JOURNAL OF THE OPTICAL SOCIETY OF AMERICA

Electronic-Recording, Time-Resolving Spectrometer*PER GLOERSENt

A erosciences Laboratory, Missile and Ordnance Systems Department, General Electric Company, Philadelphia, Pennsylvania(Received April 16, 1958)

A photoelectric means for recording an extended spectral region as a function of time has been developed.With this method, the recording procedure imposes no limitations on the time resolution other than thatdefined by the minimum detectable energy. The optical portion of the device, which includes the time-resolving mechanism, is very similar to a conventional photographic time-resolving spectrograph. However,the photographic plate has been replaced by a television camera which is used in much the same way. In fact,direct analogies can be drawn between the photographic plate and the storage surface of the television cameratube and also between the photographic developer and the scanning electron beam in the camera tube. Themethod is capable of sensitivities of the order of 103 times those available with Tri-X film in the visible. Inaddition, the usual cumbersome photographic calibration procedures are avoided since intensities arerecorded as vertical deflections, yielding records similar in appearance to those obtained from strip chartrecorders.

INTRODUCTION

THERE are several well-known means for makingT time-resolved spectral studies of transient lightsources. The photographic techniques include the use ofa rotating drum camera which moves the film relativeto a fixed spectrometer slit, a rotating mirror whichscans the spectral images of the spectrometer slit overa stationary film, or a slotted sector arrangement at theentrance slit which effectively causes the opening to thespectrometer to move relative to a fixed photographicplate.

In the photoelectric category, periodic mechanicalscanning devices similar to the one described by Bullockand Silverman' have been used to study an extendedspectral region as a function of time, but these devicesare suitable for only relatively slow events (the orderof a millisecond in duration).

Higher speed photoelectric spectral measurementsare usually confined to monochromator plus photo-multiplier techniques. This latter method is particularlysensitive when appropriate electronics2 are used topreserve the inherently fast response of the photo-multiplier without serious loss in gain. In fact, thesensitivity is sufficiently high so that the ultimatelimit of counting individual photons is encountered, atwhich point the measurements lose their significance.However, this method suffers greatly in the inability toobserve more than one-integrated wavelength intervalper detector at any given time. By virtue of thislimitation, it is difficult to observe the change of aspectrum from a continuous to a discrete nature as afunction of time (as in the case of a gas cooling rapidlyfrom very high temperatures), or the change in shape

* Based on work performed under the auspices of the U. S. AirForce Ballistic Missiles Division, Contract No. AF-04(645)-24.

t Work performed at the General Electric Research Laboratory,Schenectady, New York.

I B. W. Bullock and S. Silverman, J. Opt. Soc. Am. 40, 608-615(1950).

2 Dieke, Dimock, and Crosswhite, J. Opt. Soc. Am. 46, 456-462,(1956).

of a spectral line with time, or wavelength shifts oflines with time, to mention only a few examples. Sincethe photographic plate is many times less sensitive foran equal exposure time than the photomultiplier, itbecomes impossible to make a study of a spectrum froma sufficiently weak light source simultaneously as afunction of time and wavelength with the means de-scribed above.

With reproducible, repetitive light sources, a success-ful substitute for studying simultaneously varioustimes and wavelengths has been to dissect the repeatedinformation either timewise or wavelengthwise bymeans of a gated photomultiplier as described byCrosswhite, Steinhaus, and Dieke." 4 This method isprobably the best technique for studying the time-resolved spectrum of repetitive sources (with a frequencyof several or more events per second) since even thephoton noise is averaged out. However, it is not at allsuitable for studying single transient events.

The technique to be described can surpass the sensi-tivity of the photographic method without sacrifice ofsimultaneous wavelength and time measurements andat the same time avoids troublesome photographicintensity vs density measurements. It also appears that,with presently available components, it is possible tocome within two orders of magnitude of the samephoton noise limit encountered by the ungated photo-multiplier technique.

DESCRIPTION

The electronic-recording, time-resolving spectrometerconsists of two independent parts; the optics whichincludes the entire time-resolving mechanism and therecording electronics which consists of a televisioncamera chain,-an oscillograph, and an electronic shutter.

I Crosswhite, Steinhaus, and Dieke, J. Opt. Soc. Am. 41,299-302(1951).

4 Steinhaus, Crosswhite, and Dieke, "Short period investigationsin spark discharges," Department of Physics report, The JohnsHopkins University (June, 1952).

