Continuous-wave laser spectrometer automatically alignedand continuously tuned from 11.8 to 16.1 mm by use ofdiode-laser-pumped difference-frequencygeneration in GaSe

Roger S. Putnam and David G. Lancaster

We report a fully automated mid-IR difference-frequency spectrometer with a spectral resolution under70 MHz pumped by a pair of conventional room-temperature 800–900-nm diode lasers. 0.1 mW oftunable cw radiation is produced from incident-diode powers of 120 and 75 mW. The system hascomputer-controlled beam alignment with compact CCD cameras, motorized mirrors and positioners toobtain 0.01° crystal-angle positioning, 4-mm beam overlap at the nonlinear crystal, and automated diodelaser beam collimation. Computer-operated frequency control uses temperature tuning and currenttuning of the free-running diode lasers. The system has been demonstrated by successfully scanning,without any human intervention, 64 randomly selected acetylene absorption lines between 12 and 15 mm.Spectral scans of ammonia are also presented. This mid-IR spectrometer is suitable for fully automatedspectroscopy of an unlimited list of mid-IR frequencies and has the potential to detect any trace gas thathas an acceptable absorption line within the large tuning range. © 1999 Optical Society of America

OCIS codes: 010.1120, 120.6200, 190.2640, 300.6360, 140.2020, 300.6340.

agaasbTtgrcslars

1. Introduction

Agile and precise measurement of trace air-pollutantconcentrations and fluxes is a critical requirement forboth the elucidation and the remediation of environ-mental problems. This is true for problems rangingfrom global-scale issues such as stratospheric ozonedepletion and greenhouse gas buildup to local pollu-tion issues such as urban smog and toxic contamina-tion. Examples include SO2, NH3, H2O2, HNO3,HNO2, N2O, CH4, O3, CO2, NO, NO2, HCN, C2H2,C2H6, CCl4, H2CO, HNO4, ClO, OCS, PH3, HOCl, andchlorofluorocarbons. Trace-gaseous-species detec-tion and quantification are also important for thecharacterization and the control of combustion-driven energy and propulsion sources and key indus-

When this research was performed, the authors were with Aero-dyne Research, Incorporated, 45 Manning Road, Billerica, Massa-chusetts 01821-3976. D. G. Lancaster is now with Rice QuantumInstitute, Rice University, 6100 Main Street, Houston, Texas77005.

Received 31 March 1998; revised manuscript received 12 June1998.

0003-6935y99y091513-10$15.00y0© 1999 Optical Society of America

trial processes, including plasma and chemical vaporreactors for semiconductor device production.

Tunable IR laser differential absorption spectros-copy has proved to be a powerful technique for accu-rately determining the real-time concentrations andfluxes of trace pollutants in both point-monitoringand open-path configurations.1,2 Some of these sys-tems utilize tunable III–V diode lasers in the 0.7–2.0-mm near-IR region, but the most sensitivesystems use cryogenically cooled lead-salt diode la-sers in the 3–20-mm mid-IR region since molecularbsorption line strengths are typically 20–200 timesreater in this wavelength region where most pollut-nt molecules have their fundamental vibrationalnd rotational bands. However, these mid-IR lead-alt diode lasers have a number of operational draw-acks that restrict the utility of current instruments.hese include the need for cryogenic cooling, narrowuning ranges between mode hops, variable and de-raded tuning characteristics that are due to mate-ial diffusion, and narrow single-diode frequencyoverage which limits the system’s flexibility in mea-uring multiple trace species. Moreover, presentead-salt laser-based systems lack automated opticallignment and frequency tuning, which then oftenequire the attention of technically sophisticated per-onnel.

20 March 1999 y Vol. 38, No. 9 y APPLIED OPTICS 1513

ms1

s

g9wtlmq

pfai

1

Recently, quantum-cascade mid-IR lasers havedebuted3 in the laboratory at progressively highertemperatures4 with the expectation that room-temperature operation will become practical. Thesehave been demonstrated under laboratory conditionsat 217 to 16 °C with a 1% duty cycle to scan CH4 andN2O spectrally and to measure N2O accurately near8 mm despite a 720-MHz linewidth that is due topulsed operation.5 Also, cw operation of opticalparametric oscillators has been demonstrated in pe-riodically poled LiNbO3,6 which is limited by absorp-tion to produce wavelengths below 4.8 mm. Another

id-IR competitor is the Fourier-transform infraredpectrometer with resolutions generally greater thancm21, which is roughly 20 times the Doppler- and

the pressure-broadened linewidth that we use for op-timized laser trace-gas detection with a single ab-sorption line.

Difference-frequency generation ~DFG! was firstdemonstrated for molecular spectroscopy by Pine,7who used an Ar1 laser and a tunable dye laser in aLiNbO3 crystal. The system was tunable from 2.2 to4.2 mm, had a resolution of 15 MHz, and was dem-onstrated by scanning the absorption lines of H2O,NH3, CH4, and N2O. That system had the advan-tages of being cw and smoothly and continuously tun-able; but the size, weight, and power consumption ofthe drive lasers posed significant disadvantages forportable spectrometer applications. It is relativelysimple to obtain high power by means of DFG withpulsed-drive lasers, but the resulting broad spectraare inadequate for measuring 30-MHz-wide absorp-tion features and for separating small trace-gas ab-sorption lines from the wide absorbing wings ofpredominant species.

