Optimization of SDSU-2 CCD controller hardware and software for CCD mosaics

Transcription

1 Optimiztion of SDSU-2 CCD controller hrdwre nd softwre for CCD mosics Roert I. Kirick, Chris Wright, Steven L. Allen, De A. Clrke UCO/Lick Oservtor, Universit of Cliforni, Snt Cruz, CA USA ABSTRACT The Sn Diego Stte Universit Genertion 2 CCD controller (SDSU-2) 1 rchitecture is widel used in oth opticl nd infrred stronomicl instruments. This rchitecture ws emploed in the CCD controllers for the DEIMOS instrument commissioned on Keck-II in June In 2004, the CCD dewr in the HIRES 2 instrument on Keck-I will e upgrded to 3x1mosicofMIT/LL2Kx4KCCDscontrolled n SDSU-2 CCD controller. For ech of these SDSU-2 CCD controllers, customized versions of PAL chips were developed to extend the cpilities of this controller rchitecture. For oth mosics, custom timing ord PAL enles rpid, softwreselectle switching etween dul- nd single-mplifier-per-ccd redout modes while reducing excess utiliztion of fier optic ndwidth for the ltter. For the HIRES CCD mosic, custom PAL for the clock genertion ords provides softwre selection of different clock wveforms tht cn ddress the CCDs of the mosic either individull or gloll, without n need to reset the ddress jumpers on these ords. The custom PAL for the clock genertion ords enles method for providing differing exposure times on ech CCD of the mosic. These distinct exposure times cn e implemented in terms of series of su-exposures within single, glol mosic oservtion. This llows for more effective oserving of sources tht hve flux grdients cross the spectrl dimension of the CCD mosic ecuse those CCDs locted ner the higher end of the flux grdient cn e red out more frequentl, thus reducing the numer of cosmic rs in ech individul su-exposure from those CCDs. Kewords: SDSU-2 CCD controller, DEIMOS, HIRES, CCD mosic 1. INTRODUCTION The DEep Imging Multi-Oject Spectrogrph 3 (DEIMOS) contins n 8K x 8K pixel science mosic composed of eight 2K x 4K MIT/Lincoln L (MIT/LL) CCID-20 4 CCDs, rrnged in two rows of four CCDs ech (See Fig. 1). This ws the first CCD mosic sstem ever ssemled UCO/Lick Oservtor, nd s such, required development of our first mosic-cple CCD controller sstem. Tht sstem is sed on the SDSU-2 CCD controller rchitecture. Tht rchitecture uses VME-stle ckplne us to link together the controller ords tht comprise the sstem. There re four tpes of SDSU-2 ords used in this controller: video processing, clock genertion, utilit, nd timing. The video processing ords generte the is voltges for the CCDs nd process their video outputs. Ech ord provides two video input chnnels. Ech video chnnel contins the nlog circuitr used to perform correlted doule smpling of the CCD video signl nd high-speed (1 MHz) nlog-to-digitl converter (ADC). The digitized pixel dt is held in digitl ltch tht cn e red the timing ord vi the ckplne us. Ech ord contins two sets of jumpers used to specif tht ord s two ddresses: the switch-stte ddress nd the DAC/ltch ddress. The first is used to ddress the video processing circuitr nd the ADCs for oth video chnnels, while the second is used to ddress the digitl-to-nlog converters (DACs) tht generte the is voltges nd the digitized pixel dt ltch for ech video chnnel. Multiple ords cn e jumpered to the sme switch-stte ddress so tht the ll perform their video processing in precise snchroniztion. Further uthor informtion- (Send correspondence to R.K.) R.K.: Emil: kirick@ucolick.org; kirick; Telephone: ; Fx: ; C.W.: cwright@ucolick.org; Telephone: ; FAX: ; S.A.: Emil: sl@ucolick.org; Telephone: ; D.C.: Emil: de@ucolick.org; Telephone:

2 The clock genertion ords re used to generte the clock wveforms for the CCDs. Ech ord provides two sets of 12 clocks nd ech clock cn e progrmmed over rnge of -10 to +10 volts. These ords lso hve oth switch-stte nd DAC ddress. The first is used to ddress the nlog switches tht select etween the high nd low voltge rils for ech clock, while the second is used to ddress the DACs the generte those voltge rils. Multiple clock genertion ords cn e jumpered to respond to the sme switch-stte ddress in order to generte precisel snchronized clocks for ll of the CCDs of the mosic. The utilit ord provides vrious housekeeping functions tht include: 1) generting the signls to operte nd sense the stte of shutter, 2) monitoring temperture-sensing diodes in the CCD dewr, 3) djusting the voltge pplied to heter resistors used to regulte the dewr temperture 4) monitoring the sttus of the controller power supplies. The timing ord communictes with the other ords in the sstem vi the VME-stle ckplne us nd it performs the following five functions: 1) directs the opertion of those other ords in the sstem in response to externl commnds it receives over fier optic uplink, 2) genertes the timing ptterns tht re sent to the clock genertion ords to crete the wveforms used to clock the CCDs, 3) snchronizes