A 4x 1 GHz Array Acousto-Optical Spectrometer

for

air- and spaceborne observations in the submm-region

J. Horn(2)'(1), C. Macke (1) . F. Sch1Oder (1) , F. Schmii11ing (1) . O. Siebertz(1)and R. Schieder(1)

(1) : I. Physikalisches Institut, Universitat Köln.Ziilpicher Str. 77, 50937 Köln. Germany

(2) : University of California at Los Angeles,405 Hilgard Av, CA 90024 Los Angeles, USA

email: horn@jupiter.astro.ucla.edu

March 17, 1998

Abstract

In this paper an array acousto-optical spectrometer (AOS) with four 1 GHz bandsand a frequency channel spacing of 1MHz is presented. The test results of the arrayAOS are comparable to those of the space-qualified AOS for the Submillimeter WaveAstronomy Satellite (SWAS). Allan variance tests, longtime integration tests andinvestigations of the relative stability between the four bands are presented. The Allanvariance minimum time was found to be at least 1000s, which is one of the best resultsobserved for broadband AOSs so far. Performance tests of the different bands revealedan identical frequency response. Therefore the relative stability of two simultaneousmeasured baselines is extraordinary, so that correlated noise can be eliminated efficiently.Recently the Array AOS has been in operation for 2 months at the IRAM 30m telescope.From this test-run, several results present the competitiveness of AOSs compared tofilterbanks and hybrid autocorrelators. In an upgraded design concept the secondgeneration array AOS will be more compact and reliable. This version will be operatednot only on ground-based observatories like KOSMA or AST/RO, but also on airborneand spaceborne facilities like SOFIA and FIRST.

Keywords: Radio Astronomy, Acousto-Optical Spectrometer, Backends, Radiometers,Heterodyne Instrumentation, Array Receivers

315

Slit 4 -LmeAPert44., CCD

Scaz -°P lAca

Grattni

CylzutpLeza

Lightpen:Laser Diode

incl.Collimating

Op tics

CyLid. multi -Lens Band=rice _cell

X00

�5,00 •

300 2200Frequency [MA

1600 11100 26C024031000

I Introduction

In the recent years the worldwide develop-ment of large bandwidth, multi-frequency ormulti-beam receivers has incited a demandon corresponding backends. In the submil-limeter wavelength range and of course inthe THz region, the requirements in band-width increase drastically. At the same timeit might be necessary to have relatively highresolution at the same time, for frequencysurveys e. g. Such requirements hold for tmo-spheric observations as well, because most at-mospheric lines tend to have extended linewings together with rather narrow peaks atthe center of the lines. Future airborne orspaceborne observatory projects like SOFIA'or FIRST 2 depend crucially on versatile back-end developments in order to achieve a mosteconomic usage of the available equipmentwith optimum scientific return. At KOSMA3a new superconducting mixer array frontendwith 2 x4 channels at 490 and 810 GHz is un-der development, so that the number of back-ends needed grows accordingly.

In this paper, the recent development ofan array acousto-optical spectrometer (AOS)with 4 x1 GHz bandwidth is presented. Itsoptical layout is discussed with respect tothe required specifications for normal oper-ation at radio observatories. The laboratorytest results have proven that the specifica-tions have been achieved. To check the per-formance of the array AOS during observa-tions at a telescope, it has been installed atthe IRAM 30m telescope for a two-month testperiod this early summer. The outcome ofthese tests did also show an excellent agree-ment with the expected performance.

1 Stratospheric Observatory for Infrared Astron-omy

2 Far InfraRed Space Telescope3KOlner Observatorium fiir SubMm Astronomie

2 System Design

Figure 1: The principal optical design of theKOSMA 4x I Gliz array ADS

The principal design of the array AOS issimilar to those of the standard KOS:VAsingle-band AOSs such as the one for theSubmillimeter Wavelength Astronomy Satel-lite (SWAS), (Klumb et al., 1994). Asshown in Fig. 1, four intermediate frequency(IF) signals, are fed to the array Bragg cellwhich is made out of a single LiNb0 3 crys-tal. Through the piezoelectric transducer, anacoustic wave inside the crystal is induced.The acoustic wave produces a periodically

