of absorber A at wavelength
v v m A m
m, C is the concentration of absorber A over the total optical path, and L is the
A
total optical path through the sample region. If the concentration is non-uniform
over the path, as is the more usual case, then CAL can be replaced by the integrated
concentration over the path or, more simply, let CA represent the average concentra-
tion over the path.
Typically, C has units of grams/liter or atmospheres of partial pressure, and
L is in centimeters. A is in units to make A C L dimensionless.
m m A
If a second absorber B with absorption coefficients B is introduced into the
region, the net transmission will be the product of the transmission due to each
absorber.
(1)
- A CAL+ B C^L
m A m B
T = e
m
If the natural log of the transmission at each wavelength is taken elec-
tronically, then
S = InT --(A CAL+B C_L) (2)
m m m A m B
and the resulting signals have two convenient properties:
25
-------
1) System response to any absorber is a linear function of the
quantity (concentration x path) present
2) System response to several absorbers is the sum of the
responses to the individual absorbers.
Therefore, it the system can be designed to give a zero response to spectrally
interfering absorbers, the system will respond only to the pollutant to be measured,
and the response will be proportional to the quantity present.
Speaking more generally, the above properties define a linear 4-dimensional
vector space. Each gas is represented by a vector in this space whose length is
proportional to the concentration. This formalization permits the application of
known mathematical and statistical techniques.
Using decision theory and multivariate statistical analysis, it can be shown
that the optimum signal processing involves the use of single or multiple linear
weights. Application of a single linear weight, W, means taking n linear sum of the
signals S S ... S to give a new signal, S = W^ + W^ + ... W^. The
quantity of absorbers present can be accurately determined by examining the magnitude
of such linear sums.
Techniques may also be applied for choosing linear weights to accurately
measure the quantity of a given pollutant in the presence of known interfering
spectral absorbers, random spectral absorbers, scintillation, and other "noises".
A definitive study on spectral absorption pattern detection and estimation techniques
using linear weights appears in Reference 3. A summary of the applicable results
are given in Appendix A and B.
26
-------
For the CO laser, there Is a large number of lines from which the
^
4 wavelengths can be selected. On the basis of both analytical and experimental
work, several basic conclusions about wavelength selection can be drawn. These
are:
1) The relative success of a group of wavelengths depends directly
on the measurement problem. The pollutants to be measured,
pollutants and absorbers to be ignored, expected quantities of
absorbers and the system noise levels, all affect the choice of
wavelengths.
2) For a given problem, there will be an optimum set of wavelengths.
Increasing the number of pollutants to be estimated or ignored will
tend to increase the optimum number of wavelengths to be used;
i.e., the more complex the environment, the more wavelengths
are necessary.
3) The finer the spectral structure of an absorber, the fewer number
of wavelengths areneeded to best measure the quantities of it
present. The laser system is not limited to detecting pollutants
with fine structure such as ammonia and ethylene. In fact, it does
remarkably well in detecting or rejecting absorbers with rather
smooth spectral characteristics.
4) On the basis of past experience, four wavelengths have done an
excellent job in handling spectral recognition problems in environ-
ments representative of the real world.
The primary target gas considered in this program was ozone. Secondary
targets, which are also, to some extent, interferences for the primary target as well
as between themselves, were ethylene (C H ) and ammonia (NG ). The general
& 4 O
topics of Interferences and Optimum Linear Weights are discussed, respectively, in
Appendices A and B. A systematical approach to the wavelength selection problem
using mathematical and computer techniques is presented in Appendix C.
27
-------
2. SPECTRAL DATA
Introduction and Summary
In this section, absorption coefficient spectral data from a variety of
sources is presented. This data was subsequently used for wavelength selection
and linear weight computation as described in the following sections of the report.
Much difficulty was experienced in reconciling laser measurements with
high resolution spectrometer measurements. In general, the adage of using only
laser absorption measurements for designing a laser system has been reaffirmed.
The molecular species considered for this contract were ozone (O ), carbon
O
dioxide (CO ), water vapor (HO), ethylene (C H ), and ammonia (NH ). Reliable
L L> £> 4 o
data was obtained for all of these gases except for water vapor. This absorber is
especially difficult to characterize because of the predominant self-broadening
that occurs.
Tables I and II give a tabulation of the summarized absorption coefficient data. *
A composite spectral plot that illustrates the data in the 9. 4 micron CO band is shown
in Figure 7. Two considerations must be taken into account when attempting to
interpret this plot. First, only the deviation from a horizontal line is significant since
neutral attenuation is rejected. The second consideration is that only deviations from
the mean concentration are significant because the instrument is initially balanced. For
example, if the CO concentration only varies 32 ppm from the nominal 320 ppm
£
shown in the plot, then the actual interference level would be reduced by an order of
magnitude from the nominal curve shown.
*Blank spaces between data are linearly interpolated; blank
spaces outside of the data range are set to zero.
28
-------
J
Value
P40
P38
P36
P34
P32
P30
P28
P26
P24
P22
P20
P18
P16
P14
PI2
PIO
P8
P6
P4
R4
R6
R8
RIO
R12
R14
R16
R18
R20
R22
R24
R26
R28
R30
R32
R34
R36
R38
R40
Wavelength
Microns
9.733474
9.713998
9.684831
9.675971
9.657416
9.639166
9.621219
9.603573
9.586227
9.569179
9.552428
9.535972
9.519808
9.503937
9.488354
9.473060
9.458052
9.443325
9.428886
9.367339
9.354414
9.341758
9.329370
9.317246
9.305386
9.293786
9.282444
9.271358
9.260526
9.249946
9.239615
9.229530
9.219690
9.210092
9.200733
9.191612
9.182725
9.174070
Wave
Number
CM"'
1027.382
1029.442
1031.478
1033.488
1035.474
1037.434
1039.369
1041.279
1043.163
1045.022
1046.854
1048.661
1050.441
1052.196
1053.924
1055.625
1057.300
1058.949
1060.571
1067.539
1069.014
1070.4efficient
H20*
1.430E-04
The H2O absorption coefficient is proportional to H2O
partial pressure (see text). The values listed in the table
are for 100% relative humidity at 23 degrees C (73 degrees F)
which is equivalent to 19.8 torr.
** See text.
1.390E-04
1.360E-04
ATM'1 CM'1
C2H4
400E 00
900E-01
400E 00
600E-01
500E-01
600E-01
9.000E-01
4.400E-OI
8.700E-01
2.200E-01
1.500E-01
1.800E-01
4.600E-01
2.400E-01
Table I Absorption Coefficients for 00°1 - 02°0 (9.4 micron) CO2 Band
29
NH3
3.400E-01
9.000E-02
6.000E-02
l.OOOE-01
4.700E-01
800E-01
580E 00
600E-01
200E-01
3.400E-01
I.030E 00
3.900E-01
3.600E-01
9.000E-02
8.000E-02
3.100E-01
6.700E-01
1.900E-01
1.100E-01
2.000E-01
1.200E-01
5.000E-02
2.600E-01
3.700E-01
7.200E-01
**1.050E 01
1.400E-01
3.000E-02
.7.800E-02
7.000E-02
1.200E-01
4.200E-01
-------
J
Value
P40
P38
P36
P34
P32
P30
P28
P26
P24
P22
P20
P18
P16
P14
P12
PIO
P8
P6
P4
R4
R6
R8
RIO
R12
RI4
R16
R18
R20
R22
R24
R26
R28
R30
R32
R34
R36
R38
R40
Wavelength
Microns
10.8)1105
10.787380
10.764052
10.741113
10.718560
10.686386
10.674586
10.653156
10.632090
10.611385
10.591035
10.571037
10.551387
10.532080
10.513114
10.494484
10.476187
10.458220
10.440579
10.365168
10.349277
10.333696
10.318424
10.303458
10.288797
10.274438
10.260381
10.246625
10.233167
10.220006
10.207142
10.194574
10.182301
10.170323
10.158637
10.147246
10.136146
10.125340
Wove
Number
CM-1
924.975
927.009
929.018
931.002
932.961
934.895
936.804
938.689
940.549
942.384
944.195
945.981
947.743
949.480
951.193
952.882
954.546
956.186
957.801
964.770
966.251
967.708
969.140
970.548
971.931
973.289
974.623
975.931
977.215
978.473
979.706
980.914
982.096
983.253
984.384
985.489
986.568
987.621
Absorption Coefficient - ATM"' CM"1
co2
5.3981-04
7.037E-04
8.651E-04
1.099E-03
1.303E-03
1.539E-03
1.834E-03
2.052E-03
278E-03
500E-03
731E-Q3
756E-03
786E-03
736E-03
595E-03
335E-03
1.984E-03
1.580E-03
1.091E-03
2.725E-03
1.870E-03
2.301E-03
608E-03
867E-03
978E-03
050E-03
2.
2.
2.
2.
2.
2.
2.
2.
3.019E-03
2.870E-03
2.708E-03
2.449E-03
2.211E-03
1.955E-03
1.667E-03
1.412E-03
1.157E-03
9.449E-04
7.581E-04
5.879E-04
* The H2O absorption coefficient is proportional to
partial pressure (see text). The values listed in the table
are for 100% relative humidity at 23 degrees C (73 degrees F)
which is equivalent to 19.8 torr.
H20*
C2H4
1.800E-04
8.000E 00
2.400E 00
1.200E 00
1.800E-00
3.600E 00
5.000E 00
3.200E 01
0.
Table II Absorption Coefficients for 00°1 - 10°0 (10.4 micron) CO, Band
30
-------
0.15
0.10
u
«
n
h-
C
O3 - McClenny, EPA
CO2 -SRI
C2H4 - McClenny, EPA
H2O - Burch, Long
NH3 - McClenny, EPA
320 ppm
C02
H2O 50% RH @ 23°C
36 32 28 24
20
16 12 RIO P1012 16
00° 1 -02°0 (9.4p Band)
Figure 7. Composite Spectral Absorption in 9.4/j
31
20 24
CO2 Band
28
-------
A more refined plot showing only the deviation about the mean for water vapor and
carbon dioxide in the P branch is shown in Figure 8. As can be seen, water vapor
is a serious interferent and its characterization is critical in detecting trace con-
centrations of ozone.
Observed stability of the monitoring system that uses linear weights derived
from this spectral data has been poor under some conditions which are suspected
to be attributable to relative humidity variations. Other sources of data noted in the
following discussion of water vapor absorption coefficients also suggest significant
departure from the assumed linear continuum as well as a bizarre interrelationship
with aerosols. In Section C.5 a factor analysis of the system noise also points the
finger of suspicion in this direction.
In particular, it is now believed that the R14 line that was used Is sitting
almost on top of a water vapor absorption line. At the writing of this report the
number 1 system wavelength has been moved from the R14 line to the R16 line and
operation under this modification is being evaluated.
For all of the above reasons, a water vapor laser measurements program is
recommended for future air pollution monitoring systems that utilize laser absorption
measurements in this spectral region. A research facility such as the 980 m cell
at Ohio State University Electroscience Lab now operated by Dr. R. K. Long and
associates would be ideal for this purpose. This particular facility is the one which
was used in the past for obtaining the only reliable water vapor absorption laser
measurements known to exist (Reference 4).
Some spectral instability can be attributable to variation in the CO absorption
£t
coefficient with temperature. For the lines and linear weights that were used with the
present system, a 40 degree F change in temperature causes an apparent change in
ozone level of about 2. 3 ppb. This level of unstablllty is not a limiting factor for the
present system.
32
-------
0.015
0.010
u
a
n
H
T
0
1 j— 1
P10 P12 P14 P16 P18 P20 P22 P24 P26 P28 P30 P32
9.4fj Band
Figure 8. Composite Differential Spectral Absorption in P-Branch of 9.4yu CO2 Band
33
-------
However, if desired, this variation could be eliminated by changing linear weights
as a function of ambient temperature or by considering the first order temperature
variation as an additional interferent to be discriminated against.
A large absorption coefficient for ammonia was discovered in the 9.4 micron
CO band by direct laser absorption measurements performed by W.A. McClenny.
L*
This behavior has been previously unnoticed with spectrometer measurements which
have obscured the true peaks because of resolution limits. In fact, this absorption
coefficient is larger than the one in the 10.5 micron band which was heretofore thought
to be the largest available.
Ozone (O )
O
Ozone absorption coefficient data was obtained from direct laser measurements
(References 5 and 6) high resolution spectrometer (Reference 7) and theoretical
computations (Reference 8) using line strength data (Reference 9). Figure 9 illustrates
a comparison of the data which demonstrates very good consistency. In all cases,
the measurements were made at standard temperature and pressure. * Because of the
inherent accuracy of the laser measurements and good repeatability, McClenny's
measurements were used as a basis for the ozone data and were transcribed onto our
computer spectral library tape as listed in Tables I and H.
*It was subsequently determined that all of McClenny's measurements (References 5
and 12) were converted to 0 degrees C reference temperature by using the T 1/2
proportionality law for the line half-width. It is noted that this scaling law is only
appropriate at a line center. Moreover, the temperature dependence of the line intensity
must also be taken into account (Ref. 9 (eq. 3) and (Ref. n (ecu 77)). In anv case, the
difference is probably small compared to the spread of the other data. In addition, all
of the wavelength selection and linear weight algorithms are more dependent on the
relative ratios of absorption coefficients between wavelengths rather than the absolute
magnitude.