The availability of high-power, narrow-bandwidth,room-temperature diode lasers8 now makes possiblea compact and widely tunable mid-IR spectrometerby DFG with two diode lasers,9–12 and the emergenceof high-power diode lasers in the 600–1500-nm re-gion combined with nonlinear materials such asZnGeP2, AgGaSe2, AgGaS2, GaSe, and periodicallypoled LiNbO3 provides access to the entire 2–18-mmpectral range. Our choice of GaSe13 for the mid-IR

spectrometer results from the combined require-ments of a large nonlinearity ~54 pmyV!,14 a realtransparency at 800–900 nm for the drive beams andat 7–16 mm for the mid-IR output, and a birefrin-ence that permits angle tuning any pair of 800–00-nm drive wavelengths into a 7–16-mm idleravelength. The flaw in this choice appears to be

he GaSe crystal quality, which can limit the effectiveength of the nonlinear interaction to several milli-

eters, although an improvement in crystal-latticeuality by doping has been reported.15 Our require-

ment that crystal-angle tuning be used to ~critically!hase match any realistic pair of drive frequenciesrom the two diode lasers results in a shorter inter-ction length and a lower conversion efficiency thans available with 90° noncritical phase matching.16,17

However, our choice of angle tuning permits gener-ating the particular desired mid-IR frequency by any

514 APPLIED OPTICS y Vol. 38, No. 9 y 20 March 1999

pair of diode laser frequencies that are available andhappen to have the correct frequency difference:This is a great advantage when we are dealing withsimple diode lasers that do not have smooth temper-ature tuning but show mode hops; Fig. 1 shows thatthere are typically hundreds of frequency pairs avail-able from the two diode lasers to produce any partic-ular IR frequency when a diode temperature-tuningrange of 0–40 °C is used.

2. System Concept

The system was conceived as a prototypical universaltrace-gas monitor and laser spectrometer for themid-IR region between 7 and 16 mm and is intendedto require no human intervention while scanning anunlimited list of spectral regions, to record the spec-tra, and, in the case of trace-gas detection, to find andto extract the fractional absorption of the chosen line.

The mid-IR frequency is generated by DFG in anangle-tuned GaSe crystal driven by a pair of diodelasers at 809 and 856 nm. The temperature-tuningcharacteristics of these free-running diode lasers areautomatically recorded by the system, and the result-ing look-up tables are used in selecting the best pairof diode laser frequencies to produce any desiredmid-IR frequency. Compelling the diode lasers toproduce the chosen frequencies is a complex processand is made possible by the fact that more than 100similar pairs will give the same desired mid-IR fre-quency and that the use of angle tuning ~instead of90° phase matching! permits any such pair to beutilized ~see Fig. 1!.

The past need for technically sophisticated person-nel to tune the laser frequency over the gas absorp-tion line as well as to realign the optical systemoccasionally has driven the requirement for fully au-tomated tuning and alignment, including the task ofidentifying the correct absorption line. Our present

Fig. 1. Experimentally measured tuning data for the two diodelasers at the 130-mW power level was recorded from 0 to 40 °C.All possible combinations of the available frequencies werechecked, and the number of different pairs found that will producea given mid-IR frequency is shown in the figure. Hundreds offrequency pairs produced by the two diode lasers are available toproduce any desired mid-IR frequency.

slt

draacp0mettpt

tmcMEtcrl~;pb

oeeZata

system sets the mid-IR frequency with an absoluteaccuracy of ;50 MHz, uses pattern recognition on thepectral scan to locate the center of the absorptionine, and uses a nonlinear fitting routine to extracthe actual fractional absorption.

The correct overlap and angular alignment of therive beams in the GaSe crystal, as well as an accu-ately predicted and subsequently optimized crystalngle, are required for DFG. These requirementsre compounded by the wide tuning range that bothhanges the IR exit angle at the crystal and results inhysical movement of the lasers diodes over the large–40 °C tuning range. The optical system is auto-atically aligned by four CCD cameras controlling

ight motorized mirror mounts, by calculation ofhe frequency-dependent crystal angle and theemperature-dependent diode laser collimating-lensosition, and by optimization routines that maximizehe IR output power.

3. Experimental Setup

A. Difference-Frequency Generation Layout

A schematic of the DFG system is shown in Fig. 2,including pump ~extraordinary-beam! and signal~ordinary-beam! diode lasers that are nominal150-mW narrow-bandwidth diodes at 856 and 809nm ~SDL 5400 series!. Two Newport 700C thermo-electrically controlled mounts are driven by an IEEEgeneral-purpose-interface-bus ~GPIB! interfacedmodular dual-diode controller ~ILX Lightwave LDC-3900!. The diode outputs are collimated by Geltechaspheric lenses ~ f 5 2.75 mm, N.A. 5 0.65! that areheld in a threaded rotating mount that allows fineadjustment of the diode–lens spacing by computer-controlled linear translation stages ~National Aper-ture!: This was necessary to compensate for thevariation in the diode position as a function of thethermoelectrically cooled mount temperature. Thediode laser beams are reflected off computer-

controlled mirrors ~New Focus! and pass through op-tical isolators and ly2 wave plates to the correctpolarizations for type I phase matching in the GaSecrystal ~Eksma18!. A portion of the mixing beamsare reflected by an antireflection- ~AR! coated glasswedge to a CCD camera and to a gradient-index lensthat couples into an optical fiber. The CCD image isused for monitoring the beam collimation and align-ment. The fiber-coupled light reaches a wavemeterfor diode frequency monitoring ~Burleigh 1500; 0.001-cm21 accuracy! and a scanning Fabry–Perot spec-rum analyzer with variable mirror spacing foronitoring the diodes’ spectral characteristics, espe-

ially the relaxation sidebands spaced at 1–3 GHz.otorized shutters allow either beam to be selected.ach diode laser beam is passed through a cylindrical

elescope to reduce the horizontal dimension and toorrect astigmatism, and a vertical Gaussian fieldadius of 1.2 to 1.6 mm is used at the 25-cm focusingens, which produces a focused spot size of ;50 mmvertical field radius!. The calculated optimum is44 mm in the 5-mm crystal. The two orthogonallyolarized diode beams are combined in a polarizingeam splitter.The angle-phase-matched GaSe crystal is mounted

n a compact motorized rotation stage ~National Ap-rture! with an absolute accuracy of 0.01°. The gen-rated mid-IR idler beam is collimated by an f 5 5 cmnSe lens and is separated from the drive beams byn AR-coated Ge filter. The mid-IR beam passeshrough the 10-cm absorption cell and is focused byn f-2.5-cm ZnSe lens onto the liquid-N2-cooled HgC-

dTe detector ~Kolmar! that is optimized for 10–15mm. One diode laser drive beam is mechanicallychopped, and a lock-in amplifier is used with thedetector.