the opertion of the video processor ords s the process nd digitize the video outputs from the CCDs, 4) reds the digitized pixel dt from the digitl ltches on the video processing ords, nd 5) trnsmits those digitized pixels vi fier optic downlink to computer where the CCD imges re stored. Figure 1. DEIMOS CCD mosic nd FCS CCDs Figure 2. The DEIMOS mosic CCD controller The DEIMOS mosic CCD controller 5 is shown in Fig. 2. It emplos these four tpes of SDSU-2 ords (locted in the ottom hlf of the controller) in comintion with three UCO/Lick-designed printed circuit ords: the clock cle, is cle, nd CCD cle interconnect ords; those ords re locted in the top hlf of the controller. Connections etween these ords re vi rion cles tht re esil fricted using insultion displcement connectors. The interconnect ords simplif the signl distriution from the controller to the CCD mosic dewr collecting into single cle for ech CCD the vrious clock, is, nd utilit signls for tht device, which originte from three seprte SDSU-2 ords. The interconnect ords lso simplif reconfigurtion of the controller for different mosics or operting modes. The DEIMOS mosic CCD controller cn e configured to provide single-mplifier-per-ccd redout of the 8-CCD mosic using onl four video processing ords, or to provide dul-mplifier-per-ccd redout using eight of those ords. During the first ers of testing the mosic susstem using engineering-grde CCDs (some of which hd onl one working redout mplifier), the former configurtion ws used. Once enough science grde devices with two working redout mplifiers were otined, four more SDSU-2 video processing ords were purchsed nd the sstem re-configured to provide dul-mplifier-per-ccd redout cpilit. Upgrding the controller from single-mplifier to dul-mplifier-per-ccd redout cpilit required onl the instlltion of the dditionl video processing ords, the re-rrngement of the rion cles etween the vrious interconnect ords, nd the resetting of some jumpers on those interconnect ords.

3 In the DEIMOS mosic CCD controller, the clock genertion ords ll hve their switch-stte ddress jumpers set to the sme vlue (ddress 2) nd the video processing ords re similrl jumpered to common switch-stte ddress (ddress 0). As result, the clocks sent to ll eight CCDs of the mosic re generted in prllel nd re precisel snchronized; the video processing for ech CCD is similrl snchronized. This enles ll eight CCDs of the mosic to e red out in prllel nd voids the increse in red noise tht would occur if such snchroniztion were not mintined The mpping of video outputs to video inputs Ech SDSU-2 video processing ord provides two independent video chnnels, leled A nd B. Ech MIT/LL CCID-20 CCD hs two redout mplifiers, lso leled A nd B. When configured for dulmplifier-per-ccd redout mode, logicl rrngement is to provide one video processing ord for ech CCD nd to route the A nd B video outputs from ech CCD to the corresponding A nd B video inputs of the video processing ord for tht CCD. Since these ords lso generte is voltges, the video ord for given CCD cn e used to generte seprte is voltges for oth the A nd B sides of the device. As shown in Tle 1, this is the mpping tht ws used for the DEIMOS mosic CCD controller. Video Video Chnnel Bord Video Ltch DAC Bord CCD Address Address Chnnel CCD Amplifier 0 0 A 1 A 1 0 B 1 B 2 1 A 2 A 3 1 B 2 B 4 2 A 3 A 5 2 B 3 B 6 3 A 4 A 7 3 B 4 B 8 4 A 5 A 9 4 B 5 B 10 5 A 6 A 11 5 B 6 B 12 6 A 7 A 13 6 B 7 B 14 7 A 8 A 15 7 B 8 B Tle 1. DEIMOS mpping of CCD mplifiers Video Video Chnnel Bord Video Ltch DAC Bord CCD Address Address Chnnel CCD Amplifier 0 0 A 1 A 1 0 B 2 1 A 1 B 3 1 B 4 2 A 2 A 5 2 B 6 3 A 2 B 7 3 B 8 4 A 3 A 9 4 B 10 5 A 3 B 11 5 B Tle 2. HIRES mpping of CCD mplifiers Unfortuntel, the SDSU-2 video processing ords exhiit ver low-level crosstlk (pproximtel 1 prt in 20,000) etween the two video chnnels on ech ord. While this m not e concern for mn direct imging pplictions, it is prolemtic for multi-slit spectroscopic oservtions of fields contining mix of right nd fint ojects. In such situtions, further dvntge of this mpping of video outputs to inputs (i.e., A to A, B to B ) is tht it provides workround to this crosstlk prolem. B switching to single-mplifier-per-ccd redout clocks for such oservtions (with corresponding penlt in redout time), the mosic CCDs imges cn e red out using onl one hlf (i.e., one of the two video input chnnels) of ech video processing ord, there voiding the crosstlk prolem on these ords. For the HIRES mosic CCD controller, the decision ws mde to void this inter-chnnel crosstlk prolem using onl hlf-populted video processing ords (the B chnnel of ech ord is not used nd the components on tht hlf of the ord re not instlled). As result, ech CCD in the HIRES mosic connects to two hlfpopulted video processing ords, s shown in Fig. 2. Note tht onl the ords with even-numered DAC ddresses re used for is voltge genertion.