34
-------
0)
E
15
00° 1 -02°0 Band
Figure 9. Ozone Absorption Coefficient Data Comparison in 9.4}} CC>2 Band
35
-------
Carbon Dioxide (CO )
-------
5x10
co
4x10
3x10
o
1
"o
2x10
1x10
R14 R16 R18 R20 P12 PU P16 P18 P20 P22 P24 P26 P28 P30
9.4jj Band
Figure 10. CC>2 Absorption Coefficient Data Comparison
-------
CO
oo
Temperature
Degrees F
70
70
70
50
70
90
Pressure
Bars
1.003 !(-!%)
1.0132(nom)
1.0233(+1%)
1.0132
1.0132
1.0132
Relative Absorption Coefficient
Wavelength - microns
9.305386
(RH)
1.106621
1.106 650
1.106678
1.106080
1.106650
1 .107 239
9.503937
(PI 4)
1 .000 000
1 .000 000
1.000000
1 .000 000
1 .000 000
1 .000 000
9.586227
(P24)
0.841 477
0.841 471
0.841 464
0.816962
0.841 471
0.864891
10.532080
(PI 4)
0.747 132
0.747 138
0.747 143
0.731 860
0.747 138
0.761 726
10.674586
(P28)
0.500629
0.500720
0.500809
0.468 236
0.500720
0.532764
10.718560
(P32)
0.355 691
0.355 710
0.355 729
0.327 494
0.355 710
0.384 019
Table III. Relative CO^ Absorption Coefficient Variation with Pressure and Temperature
-------
Water Vapor (HO)
Lt
In the 8-14 micron water vapor window, the absorption is a relatively smooth
function of wavelength. This continuum is believed to be mostly due to the extreme
wings of strong collison-broadened absorption lines centered more than 10-20 cm"1
away (Reference 13). However, the effects of pressure induced absorption resulting
from forbidden transitions of unperturbed molecules and the possible existence of the
water dimmer (^0:^0) have also been suggested (References 9 and 14).
The absorption coefficient due to the continuum can be written as (References
9 and 13).
R = CsP + CbPb (3)
where Cg is the self-broadening coefficient, Cb is the foreign gas broadening co-
efficient, p is the partial pressure of the species, and pb is the foreign gas partial
pressure. As can be seen from (3), for small p, the absorption coefficient is nearly
constant which is usually the case for most atmospheric gases. On the other hand,
for large p, the first term dominates and the absorption coefficient is proportional to
p. In this case, the log transmission is proportional to -p2. It has been experimentally
demonstrated that the latter situation occurs for atmospheric water vapor absorption
under normal relative humidity conditions (Reference 4).
In the SRI computer study, (Reference 8) the water vapor absorption due to the
superposition of many lines was computed. However, the effects of self-broadening,
which is predominant for water vapor, were neglected and so these results were not useful.
39
-------
Figure 11 shows the spectral dependence of C for water vapor continuum
absorption for three temperatures that has been experimentally determined by
Burcli (Reference 1-1). n'he ratio of the coefficients at the f).rir,:> micron and lO.ri!'!
micron C()i? P20 lines is seen to be about O.S which is in good agreement \\iih
McCoy, (Reference 4). After a suitable conversion of units (2. C9 x 10 ' mnl/cm"
] atm - 7(10 tori1), the absolute value of the 10.591 micron coefficient indicated in
-I -2 -1
Figure 11 is about ll.fi .x ]0 torr Ian at 23 degrees C which is at. variance
-4 -2 -1
with the value of S. :;<) x 10 torr km reported by McCoy. In a recent report
by Trusty, (Reference 11) spcctrophone measurements of water vapor absorption
nt this .same line were noted to be about SO1/? higher than McCoy's measurements,
thus giving more credibility to Burch's measurements.
The C value for nitrogen has been measured to be C 0. 00.") C at mom
n b s
temp(jrauire by McCoy (Reference 4) and Burch (Reference 9) considers this to be
a reliable measurement.
The data in Figure 11 was used as a basis for the water vapor absorption
coefficients. The values listed in Tables I and n were computed from (:j) and
Figure 11 for a relative humidity of 100^ and at a temperature of 2.'J degrees C (?:}
degrees F), i.e. , for a 19.8 torr partial pressure of water vapor. For other partial
pressures, the absolute value of the coefficients will be scaled accordingly, however,
the relative absorption coefficient pattern will be invariant.
The wavelength selection and linear weight computations described in Sections
C.3 and C.4 were based upon the continuum values derived from Figure 11 (originally
obtained from Reference 9)duo to the unavailability of any other data at that time.
Since then, additional data concerning the line structure has been uncovered.
40
-------
10
-21
E
"o
E in'22
o ID
o
10
-23
1200
Ref. 11
296 K
K
1000
Wavenumber (cm"')
800
600
10
Wavelength (microns)
14
Figure 1 1 . Comparison of the continuum absorption coefficient at three temperatures. (From Ref. 14)
-------
Table IV lists some of the more significant water vapor line data that were
used in the SRI study (Reference 8). This data was obtained from a magnetic tape
described in Reference 9. Figures 12 and 13 show measured water vapor absorption
spectra, from Burch (Reference 13) and Hanst (Reference 7) respectively, winch
are in remarkable agreement with Table IV. It is especially noted that the R14
(1074. G5 cm"1) CO laser line that was used in the present monitoring system is
sitting almost directly on top of the 1074.430 cm"1 water vapor line. This could
explain some of the unstability that was noted.
In Reference 11, CO laser measurements with a spectrophone have demon-
strated a 2:1 variation in the water vapor absorption coefficient from PS (954.54 cm )
to P36 (929.02 cm"1). These measurements are plotted in Figure 11 and show a
significant deviation from Burch's continuum.
Another source of anomoly has been suggested by Carlon (Reference 15) which
attributes additional broadening to inelastic collisions involving water aerosols.
This effect has been experimentally verified and has been used to reconcile apparent
inequalities of absorptance and emittance in atmospheric field measurements.
In view of all of the uncertainties mentioned above, a water vapor laser
measurements program would be highly desirable for future work in this area in
order to pin down the true behavior of water vapor.
Ethylene (C)
The absorption coefficients for ethylene listed in Tables I and II were obtained
by direct laser absorption measurements (Reference 12) and represent the most
reliable information to date. The maximum absorption coefficient of 32 at the PI4
(10.532 micron) line agrees favorably with the value of 36 measured by Hanst (Reference.7).
42
-------
Wavenumber
cm
922.142
924.988
948.260
976.012
1066.200
1074.430
1091 .240
Linestrength
1 .690E-23
9.730E-24
1 .980E-23
7.700E-24
3.730E-23
1 .350E-23
2.040E-23
Halfwidth
cm'1
0.048
0.050
0.038
0.040
0.047
0.050
0..056
Table IV. Partial Tabulation of H20 Vapor Line Data (From Ref. 8)
43
-------
100
1000
Wavenumber (cm )
P14
Wavelength (microns)
Figure 12. Representative spectrum of H?O between 800 and 1250 cm . The sample is pure
0*7 O
at 14.2 torr; path length is 1185 meters; u = 5.48 x I0'z molecules/err/; temperature is 296K.
The spectrum contains a few lines due to a trace of ammonia in the sample; some of the stronger
lines are indicated. The ammonia lines can be accounted for by comparing the spectrum with
one of pure ammonia. (From Reference 13).
-------
1.1
c
o
VI
Oi
900
P14
1000
Wavenumber - CM"
P24 P14
P14
11GX
Figure 13. Spectrum of 383 meters of room air at 750 torr on a rainy day (March. 31, 1972)
in Research Triangle Park, N. C. Recorded on the digital FTS system at 0.5 cm~'
resolution using a Hg-Cd-Tel detector.. .(From Hanst, Ref. 7).
-------
Ammonia (NH )
o
The absorption coefficients listed in Tables I and II were also obtained by
direct laser absorption measurements (Reference 12). The P32 (10.7186 micron)
absorption coefficient of 8.0 is lower than the value of 32 previously measured by
Hanst (Reference 15). However, at that time, the wavelength of the P32 line was
not accurately known and the absorption coefficient was measured at 10. 717 microns.
In addition, the resolution of the spectrometer used by Hanst was not fine enough to
adequately measure the fine line structure of ammonia.
The surprising feature of the data is the large absorption coefficient (2.58)
in the 9.4 micron band at the P20 (9.5524 micron) line. A narrow peak has been
previously noted with high resolution spectrometers at that wavelength, however,
the magnitude of the fine structure was not evident.
It was recently determined that the ammonia coefficients shown in the R-branch
of Table I are displaced one J value too high, e. g., the R12 value listed is really the
RIO value. In addition, remeasurement of the true RIG coefficient has given a value of
14. 0 rather than the value of 10.5 indicated by the table. However, for consistency,
the values listed in Table I are shown as they were used in the subsequent wavelength
selection and linear weight computations.
3. WAVELENGTH SELECTION
In Appendix C, the philosophy and methodology of wavelength selection is
delineated as a two step process. First a preliminary wavelength selection is used
to reduce the number of potential lines (74) to a manageable number (say 10) by using
a computer program called LWSP. This program eliminates wavelengths of low
information by an iteration process of adjusting the power allocation. The second step
4fi
-------
in the selection process involves a combinatorial evaluation of the reduced set by
which all combinations are ranked in accordance with their performance in measuring
ozone. An existing program (MFIL) was modified to perform this operation and is
called CMFIL.
A 25 iteration LWSP run resulted in the line selection illustrated in Figure 14.
The solid lines in this figure represent the linear weights applied to each wavelength
and their length is indicative of the relative importance of each line. The X's
designate normalized ozone absorption coefficients and the +'s designate the average
interferent noise level.
The top 9 wavelengths from the LWSP output were combined with the PI4
(10. 5321 micron) ethylene line to form the basis set for CMFIL. Combinations of
this set were then evaluated and ranked using CMFIL and the output listing is shown
in Figure 15. As expected, the P12 and P14 (5 and 6) lines, which correspond to the
peak ozone absorption, appear in all of the highest rankings. The RIG, R14, and P24
(3, 4, 8) lines also predominately appear in all of the highest rankings and are therefore
indicated as good reference lines.
The combination 4, G, 8, 10 (R14, P14, P24, P14) was selected as the com-
bination that gave the highest signal-to-noise-ratio (SNR) while retaining the ethylene
line. In retrospect, this may have been a bad initial choice since the R14 line has been
shown in Section C 2 (water vapor) to fall right on top of a water vapor absorption line.
Using the RIG line instead of the R14 line would result in the 3, G, 8, 10 combination
which, as can be seen from Figure 15, is only slightly inferior to the 4, 6, 8, 10
combination.
The effect of the number of lines used to detect ozone has also been determined
and is illustrated in Figure 16. As can be seen, a 3 wavelength system (which we
essentially have at the present due to the retention of the 10. 5321 micron ethylene
line) provides near optimum performance with a minimum of complexity. A two-
wavelength ozone system results in about 1/2 the sensitivity.
47
-------
o
CN
TARGET 1
Time 10.41
Date 091173
o
CO
0)
o)
oo
o_
o
8
o 9
OJ
o
o
GO
, ^V^S. sr\ ^p s\ s{
-h -hT ^ -h
•f 4-
J- ' + +
t -1-
o
Z
_o>
'aJ
0)
_c
O)
9.12
9.20
9.28
9.36
9.44
Wavelength
9.52
9.60
9.68
9.76
Figure 14. Resultant Line Selection From a 25 Iteration LWSP Run
-------
JF lfl HAVEL6NOTHS, 4 «T * fll!',
TMROOBH
9.219690 9.282444 ,,2937,4 ,.„,„* ,.4883,4 9.503937 ,]»2^ 9™W ,!3?JM !,.,£,',.
tOS4;635l791077. 302490 lfl75;987742 K>7«.J4S4J4 (0*3.923584 f052.1»JJ2» 1046.894263 1043.1«3269 1037. 434198 949.4«0057
3-.;?. 76551 3568
349/97922 4968
3\2.1?564 346-8
3ie/8St>27 4689
315/01049 4679
0*t3i9.Tv72 1 4 £ 9
3,15/91113 4 6"' 6 10
315/U2-i6 2469
304/5H27 3 6 7 J
3,13,2?~>""5 245s
3-.13.VS429 2363
"5ni.3S2U5 2399
?">7. 517.16 1368
2J*. 7*318 3. 6 8 9
2'3.'B53lH 3 6 8 10
2^6i?4'A8 3 3 7 9
276/711*6 3456
2'5/?'333 1 6 8 9
374/450J8 4967
P74 'AT'j^H *i & & A
C'».OJ^JO J 0 9 9
274/20047 3969
273/74134 1968
272/5.1618 3589
2*9/04572 5 6 8 10
2J7/67309 1298
247/290A4 1396
245.-4r.-49 3469
244.64051 4969
263.90549 2 9 68
263/3'037 5 0 7 8
5<^5'isnii \ A • a
r.ss , 130 6/34244 i 6 -6 10
255/97542 4589
255/81315 2456
255/1S032 3 4 6 10
254/77141 1457
254:37717 2 9~ 8 9
253/81235 - 1 4 6 9
253/719-J9 1 4 6 10
253/57040 1357
2~"3/2~l642 2346
252/59559 1467
252/54690 3457
Z52/347->4 2 J 8 10
252/09652 1 2 6 8
251.99593 4 5 6 10
25l/372<»8 3567
25l/BSfl02 1679
251/27319 I 9 7 g
"250/61997 5 78 9
2"~4.'2iU5 3467
253/17777 3 5 6 10
3I9/5J192 2689
;> ' •» / 2 ? 4 H 2 1 2 4 4
21S,"34ini 1 2 4 5
2 * S / 73335 4 6 7 9
24*177027 6789
:»3/2'075 4 6 9 10
3 ?,97->00 4 8 9 10
2 V95448 2469
7. 7///047 1369
? '/4J409 1 3 6 10
V J.-S1310 2467
""T.47715 I 4 5 9
"• « / 4 G 5 7 J 2 3 4 J
243."0'H47 2 3 56
243/7U70 5 7 8 10
243/61S37 1239
243/35941 1236
242/8S4U4 4979
242/31198 2 6 8 10
242r?l8*6 4 7 8 10
2«1/S">637 2369
241.-.-1773 3579
24l/3i2.^l 1 3 6 7
2iC/"«U3 2579
24Q/JC-1I3 2359
24Q/Q7936 1 9 8 9
239T9J345 4 6 7 10
2tv:43«?l 4 5 7 10
2-»9'13-'l<16 2457
2"!9:oi':J7 2367
238/31597 2 4 6 10
238/57391 4 5 8 1"0
231.'03-t73 3679
237;7»?03 1 4 5 8
237;nO*5 3 6 9 10
233/41«-Ji 2 4 5 10
232/:>?74i 3 5 7 10
2'"2/7?'J42 2157
230/21 >33 2 3 6 10
230/07330 2 " 6 7 3
2.?9.'37144 5 8 9 10
225T27631 2 3 5 10
2?4/2=i596 3 4 5 10
223/71757 3499
222/74214 3 6 7 10
221:91457 1345
221T79184 1567
2J1.-7J933 1 3 5 10
2l9,'l37')291 "2 5 7 10
l?3/?4397 7 8 9 10
120/37747 2979
11 4/3. '204 2 6 7 10
ll3/«2-?a9 1 2 7 g
112/1U17 2 6 7 9
111. 1< -3 74 2 7 8 10
:iC.5Ti"2i 567?