The internal phase-matching angle for type I DFGin GaSe is 14.8° at an idler wavelength of 13.9 mm.This results in a 45.7° external angle because the

Fig. 2. Optical setup includes twotemperature-controlled diode lasers with auto-mated collimation at 809 and 856 nm, eightmotorized picomirrors, single-mode fiber cou-pling, optical isolators, horizontal cylindrical-lens telescopes, three CCD cameras, apolarizing beam splitter ~PBS! for combiningthe drive beams, a motorized rotary mountwith a 5-mm GaSe crystal, a deflection mirroron a swing arm, a 25-cm focusing lens, a 5-cmZnSe mid-IR collimating lens, a 2.5-cm detec-tor focusing lens, an AR-coated Ge filter, and a10–16-mm HgCdTe ~MCT! detector. FP,Fabry–Perot spectrum analyzer.

20 March 1999 y Vol. 38, No. 9 y APPLIED OPTICS 1515

1

pure GaSe crystal is soft and cannot be cut at anangle but cleaves in a micalike fashion perpendicu-larly to the c axis, which then also produces, with alarge optical index of 2.8, astigmatic elliptical beams.The external angular-phase-matching sensitivity forour 5-mm GaSe, which has an ordinary-ray idler, wasmeasured as 0.4° FWHM.

B. Electronics

The 166-MHz Pentium control computer includes aViewpoint Software Solutions framegrabber and soft-ware for the various compact CCD alignment cam-eras, a nuLogic motor controller board to adjust thetwo collimating lenses and the crystal angle througha National Aperture servoamplifier, an IEEE GPIBinterface board, and an input–output ~IyO! board foranalog output to scan the diode lasers’ currents, foranalog input from temperature sensors and the am-plified Fabry–Perot optical detector, and for digitaloutput to the various small motors and voice coilactuators that operate beam blocks and mirror beamdeflectors. A New Focus controller operates theeight motorized mirrors from the GPIB, which alsocontrols the Burleigh 1500 Wavemeter with 0.001-cm21 resolution and the ILX Lightwave LDC-3900dual-diode-laser current and temperature controller.The focused beams’ approaches to the GaSe crystalare observed on a pair of black-and-white monitors,and the mid-IR detector signal is amplified by anordinary lock-in amplifier.

4. Experimental Results

A. Diode Tuning

The frequency of the laser diodes is principally con-trolled by changes in the diode temperature and cur-rent. The single-longitudinal-mode lasers arecharacterized by somewhat unpredictable mode hop-ping and significant hysteresis behavior as a functionof temperature, as shown in Fig. 3 of a typical100-mW device ~SDL-5410!. Figure 3 also revealsconsistently smooth temperature tuning on a partic-ular longitudinal mode. An alternative to simpletemperature tuning is to turn the diode current offmomentarily, which allows the laser to restart on oneof the several available longitudinal modes. Thisboth removes any temperature-tuning hysteresis andtypically provides random access to more modes if thecurrent is repeatedly interrupted, as shown in Fig. 4.This technique is most effective if the current inter-ruption is kept below 10 ms, which reduces activethermal-tuning effects at the diode junction.

The pseudo-AR coating ~;1%! on the emitting facetof the 150-mW SDL-5400 lasers increases the lasers’sensitivity to externally reflected fields, resulting inmode hops, frequency pulling, and noisy behavior asindicated by the sidelobes at the 1–3-GHz relaxationoscillation frequencies. These are revealed by ascanning Fabry–Perot spectrum analyzer and by sub-tle, but detectable, amplitude fluctuations. We ob-served frequency pulling at reflected fractional powerlevels of ;1028. The closely spaced laser collimat-

516 APPLIED OPTICS y Vol. 38, No. 9 y 20 March 1999

ing lens provides significant optical feedback, whichgreatly determines which laser modes are availableand which are noisy. This fact is exploited to pro-vide better control over the tuning and is discussed indetail in Section 5. Removing the optical feedbackfrom the diode laser window by removing the windowchanges the erratic temperature-tuning curve in Fig.5 into a smooth and straight tuning curve.

The two diode lasers are characterized by a com-

Fig. 3. Temperature tuning of a 100-mW diode laser with apseudo-AR coating on the emitting facet shows hysteresis andrandomly appearing mode hops. The hysteresis was removedwhen the current was momentarily lowered below threshold,which permits the laser to restart: This also permitted reachingseveral different modes without changing the temperature. Thestability of these mode choices was strongly influenced by opticalfeedback and was controlled by the position of the collimating lensafter the laser window was removed.

Fig. 4. Temperature-tuning characteristic of the 150-mW SDL-5400 diode laser was measured at a constant 130-mW power withthe laser momentarily dropped below threshold to restart at eachdata point. Multiple modes are available at each temperature butnominally run one at a time. Which mode will appear after acurrent interrupt is random but is influenced by the phase of thefield reflected from the collimating lens, which we mechanicallycontrol to force the laser to a particular mode. This characteriza-tion curve, with 0.2 °C steps, is automatically recorded and is usedto find a preferred frequency pair for any required mid-IR fre-quency.

lttdc

syaa

lG

puterized tuning sequence that steps the tempera-ture by 0.2 °C from 0 to 40 °C and steps the currentaccording to a predetermined matching ramp tomaintain 130 mW. At each temperature step thecurrent is repeatedly interrupted and the various re-sulting modal frequencies recorded. Each interrup-tion also includes a change in the collimating-lensposition. The data are automatically reduced to re-move redundancies, and the result is shown in Fig. 4.The large temperature range used in the laser char-acterization process also requires automated adjust-ment of the laser lens to keep the beam collimated,and a CCD camera and New Focus picomirror to keepthe beam aligned into a single-mode fiber for thewavemeter. Gaining access to several modes ateach temperature, as shown in Fig. 4, provides denserand more consistent spectral coverage across the40 °C tuning range.