5000 Bend 1 exp. set. 40 5.4 =NV 1Bd2eq.t@44ziWBand 3 op. amt. 3.2 euWBend 4 esp. wt. 42 inV/

3500 1- /

Figure 2: The bandpasses of the four bands ofthe array AOS measured with a synthesizer

varying diffraction index inside the crystal.At the associated phase grating, monochro-matic laser light, four beams for each acoustic

450o

4000 1-

316

16970001696500 - o o C e169600016955001695000169450016941300 7 6)00

1693500 0 10000 20000 30000 40000 50000

Time [sec]

et

active zone produced by means of a holograficgrating, is diffracted under the Bragg condi-tion. The first order diffracted laser light isthen imaged onto the CCD by means of ascan optics and a cylindrical lens. The Braggcell is operated in shear wave mode, enablingefficient light scatter reduction by means oftwo crossed polarizers. The deflected lightbeams are then imaged onto four indepen-dent and parallel CCD lines. Especially thesecond cylindrical lens is responsible for theexact matching of the beams onto the corre-sponding CCD line.

With this setup, a high efficiency has beenachieved, as can be seen in Fig. 2. The IFpower, which is necessary to bring all pixelsequalized in saturation, is well below 6 mWfor all four bands, yielding a maximum to-tal power level on the crystal of less than20 inW, which is comparable to the amountof one of our single-band AOSs. Further-more, the bandpasses have a 3 dB shapeover the 1 GHz band, which could be furthersmoothed by means of an r. f. equalizer if nec-essary. The problem in such an efficient op-tical layout is the increasing amount of lightscatter collected on the CCD. Former stud-ies have shown, that a strong light scatterlevel decreases the spectroscopic stability of aspectrometer, which is measured by the max-imum total power integration time as a resultof the so-called Allan-variance plot (Schiederet al.., 1989). By introducing a slit aperture atthe beam crossing point, the scattered lightcould be nearly eliminated in all four bandssimultaneously. As a result, extraordinaryspectroscopic stabilities have been achievedwith the array AOS, as is described in thenext section.

To avoid any overlap of the acoustic beams,which would cause unwanted crosstalk be-tween the bands, the transducers are spacedby 1.6 mm. The residual crosstalk measured

is below -30 dB between the bands and purelyof electrical origin. This has been confirmedwith measurements done by the supplier ofthe cell. measuring the electrical crosstalkof the IF matching circuit (GEC Marconi.1995).

3 Performance Tests

3.1 Laboratory tests

spectroscopic channel data of pixels #1650 #2:750

0 Band 1 11}. = 220 MHzc Band 2 BF = 2.41 MHz

Band 3 13, = 2.41 MHz* Band 4 BF = 2.26 MHzcs

14-g_ A011P1

41' ti44 17.-.

3

1000

10 100 1000Integration Time [sec]

Figure 3: Data of one spectroscopic channeland the spectroscopic Allan-variance plot

One important test parameter is the min-imum time of the Allan-variance plot, orthe maximum total power integration time,which has been found to be up to 1000 s usinga noise source under thermally stable condi-tions, see Fig. 3. Compared with normal op-eration, where a minimum time of above 100 s

10000

317

— Band 2 I— Band 3 I

OA

g 0.6

0.4

0.2

0.0 2094.0 2096.0 2098.0 2100.0 2102.0 2104.0 2106.0

Frequa2cy 111014

a = 0356

2

0.018 -. Bond 3

0.015 7

0.013

0.010

0.007 1-

0.005

0.003 r

nnwilt.0 - leram.=-9.1 -

renw9/ - 14/4ran =9 1 - le-

rem.23 - le- le

IWO 2030 22D3 2403hammy 1113321

— ideal CCD performancex relative ram of leftism diemance

0111 a 10 MIS OM a= 0.30Merge CCD ilasaimdie

1-0 Wit

Figure 4: The ffitercurves of two verticallyadjacent pixels (in two adjacent AOS bands)

is enough to guarantee proper measurements,the array AOS is therefore not limiting themeasurements.

The identical performance of the differ-ent bands has been tested in various ways.First, the bandpasses have shown, that the IFpower response of the bands is equal within2 dB. The differences are partly due to thegrating and partly to alignment tolerances.The frequency response has been tested tobe identical for two bands as well. Therefore,filtercurves of the same pixel in two adjacentbands have been measured, see Fig. 4. Thestrong overlap of these curves indicates highradiometric correlation, which can be calcu-lated to be 99.8 %, (Horn, 1997).