1 •:<:.• 422Q 1 4 9 10
I0b/«1607 6 7 9 10
lflfl.7?393 1 3 9 10
106.'-!.' '-ofl 1 2 9 10
ina.7T5:)3 2 9 9 10
105 /-M-m 1 7 9 10
l!;5:i4S4i» 2 5 9 10
in2.7«'3:>4 4 8 9 10
1C1/-5705 3 a 9 10
1 ^ 0 . 3 ^ y $ fl 2 4 7 10
.'4i243 2347
94;8l743 2 " 6 9 10
93/44670 1237
92/05010 2479
89/33134 2379
89/11746 3 7 9 10
17,3H3l2 4 7 910
("6/63/00 5 6 9 10
S5/65,;?4 5 6 7 10
9i/75742 1 2 7 10
IS/C'Jf/ 1389
8 1 .' S i 3 8 2 1 4 8 9
91/4S9-JJ 2 5 6 9
*1/21553 1289
80/7?A,-|6 2 4 9 10
90/«5J"l 3 4 9 10
79/8S349 1279
79/7->6i3 2389
79/63809 2489
78/253*2 2 3 9 10
77?an06 3489
72/77499 1379
7 1 / J 7 .< •; 4 i 4 7 9
70/4U.T3 3479
46/i)«°J4 i 3 4 t
^ 44/3. 5/37 1249
45/i3147 1239
:59/-.M455 2349
•ijTi'llS 1373
53/i?9;s.j 1 3 7 1C
52.'7n->l 3 4 7 10
=! i . • t • 4 7 3 7 8 10
H. '9:21 3 1347
5i/o)0<'4 3479
47/31/U9 1 4 7 S
46/f»-9l»4 4 7 8 10
• 4_6 :?.'<'' 2 _ .1.4 7 15
1 J 7l 1 0 J '« 1 > 6 i u
?7/7;-i.» 2 5 6 10
. IS.'SP-MJ 2 7 9 !•-.
l?-'j9i'J I 3 8 15
lj/?5J)0 1 4 9 10
ll/597'j? 3 4 8 10
11.'2PJ14 1343
.'."..' 1 * 9 i) ? 1 2 4 3
'IJ.'.'. 33)1 2 4 8 10
1 'j / V / ^ j 0 1 2 3 3
IC.A13-14 2 3 8 10
1C.7-.5J3 2341
IT/37207 1 2 8 10
0, '22631 1234
3/22350 I 2 3 10
0/21490 2 3 4 10
0/20144 1 ? 4 10
07 1 3 4 10
Flfur. 15. CMFIL OUTPUT LISTING
-------
0
2 3
Number of Wavelengths
Figure 16. Signal-to-noise ratio as a function of the number of
wavelengths used for the measurement of ozone.
50
-------
4. LINEAR WEIGHT COMPUTATION (MFIL)
After the four wavelengths were selected, the optimum linear weights for
measuring each species were computed using the MFIL program. The absorption
coefficients for these lines (R14, P14, P24, P14) from Tables I and II are shown in
Table V. For lack of any better data at that time, the same variances listed in
Table C-l in Appendix C were assumed and are also listed in Table V . As more
field data is accumulated, these estimates can be revised and more accurate linear
weights can be computed.
Table VI lists the linear weights and relative SNR's that were computed by
MFIL using the input data listed in Table V. The relative SNR's shown in Table VI
are computed for an optical thickness of CL = 0.1. Therefore, the noise equivalent
concentration (NEC) is given by
NEC - (10 L SNR f1 (4)
where L is the path length in cm. For example, the present system monitors over
a 1. 34 km folded path. The NEC for ozone is then
NEC = (10 x 1. 34 x 105 x 305. 9)"1 = 2. 5 ppb
3
for the assumed noise variances.
The cross response of the linear weights in Table VI to 0.1 CL of each
species is shown in Table VII. These responses have been scaled so that the
background noise level is unity. Therefore, the diagonal terms are the same as
the SNR's listed in Table VI.
51
-------
en
to
Species
03
CO2
C2H4
NH3
H2O
Neutral
Absorption Coefficient - ATM"1 CM"1
R14
0.
4. 052363 E-03
9. 032525 E-02
3. 698675 E -01
1.379555E-04
1. 000000 E 00
P14
1 . 265892 E 01
3.661722E-03
1. 500705 E-01
3.416207E-01
1.407015E-04
1. 000000 E 00
P24
6.918318E-01
3.082101E-03
4.607009E-01
4.698566E-01
1.417035E-04
1. 000000 E 00
P14
0.
2. 7361 21 E-03
3. 197261 E 01
8.535154E-04
1 . 737966E-04
1. 000000 E 00
Variance (CL)
25 X 10"6
10.0
25 X 10"6
25 X 10"6
106
1.0
Table V. Absorption Coefficients and CL Variance of Atmospheric Species
-------
en
Species
°3
co2
C2H4
NH3
H2O
Linear Weights
R14
-0.4745
0.9302
0.6508
-0.9171
-0.9469
P14
1.0000
0.0559
0.0520
-0.0619
-0.0519
P24
-0.5286
-1.0000
-1.0000
1.0000
1.0000
P14
0.0029
0.0145
0.2962
-0.0212
0.0000
Relative SNR
305.8907
0.0175
110.3722
1 . 7920
0.0001
Table VI. Linear Weights and SNR's for Atmospheric Species
-------
Li osor
Weight
°3
co2
C2H4
NH3
H2O
Response to 0. 1 CL
03
305.8907
0.28537E 00
-0. 40088 E 00
-0.1 4985 E 00
0. 60282 E 00
CO2
0.29339E-02
0.0175
0. 67632 E-02
-0.1 5042 E-01
-0.1 6402 E-01
C2H4
-0.10789E 01
0.17755E 01
110.3722
-0. 50682 E 01
0.63966E 01
NH3
-0.20457E 01
-0.19995E 01
-0.25674E 01
1 . 7920
0.17689E 01
H2O
0.21 225 E-04
-0.56256E-04
0.83622E-04
0.4564 IE -04
0.0001
Neutral
-0.29074E-02
0.99228E-02
-0.11857E-01
-0.34612E-02
0.21823E-01
Table VII. Cross Response of Linear Weights
-------
5. FACTOR ANALYSIS OF DRIFT
An attempt was made to characterize the spectral nature of the drift/noise
associated with the monitoring system. A teletype output chart was examined for
the date of Saturday, 20 October 1973 between the hours of about 1535 PM to 1G05 PM.
During this period, the weather conditions were damp and overcast and the drift/
noise of the system was similar to the erratic behavior experienced in the mornings
of clear days. A total of 68 data points was selected out of this record and the effects
of printout round off error were minimized by sampling only nfter the last digit of
most of the channels changed simultaneously in the positive direction.
A sample covariance matrix was computed from the data and an eigenvector/
eigenvalue analysis was performed using the FACD program (Reference 17).
Eigenvectors and eigenvalues are listed in Table VIII. The eigenvectors, or factors,
represent an optimal basis for representing the spectral data in the sense of
minimizing the rms reconstruction error for the ensemble, and the associated eigen-
values are a measure of the importance of each factor in this representation. The
rms reconstruction error for the ensemble depends on how many factors are used and
is numerically equal to the square root of one minus the sum of the normalized (by G8
in this case) eigenvalues. This error, of course, diminishes to zero for four factors
since that is the dimensionality of the space.
In order to remove the effects of neutral attenuation, the factors listed in
Table vm were projected onto the neutral hyperplane, (1, 1, 1, 1), by subtracting 1/4
of the sum of the components from each vector. These vectors were then normalized
to unit length and are listed in Table DC. The absorption coefficients of known atmospheri
species were also projected and normalized in the same manner and are listed in
Table K for comparison.
55
-------
Ul
No.
1
2
3
4
Eigenvalue
6.298509E 01
3. 534071 E 00
1.051795E 00
4.290484E-01
RMS
Error
27.1%
14.8%
7.9%
0
Eigenvector
R14
-4.6076
-4.7461
-4.3835
0.1213
P14
1.3684
-1.2651
-0.0751
-0.2355
P24
-0.2501
-0.1220
0.3697
-0.9153
P14
-0.1778
-0.2743
0.4917
0.2837
Table VIII. Eigenvector/Eigenvalue Analysis of Data Record
-------
Factor/
Species
Fl
F2
F3
F4
°3
co2
C2H4
NH3
H2O
Wavelength
1
(R14)
-.295
.757
-.023
-.407
-.310
.659
-.296
.212
-.368
2
(PI 4)
-.329
-U646
.117
-.559
.865
.274
-.291
.127
-.266
3
(P24)
-.240
-.012
.653
.647
-.245
-.299
-.279
.494
-.232
4
(PI 4)
.864
-.099
-.748
.320
-.310
-.633
.866
-.833
.862
Table IX. Comparison of Factors and Atmospheric Species
Examination of Table IX reveals a similarity between the dominant factor and
both ethylene and water vapor. Since the presence of ethylene on that particular clay is
unlikely, the finger of suspicion is pointed in the direction of water vapor as a
possible atmospheric interferent. The uncertainties of the water vapor absorption
coefficients have been previously mentioned in Section C.2. However, any con-
clusions at this point would be premature considering the limited amount of statistical
data processed.
If repeatable factor analysis of data taken on other days was obtained, a
linear weight could be computed to reject this source of spectral interference no
matter what its underlying source is.
57
-------
D. SPATIAL FILTER
Experiments performed under Contract EHSD 71-8 showed that substantial
improvement in system performance could be expected from the introduction of a
cleanup aperature or spatial filter in the output beam of the laser. These results led
to incorporation of a spatial filter in the laser under this contract.
In terms of optical theory, the function of the clean up aperture (spatial filter) is
to filter the higher spatial frequency variations in the transmitted laser output beam
particularly in the far field pattern. We can visualize the far field as the Fourier
transform of the near field and vice versa. A small aperture located in the far field
provides an abrupt cutoff of the spatial frequency of the near field and at the same time
edits any non-uniform clutter in the far field. An aperture with a two-dimensional
gaussian amplitude distribution would be a better solution, but more difficult to
fabricate.
For the purposes of this experimental program, it was decided to use a
near-circular aperture like an iris in the far field. In the application of this device
it must be kept in mind that the near field will be a classical Airy disk pattern
2
(J /x) which is the two-dimensional Fourier transform of the uniformly illuminated
x
circular cleanup aperture. This Airy disk pattern is characterized by fringes in the
shape of rings that do not carry a large percentage of the laser power but are important
in shaping the far field pattern that is formed on the retroreflector. If a larger
percentage of the power in the rings is collected then the far field pattern formed upon
the retroreflector will be a disk with sharp edges (a true image of the clean up
aperture). Because the beam expander will not collect all the power exiting from the
clean up aperture then the illumination of the disk in the far field will be non-uniform
especially near the edges. The more of the rings or fringes from the clean up
aperture that are collected by the beam expanding optics the more uniform the disk
pattern and the less the effect of incoherent illumination. The tradeoff between
aperture configuration and beam expander size was a fundamental objective of
the experiment.
58
-------
The spatial filter was introduced to the CO^ laser breadboard system at :i
point between the coupling mirror and the beam splitter. The layout is shown in
Figure 17. The single mode beam diameters from the laser for this particular laser
design configuration are indicated on the figure. Note that the beam diameter, \v, at
the spatial filter is 0.67 mm. This is the nominal beam diameter, w, which is the
2
distance between the 60 percent power point or the half-width to the 1/e power point.
Figure 18 describes the amplitude and power of a Gaussian beam0 This optical con-
figuration was designed with the assistance of a computer simulation of the optical
elements and the diffraction limited Gaussian beam.
Ten optical apertures were made from 5 mil stainless steel sheet, ranging
in size from 0.1 mm to 2. 0 mm. The beam intensity at the exit of the 122 mm beam
expander was observed for each of these apertures. Some improvement was
obtained as the aperture diameter was decreased down to 1 mm. Smaller diameters
than that showed no measurable improvement in the uniformity of the beam at the
beam expander exit. Figure 19 shows the map of the beam intensity at the beam
expander exit for the two extremes. There is some spreading as well as smoothing
of the beam at the aperture.