The measured diode laser tuning curves are gen-erally repeatable with multiple frequencies ~modes!available at each temperature, but exhibit some ran-domness as well as some drift due to aging. Thedesired mode is obtained by repeated restarts of thelaser with changes in the position of the collimatinglens. The randomness is mostly avoided by thegrading of every laser frequency in terms of its sta-bility during the laser characterization stage and bythe fact that there are hundreds of acceptable fre-quency pairs available. The slow frequency driftthat is due to aging is accommodated by a change inthe diode temperature if a histogram analysis of therecently available frequencies shows a collective shift~,1.5 °C!.

B. Spectral Resolution and Scanning and DemonstrationSpectra

Our system can produce continuous spectral scansbetween 11.8 and 16.1 mm by combining sections, asshown in Fig. 6 for NH3. The resolution is betterthan 70 MHz. This is adequate for identifying and

Fig. 5. The temperature tuning characteristic of one diode laser isvastly improved by removing the laser window. The SDL-5400series 150 mW diode lasers have a pseudo AR coating on theemitting facet which increases the influence of reflected opticalfields.

discriminating most Doppler-broadened molecularresonances. This continuous-scanning ability is ac-complished despite the use of free-running near-visible lasers. The combined temperature tuning ofthe two diode lasers limited to the 0–40 °C region is300 cm21 that, with a better selection of their centerfrequencies, would allow tuning from 10.9 to 16.1 mm,imited by the 16-mm roll-off of our detector. Fur-her, the addition of a third laser could triple theuning range; for example, 790-, 850-, and 1015-nmiode lasers can pairwise provide DFG convenientlyentered at 3.6, 5.2, and 11.1 mm.

The system has also passed the definitive test ofpectrally scanning a list of 64 randomly chosen acet-lene absorption lines between 12 and 15 mm withoutny human intervention. Figure 7 shows a singlecetylene absorption line at 12.4 mm.The current tuning rate of the 150-mW SDL diode

asers running at the 130-mW power level was ;1.6HzymA. The smooth spectral scanning was ac-

Fig. 6. NH3 spectrum that was produced by combining a set ofcurrent-tuned scans, each of which was produced at a different~temperature-tuned! center frequency. The vertical displace-ments of the scans are intentionally not corrected. The HITRAN-based prediction of the spectra at the top of the figure correspondswell with our measured spectra.

Fig. 7. Example of a low-pressure acetylene absorption line at12.4 mm.

20 March 1999 y Vol. 38, No. 9 y APPLIED OPTICS 1517

cqpalscs

p

Tas

vpld8tcm

1

complished when the analog modulation inputs onthe dual-laser current controller were driven from theanalog output of the computer IyO board.

C. Tuning Range

Figure 1 shows that any frequency within the overalltuning range is accessible despite the mode-hoppingbehavior of the free-running drive lasers. This isdemonstrated in Fig. 8, for which a list of 64 ran-domly chosen acetylene absorption lines between 12and 15 mm was provided and the system automati-ally produced mid-IR sequentially at each fre-uency, spectrally scanned the absorption line,recisely located the line, and reported the fractionalbsorption without any human intervention. Noines were missed. The system checks for errorsuch as a mode hop during the scan and will thenhoose new frequency pairs until a successful spectralcan at every chosen frequency results.The absolute frequency of the generated mid-IR

ower is known with an error of ;50 MHz. Oneconsequence of this high precision is that a strongabsorption line will prevent the mid-IR from beingdetected during the alignment and optimizationsteps; therefore a small frequency shift is added dur-ing the tuning phase and is then corrected when theabsorption line is scanned.

The sequence of spectral scans is also recorded inconvenient spreadsheet form, which is in turn pre-sented by a program that permits the user to steprapidly through the spectra to identify particular fea-tures, for example, optical gain from a list of potentialbut yet untested plasma gain lines.

D. Power

The measured mid-IR output power is ;0.1 mW.he expected power is 0.5 mW. The input powersre roughly 100 mW each, which makes the conver-ion efficiency ;1026. Recently doped GaSe15 has

shown an improved conversion for thicker crystals

Fig. 8. Mid-IR power generated during the fully automated test of64 randomly picked acetylene absorption lines shows a generaldecrease with increasing wavelength that is due to the inversesquare wavelength dependence. The vertical scatter reveals theimperfect behavior of the automated power peaking system.

518 APPLIED OPTICS y Vol. 38, No. 9 y 20 March 1999

but no change for thin crystals, which suggests thatour conversion problem is a loss of phase matchingover the 5-mm length due to lattice imperfections.

Figure 8 shows the mid-IR power versus the wave-length. The automated alignment for maximumpower is revealed to be imperfect by the ~vertical!ariation in power at a given wavelength. Theower is expected to decrease with increasing wave-ength because of an intrinsic inverse square depen-ence on the mid-IR wavelength. The power in Fig.is corrected for the detector response, but the de-

ector rolls off sharply at 16 mm, which limits oururrent spectrometer to an upper wavelength of 16.1m.