To make sure, the filtercurves show suchstrong overlap along the whole frequencyband, the frequency nonlinearity, which typ-ically exist in AOSs, has to be taken into ac-count. This nonlinearity has been measuredin all four bands to be identical over the whole1 Gliz bandwidth. Furthermore, although itis very low about I frequency channel alongthe full bandwidth it can be further reducedto about 0.1 channels by means of a prism.This identical frequency response of the

Figure 5: The top graph shows simultane-ously measured noise spectra in both AOSbands at high illumination level together withthe difference spectrum. The noise of suchdifference spectra, measured at various illu-mination levels, has been scaled down to thecontribution of one single CCD relative to ra-diometric noise, as shown in the lower graph

various bands is a precondition, to elimi-nate identical noise apparent in two differentbands. To check this performance, the out-put of a noise source has been split up andfed into two bands of the array AOS. Sub-tracting both simultaneously measured noisespectra, only the detector resides. The theo-retical noise of the CCD has been calculatedto be identical to the noise of the differencespectrum. The difference spectrum in Fig. 5directly reflects the identical frequency re-sponse of the 1 GHz bands.

318

Decreasing the illumination level, that is.the IF drive power level at the Bragg cellthe detector noise of the AOS. which con-sists mainly of shot- and dark-noise of theCCD, should increase. In that sense. one candefine a noise dynamic range as the illumi-nation level, where the CCD adds less than1 dB noise to the radiometric noise.

The noise spectra described above havebeen repeated at various IF drive power levelsof the noise source. The lower plot in Fig. 5shows the noise contribution of the CCD rel-ative to radiometric noise. One can see, thatthe noise is increased by 1 dB at an illumina-tion level of about 5-6 %, yielding a noise dy-namic range of at least 12 dB. If the CCD is il-luminated below this level, the noise increasesdrastically, as is indicated by the steep slopeof the curve.

3.2 Astronomical measure-ments

1907: 1 TRC+ 10216 13C0(2-1) !RAM-3014-831 0: 15 — MAY —1997 R: 15—MAY —1997RA: 9:45:14.800 DEC: 13:30:40.0C (1950.0) Offs: 0.0 0.0 EciUnknown Tau: 0.2870 Try,: 462.0 Tine: 15.00 a: 32.54

512 KO: 258.5 VO: —27.00 De: 1.380 LSRTO: 220398.682 Df: — 1.000 228282.941

Figure 6: A measurement of 13C0(1-0)IRC+10216 to test the calibration scale of theAOS compared to the filterbank

The calibration of the array AOS has alsobeen checked, using well known sources, likeIRC+10216, where the line peak tempera-

300 ■CCtwspasy Cloaoca

6

20C 40CSca, 5062 Doc., 6,

Figure 7: The baseline quality at illumina-tion levels below the specified range for theAOS; the lower spectrum is measured withthe AOS, the upper one with the filterbank.The bandpass shows a 500 MHz band mea-sured with the AOS, on Hot Load. Satura-tion is reached at about 251(r counts, the12dB noise dynamic range limit is thereforereached at roughly 2106 counts.

ture is known. The IRAM filterbanks havebeen used in parallel to directly compare theresults. For an interpretation, one has totake into account, that the AOS has a lowerresolution (resolution bandwidth 1.5 MHz)than the filterbank (resolution bandwidth1.0 MHz). In AOS spectra, narrow line fea-tures must therefore be frequency diluted andshow a lower peak response as lines measuredwith the filterbank. This is nicely visible inFig. 6, where the curve with the lower peakresponse has been measured with the arrayAOS. The identical response in the flat part

319

r:2 cr.

, A A2:2

,.....,....", ail '.....A., — ,---.........- ,./..

1 1 1 1C r

, i ;

-2 i-, AC ,t r

iA............,,,........:, "" ''..",..—..1". 1"..................--. :

I f ,,,_!,

,

1. pa -11- -

-4 II-

,

of the line center demonstrates on the otherhand that the calibrations are identical. ascan be seen in the left zoom. The velocityrange from 0 to 100 km s zoomed. showsthat features in the baselines are also nearlyidentical. The ratio of the RMS values isconsistent with the ratio of the fluctuationbandwidths, which is 2.3 MHz for the arrayAOS and 1.4 MHz for the filterbank (Wild,1995). For the calculation of the RMS values,line window have been set as are indicated inFig. 6, so that the bad filterbank channel isnot regarded.