The solid lines in the figure are a map of the beam expander output at a single wave-
length without a spatial filter. The dashed lines are the same plot using the 1 mm aperture.
The observed improvement in the beam pattern is responsible for the majority of the
improvement in the signal-to-noise and drift of the system, over that during the previous
study program.
The original intent of the spatial filter was to place the aperture at a focal
point that represented the far field pattern of the laser, i.e., the pattern that would be
o
formed at a plane located more than w /X from the coupling mirror of the laser, [f
one places a short focal length lens at the exit of a laser with a near-collimated output.
beam, then the far field pattern will occur at the point where the beam diameter is
minimum (the waist) which would also coincide with the focal point of the lens. In the
more usual case as in the ILAMS, none of these points coincide. A waist is formed
within the 762 mm long portion of the laser cavity between the coupling mirror and
59
-------
SPATIAL FILTER DIA. - 1 mm
SIGNAL
OUTPUT
OS
o
25 mm
DOUBLET
TO
REFERENCE
APERTURE
Figure 17. Spatial Filter Experiment Layout
-------
E
a. AMPLITUDE DISTRIBUTION OF THE FUNDAMENTAL BEAM
/
i
p
1
/
1
1
f
1
0 I
1
1
1
1
\ ^
1 K
\
I
1
\ f
w
k
1
b. POWER DISTRIBUTION OF THE FUNDAMENTAL BEAM
Figure 18. AMPLITUDE & POWER DISTRIBUTION OF THE
GAUSSIAN FUNDAMENTAL MODE
61
-------
WITH CLEANUP APEKTIWE
WITHOUT CLEANUP APEKTUKK
Figure 19. CONTOURS OF EQUAL DENSITY AT X = 9.504pn WITH
AND WITHOUT A CLEANUP APERTURE
62
-------
the concave mirror at the vertex of the "V". The design called for this waist to be
positioned midway between the two mirrors, but it actually lies closer to the coupling
mirror.
The true position of the waist (364 mm from the coupling mirror) was determined
by computer simulation of the laser cavity and confirmed by measurement of the
exit beam. A new waist is formed by the first lens following the coupling mirror and
then again by the doublet as indicated in Figure 17. The position of these new waists
does not coincide with the position of the real1 image of the waist of the beam within
the laser cavity, but this fact does not reduce the effectlvenss of a clean up aperture
located at a particular waist. The aperture will strip off any undesirable side lobes
in the diffraction pattern that represent high frequency clutter in the near field. Tho
ideal optical configuration following the clean up aperture is a large transmitting
telescope that would pick up the major side lobes of the diffraction pattern of the
clean up aperture itself yielding good image of the clean up aperture on the retroreflector.
63
-------
E. SEALED ISOTOPIC LASER DESIGN
1. Introduction
An isotopic CO^ laser in ILAMS offers additional pollutant detection capability
compared to the natural isotope, flowing CO laser In current use. Figure 20 illustrates
13 16
its greater spectral coverage. Use of C C>2 In the plasma tube adds, among others,
HNO , PAN.and PBzN to the system's list of potentially detectable pollutants. With
12 18
C O2 in the plasma tube a suitable absorption coefficient for long path monitoring of
SO has been reported. (Reference 18)
The C02 laser produces high CW power at high efficiency, i.e., more than 1 kw
at up to 15% efficiency. Patel (Reference 19) demonstrated CO lasing at approximately
£1
10.6 and 9.6 microns and, subsequently, Moeller and Rigden (Reference 20) demonstrated
CO2 lasing at approximately iO. 6 and 9.6 microns and, subsequently, Moeller and
Rigden (Reference 19) demonstrated lasing in both the P and R branches of the rotational
lines up to J values of over 50. Figure 20a shows the wavelengths at which lasing has
been achieved by the investigators cited.
The use of the heavier, stable isotopes of carbon and oxygen shifts the energy
levels of the CO2 molecule. Figure 20c shows the predicted lasing wavelengths for
the isotopic form of carbon dioxide, C13O l6. These predictions were based on
Li
measurements of the molecular vibration levels of carbon dioxide by infrared
64
-------
CI2<46 CALCULATED
'AND MEASURED
12
6- c1 V
'MEASURED
Ul
CALCULATED
r
~t .
9
•P
f
r—
9.0
9.5
10.0 105
WAVELENGTH (MICRONS)
11.0
II 5
a
~
-------
CI202'8
CI20217
c'2ol7o'8
CI20I6018
c'W7
CI202'6
CI30218
CI30I60I8
CI30I6017
c'6
10
II 7
WAVELENGTH. MICRONS
Figure 21. CALCULATED CO9 ISOTOPE BAND CENTERS
-------
spectroscopy. Verification of these predictions has been achieved using a mixture
1 Q 1 f* 1 9 1 fl
of 55 to 60% C O and 40 to 45% C O . Spectral measurements of these lines
2 2
with an infrared monochrometer were made at a resolution of +0.01 micron. Figure
20g indicates the spectral region spanned by the laser using the isotopic mixture
12 18 n 16
C O and C O . Note that lasing is occurring at the allowed transitions for
Z £t
both isotopes. Figures 20f and 21 show the band centers for several isotopic forms of
carbon dioxide for both the 00 1 - 02 0 and the 00 1 - 10 0 transitions. Figure 20e
12 18
shows measurements made of the lasing lines of C O providing further verification
£t
of the validity of the analytical results, Figure 20d. These results indicate that the IR
spectrum from 8.5 to 12 microns can be swept with a CO laser filled with appropriate
1*
isotopes.
2. LASER PLASMA TUBE DESIGN CONSIDERATIONS
The primary objective in the design of the plasma tube is high gain. The
requirement for spectral lines corresponding to remote "J" values of 150 or more
using mixtures of isotopes in a sealed CO laser demands high optical gain of the
&
working plasma. In addition, the long path length (to maintain closely-spaced
longitudinal modes) with multiple path folding plus a lossy spectral tuner within the
optical cavity of the laser produces losses that must be compensated for with gain.
Since the gain of the plasma in a CO laser is exceptionally high, laser
designers are usually more concerned with output power and efficiency for a given
working volume. Power output is usually obtained through the use of large fundamental
mode diameters and correspondingly large tube diameters. Coupling coefficients are
optimized for maximum power output at the P-20 line of the laser. For the scaled
laser designed for this laser system, the power output of 1/4 watt is more than adequate.
The "V" laser concept is designed to take advantage of the fact that the gain
of a laser is inversely proportional to the tube diameter. The stable mode diameter
for the TEM QO traverse mode is proportional to the square root of the distance
between curved mirrors for near confocal systems. The advantage of the zig-zag
folding configurations of which the "V" laser is the simplest example, is that a curved
mirror can be placed at each vertex and the beam diameter (and thus the tube diameter)
can be kept to a minimum.
(57
-------
Once the smallest possible fundamental mode diameter is established, llicn
the tube diameter can be chosen. In general, the gain increases for decreasing
plasma tube diameter until diffraction losses are introduced which exceed the in-
crease in gain. These diffraction losses are not introduced by the tube walls directly,
but by the fact that the gain of the pumping plasma falls off near the walls. This
effect is shown in Figure 22 to be a function of the exciting current. In order to
establish the optimum size the laser on-axis gain was measured in an experimental
setup as a function of nominal beam diameter and plasma tube diameter for the case
of optimum plasma excitation (determined experimentally during each run). During
each experimental run, gain was measured with a circular mode stop in place to
confine lasing to a fundamental mode. A variable iris was used as a mode stop and
was set at the largest aperture size that would constrain operation to the TEM QQ mode.
The tube diameter for maximum gain was found to correspond to 1. 532 times
the stop size that would produce 0.1B diffraction loss. This stop size can be
calculated using the curves of Kogelnik and Li (Reference 21). The advantage of a
folded plasma tube configuration such as the "V" laser is illustrated in Figure 23.
Note that a gain of 6.3dB may be obtained with two, 100 centimeter tubes, whereas
a single, 200 centimeter tube produces about 4.5 dB of excess gain. The improve'ment
in gain is 1. 85 dB while incurring an additional loss of 0.1 dB because of the extra
folding mirror. Aberration losses (coma) due to off-axis operation of the confocal
mirror at the vertex are negligible because of the small beam diameter (low f number)
within the plasma tube.
Because the glass envelope of the plasma tube is highly reflective at grazing
incidence it is unfortunately easy for lasing to occur over paths not anticipated in
the original design. To inhibit these "whisper" modes constrictions are introduced
-------
TUBE WALLS
Oi
to
r I I r I- I I, i ii '•
1.0 .8. .6 .4 .2 0 .2 .4 .6 .8 1.0.
RADIAL POSITION ~ CM
Figure 22. RADIAL GAIN PROFILE FOR A MIXTURE OF 1.8 TORR CO 2, 2.0 TORR N2
AND 4..6 TORR He AT WALL TEMPERATURE OF 13° AND DISCHARGE CURRENTS
OF : a.10 mA; b.20 mA; c.30 mA; d.50 mA
-------
EXTRAPOLATED
EXPERIMENTAL
GAIN/METER
TUBE DIAMETER
FOR MAXIMUM GAIN
OPTIMIZED GAIN
OF SINGLE CONFOCAL CAVITY
50 75 100 125 150 175
LENGTH OF CONFOCAL CAVITY (~ CENTIMETERS)
Figure 23. UNSATURATED GAIN OF A CO2 LASER AT A J-40 TRANSITION
CAVITY VS LENGTH ASSUMING THE TUBE DIAMETER IS OPTIMIZED
FOR MAXIMI IM r;AIN
-------
in the glass tubing every few Inches. These rings in the glass have an internal diameter
of one or two millimeters less than the I. D. of the bulk of the tubing; however they do
not increase diffraction losses because they are spaced too far apart to shrink the
diameter of the active plasma in the tube and are large compared to the mode stops
deliberately introduced to inhibit higher order modes.
3. STABILITY CONSIDERATIONS
Laser stability places a limit on the system performance. Angular riucliiMtions
and frequency fluctuations produced by changes in the plasma or in the option I
alignment introduce spectral effects and measurement errors.
Dimensional changes in the laser cavity are introduced by temperature .md
mechanical vibration. These changes shift the laser's frequency, amplitude, ;md modi
pattern. The percent change in laser frequency Is proportional to the percent change
in cavity length which is small (a few parts in 10 ) compared to the spectral bandwidth
of the gases we are trying to detect. Amplitude fluctuations produced by changes in
alignment, however, arc of major concern when they lead to mode hopping that mny
occur at some spectral lines and not others. Table X gives some indication of the
alignment tolerances that must be maintained to achieve satisfactory mode stability.
Mirror Tilt
Milliradians
-
0.16
0.33
0.66
1.64
TEM QQ % Losses
2, 3%
2.5%
4.0%
6. 8%
35.0%
TEMQ1 % Losses
28%
28%.
32%
40%
75%
TEMnj Rejection Mode
12:1
1 1:1
8:1
6:1
2:1
Table X. Losses versus Mirror Tilt for a Confocal Resonator
71
-------
Frequency and gain can also be influenced by the excitation; however, the
effects on frequency stability as a result of changes in plasma current are con-
siderably less than that contributed by other sources. The CO laser will have a
Ci
frequency drift in the range of 0.5 to 0.9 MHz per milliamp change in excitation
current. A typical current for a cooled laser is about 30 ma; thus, for a frequency
tolerance of 0.5 MHz the required regulation on the source current would only be
about 3.7%. A typical setup for lasers uses large series ballast resistors. Prior
to firing, the voltage at the electrode exceeds the maximum value needed for
ignition; however, as current begins to flow, the voltage drop across the ballast
resistor increases rapidly thereby lowering the voltage across the electrodes to the
appropriate operating level. The ballast resistor is chosen to accommodate the
desired voltage and is very much larger than the dynamic impedance of the laser (a
factor which improves current regulation by causing the voltage source and the
resistor to appear as a current source across the laser electrodes). If the use of
ballast resistors is objectionable, because of power dissipation in the ballast or
because of stability requirements, then an active current regulation system may be used.
4. INTRACAVITY WINDOWS
Practical design considerations make necessary at least one window within the
laser optical cavity to seal off the plasma portion of the laser from the spectral tuner.
Considerable effort in this design study was expended on the Brewster window problem.
The requirement for wide spectral range tunability from the laser means that neither
anti-reflection coated windows nor Fabry Perot etalons will suffice as a window. Metal
halides having adequate transmission at 10.5 micrometers are unsatisfactory because
they are hydroscopic, birefringent, and have been found to have poor surface quality.
Germanium and gallium arsenide have been the choice as Brewster windows to date,
but they have some disadvantages. The index of refraction of these materials is high
(4.0 and 3.28 for Ge and GaAs respectively). They are opaque to visible light.
72
-------
The absorption of laser radiation is sufficiently high that the change of index ol n>~
I'rnction with temperature distorts the beam as the laser conies up i.o a stable*
operating temperature.
Until recently we felt that thin GaAs windows offered the best compromise.
These windows can be polished to one (1) mm thickness maintaining a wedge angle of less
than 40 arc seconds between the front and back surfaces. In practice they were cemented
to a kovar flange which was then cemented to the borosilicate glass laser tubing rut ;ii
the Brewter angle. The kovar steel flange serves as a he:it sink to mmimi/e the
thermal gradients across the window which produce defocussing of the beam. Kovar
was chosen because its thermal coefficient of expansion lies midway between that ol
gallium arsenide and borosilicate glass. Thermal flexing is the major source of |xist
sealing leaks in the laser system. The-sealing material was Torseal high vacuum
epoxy which was then overcoated with General Electric HTV-lls. The RTV is used us
insurance against hairline cracks that may develop in the seal with very large arnbimi-
temperature changes.