4 2. TRANSMISSION OF ALTERNATE CHANNELS In the first genertion SDSU rchitecture, progrmmed I/O ws used to trnsmit the digitized pixel dt from the CCD controller to the downstrem computer. Softwre on the timing ord would first red the digitized pixel dt from the ltch on ech video processing ord nd then cop tht dt into the trnsmit register for the fier optic downlink. While this rchitecture enles the timing ord to selectivel trnsmit the pixel dt from n ritrr suset of video processing chnnels, it consumes significnt frction of the per-pixel processing time in multi-chnnel CCD mosic sstem, ecuse the timing ord cnnot overlp other functions (e.g., genertion of seril clocks or processing of CCD video) with pixel trnsmission. While the SDSU-2 rchitecture retins this progrmmed I/O cpilit, one of its mjor design improvements is its ilit to overlp the trnsmission of digitized pixel dt (over the fier optic downlink) with these other pixel-processing functions. This is ccomplished in hrdwre mens of two progrmmle rr logic (PAL) circuits, U12 nd U17, on the SDSU-2 timing ord. To utilize this feture, the softwre on the timing ord stores into control register in the PAL the strting nd ending ltch ddress numers for the inclusive rnge of video chnnels whose ltched dt is to e trnsmitted vi the fier downlink. The ct of writing tht control register triggers the relevnt circuitr in these PALs to sequentill ccess nd trnsmit the digitized pixel dt from the specified rnge of video chnnels. The DEIMOS mosic CCD controller mkes extensive use of this overlpped-trnsmission feture. At the end of ech pixel processing time slot, the softwre on the timing ord writes into the control register the desired rnge of video chnnels to e trnsmitted. Once tht control register hs een written, the timing ord softwre cn egin the per-pixel processing for the next pixel. As result, the trnsmission of the pixel dt (for the specified video chnnels) cquired t pixel time slot N is overlpped with the pixel processing (i.e., seril trnsfer, video processing, nd digitiztion) for pixel time slot N + 1. Unfortuntel, s originll implemented, this overlpped-trnsmission functionlit did not provide sufficient flexiilit for the vrious redout modes of the full-configured DEIMOS mosic CCD controller. As cn e seen from Tle 1, the DEIMOS mosic nd its controller hve een configured so tht the A mplifiers from ech CCD occup the video input chnnels hving even-numered ltch ddresses while the B mplifiers from ech CCD occup those with odd-numered ltch ddresses. Accordingl, for the single-mplifier-per-ccd redout modes, it would e desirle to trnsmit onl the video chnnels with even-numered ltch ddresses (if using onl the A mplifiers) or those with odd-numered ddresses (if using onl the B mplifiers). One pproch to this prolem is to lws trnsmit the pixel dt from oth the A nd B video chnnels (even if onl one hs useful dt, s would e the cse when reding out in single-mplifier-per-ccd mode) nd to discrd the unwnted pixels t the receiving end of the fier downlink. This pproch is fesile s long s: 1) there is sufficient processing power t the receiving end of the downlink to keep up with the higher pixel rte nd 2) the effective ndwidth of the downlink is sufficient to sustin tht trnsmission rte. This pproch ws used implemented modifing the softwre in the fier optic interfce ord (VMEINF- 2) tht is ttched to the receiving end of the downlink fier; tht ord resides in the computer (in this cse, VME crte running VxWorks) tht receives nd uffers the CCD mosic imges. The softwre for tht interfce ord ws modified so tht it could e commnded to either ccept ll pixels or to discrd either even-numered or odd-numered pixels rriving on the fier optic downlink. The first mode (ccept ll pixels) ws used when reding out the mosic in dul-mplifier-per-ccd mode. When reding out in single-mplifier-per-ccd mode, the discrd odd pixels mode ws used if reding from the A mplifiers of the CCDs in the mosic, nd the discrd even pixels mode ws used if reding from the B mplifiers. This scheme ws used to support oth the single- nd dul-mplifier-per-ccd redout modes during the period tht the DEIMOS mosic CCD controller ws operting the first engineering nd science mosics composed of MIT/LL lot 9 nd 10 devices. Due to prolems tht occurred during wfer processing, these devices hd mrginl seril chrge trnsfer efficienc (CTE) when operted with seril clocks hving nrrow (i.e., less thn 1 microsecond) overlp etween seril phses. As result, these mosics were operted with reltivel slow seril clocks, resulting in pixel processing time of out 10 microseconds. Such slow pixel processing times txed neither the processing speed of the fier optic interfce ord nor the effective ndwidth of the fier optic downlink, so this scheme of discrding pixels t the fr end of the fier downlink proved dequte.