712

VOLUME 48, NUMBER OCTOBER, 1958

TIME-RESOLVING SPECTROMETER 713

Optics

A schematic diagram of the entire apparatus isshown in Fig. 1. An extended source of light is magnifiedsufficiently with an appropriate lens so that a uniformportion of the source fills the entire length of theentrance slit between two adjacent openings in the time-resolving sector. The sector slots become perpendicularto the slit opening as they approach the midpoint of theslit. The flat sides of the pairs of wedges forming boththe sector slot jaws and the slit jaws face each other, sothat the horizontal and vertical edges of the resultingopening to the spectrometer can be made very nearlycoplanar. This combination of sector and slit performsthe entire task of time resolving and, in fact, has beenused with the photographic method mentioned pre-viously. In both these methods, if the sector speed isknown, then the position of the slot along the lengthof the slit and hence the vertical position along anyspectral line is a known function of time. Since straightslits were used in the experiments, the time dependenceof position along the slit will be a sine function ratherthan linear, but the difference is negligible for thepresent purposes.

The number of slots in the sector varied from aminimum number determined by the length of thespectrograph slit to a number determined by the small-ness of the source or the desired demagnification on thedetector. Thus, one portion of the spectrometer slit oranother is always exposed in order to avoid the need forsynchronization with the transient event underobservation.

For geometrically small or nonuniform sources oflight which cannot be suitably magnified to cover theentire slit evenly, a more satisfactory arrangementwould be a rotating mirror used to sweep a thin-hori-zontal image of a portion of the source across thevertical slit.

The fastest sector used in these experiments iscapable of 15 000 rpm in air and considerably more in avacuum. The maximum time resolution obtainable isdetermined by the minimum attainable slot opening inthe sector as well as the maximum attainable sectorspeed. The minimum practical slot opening is deter-

SOURCE SPECTROMETERLENS SLIT

FIG. 1. Block diagram of the electronic recording,time-resolving spectrometer.

TABLE I. Minimum resolvable times ultimately observable withthe f/24 Bausch & Lomb (B & L) spectrometer, assuming perfectsource, detector, and optics. Minimum resolvable times for otherspectrometers can be calculated by dividing the various productsof a mrin from below by the a appropriate to the other spectrome-ter, e.g., a- 5 X 1O 5 for the f/4 spectrometer referred to in thetext.

f/24 B & L

D= 100 A/cma = 10-2 cm

a= (5D62/4f2) X 10-2 a= 2X 10-7

T0K Energy emitted by Energy transmitted Minimumsource (Wx) in pho- by spectrometer resolvabletons ptsec' cm-2 A-1 time

(X = 5500 A) E=aWx Tmin (sec)

(photons//isec)

3000 3 X 1010 6X 103 0.025000 1X1012 2X10 5 0.0005

10000 2X 1013 4X 103 0.00003

mined by the optical resolution of the system. It wasnever necessary to go to this limit under the presentsource conditions, however. A 0.020-inch wide slot wasused. Since the sector disk is seven inches in diameter,this corresponds to a time resolution of approximatelyfour microseconds at 15 000 rpm. By operating thesector in a vacuum chamber, a time resolution of about0.5 microsecond can probably be achieved with thisparticular mechanical system.

The photon noise limit must now be considered, sincethe ultimate time resolution may not be determined bythe above mechanical and optical considerations butinstead by the intensity of the source, the resolution(6), aperture (f), and dispersion (D) of the spectro-graph, and the optical resolution of the televisioncamera chain. The camera tube photosurface is assumedto be a mosaic of small, independent photosurfaces. Themosaic size is determined either by the geometricalresolution of the photosurface or by the band width ofthe video amplifier, whichever gives the larger element.It is further assumed that the reduced image of thespectrometer opening on the camera photosurface isapproximately the size of such a mosaic element.Table I shows approximately at what time resolution(smin) this photon noise limit occurs under variousconditions and under the following idealized assump-tions: (1) The use of a perfect detector, i.e., the signalsfrom single photons are not lost in the noise, (2) theaverage number of photons per detector mosaic elementmust be of the order of 100 in order to obtain acceptablespectral line contours, (3) the source is a blackbody ata temperature T (for the sake of comparison), (4) theoptics and detector have unity efficiency, and (5) theslit width, 3, is 10-2 cm. The slit height, or slot width(5 X 10-2 cm), has already been included in a of Table I.