The recent introduction of vapor-deposition-fabricated /inc selonido windows
has changed our philosophy somewhat with respect to sealed laser system design for
spectrally tunable CO lasers. The alignment procedure, which requires that all ihr
mirrors be lined up within one-half milliradian before lasing can occur, is quite
complicated when using an opaque Brewster window with even small \vodftv angles.
However, in the proposed design, zinc sclenide windows which transmit in the visible,
and a coupling mirror external to the laser plasma tube will allow relatively easy
alignment with a HeNe (G328A) laser. The index of zinc sclcnido is 2.4 and its
absorption coefficient is lower by more than 4:1 than cither germanium or gallium
arsenide. The smaller index yields a smaller Brewster angle and thereby somewhai
lowers the tolerances on the optical surface flatness.
73
-------
5. SPECTRAL TUNER DESIGN
Because of its high dispersion, relatively low losses, and easy maintainability,
the reflectance diffraction grating has been the choice as a spectral tuning element
for use within the laser cavity. Other possible approaches such as metal halide
prisms, Fabry-Perot etalons and dielectric fibers have been evaluated under earlier
study programs (Reference 22).
Used as an end mirror in the laser cavity, a blazed reflectance diffraction
grating designed for 10 microns has a dispersion of approximately 100 milliradians/
micron. The measured reflection efficiency of a Bausch and Lomb aluminized replica
grating, to CO^ laser radiation at ten microns, was 95 percent at the proper polarization.
There is some change in properties with temperature at moderate power levels near
2
100 watts/cm .tending to increase losses.
The efficiency of a blazed, reflection, diffraction grating varies as a function
of wavelength and angular orientation of the grating with respect to the incident radiation.
Except for approximately 3% absorption and scattering losses (using a gold overcoat)
all the losses from the first order reflection go into the zero order reflection. (The
direction of the zero order reflection is such that the angles of incidence and reflection
are equal about the normal to the grating surface). This zero order out-coupling can
be adjusted by selection of blaze angle, grooves per mm, and angle of orientation, and
varies strongly with wavelength.
The grating must be oriented with the grooves horizontal (with path folding
in the horizontal plane) so as to minimize the zero order reflection losses for both
ingoing and outgoing reflections. The first order reflections from the grating arc
dispersed in elevation with wavelength. The shorter wavelengths, for example, are
directed slightly downward (the grating is facing slightly downward) and the longer
74
-------
wavelengths reflect upward. The separate beams are collected in groups by the Hat
mirrors and redirected through the chopper wheel In a convenient orientation for
sequential shuttering. The shorter wavelengths strike the lower mirrors and the
longer wavelengths reflect off the upper mirrors. Because of the redirection ol' tin-
beams by mirrors, the position of the beams at the chopper and the wavelength
separation end mirrors is not proportional to wavelength. The number and location
of the mirrors following the diffraction grating and the respective position of the holes
in the chopper wheel both depend upon the wavelengths selected for the particular
target detection problem.
The optical design of the laser cavity to include the grnting clement as a
spectral tuner involves several, sometimes conflicting, requirements:
1. High spatial resolution between adjacent spectral lines at the end
mirrors so that adjacent spectral lines may be used in the same
laser system.
2. Large angular resolution between adjacent spectral lines so that un-
wanted spectral lines may be tuned out by angular alignment of the
end mirror.
3. Low aberration losses from curved folding mirrors which must be
operated off axis.
4. Small losses from the grating which must also be operated off axis.
2
5. Less than 100 watts/cm power density on the grating surface.
The spatial resolution of the spectral tuner is dependent primarily upon the
diameter of the laser beam incident upon the grating and the dispersion of the grating.
Spatial resolution is measured by the spacing between adjacent spectral lines
( A^ = 0.007 to 0.026 urn) in spot diameters at the end mirrors. If the grating has
a dispersion ft then the resolution is given by:
75
-------
R - /$ 7/w diameters/micrometer (5)
A
where w is the nominal diameter of the beam at the diffraction grating. For
example, if the spectral separation between two lines at 9.2 micrometers is
0.01 micrometers and ^3= 0.1 radians/micrometer, then in order to get a spectral
separation of four diameters, the beam diameter on the grating must be:
4 A 4 x 0.0092 mm
w= = =11.1 mm (6)
0.1 if x .01 .OOlrr v '
In order for the grating to accommodate this beam without introducing appreciable
diffraction losses within the laser cavity, it should be four diameters wide, which
is 46. 8 mm or 1. 84 inches in diameter.
The use of a large beam on the grating poses another difficulty. Because the
surface of the reflectance grating is inclined with respect to the vertical, beams with
high numerical apertures (small f numbers) introduce large aberration losses. For example
the "V" laser used in this program uses a spherical mirror of one meter focal length
to focus the beam on the end mirrors via the grating. The beam diameter at the
grating is 2. 9 mm yielding a numerical aperture of NA = 1/2 x 2.1/1000 = .00145 for
which the losses are negligibly small. However, in order to obtain the resolution
indicated above, the numerical aperture for the same focal length mirror would be:
NA = 1/2 x 46. 8/1000 = . 0234 (7)
Measurements with the laser system using just this numerical aperture on the
diffraction grating resulted in diffraction losses in excess of 3dB making the laser
virtually inoperable for this configuration on any but the strongest transitions.
In addition to the requirement for spatial resolution, it is necessary for the
spacing from the grating to the end mirrors to be large in order to preserve the
angular resolution. The angular resolution is given by:
R = p (1 -a/F) (8)
76
-------
where
p is the dispersion of the diffraction grating
a is the spacing from the grating to the last concave mirror
or focusing mirror, and
F is the focal length of the mirror.
The conflict between requirements is indicated in Figure 24. The nngle, A,
between mirrors Ml and M2 should be kept small to minimize zero order losses
from the grating and the angles C and D at the mirrors must be minimized Lo reduce
off axis aberration. The off axis image is degraded by coma, which is given by:
a = 6
T 16 (f No.)2 (9)
where aT is the third-order tangential coma expressed in the snmc units of angle
as u, which is the semi-field angle. Note that for a given f number the coma increases with
the square of the beam diameter and linearly with the off axis (semi-field) angle.
The only way to reduce "C" and "D" is to reduce the beam diameter and make
dimension "a" small. The beam diameter is set by'the requirement for sprttinl
resolution, and "a" is necessarily small because mirror M2 must intercept all the
spectral lines dispersed by the grating. Alternate approaches are:
1. Forego the use of spectral lines that are closely spaced in wavelength
and use small diameter beams of low numerical aperture at the
diffraction grating. This approach is used in the General Electric
ILAMS System.
2. Use off axis parabolas with long focal length (greater than 2.5 meters) in
the configuration shown in Figure 24. This will permit the dimension "a"
to become as large as O.G meters and keep angle "A" small. The axial
image formed by a paraboloid is well known to be free of geometric
aberrations. This method is recommended for the proposed sealed
system design. f
77
-------
-0
00
LASER BEAM
FROM PLASMA TUBE
,/xBBB
WAVELENGTH
SELECTION
END MIRRORS
"CHOPPER WHEEL
Figure 24. LAYOUT OF LASER OPTICAL SYSTEM
-------
6. SEALED LASER DESIGN CONSIDERATIONS
The problem of building a CO laser using the less common isotopes of
£t
carbon and oxygen is to construct a sealed plasma tube that will last from two to
five thousand hours without serious degradation in gain. The techniques for
accomplishing this require a marriage of the arts of vacuum technology and
laser physics.
Because of the high cost of carbon-13 and oxygen-18, the COf laser must
be operated as a sealed (non-flowing) system. The gain, power output, and life
time of a sealed CO Jaser are a function of the gas mixture, the tube diameter,
L*
the proximity of the path taken by the laser beam to the walls, the means of
excitation and the electrode structure and materials. The sealed CO laser shows
&
a gradual decay in gain and hence in power output with operating life, m order to
have an operating laser detection system, the gain within the plasma envelope must
be kept greater than the optical cavity's fixed losses and including those of the output
coupling mirror, while maintaining an internal power density compatible with the
desired output power. This minimum gain requirement must be met throughout the
operating life of the plasma tube and at the weakest laser transitions that are to be
used for gas detection.
This program included a detailed study and experimentation of the effects
of gas fill mixtures, plasma tube configuration and excitation on la.ser performance
and lifetime. A large background of work has already been accomplished by a variety
of investigators in the areas of laser gain and lifetime using the most common isotopic
form of carbon dioxide, i.e., C O 1U (References 23, 24, 25, 2f>, 27 and 28).
£1
Witteman (Reference 23) achieved 10,000 hours of continuous operation
with a forty watt CO laser operating at 10. G micron wavelength. The results of this
work have been extremely valuable to us, but there are significant differences
between the 10. G micron laser and the laser designs being considered. The isotopic
79
-------
CO .aser must maintain high gain over a large number of wavelengths at high J
12 18
values. The purchased C O isotopes will necessarily be contaminated with
Lt
more than ten percent CO .At each wavelength corresponding to energy
z
transitions in both of the CO gases, the gain-saturation characteristic is
u
different. The gain-saturation characteristic at a wavelength corresponding to on
energy transition in one of the isotopic forms is influenced by the presence of
other species because of the competition effects. The optimum ratios of carbon
dioxide to the other gases in the system are not the same for all isotopes of mixtures
as for ordinary CO .
&
The lifetime of the CO laser is directly related to the gas mixture, the
Lt
cleanliness of the internal surfaces of the plasma tube and the flow rate from out-
gassing and leaks. A gas filling station must be used to bake out the plasma tube
under high vacuum, to fill the plasma tube with the desired mixture of the isotopic
forms of CO , N , He, Xe, and HO, and to monitor the partial pressure of each
L* & £t
gas. (See Figure 25).
When direct current excitation of the plasma is used the electrode material
and the configuration of the electrodes are extremely important to the laser lifetime.
The electrode material may act as a sponge to some of the gases in the laser, it
may act as a catalyst for the production of undesirable compounds and it may sputter
away material from the cathode which may in turn react with the gases in the laser
or batter them against the glass walls of the tube. Platinum is one of the better
choices as an anode material because it does not encourage the production of
compounds (carbonyls, etc.) that Interfere with lifetime, nor does it oxidize to use
up the oxygen in the CO . However, platinum is a notorious sputterer and therefore
£
not suitable as a cathode material. Cathodes made from high purity nickel and
designed to operate at from 300 to 500 degrees Centigrade have been tested and found
to be acceptable.
80
-------
oc
LASER
FLEXIBLE
COUPLING
ULTRA-HI VACUUM
BAKEABLE VALVES
CAPACITANCE
MANOMETER
TO
ROUGH
PUMP
ABSOLUTE
PRESSURE
MILLIMETERS
ION
GAG!
TUB!
TO
ION GAGE
CONTROLS
STAINLESS STEEL
FLEX TUBE
NUPRO
BELLOWS
VALVES
SI L1C ON h
RUBBER
SEPTUM?
TO DIFFUSION
*.- ROUGH VACUUM
PUMPS
\VATEK
INJECT
Figure 25. MANIFOLD FOR HIGH VACUUM FILL STATION
-------
Another essential ingredient for long lifetime in a sealed laser is to maintain
a partial pressure of approximately 1.5 torr of water vapor in the gas mixture.
Because the electrodes and the glass walls tend to absorb HO, it is necessary to
u
add to the laser several times the amount of water needed to fill the plasma tube
volume to 1.5 torr. The water vapor is introduced into the plasma tube through a
double septum of silicon rubber with a rough vacuum held between the two septums.
This approach prevents the introduction of atmospheric gases into the plasma tube.
The system is allowed to come to stable equilibrium with the water vapor following
each injection and the amount of injected HO is adjusted for peak laser power output.
£
A small container of Linde 4-a Molecular Sieve material Is tied to the plasma tube
as a reservoir of water vapor to compensate for any long cleanup by the electrodes
and the glass walls.
A second reservoir consisting of a large bottle attached to the laser plasma
tube is used simply to increase the total gas mixture volume available to maintain
maximum sealed lifetime. If the gain of the laser falls below threshold because of
normal degradation of the fill gas mixture then the refilling process can be accomplished
fairly quickly at the gas filling station. If a leak is the cause for laser failure, however,
then the internal surfaces will have absorbed gases from the air which will have to be
removed by bakeout of the glass and high temperature bakeout of the electrodes with
an induction heater. The bakeout process requires several days.
Determining the laser gas mixture for optimum unsaturated gain is a tedious
business because of the large number of variables. The optimum mix determined
12 18
for use with C O is:
Torr
Carbon Dioxide 4.5
Helium 9
Nitrogen 2
Xenon 1
Water Vapor 1. 5
82
-------
12 18
Given this mix as a starting point, the optimum mix using ninety percent C O
12 16 ^
ten percent C O can be determined by trial and error monitoring output power.
jL
7. PROPOSED SEALED LASER DESIGN
The critical elements in this sytem are the seals, reservoirs, and the gas
fill mixture. Bakeout with a good, high-vacuum filling station is essential. The
proposed sealed system design is illustrated in Figure 2f>. It is a water-cooled "V"
shaped plasma tube with two Brewster windows. The coupling mirror will not be
connected directly to the plasiiia tube as in the present ILAMS system laser. This
avoids the problems associated with a flexible seal between the coupling mirror and
the plasma tube, which was necessary in order to obtain adequate alignment of the:
optics. Such a seal presents a high vacuum problem and must be. insulated against
the high voltage between the plasma tube and the support frame. The tube is necked
down by two millimeters at regular intervals to inhibit whisper modes from developing.
