Real-Time Computer for Monitoring a Rapid-ScanningFourier Spectrometer

Guy Michel

A real-time Fourier computer has been designed and tested as part of the Lunar and Planetary Labora-tory's program of airborne infrared astronomy using Fourier spectroscopy. The value and versatility ofthis device are demonstrated with specific examples of laboratory and in-flight applications.

1. Introduction

A real-time computer for Fourier transform spectros-copy has been developed at the Lunar and PlanetaryLaboratory for use as a monitoring device with a rapid-scanning interferometer. Its first application was inNASA's program of spectroscopic observations ofMars from the CV-990 aircraft in August 1971. Wefound real-time monitoring invaluable for checkingon the quality of the data and for analyzing the be-havior of the complex systems constituting our Fourierspectrometer that are particularly sensitive to thehostile environment encountered aboard aircraft. Thisability to evaluate objectively our experiment whilestill in a position to modify its goals allowed us to re-cord data superior to that previously acquired in air-borne observations of Mars.

II. Computer Organization

The basic computer' is of the type developed andused at Laboratoire Aim6 Cotton (LAC), CNRS, Paris,France, with very high resolution stepping interfer-ometers.2 Its versatility and performance have beenincreased to handle interferograms generated by thecommercially produced, rapid-scanning interferometersused in LPL's programs. We will briefly recall itsprinciples of operation and present the new featuresintroduced in the LPL computer.

This special-purpose digital computer (Figs. 1 and2) is hard-wired to perform a discrete Fourier trans-form. Each. interferogram sample (10-bit + signmantissa, 4-bit exponent) is multiplied by a sine func-tion provided through an address generator and a readonly memory (ROM, 1024 ten-bit words) in which isstored a quadrant of a sine table. The result of the

The author was with the Lunar & Planetary Laboratory, Uni-versity of Arizona, Tucson, Arizona 85721, when this work wasdone; he is now with Laboratoire Aim6 Cotton, 91-Orsay, France.

Received 3 April 1972.

multiplication is then scaled according to the exponentand fed into an adder connected to a circulating memory(1024 twenty-three-bit + sign words) storing the outputspectrum points. The spectrum is displayed on anoscilloscope through a digital-to-analog converter.

The operation sequence is timed by a program gen-erator. The number of modes or programs availablefor the transform calculation has been extended tothree: sine transform, cosine transform, and bothsine and cosine transforms with power spectrum dis-play. This latter mode is especially useful with fastscanning interferometers where the zero path pointcannot be predetermined, thereby precluding simplesine or cosine transforms. After selecting the trans-form mode, the operator dials the other initial settings:resolution in the spectrum, first point of the spectralwindow to be displayed.

To maximize the speed we used the same techniquesdeveloped at LAC for the first real-time computer,i.e., parallel computations and simultaneous operations,but thanks to a new technique of multiplication thisspeed has been increased by a factor of 2. The com-putation time expressed in terms of spectrum pointper interferogram point is now 500 nsec. If we samplethe interferogram at 1 kHz, the computer is able toproduce 2048 spectrum points in the SIN or COS modesand 1024 points in the SIN-COS mode. Technologi-cally, the LPL computer is quite different from theoriginal LAC model. Built two years apart, we havebenefited from the fast-growing field of integratedcircuits. The LPL computer is an all TTL-MSI de-vice in place of ECL and TTL. The use of a singleIC family greatly simplified the design. We introducedas an array multiplier a very interesting circuit forapplications where speed is at a premium. 4 In thistype of multiplier the multiplication time of 2 N-bitwords is proportional to N instead of to N2 with theclassical technique of partial products and shifts. Theprice paid for the increase in speed is that the number ofcomponents is proportional to N2. The actual multi-plication time of two 10-bit + sign words is less than

November 1972 / Vol. 11, No. 11 / APPLIED OPTICS 2671

Fig. 1. Block diagram of real-time computer.

Fig. 2. Complete computer system. Top chassis is the arith-metic unit and read-only memory. Middle chassis houses theoutput spectrum memory and function generators. Power sup-

plies are contained in the bottom unit.

ently the successive sine and cosine transforms.With faint sources we can see the improvement of theSNR in the spectrum vs the number of scans. Withstrong sources, where a single scan is sufficient, we getfrom the very start an estimate of the SNR and watchthe improvement in resolution.