The actual minimum resolvable times observed witha sufficiently sensitive detector will be at least an order

OctoberI958 ELECTRONIC-RECORDING,

PER GLOERSEN

of magnitude less favorable than those given in Table I,when the efficiency of the spectrometer and the non-blackbody characteristics of the source are taken intoaccount, and several orders of magnitude less favorablewhen using currently available camera tubes, whichcannot detect individual photons.

Various spectrometers have been used for theseexperiments. The only important requirements arethat the spectrometer be stigmatic and that its focalplane be perpendicular to its optic axis. This latterrequirement eliminates most prismatic spectrographs.The results shown here were obtained with either of twospectrometers, an f/24 Bausch & Lomb Ebert typespectrograph or an f/4 transmission grating spectro-graph with two 7X50 binocular lenses as collimatorsand a 30X32 mm ruled area 600 groove/mm grating.

A field lens is placed in the position normally occupiedby the photographic plate in order to image the gratingon the lens of the television camera. Without the fieldlens, the rays emerging at the various wavelengthswould be diverging and the wavelength region sub-tended by the television camera lens would be verysmall. The focal length chosen for the field lens dependson the desired degree of demagnification of the spectrumor distance of the television camera from the photo-graphic focal surface. The lens in the television camerawas a standard 16-mm movie camera lens of f/1.9aperture.

Recording Electronics

Any television camera can be used in this application,the more sensitive the better. The choice of the Vidicontype camera for these experiments was on the basis ofavailability. Given the operating characteristics of anyparticular camera, the performance of the system witha more sensitive (and elaborate) camera can be reliablypredicted. The time response of the photosurface doesnot enter in any way with this system, since the cameratube surface is in fact used for storage of the time-resolved spectrum until the electron reading beam canconvert the stored image into a video signal. The videosignal then is displayed on an oscilloscope screen afterthe light emitting event has occurred. The analogy canbe made of comparing the camera tube surface to aphotographic plate and the electron reading beam to aphotographic developer.

In order to determine more easily the magnitudeof the video signal, it is placed on the vertical amplifierof the oscilloscope rather than being used for Z-axismodulation as in normal television receiver practice.The vertical sweep sawtooth is superimposed on thevideo signal so that each line element from the cameratube will appear at different heights on the oscilloscopescreen. A type 535 Tektronix oscilloscope with a type53/54 G differential preamplifier is used for recordingthe video signal. The differential preamp has independ-

ent gain controls for each of its two inputs whichallows the sawtooth signal to be amplified independentlyof the video signal. The sawtooth gain was usuallychosen so that about 50 of the 525 horizontal linesappeared in the oscillograph screen field.

In order to take a permanent photographic recordof an electronically-recorded time-resolved spectrumfrom a transient light source, it is necessary to include ameans for keeping the oscilloscope screen dark whenthere is no video signal superimposed on the horizontallines. This function is performed by an electronicshutter which consists of the photocell pickup andpulse shaper shown in Fig. 2. The transient sourcecauses the photocell to send an impulse that triggers aone-shot multivibrator, which puts out a square wave1/60 second wide and of sufficient height to activate thewriting beam of the oscilloscope. The multivibratorsignal is applied to the intensity grid of the cathode raytube through a 0.5-,uf 2000-v condenser. The delay of thesystem has not been measured, but it is certainly lessthan 1/60 of a second. Since a complete televisionframe requires 1/30 of a second and the vertical field isscanned twice (interlaced) in this time, a half-raster canbe lost without any effect before the square waveintensifies the oscilloscope beam. Only a half-raster isrecorded during the 1/60 second of intensification; thusthere is no record made of horizontal lines from whichthe stored time-resolved spectrum has been erased.

OBSERVATIONS

A tungsten ribbon lamp with a true temperature of2750'K was used as a sensitivity calibration source for

< X ] 2 1K 1500K 810K ISOK to t 450 V

200K SIX4LIM 180K-

I0K2 60 450V 2K r

01 51001 ~~1 7470K -57K1 fl77¶S Fl F

I0K 500K ' 2

FIG. 2. Circuit diagram of the electronic shutter.