The Brewster windows are of zinc selenide and are cemented directly to
the laser tube with high vacuum epoxy Torseal. The epoxy is then over-coated
with RTV-118 as a seal against hairline cracks that may appear in the epoxy as :i
result of age or large temperature changes. The laser tube and associated glass
plumbing is made of pyrex type borosilicate glass (Glass Code Number 7740). At
the two ends where the Brewster windows are sealed this glass is graded to a soda
lead glass (Number 0120) which is a better match for the thermal coefficient of
expansion of zinc selenide. The glass is cut to the approximate Brewster angle
and then hand ground to meet the Brewster angle and to provide a smooth, flat
mating surface for the window.
The length of the tube is approximately 1.1 meters with a convex one-meter
radius-of-curvature gold mirror at the vertex. The glass tubing at the vertex is cut
;md hand ground to prealign the mirror which is cemented directly to the glass. This
83
-------
2-
CONCAVE MIRROft
SEALED TO GLASS
XENON RETURN
NECKDOWN
EVERY 30 cm
PLATINUM
\ANODE
LINDE
4-a RESERVOIR
22.6 ANGLE
FOR BREWSTE
WINDOW
Figure 26. Proposed Sealed Isotopic Laser Design
-------
again, is to avoid sealing and high voltage problems. Water cooling of the tubes out
beyond the electrodes provides sufficient temperature stability so that alignment is
maintained. A circular metal aperture is cemented to the center of this mirror as a
stop for higher order modes.
Xenon being a heavy molecule tends to drift in the plasma toward the cathode
so a return or bypass is provided for each leg of the "V". This bypass must have
smaller diameter and/or longer length than its associated plasma tube in order to
prevent the discharge from firing along the bypass.
A gas reservoir is attached to one of the legs of the "V" and the inlet valve is
in turn connected to the reservoir. This is a high vacuum bellows type valve with
sufficiently high conductance to allow bakeout of the system in less than a week at
temperatures of 250 degrees Centigrade. A second reservoir of molecular sieve
material is used to keep the water vapor concentration fixed in spite of the clean up
effects of the internal surfaces of the system. The capacity of the molecular sieve
material is large compared to that of the inner surfaces at the same vapor pressure.
The anode is platinum and the cathodes are made from Number 270 Nickel.
They are 6 mm cylinders with a 2.6 mm hole drilled in one end. The other end is
welded to a kovar wire for the glass seal.
85
-------
REFERENCES
1. L. R. Snowman and D.R. Morgan, "Studies on An Isotopic CO Laser LOPAIR
£t
System," Department of the Army Edgewood Arsenal DDEL, First Quarterly
Report, Contract DAAA15-72-C-0359, February 1973.
2. L.R. Snowman, D.A. Ware, and D.R. Morgan, "Gas Laser Detector", Air Force
Armament Lab., Eglin AFB, Florida, Ftnal Report, Contract F08635-68-0116,
September 1969; also, General Electric Company, Electronics Laboratory, Syracuse,
New York, Report No. R69ELS-1, September 1969.
3. D.R. Morgan and D.A. Roberts, "Computer Signal Processing Study", Dept. of
the Army, Edgewood Arsenal DDEL, Final Report, Vol. 1: Analytical Results,
DAAA15-71-C-0186, September 1972.
4. J.H. McCoy, D.B. Rensch and R.K. Long, "Water Vapor Continuum Absorption of
Carbon Dioxide Laser Radiation Near IQn" Applied Optics Vol. 8, No. 7, pp 1471-
1478, July 1969.
5. W.A. McClenny, EPA, Private Communication, August 1973.
6. E.H. Christy, Tulane University, Private Communication, June 1973.
7. P.L. Hanst, EPA Research Triangle Park, North Carolina, Private Communication, 1972
8. L. Gaslorek, Stanford Research Institute, Private Communication, August 1973.
9. R.A. McClatchey, et al., "AFCRL Atmospheric Absorption Line Parameters Comp-
ilation", Air Force Research Laboratories, Bedford, Massachusetts, AFCRL-
TR-73-0096, January 1973.
10. R.K. Long, Ohio State University, Electroscience Laboratory, Private Comm-
unication, September 1973.
11. G. L. Trusty, "Absorption Measurements of the 10.4 Micron Region Using a CO
Lt
Laser and a Spectrophone" Air Force Avionics Laboratory, Wright-Patterson AFB,
Ohio, AFAL-TR-72-413, January 1973.
12. W.A. McClenny, EPA Research Triangle Park, North Carolina, Private Comm-
. unicatlon, September 1973.
13. D.E. Burch, "Semi-annual Technical Report; Investigation of the Absorption of
Infrared Radiation by Atmospheric Gases" Phllco-Ford Corporation, Aeronaut-
ic Division, Contract No. F19628-69-C-0263, U-4784, January 1970.
86
-------
14. D. E. Burch, "Radiative Properties of the Atmospheric Windows" Conference on
Atmospheric Radiation, pp 61-68, August 7-9, 1972, Fort Collins, Colorado;
published by AMS, Boston, Massachusetts.
15. H. R. Carlon, "Model for Infrared Emission of Water/Aerosol Mixtures" Applied
Optics, Vol. 10, No. 10, October 1971.
16. P. L. Hanst, NASA Electronics Research Center, Cambridge, Massachusetts,
Private Communication, 1970.
17. D. R. Morgan and D.A. Roberts, "Computer Signal Processing Study" Final Report,
Vol. 2: Computer Programs, Dept. of the Army, Edgewood Arsenal, Contract
No. DAAA-71-C-018G, September 1972.
18. R. T. Menzies, "Remote Detection of SO and CO with a Heterodyne Radiometer"
2 2
Appl. Phys. Letters, Vol. 22, No. 11, p592, 1973.
19. Patel, Phy. Rev. 136, 5A, November 1964, pAH87.
20. G. Moeller and J. Ridgen, Appl. Phys. Letters Vol. 8, No. 3, p 68, 1966.
21. H. Kogelnik and T. Li, "Laser Beams and Resonators" Proceedings of the IEEE,
Vol. 54, No. 10, October 1966.
22. L. R. Snowman, "Laser Coincidence Absorption Measurements", General Electric
Co., Electronics Laboratory Report No. R72ELS-15, March 1972.
23. W.J. Witteman, "Sealed-off High-Power CO Lasers" Phillips Technical Review,
Vol. 28, Nov. 10, 1967.
24. R.J. Carbone, "Characteristics of a Single-Frequency Sealed-off CO Amplifier"
IEEE Journal of Quantum Electronics, January 1969. 2
25. V. Hoffman and P. Toschek, "One-year Operation of Sealed-off CO Laser"
IEEE Journal of Quantum Electronics, November 1970.
26. R.J. Carbone, "Continuous Operation of a Long-lived CO Laser Tube" IEEE
Journal of Quantum Electronics, March 1968. 2
27. H.W. Mocker and H. A. Gustafson, "New Contender for Space Communication"
Laser Focus, October 1970.
87
-------
28. W. J. Wltteman and H. W. Werner, "The Effect of Water Vapor and Hydrogen on
the Gas Composition of a Sealed-off CO Laser" Physics Letters, Vol. 26A, No.
April 10, 1968.
29. M.M. Whatiey and D.A. Smity "Atmospheric Effects on Digitally Modulated Laser
Transmission" U.S. Army Electronics Command, Fort Monmouth, New Jersey,
Tech. Rep. ECOM-3005, July 1968.
30. R. Paulson, E. Ellis and N. Glnsburg, "Atmospheric Noise Measurements" Air
Force Cambridge Research Laboratories, Tech. Rep. AFCRL-62-869,
AD 287 517, August 1962.
31. W. L. Wolfe, Handbook of Military Infrared Technology, Office of Naval Research,
Dept. of the Navy, Washington, D.C., 1965.
88
-------
APPENDIX A
INTERFERENCES
Anything that frustrates the detection of the IR absorption pattern associated
with a particular gas is considered as a source of interference.
System Noise
System noise is composed of detector, optical, and other noise-
sources that arise in the measuring apparatus. This type of noise usually consists
of fairly rapid zero-mean fluctuations of the received energy.
An actual instrument will time-average the demodulated signal. Therefore,
if response time is not a consideration, then the effect of system noise may be
reduced to an arbitrarily small level by choosing a large enough integration time.
Of course, any real-time measurement problem does necessitate a finite response
time and so there are limitations on how much noise reduction may be attained in
this manner.
A-l
-------
Detector noise is additive and usually uncor related between wavelengths for
a scanning-type system. The amplitude distribution of detector noise is Gaussian
and its variance depends on the particular type of detector that is used. The spectrum
of the noise usually consists of a uniform thermal noise components and a 1/f
"flicker noiseM component that dominates at lower frequencies. Detector noise will
be negligible if the source strength is large enough.
Optical noise or "scintillation" is a modulating type of noise that results from
inhomogeneities of the index of refraction in the atmosphere. ; This type of noise
is strongly correlated between wavelengths over a large range because the refractive
index does not vary much with wavelength. The amplitude distribution of optical
noise is log-normal (Reference 29) and its variance is a function of aperture, range,
and meteorological factors (Reference 30). Its spectrum shape is of a 1/f nature
(Reference 28) although the exact value of the exponent varies, depending on conditions.
Unlike detector noise, the relative magnitude of scintillation noise does not depend on
Lht: source strength, since it is a multiplicative type of noise.
Laser systems may be either optical or detector-noise-limited. Past experience
indicates that, in most cases where a cooperative reflector is used, enough laser
energy is available so that detector noise can be neglected. Under these circumstances,
the laser system is optical noise limited. Typical detector and scintillation noise
densities as a function of frequency are plotted in Figures A-l, and A-2, respectively.
For a system corrupted by additive detector noise and multiplicative optical
noise, the measured absorption at wavelength i is given by:
-
A. --•- -log(MP.e i + N) + log P. (A-!)
A-2
-------
100
Detector
>
c
0)
in
'5
Z
o
0>
10
10
Preamp
-1 1 L.
I _ I _ I • I
Frequency - Hz
Figure A-l. DETECTOR NOISE VERSUS FREQUENCY
IK
3 Mile Path
6" Aperture
"Optical Noise Measurements, " Report #AD 287517,
U .S . Dept of Commerce
= 5.4
f 1.25
100
Frequency - Hz
IK
Figure A-2. OPTICAL SCINTILLATION NOISE VERSUS FREQUENCY
A-3
-------
where P. is received power,<*. the absorption coefficient at wavelength i and CL
is the optical thickness or mass per unit area of target gas. The optical modulation
noise M is assumed to be distributed log-normal with unit mean and variance cf 2
M
cycle of bandwidth. The detector noise N is assumed to be Gaussian distributed and
is specified by the noise equivalent power NEP of the detector which is defined as the
amount of optical power necessary to produce a signal-to-noise-ratio of one at the
detector output in a bandwidth of one Hz. Here, the svstem is assumed to have been
initially balanced by adding the quantity log P. corresponding to CL = 0.
If detector noise is small, then for small absorptions, (A-l) reduced to:
A =o-CL - log M - N/MP., (1 - 0" ) P » NEP (A-2)
11 i M i ' '
Each noise component in (4) is now additive and linear mean-square estimation
theory can be applied.
If detector noise is not negligible, then some pre-log integration (filtering)
is necessary in order to reduce the noise level so that the approximation in (A-2)
can be made.
The first noise component in (A-2), log M, is now zero-mean Gaussian-distributed
2
with variance log (1+0'M )• The second noise term is somewhat more unwieldy in
distribution. However, if <7 ^» NEP and the second noise
term can be neglected.
A-4
-------
Specific Interferents
An interfered is defined as any spectral absorber that may be present in
sufficient quantity as to interfere with the detection of the target gas absorption
pattern. Unlike system noise, this type of interference does not necessarily consist
of rapid fluctuations and therefore is not appreciably reduced by increasing the
integration time of the measurement.
Neutral Attenuation
One of the most dominent interferents in any system is neutral attenuation.
This effect is caused by:
1) Unfavorable atmospheric conditions such as rain or fog,
2) Changes in the transmission or reflectivity of optical components
(e.g. , dust accumulation on mirrors), and
3) Gain variations of electronic components (e.g. , respr.nsivity variations
of a thermistor bolometer due to ambient temperature changes).
An automatic gain control (AGC) amplifier is used in the system to control
large average variations of signal level and thereby reduce the dynamic range
requirements of the signal processor. However, neutral attenuation must still be
considered as an interfered since it perturbs the absorption pattern. Specifically,
the effect of the AGC amplifier is to adjust the system gain so that the average
signal level is constant. For example if there is a 4% absorption at A , then the
AGC will boost the average signal which will result in a 3% absorption at A and
a -1% absorption at A 2> Agi and ^ This effect is equivalent to the combination
of the 4% absorption at A 1 and a -1% neutral attenuation.
A-5
-------
Other Specific Interferents
The amount of an interferent that is present may be' regarded as a random
quantity. The absorption pattern of a particular interferent is of the form:
S. = I. CL, i = 1, 2, ..., n
where I. is the absorption coefficient of the interferent at the i wavelength and C
is the average concentration of the interferent over the path length L. This inter-
ferent constitutes a noise source whose covariance between wavelengths is given by:
cov (S.S.) = I.I. var (CL) /A_4)
In the absence of detailed statistical information, the variance of the CL factor
may be bounded by the square of the largest CL factor that has been observed in
the field. The danger in using such coarse statistical measures is that the actual
probability distribution is obscured. This leads to difficulties if the system is
optimized by treating system noise and interferent noise on an equivalent rms
statistical basis, that is, by summing their covariances. A more detailed discussion
is given In Appendix B.