E. Automation

The system is fully automated, including the selec-tion of the optimal drive laser frequencies, tuning thediode lasers to the correct frequencies, aligning thebeams in the nonlinear crystal, setting the crystalangle, detecting the IR signal, optimizing the mid-IRpower, spectrally scanning the region, and locatingand evaluating the absorption line found in the scan.Errors in optical alignment are corrected from thebeginning of the system to the end of the system, fromthe near-visible diode laser drivers to the mid-IR de-tector. The automation of the whole system wasthoroughly exercised during the test scan of the 64acetylene absorption lines between 12 and 15 mm,which was accomplished without error.

This successful automation gives the system itsprimary virtue, the ability to perform an unlimitedseries of laser-based spectroscopic tasks more rapidlyand probably with a much greater chance of successthen can be accomplished by human control of thesystem. This virtue permits testing, for example, allthe possible energy-level pairs in a He–Ne plasma foroverlooked IR gain lines, a task undoubtedly previ-ously rejected in part because of the amount of hu-man labor involved. The successful automation nowmakes possible many such previously unapproach-able problems. The system is also a large step to-ward a universal laser-based air-pollution monitor,which sequences through a large set of known orsimply suspected trace gases. The present status ofhigh-sensitivity laser-based detection of multipletrace gases is limited to, at best, two or three differentmolecules that are conveniently close in frequencyand can be captured by a single current scan of alead-salt laser or by multiple scans of multiple la-sers.19

5. Algorithms

In this section the LabView software language, auto-mated optical adjustments, tuning the free-runningdiode lasers, and processing of the spectral-scan dataare discussed.

A. LabView Software

The spectrometer was built with LabView, which isan object-oriented programming language in whichicons that represent subprograms have ~electriclike!

4ticdfspp1

uorcsspoffrbaltgdt

pGuew

f;bicebbaahem

2Tarp2iwpblm

tisitiptvawts

onqoTlusaFcs

aterrtt

terminals that are wired together to indicate vari-ables and data flow. This differs from a command-line language in that program steps can occur in anysequence, waiting only for the necessary new data toappear. This data-dependent language was ex-tremely effective and certainly seemed to simplifyand make possible, in the limited time span, the de-sign and the testing of the complicated spectrometerfor which 300 programs were written. The languagealso greatly facilitated data manipulation and graph-ing, communication with laboratory instruments,and the IyO of analog signals.

B. Automated Optical Alignment and Optimization

1. Drive-Beam Alignment into the NonlinearCrystalFour motorized mirrors are used to set the positionand the angle of the two input beams. Figure 2shows the focusing lens separated by one focal dis-tance from the nearest picomirror: this spacing per-mits this mirror to move the position of one beam atthe crystal without changing its angle of approach.The optimum-approach angles of the two drive beamsare different primarily because of the different opticalindices for the ordinary and the extraordinary beamscombined with the large approach angle ~roughly5°!. Also, the two beams are displaced at the crys-al entrance to partially correct for the walk-off ~typ-cally 48 mrad! of the extraordinary beam inside therystal. The second adjustment mirror for eachrive beam is several focal distances away from theocusing lens to minimize its effect on the beam po-ition at the crystal while adjusting the angle of ap-roach. The same angle of approach and beamositions are used throughout the tuning range of1.8–16.1 mm.The compact CCD cameras ~without lenses! are

sed to gauge the position and the angle of approachf the two drive beams. A motorized deflector mir-or is inserted into the focused beam path, and oneamera is positioned at the equivalent ~focused! po-ition of the crystal to gauge the beam position. Theecond camera is imaged with a simple lens to aosition 60 mm before the crystal to gauge the anglef approach. The deflector mirror is constructedrom a magnetic-head drive mechanism as extractedrom a failed computer hard drive and has excellentepeatability. The CCD cameras feed a framegrab-er board inside the control computer, and the im-ges are processed to determine the diode laser beamocations. The processing consists primarily ofhresholding and low-pass filtering to remove back-round noise and the extended wings of the imperfectiode laser beam shapes that can significantly affecthe two-dimensional center-of-mass calculation.

The beam-positioning corrections are passed to theicomirror controller ~New Focus! through an IEEEPIB, and the cycle is repeated, typically six times,ntil one-half pixel errors result ~;4 mm at the cam-ras!. The complete two-beam-alignment process,hich uses a 166-MHz Pentium CPU and a slow

ramegrabber that transfers 1 byte per cycle, takes2 min. No special signal processing for this feed-ack loop is necessary since no time constants arenvolved, other than paying attention to the size oforrection ~gain! during each cycle and the coupledffects of the tandem adjustment mirrors on eacheam. Since neither hysteresis nor nonmonotonicehavior is a problem in this particular beam-lignment feedback system, we can correct noise orctuator resolution problems by loosening the one-alf pixel requirement, by reducing the fraction of therror that is corrected during each cycle, and by usingore data averaging.

. Diode Laser Collimation-Lens Adjustmenthe 0–40 °C span used to temperature tune the 809-nd the 856-nm diode lasers excessively changes theelative position of the collimating lenses in the New-ort diode laser mounts. This was measured as0.47 mmy°C and was compensated for by a motor-

zed swing arm mounted on the rotating lens mount,hich was set based on the known diode laser tem-erature. The frame temperature, despite beingolted to an optical table, also affects the collimating-ens spacing and was measured to have a rate of 0.74my°C.The collimating-lens mount uses 80-turnsyin.

hreads and exhibits a rocking motion that is signif-cantly reduced when the threads are filled with atiff, nonmelting high-vacuum grease. The remain-ng rocking motion, plus any flaw in the lens concen-ricity with the laser, deflects the exiting beam, whichs corrected with one common CCD camera and oneicomirror each for the two diode lasers. This keepshe drive beams aligned as they subsequently tra-erse the optical table, as well as aligning them intocommon single-mode optical fiber that feeds theavemeter for absolute frequency information and

he scanning Fabry–Perot spectrum analyzer foridelobe-noise information.The collimating lens is also adjusted in increments

f ;0.125 mm to reduce the diode laser’s sidelobeoise ~typically at the relaxation oscillation fre-uency, 1–3 GHz! by a change in the phase of theptical reflection returning to the laser from the lens.his effect is significantly more pronounced when

ow-facet-reflectivity 150-mW SDL diode lasers aresed instead of natural-reflectivity 50-mW Sharp la-ers. The sidelobe-noise level is extracted from thenalog data stream that comes from the scanningabry–Perot by identification of the primary spectralomponent and a comparison of it with other peakspaced appropriately from the strongest peak.Finally, one infrequently used program correctly

djusts the collimating-lens position without regardo the expected temperature relationship by repeat-dly fitting a Gaussian profile to the beam shape andesetting the collimating lens until the proper beamadius is obtained. This program then updates theemperature-dependence formula that is used rou-inely to predict the proper lens position.