In our rapid-scanning interferometer no provisionexists to locate the zero path point. This is not adifficulty when the interferogram is transformed laterwith a general-purpose computer. One usually startsthe interferogram far enough before the zero pathpoint to include the main lobes of the interferogram.Phase correction followed by a cosine transform is oneof several possible treatments that can then be usedwith a large, general-purpose computer. With a real-time computer this is not practical, so we have to per-form a power transform where the phasing of the sam-pled points with respect to zero path need not be ex-plicitly known. A power transform theoreticallyrequires recording the interferogram over the interval(-L, +L) with the disadvantage of having to doublethe path difference L and the number of points to betransformed for a given resolution. To avoid thesecomplications we checked by simulation on a general-purpose computer that we do get a correct power trans-form without noticeable phase distortion if we start theinterferogram a few hundreds of samples before zeropath. To do so, the white light interferogram is opti-cally phase-shifted and its main lobe is level-detectedto produce a flag pulse used to trigger the computationsequence. The computer performs the sine andcosine transforms and displays the power transformby taking the square root of the sum of the squaresof both transforms with fast analog circuits.

IV. Preliminary Results

The following examples illustrate ways in whichwe have employed the computer in preparing and

200 nsec. The circulating memory was built withMOS dynamic shift registers instead of magnetostric-tive delay lines. Present memory size is 1024 wordswith provision for extension to 8192 words. An analogapodizing interpolator using the technique of tappeddelay line filters has been built but not yet incorporatedinto the system.

111. Real-Time Computer Mode of Operation

Figure 3 shows the connection of the computer toan experiment including the rapid-scanning interfer-ometer, a programmable gain amplifier to handle thedynamic range of the interferogram, an analog-to-digital converter, and a buffer memory necessary todrive the digital tape recorder with a constant datarate. The computer is merely connected in parallelwith the recorder input and computes at full resolutiona slice of the spectral range covered by the spectrom-eter during each scan. The computer adds coher-

Fig. 3. Connection of computer to Fourier spectrometer. Thisis a classical digitizing and recording system for Fourier spec-troscopy. The sampling rate is always affected by flutter due tovibrations or residual solid friction in the moving mirror's slide.A buffer memory is added to drive the digital tape recorder at aconstant data rate. The buffer memory includes two sections of100 words each. When one section is accepting interferogramsamples as input, the other is reading out at a clock rate determined

by the tape recorder.

2672 APPLIED OPTICS / Vol. 11, No. 11 / November 1972

DATA IN~

cos

sin

power

0-

go NOe 4149

500 points

Fig. 4. Transform of built-in test function showing the threemodes of calculation available with the real-time computer.

a)

100

4292 -0-cm

Fig. 5. Laboratory test of spectrometer using CO band at 2 .35-A.Ten power spectra were coadded with resolution limit of 4.0 cm-l.

200 300 -O Hz

.25

13)

100 300 -+'Hz200

Fig. 6. Aircraft vibration spectra produced by computer. The series of spectra in (a) and (b) were produced from an accelerometer lo-cated, respectively, before and after the shock mounts used to isolate the interferometer from the aircraft. The accelerometer axis wasparallel to the main axis of the aircraft and to the translation axis of the moving mirror of the interferometer. Acceleration is in arbitrary

units; note the change in scale between series (a) and (b).

executing our airborne experiments. To check onits own internal functioning the computer has a built-intest generator producing the following sequence ofsamples: 0, 1, 0, -, 0, etc., which is merely thesampling of a sine wave. The transform of this func-tion represents the theoretical instrument function forthe resolution selected. Figure 4 shows the transformsof this test function in the three modes of computa-tion.

Several experiments were conducted in the labor-atory to evaluate the performance of our interferom-eter in the absence of vibrations. Real-time analysiseliminated the need to send interferograms to a large,general-purpose computer with consequent delays inreviewing the spectra. Figure 5 shows an absorption

band of CO used to verify the resolution achieved bythe interferometer. In this spectrum ten scans, eachof several seconds' duration, were coadded.

During flights on the NASA CV-990 the computerwas used in several ways to provide important docu-mentation on the performance of our interferometerin the presence of vibrations. The first applicationconcerned the problem of vibration isolation of theinterferometer. We designed at LPL shock mountsto attenuate those vibration frequencies known to besevere on the aircraft. The computer provided acheck on their effectiveness by serving as a real-timeaudiofrequency analyzer.' The signal from an ac-celerometer located above and below the shock mountson our experiment was transformed, providing a real-

November 1972 / Vol. 11, No. 11 / APPLIED OPTICS 2673

-

o

o

4935 5007 - cm-1

Fig. 7. High resolution test of spectrometer in flight. Singlescan (24.5 see) of the sun showing CO2 band at 2 A with a resolution

limit of 0.67 cm-'.

time display of the aircraft vibration spectrum and theresidual vibrations seen by the interferometer. Figure6 shows such vibration spectra verifying the generalattenuation of all aircraft frequencies and the absenceof any resonant vibration in the interferometer's mount.