5 However, this scheme ws not dequte to support the fstest single-mplifier redout modes for the finl DEIMOS science mosic tht ws composed of lot 14 MIT/LL devices; those devices chieved excellent seril CTE when operted with seril clocks hving onl 0.1 microseconds of overlp etween seril phses. Those nrrow overlps mde possile per-pixel processing times of 5.3 microseconds nd tht is shorter thn the time required to trnsmit 16 pixels of dt down the fier optic downlink. Becuse this trnsmission rte exceeded the ndwidth of the downlink, we could no longer fford the luxur of discrding the empt pixels t the fr end of the the downlink. Figure 3. Schemtic for revisions to Timing Bord PAL U12 In the revision 4B timing ords used in DEIMOS, the fier optic downlink opertes t 50 MHz, or 20 nnoseconds per it. Ech digitized pixel is sent s 16-it unsigned integer nd thus requires 320 nnoseconds for trnsmission down the fier. Add to this two DSP clock ccles of 40 nnoseconds ech: the first to fetch the pixel dt from the digitl ltch on the respective video processing ord vi the VME-stle ckplne us nd the second to write tht dt into the fier optic trnsmitter register. Thus, this overlpped-trnsmission feture requires 400 nnoseconds ( ) for ech pixel trnsmitted, ielding n effective downlink ndwidth of 2.5 megpixels/second. The time required to trnsmit 16 pixels of dt is 6.4 microseconds.

6 To ddress this limittion, we sked Astronomicl Reserch Cmers, Inc., to explore the possiilit of modifing the pixel trnsmission PAL logic to enle the selective trnsmission of pixels onl from video chnnels with either even-numered or odd-numered ltch ddresses. The developed such modifiction to PAL circuits U12 nd U17 on the Rev. 4B timing ord (See Figs. 3 nd 4). Since these PAL circuits re socketed devices, the re esil chnged on the ords in question, enling the upgrded PALs to e instlled t the telescope. Figure 4. Schemtic for revisions to Timing Bord PAL U17 This modified functionlit is ccessed vi its 10 nd 12 of the PBD port of the Motorol DSP on tht timing ord; the opertionl definition of those two its is provided in Tle 3. Bit Bit Functionlit 0 0 Trnsmit onl pixels from even-numered ltch ddresses 1 0 Trnsmit onl pixels from odd-numered ltch ddresses X 1 Trnsmit pixels from oth even- nd odd-numered ltch ddresses Tle 3. New it definitions for PBD its

7 The ilit to selectivel trnsmit the pixel dt from video chnnels with either even-numered or oddnumered ltch ddresses proved prticulrl useful for the HIRES mosic. As noted in section 1.1, the HIRES mosic CCD controller emplos onl hlf-populted SDSU-2 video processing ords. As such, onl the video chnnels with even-numered ltch ddresses contin n dt (see Tle 2). Accordingl, the HIRES CCD controller lws opertes with the PBD register configured for even-onl trnsmission, nd the pixel dt from oth the A nd B mplifiers of ech CCD re lws sent down the fier optic downlink. If the HIRES mosic is red out in single-mplifier-per-ccd redout mode, then the unwnted pixels re discrded t the fr end of the downlink. Since the HIRES mosic hs onl 3 CCDs (nd hence onl 6 mplifiers), the ndwidth of the fier downlink is not concern. 3. SELECTIVE READOUT OF CCDS WITHIN A MOSAIC The HIRES CCD mosic (slted for instlltion t Keck during the summer of 2004) consists of 3 x 1 mosic of MIT/LL CCID-20 2K x 4K pixel CCDs; these CCDs hve 15 micron pixels. This mosic will replce the existing Tektronix 1K x 1K (24 micron) pixel detector, which hs een prt of the instrument since it ws commissioned in The mosic will provide significntl incresed quntum efficienc, especill in the lue. Together with the new mosic CCD controller, it will reduce the redout time nerl fctor of 5 nd cut the redout noise nerl fctor of 2. The lrger effective detector re will enle most of the HIRES Echelle formt to e cptured in single exposure. A gol (lthough not requirement) of this detector upgrde project ws to provide the hrdwre hooks tht would llow the individul CCDs of the mosic to e red out seprtel. Provided tht the CCDs in tht mosic do not exhiit n pthologicl ehvior (e.g., glowing mplifiers or seril register pixels) when the seril clocks re operted without corresponding prllel clocks, this gol cn e chieved if the CCD controller cn e configured so s to generte independentl the prllel clocks for ech CCD of the mosic. However, tht cpilit needs to e implemented in mnner tht does not compromise the performnce of the sstem when ll of the CCDs of the mosic re red out snchronousl. As mentioned in Sect. 1, one of the ke fetures of the SDSU-2 rchitecture (tht enles it to perform efficient clocking of CCDs) is the ilit to set the switch-stte ddresses for ll clock genertion ords to the sme ddress. B configuring the ords in this w, ech clock trnsition cn e sent simultneousl to ll of the CCDs of the mosic, thus significntl reducing the redout time compred to the cse where ech such trnsition is seprtel commnded to ech clock genertion ord. But if one sets ll of the clock genertion ords to respond to the sme switch stte ddress, then it is no longer possile to generte the prllel clocks independentl for ech CCD, s is needed to chieve the specified gol. One possile solution would e to ssign the seril clocks (for ll of the CCDs of the mosic) to set of clock genertion ords whose switch-stte ddress jumpers were ll set to the sme ddress, while ssigning the prllel clocks for ech individul CCD to seprtel-ddressle clock genertion ord. This would enle the seril clock wveforms (which need to e generted s rpidl s possile) to e sent simultneousl to ll of the CCDs of the mosic, while enling the prllel clocks (which operte much more slowl) to e generted seprtel for ech CCD. While such n pproch is possile using the stndrd SDSU-2 rchitecture, it is incomptile with the existing design for our clock cle interconnect ords nd cling. Also, for CCD mosics with n even numer of CCDs, this pproch would require more clock genertion ords nd more slots in the VME-stle ckplne of the CCD controller. Further, to simplif spring nd trouleshooting, we wnted to use the sme interconnect scheme s ws used in three other SDSU-2 CCD controllers lred in use t Keck Oservtor (the ESI CCD controller, the DEIMOS mosic CCD controller, nd the DEIMOS FCS CCD controller). In ddition, mintining the identicl mppings of CCD clock signls to clock genertion ords simplified documenttion nd enled re-use of identicl MIT/LL CCD wveforms s used in the ESI nd DEIMOS sstems. The SDSU-2 clock genertion ords provide 24 clocks per ord. These re divided into two nks of 12 clocks ech. Ech set cn e individull ddressed, or oth nks cn e ddressed in prllel. Our interconnect ord design ssigns the 12 clocks for ech MIT/LL CCD (9 seril clocks plus 3 prllel clocks) to one of these two nks. Thus, ech clock genertion ord cn generte the clocks for two MIT/LL CCDs.