Random Interferents
Besides the known interferents, there may be certain other intcrferents of
an unknown absorption pattern. One model that is both intuitively appealing and
mathematically tractable assumes a Poisson distribution of Lorentz-broadened lines
(Reference 31).
A-6
-------
It is shown in Appendix A of Reference 2 that the spectral autocovarianec
of such a model is given by
C(0) (
where C(0) is the variance of the noise, and a is the half width of the lines. At
atmospheric pressures, a is on the
to some extent on the absorbing gas.
atmospheric pressures, a is on the order of 0. 1 cm , although the value
Thus, the random interference model is equivalent to a for related noise- soui-rc-
and can be lumped together with the other sources of noise.
Other Interferents
If a particular interferent is not well-defined or is of an unstable nature, for
example, dependent upon meterological and/or environmental conditions, then a
factor analysis of this behavior pattern must be performed. This implies that am
particular variation of the interferent ensemble can be suitably approximated by a
linear combination of a set of factors. Each factor is in turn considered as in
interferent in its own right and is used in the design of the linear spectral weights.
To the extent that the subspace spanned by the factors represents the interferent
ensemble, a design based upon such an approach will be successful in rejecting this
interferent.
Such an approach was taken in a previous study (Reference 2) in which dust
particulates were a serious source of interference. For the present program it
appears that water vapor may be of such a nature as to require a similar treatment.
This will depend a great deal on the results of the SRI simulation study which are
expected shortly.
A-7
-------
APPENDIX B
OPTIMUM LINEAR WEIGHTS
Introduction and Notation
In this Appendix, techniques for the design and evaluation of optimum linear
weights for detecting and estimating quantities of spectral absorbers and discussed.
They are a summary of the results derived in Reference 2. The purpose of
determining optimum systems is to establish a theoretical limit on the performance
that may be obtained. This limit can then be used as a goal in the design of actual
systems, that is, a frame of reference by which tradeoffs between complexity and
performance may be reasonably conducted. In addition, optimum solutions often
serve as a guide in designing an actual system. If a solution is arrived at by this or
any other means and it is close to optimum, then there is no need to search further.
Spectral information will be contained in the transmission values T , T .... T
12 n
where n is the total number of wavelengths used. If a single absorber is placed in the
sample region, the transmission at each wavelength will be of the form
•
(B-l)
T =exp (-A.C L) , i = I, 2 n
1 L f\
where A is the absorption coefficient of absorber A at wavelength i, C is the
f\
average concentration of absorber A over the total optical path, and L is the total
optical path through the sample region. Typically, C has units of grams/liter or
atmospheres of partial pressure, and L is in centimeters. A is in units to make A
_ i i
C L dimensionless.
/\
In vector notation, eq. (B-l) becomes
(B-2)
T= exp(-A CL)
B-l
-------
where A - (A A ,...,A )' and T = (T T ... ,T )' are n-dimensional column
i ^ n i L n
vectors representing the absorption coefficients of absorber A and the transmission
values of the sample region, respectively, and the prime (') denotes transpose. If
a second absorber B with absorption coefficients J3 is introduced into the region, the
net transmission will be the product of the transmission due to each absorber; thus:
T = exp (-A C L - B C_L). (B-3)
A B
First, some discussion of possible patterns is necessary. Figure B-.1 shows
possible transmission patterns for an absorber A characterized by absorption co-
efficients, A^^ = 1 liter/gm-cm and A = 5 !iter/gm-cm, and a second absorber B
characterized by absorption coefficients, B = B = 1 liter/gm-cm. Transmission
.L. 4J
patterns are plotted for various amounts of each single absorber. Some mixtures
of the two absorbers are shown in Figure B-2. All possible mixtures of A and B will
be a smooth distribution lying between the contours for each pure absorber. For
small absorptions, i.e., 1- T. « 1, eq. (B-3) becomes
T=1-ACL-BCL, (B-4)
where 1 - (1, 1 1)'.
If the measured transmission is subtracted from a reference level at each wavelength,
the signal is
S= 1 - T~AC L4 BC_L. (B-5)
- t\ — D
Under these circumstances, the absorption coefficients add vectorially and a linear
space is defined.
Another method of displaying transmission patterns is to plot the natural logs
of the transmissions. Let S = -In T. If T is due to several absorbers, then
S = A CL+ B CL+ ... B-<1
B-2
-------
A] = 1 liter/gm-cm
A2 = 5 liter/gm-cm
B] = 1 liter/gm-cm
CL = 0, 1, 2, 3 gm-cm/lifer
Figure B-l . POSSIBLE TRANSMISSION PATTERNS FOR SINGLE ABSORBERS
Figure B-2. POSSIBLE TRANSMISSION PATTERNS FOR MIXTURES OF ABSORBERS
B-3
-------
which is the exact version of eq. (B-5). Figure B-3 is the equivalent of the patterns
of Figure B-2. For this method of display, a linear space is defined for all
absorption levels, and so the vector for each absorber is constant in direction,
has a magnitude proportional to CL, and obeys vector addition. Visualization of
possible transmission patterns is also easier in this manner.
The linear model adequately represents the actual process and hereafter,
it is assumed for this discussion that the linear vector space model adequately
represents the actual process over the range of interest either by virtue of small
absorptions or by a log transformation. The system can therefore be considered
as a black box with an output column vector X that is composed of the sum of a
signal vector A and a noise component N that represents all sources of interference.
The signal vector represents the absorption coefficients of the target gas over a
selected set of wavelengths.
A detailed mathematical derivation of optimum linear weights is given in
(Reference 32) and the major results are summarized in this section. The optimum
linear weights maximize the signal-to-noise-ratio
0 - W'A/(Var W'X) 1/2 (B~7)
for various constraints.
Unconstrained Weights
The direction of the optimum unconstrained weight for a single target in a field
of correlated noise is derived as
W = Z ~1A
B-4
-------
0
Figure B-3. POSSIBLE PATTERNS FOR THE NATURAL LOG OF THE TRANSMISSION
B-5
-------
where E is the covariance matrix of the noise vector and A is the signal vector
which represents the absorption coefficients of the target gas. The optimum signal-
to-noise ratio is given by
t MA'E'
max - -
2
If the noise is uniform and uncorrelated.then £ =0 1 and W is the "matched filter"
for the signal vector.
The composition of C takes into account the interferents as well as other
sources of correlated or uncorrleated noise. If an interferent is of the form P \
where/»is a random variable and T = (I I . ,i ) is a fixed pattern vector, then
i L. n
its contribution to the total covariance matrix is
O" .. = 11. var (P\ (B-10)
i] ij
For a spectrally absorbing interferent gas, the quantity/? represents the CL factor
or mass per unit area. In addition, the covariance matrix provides a convenient
means of handling random interferents since this is just another correlated noise
source. A detailed description of interference was covered in the Appendix A.
The disadvantage of a signal-to-noise optimization is that all interferences arc
characterized only by second-order statistics. This leads to difficulties if system
noise and interferent noise are treated on an equivalent rms statistical basis, that is,
by summing their covariances. The reason for this is that while the system noise
probability distribution is fairly well-defined, an interferent distribution may be of a
very non-stationary and erratic nature; e.g., a passing truck stirring up a dust cloud.
For these reasons, it is often desirable to design a system in which certain "well-
behaved" interferents are described by the covariance matrix and a null response is
required for the more irregular distrubances.
B-6
-------
Orthogonal Weights
Orthogonality constraints of the type
W. J. =0, j = 1, 2,..., m < n (B-li)
are now considered. That is, a zero response is required for certain interferents
of a perverse nature. It is shown that the direction of the optimum weight is given
by
W = PA .
where
P = E I I - Q(Q' £ ~ Q) Q1 £ ~
and
Q= (I., L I ) (R-14)
i ^ m
is an n x m partitioned matrix whose columns consist of selected interferent vectors.
The optimum SNR in this case is given by
max
=(A'PA,1/2
— —'
If the noise is uniform and uncorrelated, eq. (B-13) reduces to
-i (B-1G)
P=I-Q(Q'Q) Q' .
This form has the simple geometric interpretation of dropping a perpendicular from
the signal vector down to the subspace spanned by the m interferents.
On the other hand, if there are no orthogonality constraints, Q = (), and eqs.
(B-12) and (B-13) reduce to eq. (B-8) as expected.
B-7
-------
An alternative method of imposing orthogonality constraints is to assign an
equal amplitude variance n to the m interferents and to include them in the covariance
matrix using eq. (B-10) it can be shown that as /z becomes large, the optimum weight
calculated by eq. (B-8) will converge to a vector that is orthogonal to the m selected
interferents. An alternative method of computation is therefore available. The choice
of methods will depend on the particular problem at hand. For example, if the noise
is uniform and uncorrelated, then eq. (B-16) applies which only involves the inversion
of an m x m matrix as opposed to the n x n Inversion required in eq. (B-8). On the other
hand, for correlated noise, the alternative method may be preferable.
Multiple Weights
In some applications, several linearly independent targets can be present
simultaneously. If it is desired to measure or detect each target independently,
then the linear weight for each target is required to be orthogonal to the other
targets, i.e. ,
In matrix notation, eq. (B-17) becomes
W'A = I (B-ls)
where W = (W , W ..... W ) and A = (A , A ..... A ) are n x m matrices whose
L £. m ~ ~ L £ m
whose columns are the W. and A. vectors respectively.
The optimum constrained weight matrix is given by
and the SNR for each target is given by
d -1/2
(B-20)
B-8
-------
where the r..'s are diagonal elements of the matrix
u
It is noted that R is the cross response matrix in the absence of orthogonality
constraints.
r -1* u
For m = 1, eq.(B-19) reduces to the usual form W^ = at)_ A^ where a \s a
scalar constant. On the other hand, if the number of targets is equal to the dimension-
ality of the space, i.e., m = n, then
(B-22)
W' = A
For the special case when the noise is uniform and uncorrelated, i.e. ,
2
£ =cr I, eq. (B-19) reduces to
-1 (B-2.1)
W - A(A'A)
This expression is the well-known "generalized inverse" of the matrix A and W'Y
is the "least squares" solution to the over deter mined system of equations AX^Y.
Application
For the present problem, the number of targets is m = 3. Since neutral
attenuation is a dominant interference and is not very statistically well-behaved
or modeled, the weights will be orthorgonally constrained to reject this interfered,
i.e.,
W + W + W + W = 0
J_ Lt O *
(B-24)
for each target. This will be indirectly accomplished by considering neutral
attenuation as a 4th "target" in the formulation of equations (B-17) through (B-21).
All other interferents will (by necessity in this case, since we only have 4 wavelengths)
be considered to be "well behaved" and will be described by the noise covariance
matrix. At least one such interferent, HO vapor will be included in this set.
Lt
B-9
-------
APPENDIX C
WAVELENGTH SELECTION
Introduction
A large number of spectral lines are typically available using the CO^ laser.
However, practical considerations limit the number of lines used to about 4 to 8.
A selection of the best subset of available lines is based on maximizing the signal to-
noise-ratio (SNR) for a specified set of targets and interferents.
•lo 1C
For the present 4 wavelength C O laser system, there are about 74 possible
£4
lines from which to select. A direct combinatorial approach to selecting the best 4
74
wavelengths out of the available 74 would require (4 ) = 1, 150,626 combinations; an
exceedingly large number of computations. Each SNR computation involves an express-
ion of the form of (B-21). This eolation Involves the construction and inversion of a 4 X
4 matrix, 20 multiplications, and 20 adds to determine the SNR for ozone alone - the
target we are optimizing. Consequently, even if each computation were on the order
of one cent, the required total cost would be prohibitive.
Fortunately, a viable and expedient alternative exists that can at least narrow
down the number of wavelengths to be considered to a manageable number. As a bonus,
the method can also determine the increase in performance that could be attained if
C-l
-------
more than 4 wavelengths were used. This technique was developed (Reference 1)
as an outgrowth of Reference Sand is referred to as the "squeeze method" of wave-
length selection.
Squeeze Method
For a scanning system, the noise O" . associated with a measurement at each
wavelength \ . is indirectly proportional to the time T. spent at that wavelength:
O O / /f~> 1 \
rr \ - v /rr (^-l)
Ol - K. 1.
l L
where K. is a constant determined by the system noise sources. For a fixed scan
rate, the sum of the time intervals is fixed, i.e. ,
T + T + ... + T = T. "^
1 2 n
Assuming uncorrelated noise, the variance of a weighted sum of measurements
is given by
n
0- 2= Vffi 2 |W 1 2 (C-3)
/ i
i = 1
where W. is the weight attached to the measurement at A ..
By using a variational argument, (Reference 3, Appendix E) it can be shown that the
time intervals that minimize (C-3) subject to the constraint (C-2) are given by
n
/«
T - TK
i i
i
j = 1
C-2
-------
i.e., the optimum time intervals are proportional to the product of the noise
constant and the absolute value of the weight. Substituting (C-4) into (C-l)
n
_ 2 K TII
glves CT< = i \K. |W|, (C_rj)
TJW. | L_ ]
and determines the system noise component of the noise covariance matrix used
in computing the optimum linear weights.
In Appendix B, procedures were developed for selecting optimum weights.
These procedures were dependent on the noise covariance matrix. If the optimum
time intervals are selected according to Eq. (35) , then the covariance matrix is
of the form
where D is a diagonal matrix, representing system noise, whose elements are given
by (C-5) and is a function of W, and .A. represents the noise due to other interferents. Since D
and hence T. are themselves functions of W, the solution of W is in implicit form. An
iteration procedure is therefore indicated. An initial guess of the solution, say WQ' =
(1, 1 ..... 1) , is substituted into Eq. (C-6) and the optimum weighting vector W1 is
calculated by one of the techniques given in Appendix B. For the next iteration, J^
is substituted into eq. (C-6) and W is calculated. In this manner, a sequence of weights
u
will be determined which will converge to the desired optimum weight.