714 Vol. 48

October1958 ELECTRONIC-RECORDING, TIME-RESOLVING SPECTROMETER 715

the device. According to published data on the type6198 Vidicon tube, about 105 photons/frame (1/30 sec)incident on a single mosaic element are required for areadable signal, so that more than enough energy isavailable to calibrate the device, according to Table I.For the purposes of these order-of-magnitude calcula-tions, a single mosaic element of the camera tube iswell enough approximated by the area of the de-magnified image of the spectrometer entrance openingon the camera tube surface. The actual signals obtainedfrom this source were consistent with these calculations.The sensitivity of the Vidicon camera was comparedwith Kodak Tri-X film by exposing both detectors tothe short period light from a GE FT106 xenon flashlamp under comparable conditions. For this purpose aroll film holder was substituted for the field lens in thef/4 spectrometer. It can be deduced from the compari-sons made in Fig. 3 that the doubly developed Tri-X isabout five times more sensitive than the Vidicon cameraas a whole. However, the Tri-X film was not used at the

(a)

(b)

(c)

(d)

FIG. 3. Comparison of sensitivities of the Vidicon camera andKodak Tri-X film. (a) Time-resolved spectrogram of a GE FT106xenon flash tube on Kodak Tri-X film, doubly developed (f/4spectrometer). Wavelength increases toward the right; timeincreases downward. The xenon lines lie around 4700 A. (b) Micro-densitometer trace of (a) at around 2 of the total lapsed time.(c) Vidicon camera record taken under identical conditions as (a).(d) Vidicon camera record; taken at twice the dispersion as (c).

(a)

(b).._ _

FIG. 4. Vidicon records of spectra from behind a reflected shockwave in xenon (f/24 spectrometer). Wavelength increases towardthe right, time downward. Total duration of the reflected shocklight is about 500 usec. (a) =4350-4950 A. The lines super-imposed on the continuum belong to XeI. (b) X=4900-5500 A.The line appearing initially belongs to XeI (4916, 4925 A, un-resolved). The three lines appearing later belong to CuI (5105,5153, and 5218 A).

limit of its resolution, as was the Vidicon camera tube.When this is taken into account together with imagedemagnification and lens losses in the camera, Tri-Xappears to be approximately twenty times more sensi-tive than the Vidicon camera tube surface itself.

Two typical records of the light from behind areflected shock wave in xenon are shown in Fig. 4 as anillustration of what can be seen with this sort of device.

CONCLUSIONS

The Vidicon camera chain used for these experimentsis about 106 times less sensitive than required in order tosee individual photons. However, it must be said thatthe failure of the Vidicon camera chain to operatesatisfactorily is largely due to the video amplifier,which is the main source of the noise. No video amplifierwas available that could see the photoconductor noiseitself at appropriately low-dc signal electrode voltages.This electronic recording technique will have an over-whelming advantage over the photographic methodwhen an image orthicon camera chain is used. From acomparison of the published light transfer character-istics of the 6198 Vidicon with the 6849 image orthicon,a detectivity advantage of around 10-104 can beinferred for the image orthicon. In addition, the responseof the image orthicon is linear over a large range ofillumination, in contrast to that of the Vidicon.

Another possibility is to use either a solid-state orelectronic light amplifier in conjunction with the tele-

PER GLOERSEN

vision camera. Some of the electronic light amplifiers(or image converter tubes) provide a means of rapidlymoving the intensified image along the viewing screenand would, therefore, provide a means of avoiding themechanical and optical limitations of the present time-resolving mechanism in addition to an increase in thesignal level.

The optical apparatus used for this investigation isadmittedly crude for the purposes ultimately intended.It would be, for instance, better to have a circularlycurved entrance slit in order to have a linear time scalewith a slotted time-resolving sector disk. Such a slitperforms very well in an Ebert type spectrometer Inaddition it would be preferable to use reflecting opticsthroughout, including the television camera, for

6 Fastie, Crosswhite, and Gloersen, J. Opt. Soc. Am. 48, 106-11(1958).

minimum losses and for constant chromatic focalproperties.

A preliminary experiment has been carried out inwhich an image orthicon camera chain was used forrecording the spectrum from behind an incident shockwave in xenon. Numerous attempts at photographingthis particular spectrum with the f/4 spectrometerfailed, even with tenfold exposures and quadrupledevelopment. No trace of emulsion darkening appeared.The image orthicon camera, however, gave definiteindications of visible spectral lines behind the incidentshock.

ACKNOWLEDGMENTS

The author wishes to thank H. R. Day, J. R.Eshbach, P. D. Johnson, R. W. Redington, and W.Roth for their assistance and some enlighteningdiscussions.

716 Vol. 48