8 3.1. Brodcst ddress for clock genertion ords Our solution ws to propose chnge to the PAL circuit (U100) tht performs the switch-stte ord-ddress decoding for the clock genertion ords (See Fig. 5). This modifiction defines clock genertion switch-stte ord ddress 12 to e wild-crd or rodcst ddress. With the modified version of the U100 PAL instlled, clock genertion ord will respond if the switch-stte ddress currentl sserted on the VME-stle ckplne (input signls SS12 through SS15) mtches: 1) the vlue of the switch-stte ddress jumpers on tht ord (input signls SADD5 through SADD7), or 2) switch-stte ddress 12 (the rodcst ddress). (#13) (#16) TIM_A_ENCK TIM_A_CDAC x x U12 U13 CLKEN (#27) DAC_CLR (#26) (#3) SS12 (#4) SS13 (#5) SS14 (#6) SS15 (#9) SADD5 (#10) SADD6 (#11) SADD7 (#12) MODE U4 A(0) A(1) A(2) B(0) B(1) B(2) U5 A(0:2) B(0:2) U1 c A(0:2) B(0:2) U3 OE COMP3 U2 U14 x ADD_E U7 U8 c (#23) U6 U9 U10 s d q U11 r s d q U15 r U17 GND U18 x U19 x U16 OUT_LTCH_A (#19) OUT_LTCH_B (#18) (#2) TIM_A_WRSS Figure 5. Schemtic for revisions to Clock Genertion PAL U Different ddressing modes for prllel nd seril clocks With this modified U100 PAL circuit instlled, the two clock genertion ords in the HIRES mosic CCD controller hve their switch-stte ddress jumpers set to distinct vlues, enling ech CCD in the mosic to e clocked individull, if desired. However, defining clock wveforms tht reference the switch-stte rodcst ddress (12), ll of the CCDs in the mosic cn e clocked in unison. Accordingl, the prllel clock wveforms for the HIRES CCD mosic re defined in terms of the individul switch-stte ddresses so tht the prllel clocks cn e sent independentl to single CCD or to suset of the CCDs in the mosic. The seril clock wveforms for the mosic re defined in terms of the switch-stte rodcst ddress so tht the the seril clocking of ll of the CCDs in the mosic occurs in unison. Since the switch-stte ddresses for the video processing ords re lso set to common ddress (ddress 0), the video processing for ll of the CCDs in the mosic lso occurs in unison. Currentl, the switch stte ddress for ech ord cn e set to n even-numered vlue etween 0 nd 14. The top nk of 12 clocks on ech ord is ddressed vi the even-numered switch stte ddress vlue encoded the setting of tht ord s ddress jumpers, while the ottom nk of 12 clocks is ddressed vi the next higher (odd-numered) ddress. However, if it 1 of the timing ord s WRLATCH register is set to 1, then oth the top nd ottom nks of clocks re ddressed simultneousl whenever the even-numered switch stte ddress is referenced. The clocks for CCD 1 nd 2 re generted, respectivel, the top nd ottom nks of clocks on the clock genertion ord t switch-stte ddress 2, nd the clocks for CCD 3 the top nk of clocks on the ord t ddress 4. Thus, the clocks for ech of the three CCDs in the mosic cn e ddressed individull using switch stte-ddresses 2, 3, nd 4.