20 March 1999 y Vol. 38, No. 9 y APPLIED OPTICS 1519

smscapa

ilmplcrolpa

5Mpppesetpddacco

fioid

tpmnmpd

1TotpsdmfaTtstmt

2Taw

1

3. Automatic OptimizationThe laser spectrometer uses optical optimization rou-tines on only the crystal angle and the alignment ofthe mid-IR detector to maximize the infrared power.

A critical issue in optimizing a mirror or the crystalangle is the signal-to-noise ratio ~SNR!. For exam-ple, the automated observation that there is in factany mid-IR signal is accomplished by the require-ment of a minimum SNR, measured by a programthat samples the signal stream, operates a motorbeam block, samples the zero reference signal, andcalculates the standard deviation. This in fact mim-ics the human technique. If the SNR is too low, theintegration time is increased to a maximum of 60 s.Once the signal is found, a minimum integration timeis calculated and used while the optimization is per-formed. For example, if the signal level changes byless than the integrated-noise level, it is then treatedas a zero change so as not to provide confusing feed-back.

4. Optimizing the Crystal AngleThe angle-tuning sensitivity shows a FWHM of;0.4°. The phase-matched crystal angle is calcu-lated based on improved Sellmeier coefficients17 andet with a resolution better than 0.01° by the compactotorized rotary mount ~National Aperture! and a

equence that reduces mechanical hysteresis by firstausing the rotary mount to overshoot the desiredngle by a fixed amount and then second by ap-roaching the desired angular position slowly andlways from the same direction.We optimize the crystal angle by stepping the angle

n one direction until a believable decrease in signalevel takes place and then changing direction. A

inimum step size of 0.05° is used to reduce therobability of a false reversal that is due to a smallocal maximum. The measured signal levels andorresponding crystal angles are then fitted to a pa-abola, and the peak of this parabola is used as theptimum angle. This permits us to use a relativelyarge step size while also accurately locating theeak. Finally the formula used to predict the crystalngle is updated if its error is greater than 0.08°.

. Optimizing the Mid-IR Beam into the Detectoraximizing the detected signal by use of the two

icomirrors in tandem that feed the detector is com-licated by the fact that the signal shows multipleeaks because of an imperfect beam shape and etalonffects in the detector window. Our solution is totep the mirror in one direction, with a step size largenough to not be trapped by most local maxima, untilhe peak is passed, and to backstep once. This sim-le algorithm was found to be superior to fitting theetected signal and mirror position to a parabola as isone with the crystal angle, in part because, withoutposition encoder, the picomirror position is not ac-

urately known because of its randomly differentlockwise and counterclockwise speeds and becausef the strong local maxima that disturb the parabolic

520 APPLIED OPTICS y Vol. 38, No. 9 y 20 March 1999

t made with a limited number of data points. Theverall result is an imperfect optimization, as shownn Fig. 8 by the vertical scatter in the mid-IR powerata points at any particular wavelength.

C. Tuning

Tuning the two free-running 800–900-nm diode la-sers with an accuracy of 0.01 cm21 is a complicatedprocess and is described below. In short, two fre-quency look-up tables that give the optical frequen-cies of the diode lasers as functions of temperatureare compared and an exhaustive list is compiled ofpairs of frequencies that give the desired mid-IR fre-quency; the best frequency pair is then chosen interms of stability, which includes the current- and thetemperature-tuning ranges without a mode hop andfrequency repeatability, as well as sidelobe noise,probability of acquiring those frequencies, the small-est error in generating the desired mid-IR frequency,and the smallest required temperature tuning; thenthe lasers are set to the chosen temperatures andcurrents; the lasers’ frequencies are measured by thewavemeter ~0.001 cm21!, the laser current is momen-tarily kicked down below threshold to restart the la-ser, the collimating lens is moved in steps of 0.125 mmo change the optical feedback, and the desired out-ut frequency is tested for current tuning to resistode hops, for frequency stability, and for sidelobe-oise energy; and finally current tuning is used toove the two diode laser frequencies until the ex-

ected difference frequency is within 0.01 cm21 of theesired mid-IR frequency.

. Frequency Look-Up Tableshe look-up tables are previously generated by a scanf the diode laser temperature by 0.2 °C steps from 0o 40 °C while the current is changed to keep theower approximately constant at 130 mW. The la-er collimating lens is stepped six times, and with theiode current kicked below threshold before eacheasurement, the optical frequency, sidelobe-noise

raction, temperature-tuning range ~60.2, 60.5 °C!,nd current-tuning range ~62, 65 ma! are recorded.hese data at each temperature location are reducedo remove redundancies and to give a stability mea-ure, best sidelobe-noise behavior, and the number ofimes a particular frequency appears. This auto-ated laser characterization program also records

he data in the convenient spreadsheet form.

. Find Frequency Pairshe two frequency look-up tables for the two lasersre processed to find all combinations of frequencieshose difference is within 0.1 cm21 of the desired

mid-IR frequency. There are typically greater than100 such pairs, as shown in Fig. 1. A new list inspreadsheet form containing these pairs is assembledand handed to the program, which picks the best pair.