The sensitivity of the moving mirror to residualvibrations cast doubt on the ability of the interfer-ometer to perform efficiently at maximum resolutionaboard the aircraft. Again, the advantages of real-time analysis permitted us to study this problem inflight with subsequent modifications to the experimentthat resulted in most effective use of our quite limitedand very expensive observing time. By examiningthe reproducibility of high resolution features of singlescans of the sun recorded in flight we were able to selecta resolution that guaranteed the best return from ourMars observations. Figure 7 shows a resolved CO2band at near-maximum resolution (0.67 cm-') froma single scan of the sun used for this purpose.

Finally, the computer was used in the actual observ-ing runs, providing a continual check on the acquisi-tion of good data. Figure 8 contains portions of thespectrum of Mars seen by the real-time computershowing strong CO2 bands in the 2-,4 region. Aftercoadding just ten scans, each of 9-sec duration, thecontinuum level has been established although only asuggestion of the CO2 absorptions exists. After 600scans the SNR has increased as expected, and the CO2triad at 2 4 is now quite evident. In addition, theH20 band at 1.9 pt has emerged from the noise. Bywatching the real-time development of the Mars spec-trum during our first flight we were able to predictwith accuracy the averaged result of all four scheduledflights. We decided that following the flight scheduleas planned would lead to useful spectra, a conclusionlater verified by complete data reduction at LPL.

V. Conclusions

The effectiveness of the real-time analysis offeredby this small, special-purpose computer has been dem-onstrated through actual use in an airborne spectros-copy program. The experience acquired in these

b)

Fig. 8. Real-time monitoring of spectrum of Mars. The windowselected includes the strong CO2 absorptions at 2 ,u and the watervapor band at 1.9 u. Upper trace is the coadded result of just 10scans (80 see of observing). Lower trace includes 600 scans and

exhibits the expected improvement in SNR.

first experiments will result in improvements to boththe computer and the interferometer, creating an evenmore effective combination. In addition, this com-puter will be attached to another spectrometer, thefirst high resolution stepping interferometer, builtby Pierre Connes6 and now in use by LPL staffat the Steward Observatory 90-in. (225-cm) telescopeon Kitt Peak. The availability and versatility of thiscomputer effectively eliminates a frequently voicedcriticism of Fourier spectroscopy-the inability to seethe spectra until some time, often days, after the re-cording of the interferograms.

I thank G. P. Kuiper, director of LPL, whomade possible, my one-year stay in his laboratory.The computer project was one contribution to theairborne astronomy program supervised by H. P. Lar-son and U. Fink, who provided constant support. Iespecially thank M. Durand of LAC for preparing thelarge PC boards, Tex Belschner of LPL for his carefulassembly, and J. Percy for providing his services andthose of the LPL electronics shop. Financial supportof this project was provided through University ofArizona funds (NASA Institutional Grant No. 83)and NASA Grant NGL-03-002-002.

2674 APPLIED OPTICS / Vol. 11, No. 11 / November 1972

a)

4 14578 cm 1

I5722 cm-I

References 3. A. Habibi and P. Wintz, IEEE Trans. Electronic ComputersC-19, 153 (1970).

1. P. Connes and G. Michel, AFCRL-71-0019 Special Reports No. 4. H. G. Kingsbury, Electron. Lett. 7, 277 (1971).114,313-330(1971). 5. T. Bially, IEEE Trans. Audio Electroacoustics AU-18, 201

2. P. Connes, AFCRL-71-0019 Special Reports No. 114, 121- (1970).125 (1971). 6. J. Connes and P. Connes, J. Opt. Soc. Am. 56, 896 (1966).

TWO SUPPLEMENTSto Optics

and Spectroscopy- LUMINESCENCE SUPPLEMENT is a collection of some 60 papers report-

Ii (~~N ing a wide variety of work in electronic spectroscopy and luminescence.Topics include experimental determination of the stimulation of ZnS phos-

optics phors by infrared radiation, x-ray electroluminescence from alkali halide

spectroscopy single crystals, the spectroscopy of crystalline solutions of diphenylpoly-

enes, and theoretical treatments of transitions in complex molecules andthe relationship between recombination luminescence and exo-emission.

.... .... 192 pages $15

^ _ 5 MOLECULAR SPECTROSCOPY SUPPLEMENT includes some 63 papersranging in topic over electronic transitions in N2 and Al, EPR spectra ofmacroradicals, vibration spectra of silicates, strontium nitrate, cyclohex-

optics Js i ane, cyclopropane derivatives, paraffins, and hydrogen bonded substances.spectroscopy Theoretical treatments include a consideration of nonadiabatic transi-.U.pspl..n ,, tions, optical rotatory dispersion, temperature effects, normal coordinate

calculations, and absolute intensities in vibration spectra.

l ............... ...... : 180 pages $15

ORDER FROMAMERICAN INSTITUTE OF PHYSICS 335 East 45th Street, New York, N.Y. 10017

November 1972 / Vol. 11, No. 11 / APPLIED OPTICS 2675