C-3
-------
A consequence of this procedure is that weights corresponding to wavelengths
of low information will converge to zero and the required wavelength seleetion process
is thereby attained. In general, m + p wavelengths will be retained in the limit,
where p is the number of independent agents to be detected and m is the number of
intcrferents (assuming n > m <- p).
If several targets are to be simultaneously optimized, a suitable averaged
weight can be used in the feedback loop.
A computer program (LWSP) was developed for implementing the above
procedure (Reference 1) and is described in Appendix D. It assumes a detector-
noise-limited system so that K. = NEP/P. where NEP is the noise-eqmvalent-power '
of the detector and P. is the received power at A .. The program is restricted to
computing weights by (B-ll) and (B-12) with a diagonal covarlance matrix, i.e., it
assumes that enough wavelengths are available so that all interferents can be com-
pletely rejected and that all system noise is uncorrelated.
Application
In order to use the LWSP program for the present application, ozone was
considered as a single target and all other gases were considered as interferents.
In addition, the present application was expected to be optical-noise-limited rather
than detector-noise-limited and so, the P.'s were all set equal except for those
that are so low as to preclude reliable operation. In this manner, the 74 available
lines were reduced to a set of 10 from which a final selection was based.
The final selection uses another program (CMFIL) in order to compute the
performance of all ( 4 ) = 210 combinations of wavelengths using equations (B-20)
and (B-21). This program was adapted from another existing program (MFIL).
C-4
-------
LWSP
A description of the LWSP program which implements the "Squeeze Method"
as previously noted is given in Appendix D. This program essentially eliminates
wavelengths of low information by an iteration process of adjusting the power allocation
according to an optimization algorithm. In this manner, the power allocated to un-
important wavelength coverages to zero thus accomplishing the selection process.
The present configuration of the LWSP program operates in the orthogonal
mode, i. e., the linear weights are constrained to completely reject all of the gasses
in the prescribed interferent set. In this mode of operation, the number of significant
wavelengths will converge to the sum of the number of linearly independent targets and
the number of linearly independent interferents.
For the present application, the detection of one target, O , was considered
«J
in the presence of 5 interferents: CO , HO, C H , NH , and neutral attenuation.
Z Lt Z 4 o
With this target and interferent set, the number of significant wavelengths will then
converge to 6.
A 25 iteration LWSP run with the above target and interferent set resulted in
the line selection illustrated in Figure 14. The solid lines in this figure represent
the linear weights applied to each wavelength and their length is indicative of the
relative importance of each line. The X's designate normalized ozone absorption
coefficients and the +'s designate the average interferent noise level. As can be seen,
the number of significant wavelengths converged to 6 as advertised.
CMFIL
A four-wavelength system was selected on the basis of compromising per-
formance and complexity. This choice is later justified as explained in the introduction
and illustrated by Figure 16. In this case, the number of wavelengths used is not
C-5
-------
sufficient to null out all of the interferents. Therefore, the strength (variance) of '
each inh'rfoivnt must be estimated and :i linear weight is computed Hint gives llu-
best performance on UK- average. The di-tails of this nu-Uuxl ;nv di-si-rUn-d in
Appendix B. A program (MFIL) that computes the linear weights and
SNR's for a given set of wavelengths using this algorithm is described in Reference
17. This program was modified (CMFIL) in order to evaluate all combinations of
N wavelengths taken M at a time and to order the combination according to SNR.
Table V shows the input data that was used for the CMFIL program. The
wavelength set was obtained from the top 9 wavelengths of "the LWSP output combined
with the ethylene line. The interferent CL variances listed in Table I were estimated
from the best guesses available of the environment. The CMFIL output listing that
was obtained with the input data of Table V and the spectral data in Tables I and II
is shown in Figure 15. As expected, the P12 and P14 (5 and 6) which correspond
to the peak ozone absorption, appear in all of the highest rankings. The RIG, R14,
and P24 (3, 4, 8) lines also predominately appear in all of the highest rankings and
are therefore indicated as good reference lines. The combination 4, (5, 8, 10 (R14,
P14, P24, P14) was selected as the combination that gave the highest SNR while
retaining the ethylene line.
By examining the linear weights associated with the output listing, the
performance of 2 and 3 wavelength systems was estimated and is shown in
Figure 16. As can be seen, a three-wavelength system (which we essentially
have at the present due to the retention of the 10. 5321 micron ethylene line)
provides near optimum performance with a minimum of complexity. A two-
wavelength ozone system results in about 1/2 the sensitivity.
C-6
-------
WAVELENGTH SET
O
i
001-020
001-100
TARGETS (RECORD
R30
R18
R16
R14
P12
P14
P20
P24
P30
P14
NO.)
9.219690 microns
9.282444
9.293786
9.305386
9.488354
9.503937
9.552428
9.586227
9.639166
10.532080
Ozone 03 (244)
INTERFERENTS (RECORD NO.)
Neutral (201)
Carbon Dioxide CO2 (243)
Ethylene C2H4 (245)
Ammonia NHg (246)
Water Vapor HjO (244)
DETECTOR NOISE VARIANCE
10
-5
VARIANCE (CL) COMMENTS
1.0 If T log-normal, 4:1 variation 50% of time
10cO 10% x 320 ppm x 1 km
25 x 10'6 50 ppb x 1 km
25 x 10"6 50 ppb x 1 km
)06 -^58 - 100% RH at 73 degrees F x 1 km
Table C-l. CMFIL Input Data
-------
APPENDIX D
LWSP - LASER WAVELENGTH SELECTION PROGRAM
PURPOSE
To select those laser lines from the CO laser bands which arc best for
£
etection of agents in the presence of noise.
GIVEN
a. Spectra of-C12O l6 laser, including relative power at each wavelength
Lt
b. Spectra of agents and interferents.
c. Relative importance of each agent.
COMPUTE
The weight which should be applied to each line for optimum S/N and the time
fraction (power) to be allocated to each line.
DATA FORMAT
Same as MFIL (Reference 17). Laser data can be input by either cards or
tape. Individual and combined spectra are in the library.
D-l
-------
PROGRAM OPERATION
Deck setup and operation are almost identical to that for MFIL. Fewer
cards are required and a laser must be specified, along with the usual targot
and interferent list.
There are two modes of operation. One uses independent wavelengths for
each target and optimizes the weights for each individually. The second me!,h">d
uses a common set of wavelengths and involves the use of an average weight in the:
iteration loop which causes the weights for each target to move move nearly together.
If this option is chosen, the relative importance of each target is specifier! by use
of the "AVG" card.
Five iterations will be made unless otherwise specified on the "G" card. If
cards are used for laser data, they must contain wavelength and power, in th;it or.hr,
in (7X2F8.4) format. A blank card (wavelength = 0) terminates reading of the laser
data. These cards must immediately follow the "LASER/CARD" command.
COMMAND LIST FOR LWSP
COMMAND TYPE XI X2 X:3 X4
X5
LASER
LASE R
TRGTS
IFNTS
NEVT
AVG
RESET
G0
END
Li b. Scan No
CARD (Follow by A, pin (7X2F8.4) format, term, with blank t-irclj
' Same as MFIL (Reference 17)
w w
-L £
w.
(ZER0) K-plot K-A/N K-iter NDP
W
D-2
-------
MATHEMATICAL METHOD
1. Construct Interferent Matrix
QMIj, I2 ..... 1tt) (n x m) (D-])
where the columns are n- dimensional interferent absorption coefficient vectors, n
is the number of wavelengths available, and m < n is the number of interferents.
2. Initialize Average Weight and Weight for Each Target
(0) (0) =
W =W.
lt P2 ..... Pn)' , i = l, 2 ..... p (D-2)
.1.
where p is the number of targets and P. is the relative power at wavelength j
3. Compute the Diagonal Matrix, Scale Factors, and Time Intervals
for each iteration k=0, 1, 2,..., Kas follows:
a. Independent Wavelengths
Compute the diagonal matrix
v.. (n x m)
1 J j • 0 •
and scale factor
D-3
-------
0
- Y |W
L i
J 1
for each target i = 1, 2 p where W. (k) = (W , (k), W „ (kl W (k))'
~ i il i2 in
th
is the k iteration weight vector for target i. Also, compute the time fraction
(power) allocated to wavelength j
T I /P F "<>
U ' U. • ' j >.
for each target i = 1, 2, ... , p.
b. Common Wavelengths
Compute the diagonal matrix
D(k) - dia P. W. (k)
diag (P. W. )
and scale factor
n
y
f—
Fx
.1 3
where W = (W , W W )' is the kl iteration average weight.
— i £ n •
D-4
-------
In this case, it is only necessary to perform the computation once for each
iteration. The common time fraction allocated to wavelength j is computed as
T (k) = w (k) /P F . (D-S)
J J J j
4. Compute Next Weight Vecto-r
W. . D.
for each target absorption coefficient vector A. , i - 1, 2 ..... p. (In the common
mode, D. is replaced by D in the above).
5. Compute Relative Signal -To -Noise Ratios
SNR =(W'(ktl)A /F) 1/2 <
i ~i -i i
for each target i. (Again, in the common mode, F. is replaced by F in the above)
C. Normalize Each Weight to an Absolute Peak of ].
7. F>rint Weights. SNR's. and Time Fractions for each Target
D-5
-------
-~- Cotviputo Average Weight
P
w(k ' n - r- K.|W. (k ' 1}| (
i ---- 1
where the K.'s are weighting factors corresponding to the relative importance of
each target.
i>. Print _Sum ma ry_and_ Comp u te_Next__Ite_rati_on
Go to Step 3 and repeat until k - K.
10. Compute Target Cross-Response Matrix R (p x p)
with elements
R - GW (K)A L i - 1 2
lx^j [-J - i' »J i> Z
where
Gr SNRi ("7V"V "I-2 f
JLL-_C^mjgjjte Interfercnt Cross-Response Matrix U (p^ xjlll_
with elements
j = 1, 2, ... , m
12. Compute Next Iteration (Go to Step 3)
A flow chart describing these operations is shown in Figure D-l.
D-G
-------
Step
Loop
Input Data
*
(i)
Read Wavelengths, Powers,
Targets, Interferents
(2)
Initialize Weights
(3)
Compute Diagonal Matrix,
Scale Factors, Time Intervals
Mode
Switch
(4)
Compute Next Weight
Common
(5)
Compute SNR
(6)
Normalize Weights
i
(7)
Print Weights, SNR's, Time Fractions
(8)
Compute Average Weight
(9)
Print Summary
(10, II)
Compute Cross-Responses
Print Absorption Coefficients,
Cross-Responses
Stop; End
Independent
Target
Loop
Plots
Figure D-l. FLOW CHART OF LWSP PROGRAM
D-7
-------
TECHNICAL REPORT DATA
(Please read fnuruftions on the reverse before completing)
I. REPORT NO.
EPA-650/2-74-046-a
A. Tl VLE AND SUBTITLE
Development of a Gas Laser System to Measure Trace Gases
by Long Path Absorption Techniques, Vol. I: Gas Laser
System Modifications for Ozone Monitoring
. AljT
HOR(S)
8. PERFORMING ORGANIZATION REPORT NO.
S.E. Craig, D.R. Morgan, D. L. Roberts, L.R. Snowman
9. PERFORMING ORGANIZATION NAME AND ADDRESS
General Electric Company, Ordnance Systems,
Electronic Systems Division, 100 Plastics Avenue, Pittsfield,
Mass.. 01201
12. SPONSORING AGENCY NAME AND ADDRESS
U.S. Environmental Protection Agency
National Environmental Research Center
Research Triangle Park, North Carolina, 27711
3. RECIPIENT'S ACCESSION-NO.
5. REPORT DATE
June 1974
6. PERFORMING ORGANIZATION CODE
OS 74-13
10. PROGRAM ELEMENT NO.
1A1010 (26ACX)
11. CONTRACT/GRANT NO.
68-02-0757
13. TYPE OF REPORT AND PEHIOD COVLRtO
Final
14. SPONSORING AGENCY CODE
15. SUPPLEMENTARY NOTES
Volume II of the set is being published as EPA-650/2-74-046-b
16. ABSTRACT '• ~~~~
Modifications of a gas laser system for long path monitoring of trace atmospheric
constituents by infrared absorption are described. Modifications were made in preparation for
an ozone field measurement program reported in Volume II wherein path monitor data were
compared with those from a point monitor moved along the optical path. System modifications
included incorporating a digital signal processor in the system and a spatial filter in the laser
beam. Spectral studies of ozone, carbon dioxide, water vapor, ethylene and ammonia are
presented in connection with the selection of laser wavelengths used in the system to discrimi-
nate ozone effects from interferences. Design considerations and a proposed configuration
for an isotopic C©2 laser are presented.
7.
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
Lasers
Atmospheric Absorption
Ozone
Air Pollution Monitoring
3. DISTRIBUTION STATEMENT
Release Unlimited - Copies Available from
NTIS; APTIC (EPA)
1). IDENTIFIERS/OPEN ENDED TERMS K. COSATi i-'icld/din
I LAMS
Methodology for Point
Monitor, Path Monitor
Comparisons
VJ. SECURITY CLASS (This Krptirl>
Unclassified
1705
20. SECURITY CLASS (This page)
Unclassified
I 21. NO. Oi- f'AGES
123
22*. PRICE
EPA Form 2220-1 (9-73)
D-8
------- |