3. Choose Best Frequency PairThis program attempts to find a pair that matches allthe preferred criteria, but backs down the require-

c

dfcifsdctaftpcm

fda

lN

ments as needed in a predefined sequence. Thereare five levels of stability, two levels of mid-IR fre-quency accuracy, two levels of sidelobe noise, twolevels of frequency repeatability ~pick a mode that isdominant!, two levels of required temperaturehange ~stay near 25 °C if possible!, and two levels of

frequency deviation from the linear approximation tothe frequency–temperature map ~stay near the mostlikely center frequency for a given temperature!.

4. Force Lasers to Chosen FrequenciesThe free-running diode lasers with Fabry–Perot cav-ities can typically operate at several different longi-tudinal modes for each temperature and fixedcurrent. Our technique is to kick down the diodelaser current below threshold momentarily to permitthe laser to restart. The collimating-lens position isalso sequenced to improve the stability ~and proba-bility of acquisition! at the desired frequency. If the

esired frequency is acquired, the laser is checked forrequency wander, sidelobe-noise level, and stableurrent tuning. If the frequency is not acquired dur-ng the 48 attempts, a histogram analysis is per-ormed on the frequency errors to determine if amall temperature tuning ~,1.5 °C! could reach theesired frequency, assuming that the correct modean be reacquired; this accommodates a slow shift inhe frequency-tuning curve that appears to be due toging. Failure here to obtain the correct diode laserrequencies involves rerunning the program to choosehe best frequency pair, with the failed frequencyair~s! removed. Success here concludes with a fineurrent tuning of both lasers to achieve an initialid-IR frequency error below 0.01 cm21.

D. Spectral-Scan Processing

Present high-resolution trace-gas monitoring sys-tems that use lead-salt lasers require sophisticatedpersonnel to exactly set the laser frequency, which istypically accomplished by observation of a knownpattern of nearby absorption lines, and to position theprimary absorption line such that the data-processing algorithm can remain locked. Our tun-able mid-IR spectrometer performs these functionsautomatically.

Our mid-IR source is first set to be accurate to 0.01cm21 or 300 MHz, although it can be set to a higherresolution if needed. The frequency scan used in theacetylene demonstration of 64 absorption lines was;4.25 GHz each, which guarantees that the intendedabsorption line is within the scan and centered withan accuracy of ;7%. The precise position of the ab-sorption line in the scan data is determined by aFourier-transform matched filter, which is quite ef-fective under noisy conditions, especially if the widthof the absorption line used in the design of thematched filter is set to be equal to or, if not preciselyknown, to be larger than the expected width. Thisidentification and position-determining step alsoyields estimates of the background-signal level, back-ground tilt, and depth of the absorption line, whichare used as initial guesses in the subsequent nonlin-

ear fitting routine. Fixing the location of the absorp-tion dip by the matched filter greatly improves thestability of the nonlinear fitting routine, which ad-justs the following parameters: absorption line-width, absorption depth, background-signal level,and background tilt. The effect is a robust algo-rithm for extracting the fractional absorption fromnoisy scans. We can further improve the routine interms of noise resistance by more accurately ~50MHz! setting the initial mid-IR frequency and thenby allowing only a small correction to the preciseposition of the absorption line to be made by thematched filter. This would be especially attractivefor automating the measurement of weak absorptionlines in a noisy background in which random-noisedips might otherwise be incorrectly identified as theabsorption line.

The spectral scans and the extracted fractional ab-sorption were recorded in a spreadsheet format, anda LabView program was prepared to permit the userto step through a graphical presentation of the fin-ished spectral scans rapidly and conveniently.

6. Conclusion

IR absorption is a widely applicable method of sensi-tively detecting small quantities of gases in the at-mosphere since nearly all molecules absorb in themid-IR 3–20-mm region. The conventional approachor measuring these trace gases is with a tunable IRiode laser. These lasers are somewhat unreliablend typically require liquid-N2 cooling. Also, these

systems generally have a very narrow tuning rangethat limits them to measuring, at best, a few differentmolecules. The final and most critical problem isthe lack of automated alignment including detectorsand the multipass cell, and precise and repeatablefrequency tuning of the laser, which unfortunatelythen requires the attention of technically adept per-sonnel who are trained on the specific system.

Our mid-IR spectrometer reported here also mea-sures IR absorption features but uses dependable,near-visible room-temperature diode lasers to make acw narrow-bandwidth source that is continuouslytunable from 11.8 to 16.1 mm and has fully automatedoptical alignment, frequency tuning, and data pro-cessing. Our IR source is based on nonlineardifference-frequency mixing of the two diode lasersand is rapidly and smoothly tunable over ;2 cm21

and, when spectral sections are patched together, canbe continuously tuned over 300 cm21. Computer-operated frequency control uses temperature andcurrent tuning of the free-running diode lasers bymeans of previously recorded reference tables with acomplex tuning algorithm, including absolute fre-quency data from a wavemeter and spectral-noiseinformation from a scanning Fabry–Perot spectrumanalyzer. The system has been demonstrated bysuccessfully scanning, without any human interven-tion, 64 randomly selected acetylene absorption linesbetween 11.8 and 16 mm. A scan of a single acety-ene absorption line and a composite spectral scan ofH3 with a 70-MHz resolution were also presented.