9 4. SOFTWARE FOR TAKING SUB-EXPOSURES WITHIN A MOSAIC When oserving reltivel right oject with steep spectrl grdient (e.g., one tht is significntl righter in the red thn in the lue), if one exposes long enough to chieve high signl-to-noise t the lue end, the red end m e overexposed. Conversel, if one sets the exposure time to void sturtion in the red, the lue end m e insufficientl exposed. In other cses, n oject m e sufficientl fint tht over-exposure is not concern on n CCD, ut the exposure t one end m integrte ove the redout noise more rpidl thn on the CCD t the other end. In either cse, when integrting on such ojects, it m e dvntgeous to e le to red out some of the devices of the mosic more often thn others. The spectrl distriution cross the HIRES mosic is shown in Fig. 6. The mosic is positioned such tht the lue end of the HIRES Echelle formt flls on the CCD in position 1 nd the red end flls on the device in position 3. The nti-reflection coting on ech CCD hs een optimized to provide the est trnsmission for the wvelength rnge tht will fll on ech ech detector. Hence, the CCD in position 1 is referred to s the Blue, or B CCD, the device in position 2 s the middle, Green, or G CCD, nd the device in position 3 s the Red or R CCD. PIXEL 1,4096 FOCAL PLANE Y PIXEL 6144,4096 FOCAL PLANE X POSITION 1 POSITION 2 POSITION 3 PIXEL 1,1 A SERIAL READOUT B A SERIAL READOUT BLUE CCD MID CCD RED CCD B A SERIAL READOUT B PIXEL 6144,1 3 MIT/LL 2048x m PIXEL CCD'S Figure 6. HIRES CCD mosic Figure 7. The MSE su-pnel on the HIRES GUI The ilit to generte distinct prllel clocks for ech CCD of the mosic enles new tpe of exposure mode which we refer to s sequence of multiple su-exposures (MSE). At the end of such n MSE sequence, the totl exposure time for ech CCD of the mosic is the sme. However, during the sequence, some CCDs re red out more often thn others, depending on the spectrl grdient cross the mosic. For exmple, ssume tht the HIRES mosic is eing used to otin the spectrum of n oject which is much righter in the red thn in the lue. Further ssume tht 3600 s exposure is needed to otin sufficient signl to rech ove the redout noise of the Blue CCD, tht onl 1800 s exposure is needed to get ove the redout noise of the Green CCD, nd tht onl 1200 s exposure is needed for the Red CCD. B tking n MSE sequence consisting of three 1200 s exposures on the Red CCD, two 1800 s exposures on the Green, nd one 3600 s exposure on the Blue, one could perform more effective cosmic r removl on the Red nd Green devices, since the would e red out more frequentl. At the end of the MSE sequence, ll three CCDs of the mosic would hve integrted for totl of 3600 s on the given oject. Tle 4 provides detiled rekdown tht shows how the exmple MSE sequence (descried ove) would e crried out. This rekdown ssumes n erse ccle time of 7 seconds nd redout time of 35 seconds. Note tht four seprte exposures re tken in this sequence, nd tht for ech exposure, onl suset of the CCDs in the mosic re ersed or red out. Becuse four seprte exposures re tken, the totl MSE sequence tkes 3768 seconds versus the 3642 seconds required to otin single 3600 s exposure over the full mosic, or out 3.5% increse in the time required to conduct the oservtion.

10 Glol Su- Elpsed Shutter Exposure Exposure Red Green Blue Time Stte Time Time CCD CCD CCD 0 Closed 0 0 Ersing Ersing Ersing 7 Open 0 0 Exposing Exposing Exposing 1207 Closed Reding Pusing Pusing 1242 Closed Ersing Pusing Pusing 1249 Open Exposing Exposing Exposing 1849 Closed Pusing Reding Pusing 1884 Closed Pusing Ersing Pusing 1891 Open Exposing Exposing Exposing 2491 Closed Reding Pusing Pusing 2526 Closed Ersing Pusing Pusing 2533 Open Exposing Exposing Exposing 3733 Closed Reding Reding Reding 3768 Closed Idle Idle Idle Tle 4. An exmple sequence of multiple su-exposures Commnds to the HIRES instrument re expressed in terms of KTL 6 keword opertions. 7 All Keck opticl instruments currentl use common set of kewords to initite CCD exposures. These kewords cn e hidden from the oserver providing grphicl user interfce (GUI) which writes the relevnt kewords in response to user input, such s clicking on n EXPOSE utton on the GUI. Alterntivel, kewords cn e written from script or directl from the commnd line of the user s shell. For exmple, the two primr kewords used to initite CCD exposures for Keck opticl instruments re TTIME nd EXPOSE; to strt n 1200 s exposure, one would set TTIME = 1200 nd EXPOSE = true. To implement the MSE sequence mode, onl few new CCD kewords needed to e defined. The MOSMODE keword is red/write keword tht specifies which of the three CCDs of the mosic will e red out t the end of the current su-exposure. The ERAMODE keword is red-onl keword tht the CCD controller sets to the current vlue of MOSMODE t the strt of ech redout; ERAMODE indictes which CCDs of the mosic will e ersed t the strt of the next su-exposure. If n su-exposure is orted, the entire MSE sequence is orted, nd oth kewords re reset to their defult vlues. When