During normal observations after installa-tion of the array AOS, B. Lazareff et al. havegenerated some radiometric test data at 500MHz bandwidth on Venus, again using thefilterbanks in parallel, see Fig. 3.2. One cansee excess noise at the left edge of the arrayAOS spectrum, whereas the Filterbank hasonly little of this excess noise.. The reason,that this only becomes visible at the edgesof the spectra is due to the low IF powerat the edges of the spectrum, which is fedinto the „ADS as is shown in Fig. 3.2. Thebandpass for the 500 MHz band is shown inthe lower graph. The 12 dB noise dynamicrange limit is reached at about 2 10 6 counts.This limit has been calculated, assuming3dB difference between Hot Load and skyand about 25 106 counts for saturation. Theexcess noise indicates the limits of the CCD,which has to be illuminated above the 12 dBnoise dynamic range limit, in the way it isoperated at the moment. One could thinkof improvements in the CCD itself, in termsof looking for a device with better noisecharacteristics. But then the 12 Bit Analogto Digital Converter (ADC) used is thelimiting factor, as the inherent differentialnon-linearity (DNL) then contributes to thenoise. The effect of the DNL is alreadyobvious at the very left edge, below I 106

counts of the bandpass. or within the first20 frequency channels, due to the powerdifference between On- and Off-position onVenus. (iarnot R. F.. 1988). The effect ofDNL can easily be reduced by inserting 14or 16Bit ADCs instead of the 12Bit ADCused in this prototype version. A detaileddiscussion of the impact of the D.NL on thenoise performance at low illumination levelswill be published soon. Furthermore. theDNL affects the baseline only when thereare power offsets between the On and theOff position. Depending on how strong theseoffsets are, the DNL might already becomea significant contributor to excess baselinenoise next to the CCD. Sensitive measure-ments therefore require higher Bit resolutionADCs. When using 14 or 16 Bit ADCs,it would also make sense to improve theCCD characteristics or the readout speed,so that a noise dynamic range of 15 dB ormore could be achieved. Furthermore one

2636. I SCRCAOS :3C0(: -0) Cali -WM-BSI 0 :7-MAY- 1947 R. 17-MAY- 90,RA: 17:4/ 23.68: DEC: -291422.0C C1950.0: on- -0250 0.0 EgUniczumsa Tat, 7.5166E-02 'rm. 302.2 Time, 10.00 EL- 23.'7CIN: 50; IC: 251.0 VO: C.0000E4-00 -2129

70: 110201.370 D: 1.040 1 I 3 /117.118

-400 -200 0 200 400Velocity (km/s;

Figure 8: Simultaneous measurements at500 MHz bandwidth with the array AOS, El-terbank, and hybrid autocorrelator

can think of introducing an r. f. equalizer toreduce the bandpass variation of the AOSfrom 3 dB at the moment to about 1 dB.Filterbank spectra would also be affected

320

about 500 KHz. 675 lans4L-2.5 962 AOS Z rms 57 inK

t:zg,EOw(AOS) = 1.82 • Rw(IXHZ)

-> es.(1101.) 126 Wiz (for Efe.(AO9)42.33 mK (Be■2.31C1lz. bat.. 32142t. 14,5.5:72 X)

by DNL if the digitization is done withADCs. At the TRAM filterbanks however,Voltage Controlled Oscillators with followingCounters are in use. The exess noise at theedges is therefore only detector noise.

Further measurements on Sgr C using500 MHz bandwidth have been performedwith the array AOS, filterbank, and hybridautocorrelator in parallel, see Fig. 8. As inFig. 6, one can see the frequency dilutionof the array AOS due to its lower resolu-tion compared with the filterbank. This be-comes visible in the absorption feature at 0km s-1 as well as at the IF spike at -NOkm s- 1 . Besides platforming, the autocorrela-

2: 2 SIOS 12C0(2-1) IRAli-3011-862 0: 13-JUN-192, Ft 13-SUN-2s97RA: 20:25:33.801 DEC: 37:12-10.00 (1950.0) Offs: .4.020 4..917 E,Unknown Tatt: .5100 Days: 617.3 Time: 535.7 M: :7.19