20 March 1999 y Vol. 38, No. 9 y APPLIED OPTICS 1521

6. W. R. Bosenberg, A. Drobshoff, J. I. Alexander, L. E. Myers,

1

1

1

1

1

1

1

1

Tunable mid-IR diode laser absorption spectros-copy has developed over the past 20 years into asensitive trace-species-detection method for environ-mental, combustion, and industrial process problems,as well as fundamental spectroscopy studies. How-ever, it has been largely confined to research appli-cations because of the high degree of expertiserequired for operating cryogenic lead-salt diode la-sers. Our mid-IR spectrometer demonstrationshows that it is possible to extend the range of appli-cations beyond specialized research areas and makethis technique feasible for routine monitoring of mul-tiple trace-gas species simultaneously by technicallyunsophisticated personnel: Potential applicationsinclude environmental air quality and atmospherecomposition measurements, toxic-gas detection, pol-lution emission source monitoring, soil gas composi-tion determination, and process control in plasmaand chemical vapor deposition reactors. The suc-cessful automation gives our system its primary vir-tue: the ability to perform an unlimited series oflaser-based spectroscopic tasks more rapidly then canbe accomplished by human control.

The authors and their research were supported bythe U.S. Department of Energy, contract DE-FG02-94ER-81698, through Aerodyne Research, Inc.

References1. C. E. Kolb, J. C. Wormhoudt, and M. S. Zahniser, “Recent

advances in spectroscopic instrumentation for measuring sta-ble gases in the natural,” in Biogenic Trace Gases: Measure-ment Emissions from Soil and Water, P. A. Watson and R. C.Harriss, eds. ~Blackwell Science, Oxford, U.K., 1995!, pp. 259–290.

2. D. J. Brassington, “Tunable diode laser absorption spectros-copy for the measurement of atmospheric species,” in Spectros-copy in Environmental Science, R. E. Hester and R. S. Clark,eds., Vol. 24 of Advances in Spectroscopy Series ~Wiley, NewYork, 1995!.

3. J. Faist, F. Capasso, D. L. Sivco, C. Sirtori, A. L. Hutchinson,and A. Y. Cho, “Quantum cascade laser,” Science 264, 553–556~1994!.

4. R. Q. Yang, B. H. Yang, D. Zhang, C.-H. Lin, S. J. Murry, H.Wu, and S. S. Pei, “High power mid-infrared interband cascadelasers based on type II quantum wells,” Appl. Phys. Lett. 71,2409–2411 ~1997!.

5. K. Namjou, S. Cai, E. A. Whittaker, J. Faist, C. Gmachl, F.Capasso, D. L. Sivco, and A. Y. Cho, “Sensitive absorptionspectroscopy with a room-temperature distributed-feedbackquantum-cascade laser,” Opt. Lett. 23, 219–221 ~1998!.

522 APPLIED OPTICS y Vol. 38, No. 9 y 20 March 1999

and R. L. Byer, “93% pump depletion, 3.5-W continuous-wave,singly resonant optical parametric oscillator,” Opt. Lett. 21,1336–1338 ~1996!.

7. A. S. Pine, “Doppler-limited molecular spectroscopy bydifference-frequency mixing,” J. Opt. Soc. Am. 64, 1683–1690~1974!.

8. U. Simon, F. K. Tittel, and L. Goldberg, “Difference-frequencymixing in AgGaS2 by use of a high-power GaAlAs taperedsemiconductor amplifier at 860 nm,” Opt. Lett. 18, 1931–1933~1993!.

9. R. S. Putnam, “Diode laser difference frequency detects am-monia and chlorofluorocarbons at 11.5 microns,” in Conferenceon Lasers and Electro-Optics, Vol. 9 of 1996 OSA TechnicalDigest Series ~Optical Society of America, Washington, D.C.,1996!, pp. 13–14.

0. R. S. Putnam, “Applying room temperature diode lasers todifference frequency generation,” in Intracavity and Extracav-ity Control of Laser Beam Properties, R. L. Facklam, K. H.Guenther, and S. P. Velsko, Proc. SPIE 1869, 242–249 ~1993!.

1. U. Simon, C. E. Miller, C. C. Bradley, R. G. Hulet, R. F. Curl,and F. K. Tittel, “Difference-frequency generation in AgGaS2

by use of single mode diode laser pump sources,” Opt. Lett. 18,1062–1064 ~1993!.

2. K. P. Petrov, L. Goldberg, W. K. Burns, R. F. Curl, and F. K. Tittel,“Detection of CO in air by diode-pumped 4.6-mm difference-frequency generation in quasi-phase-matched LiNbO3,” Opt. Lett.21, 86–88 ~1996!.

3. N. C. Fernelius, “Properties of gallium selenide single crystal,”Prog. Cryst. Growth Charact. 28, 275–353 ~1994!.

14. A. E. Siegman, Handbook of Nonlinear Optical Crystals, ed.,~Springer-Verlag, New York, 1991!, pp. 87–88.

15. D. R. Suhre, N. B. Singh, V. Balakrishna, N. C. Fernelius, andF. K. Hopkins, “Improved crystal quality and harmonic gen-eration in GaSe doped with indium,” Opt. Lett. 22, 775–777~1997!.

16. U. Simon, Z. Benko, M. W. Sigrist, R. F. Curl, and F. K. Tittel,“Design considerations of an infrared spectrometer based ondifference-frequency generation in AgGaSe2,” Appl. Opt. 32,6650–6655 ~1993!.

7. W. C. Eckhoff, R. S. Putnam, S. Wang, R. F. Curl, and F. K.Tittel, “A continuously tunable long-wavelength cw IR sourcefor high-resolution spectroscopy and trace gas detection,” Appl.Phys. B 63, 437–441 ~1996!.

8. Source of GaSe crystal: Eksma Co., Vilnius, Lithuania,eksma@auste.elnet.lt.

9. J. Wormhoudt, M. S. Zahniser, D. D. Nelson, J. B. McManus,R. C. Miake-Lye, and C. E. Kolb, “Infrared tunable diode lasermeasurements of nitrogen oxide species in an aircraft engineexhaust,” in Optical Techniques in Fluid, Thermal, and Com-bustion Flow, S. S. Cha and J. D. Trolinger, eds., Proc. SPIE2546, 552–561 ~1995!.