n MSE sequence is not eing tken, the MOSMODE nd ERAMODE will oth e set to their defult vlues of B,G,R, which mens tht ll of the CCDs will e ersed in unison t the strt of the next exposure nd tht ll of the CCDs will e red out in unison t the end of tht exposure. To otin the MSE sequence descried in Tle 4 ove, the following sequence of keword writes would e performed: (MOSMODE = R TTIME = 1200 EXPOSE = true); (MOSMODE = G TTIME = 600 EXPOSE = true); (MOSMODE = R TTIME = 600 EXPOSE = true); (MOSMODE = B,G,R TTIME = 1200 EXPOSE = true). To simplif the setting of these kewords, script hs een written which tkes s prmeters the glol exposure time (e.g., 3600 s in this exmple), nd the numer of su-exposures to tke for ech of the three CCDs of the mosic (e.g., in this exmple, 3 for the Red CCD, 2 for the Green, nd 1 for the Blue). To further simpl use of this oserving mode the stronomer, su-pnel hs een dded to the HIRES CCD exposure control GUI. 8 Text entr fields on tht su-pnel enle entr of these 4 prmeters. The GUI su-pnel lso provides grphicl displ of the current progress of the MSE sequence (See Fig. 7). At the right of the su-pnel re rectngles representing ech of the three CCDs of the mosic. Ech rectngle is filled with different color depending its current stte: gr if ersing, green if exposing or pusing, gold if reding out, nd empt (unfilled) if idle. The chllenge ws to represent the more complicted stte of detectors in the MSE mode, since the re not merel ersing or exposing or reding out; rther, the cn lso e not-ersing, or dirt (preserving chrge from previous exposure) if the hve not een selected for ersure on this ccle, nd not-reding-out if the hve not een selected for redout. We settled on old digonl stripe effect to mrk chips in this stte. To show which CCDs will e red out t the end of the current su-exposure, gold ox ppers round the identifing initils of the CCDs tht will e red.

11 5. UTILITIES FOR COMBINING SUB-EXPOSURES INTO A SINGLE FITS FILE A complete implementtion of ll the fetures for the multiple su-exposure (MSE) cpilit requires significnt chnges in the existing softwre, oth on the HIRES CCD VME crte nd in the imge cpture processes. Becuse these fetures were not required for deliver of the HIRES mosic not ll of the necessr chnges hve een implemented. The most rchivll significnt omission is tht the pixels from the CCD mplifiers involved in ech su-redout re written into seprte multi-extension FITS (MEF) file. The kewords in the primr heder/dt unit (PHDU) of these seprte FITS files document the sitution onl t the time of the shutter opening nd closing of the most recent su-exposure. While the exposure time kewords re correct for n CCDs which were ersed just efore the most recent shutter opening, the re not correct for n CCDs which were previousl prtl exposed nd not ersed. Nevertheless, the PHDU kewords do contin enough informtion to reconstruct correct vlues for lmost everthing. Therefore it is possile to construct post-processing script to remed the rchivl sitution. The kewords which contin the most esil recognized indiction of MSE sequence re the string-vlued MOSMODE nd ERAMODE kewords. For the HIRES mosic, MOSMODE nd ERAMODE cn hve the vlue of B, G, R, n other comm-seprted suset of these letters, or NoCCD. The letters correspond to the spectrl lout of HIRES cross dispersion, where B is the CCD receiving the luest light, R is the CCD receiving the reddest light, G is the CCD in the middle. ( NoCCD is provided for completeness to indicte n unusul sitution where none of the CCDs ws red or ersed.) The MOSMODE vlue indictes which of the CCDs will e (or were) red t the end of the next (or most recent) exposure. The ERAMODE vlue indictes which of the CCDs were ersed t the eginning of the most recent exposure. When MOSMODE nd ERAMODE re oth B, G, R, it is resonle to infer tht the exposure is not prt of MSE sequence. When ERAMODE is B, G, R nd MOSMODE is not, tht FITS file cn e inferred s the eginning of MSE sequence. The end of MSE sequence is implied when MOSMODE hs the vlue B, G, R nd ERAMODE does not. In order to provide more generic nd more introspective view of the exposure chrcteristics, there re other kewords indicting the sttus of ech CCD mplifier. The integer-vlued VIDINBEG nd VIDINEND kewords correspond, respectivel, to the ERAMODE nd MOSMODE kewords. The its in ech keword indicte the mplifiers whose CCDs were ersed or red. Bit 0 corresponds to the mplifier nmed 1 on the engineering drwing of the mosic, nd for HIRES this sequence continues such tht it 5 corresponds to mplifier nmed 6. The HIRES mosic is currentl constrined such tht prllel shifts must occur for oth of the mplifiers on single CCD. Therefore the its lws gree in djcent pirs, nd n imge not prt of MSE sequence hs VIDINBEG nd VIDINEND vlues of 63. The dt cquisition softwre inserts one other keword, VIDINACT, with identicl chrcteristics. The it vlues in VIDINACT indicte which CCD mplifiers corresponded to the pixels trnsmitted from the CCD susstem during