N: 501 10: 251.0 VO .0000E+0C De: -1.35/ UnicaTO: 230793.990 Of: 1.040 236771.132

Sum of 90 OTF SCO133 On 3106 (about 9 hours). no baselines subtracted. 3.order baseline plottedRest Frequent.; (KHz)

2.31 10 2.309 IO 2.308 10 2.307 /03 2306 ze-/.5

-20C 0 ZOCVelocity (kom/s)

Figure 9: Simultaneous On-the-fly (OTF)measurements at 500MHz bandwidth withthe array AOS (top) and the filterbank (bot-tom)

tor shows much lower resolution as the nom-inally expected 1.25 MHz. One can see thisat the absorption feature, which does not be-come visible in this spectrum, as well as at theabsorption spike. This is probably caused bythe process of smoothing the data with someapodization to suppress the sidelobes, whichbecome visible if narrow lines are observed.

Schneider et al. mapped S106 on-the-fly(OTF). Summing up all 90 OTF scans, yields

to the spectrum shown in Fig. 9. After a totalintegration time of about 9 hours. the filter-bank spectrum shows several bad channels.The AOS spectrum shows some excess noiseat the right edge, where the illumination levelis below the specified 12 dB. otherwise. thebaselines would be exactly identical at thisedge as they are for the rest of the baseline.

ARRAY AO SNEXT GENERATION

4Z?

Figure 10: The mechanical concept of thenext generation array AOS. The optical pa-rameters are the same as with the first ver-sion.

The array .AOS presented herein served asa prototype and showed already very good re-sults. Nevertheless, the next versions e. g. forAST/R0 or KOSMA, are being redesigned.to make them more compact and to have al-ready a kind of 'engineering model' for mis-sions like FIRST or SOFIA. The prinicpaloptical setup of the next generation AOS isshown in Fig. 10. It differs from the firstversion only in the folded optical layout andsome means to facilitate the alignment.

4 ConclusionsThe performance tests have shown, that theconcept of the KOSMA array AOS is suit-able for radio astronomical applications atmillimeter and submillimeter observatories.Even for heterodyne receiver arrays in the

321

THz region, which can be implemented atSOFIA, the array AOS provides sufficient fre-quency coverage to allow the detection of ex-tragalactic signals. At 2 THz, the bandwidthof 1 GHz corresponds to 150 km s- 1 , which isstill sufficient for most galactic observations.If the velocity range needs to be larger. theAOS bands could also be joint together asdone with hybrid autocorrelators. The iden-tical noise performance of the different bands,if illuminated properly, are a decisive precon-dition to avoid different noise levels in thehybrid band. As there are no obvious differ-ent drifts within the two bands, platformingshould not arise. At the moment, an IF pro-cessor is under development, to split a 2 GHzIF band and to feed the signals into two AOSbands. The performance of hybrid spectracan then directly be tested.To improve the noise dynamic range of theAOS from now 12 dB to at least 15 dB are onthe way, enhancing the Bit resolution of theanalog to digital converter and the readoutspeed of the CCD.

5 AcknowledgementsWe like to thank TRAM and the 30m staffto realizing the test-run with the arrayAOS, especially B. Lazareff and G. Paubertas well as N. Schneider, C. Kramer et al. andH. Wiesemeyer for providing additional testdata. The array AOS has partly been fi-nanced through grant SFB 301 of the DFG(Deutsche Forschungsgemeinschaft).

6 References

GEC-Marconi, Test Results of the arrayBragg Cell, Device Identity: Y-36-7436-02,Serial No.: MRQ IS001 (1995)

Horn J.. The development of an arrayacousto-optical spectrometer. Ph. D. thesis.Univ. Köln (1997)Jarnot R. F.. Technical Note 1. JPL Mi-crowave Atmospheric Sciences Group (Aug.18.. 1988)Klumb M.. Frerick J.. Tolls V.. Schieder R..Winnewisser G., The SiiAS Acousto-OpticalSpectrometer. SPIE Vol. 2268. p305 (1994)Schieder R.. Tolls V.. Winnewisser G.The Cologne Acousto-Optical Spectrometers,Exp. Astron. 1. 101-121 (1989)Schneider N., Kramer C. et al., IRAM project39/97 (1997)Wild W., The 30m Manual, Version 1.0.TRAM Granada (1995)

322