the redout. The HIRES mosic sstem currentl produces onl vlues tht correspond to ll three A mplifiers, ll three B mplifiers, or ll six mplifiers. We hve prototpe post-processing script nmed joinsuexp which identifies nd comines the FITS files produced MSE sequences. When given director full of HIRES mosic FITS files, joinsuexp first exmines the kewords in the PHDUs to identif which, if n, of the files were memers of MSE sequences. The script then comines ll of the IMAGE extensions from ech MSE sequence into single, ver lrge, MEF file. Unlike the originl FITS files from the dt cquisition sstem, the PHDU of the comined file is lrgel void; it does not contin the kewords descriing the instrument nd telescope configurtion. (No single set of keword vlues in the PHDU of the comined file cn dequtel descrie the informtion from the mn files in the MSE sequence.) After coping ech IMAGE extension into the comined file, the script copies the kewords from the PHDU of the originl file into the new IMAGE extension. Finll, the script reconstructs the correct vlues for ech of the time-sensitive FITS kewords nd writes them into the individul IMAGE extensions. The time-relted kewords tht require rewriting fll into four ctegories: Begin, Sum, End, nd Histor. The Begin ctegor contins kewords whose vlues should reflect the egin time for the exposure in prticulr IMAGE extension. The Sum ctegor contins kewords whose vlues should reflect the sums of the vlues of those kewords in ll of the originl PHDUs of the MSE sequence. The End ctegor contins kewords whose vlues should reflect the end time for the exposure in prticulr IMAGE extension. The Histor ctegor

12 contins kewords whose vlues re not reconstructile or not relevnt to the comined MEF file. Kewords in the Histor ctegor re copied into the comined IMAGE extensions s FITS HISTORY crds. Although the comined MEF file produced joinsuexp conforms to the FITS stndrd, it devites from previous locl conventions. Such files rise unnswered questions out how the will e tolerted dt reduction nd rchiving sstems. 6. CONCLUSION B mking minor modifictions to the socketed PAL circuits on the SDSU-2 timing nd clock genertion ords, we hve een le to extend the cpilities of the SDSU-2 rchitecture in ws tht mke new oserving modes possile for CCD mosics. The PAL modifictions descried here enle CCD mosics to e red out in vriet of different configurtions nd with optiml redout efficienc. However, the mechnisms for ddressing the nks of clocks on the clock genertion ords nd for ccessing the digitl ltches on the video processing ords re still not s generlized s one would like. For exmple, these mechnisms still do not provide n efficient method for ddressing mix of A nd B mplifiers when operting in the single-mplifier-per-ccd redout mode. Although the existing SDSU-2 rchitecture enles such mix of mplifiers to e red in tht mode, it cnnot e done without either incurring penlt in redout time or requiring hrd-wired exchnge of seril clock phses on clock interconnect ord. More flexile ddressing schemes should e considered in future itertions of the pixel trnsmission PAL logic (U12 nd U17) on the timing ord nd the switch-stte ord ddress decoding logic (U100) on the clock genertion ord. ACKNOWLEDGMENTS We would like to thnk Bo Lech nd the stff t Astronomicl Reserch Cmers, Inc. for developing the modified PAL circuits descried here nd for his help in deugging some of our unique timing requirements. REFERENCES 1. R. Lech, F. Bele, nd J. Eriksen, New genertion CCD controller requirements nd n exmple: the Sn Diego Stte Universit genertion II controller, in Opticl Astronomicl Instrumenttion, S. D Odorico, ed., Proc. SPIE 3355, pp , S. S. Vogt et l., HIRES: The High Resolution Echelle Spectrometer on the Keck Ten-Meter Telescope, in Instrumenttion in Astronom VIII, D. L. Crwford, ed., Proc. SPIE 2198, pp , S. Fer et l., DEIMOS spectrogrph for the Keck 2 telescope: integrtion nd testing, in Instrument Design nd Performnce for Opticl/Infrred Ground-Bsed Telescope, M. Ie, ed., Proc. SPIE 4841, pp , B. E. Burke et l., CCD imger technolog development t Lincoln Lortor, in Opticl Detectors for Astronom II: Stte-of-the-Art t the Turn of the Millenium, P. Amico nd J. W. Beletic, eds., p. 187, Kluwer, (Boston), C. Wright, R. Kirick, nd B. Alcott, CCD imging sstems for DEIMOS, in Instrument Design nd Performnce for Opticl/Infrred Ground-Bsed Telescope, M. Ie, ed., Proc. SPIE 4841, pp , W. F. Lupton nd A. R. Conrd, The Keck Tsk Lirr (KTL), in Astronomicl Dt Anlsis nd Softwre Sstems II, R. J. Hnisch, R. J. V. Brissenden, nd J. Brnes, eds., ASP Conference Series 52, pp , A. R. Conrd nd W. F. Lupton, The Keck Keword Ler, in Astronomicl Dt Anlsis nd Softwre Sstems II, R. J. Hnisch, R. J. V. Brissenden, nd J. Brnes, eds., ASP Conference Series 52, pp , D. Clrke, Dshord: knowledge-sed rel-time control pnel, in Proceedings of the 5th Annul Tcl/Tk Workshop, pp. 9 18, The semntics of ech of these kewords is ville online vi the dictionr undle links t sl/fits/