under characteristic angle Cherenkov light emission

Ultra-fast FADC Data Aquisition System for

the MAGIC Telesope

H. Bartko a;

, F. Goebel a

, R. Mirzoyan a

, W. Pimpl a

,

M. Teshima a

a

Max PlankInstitute for Physis, Fohringer Ring 6, 80805 Munih,Germany

Abstrat

Ground-based Atmospheri Air Cherenkov Telesopes (ACTs) are suessfully

used to observe very high energy (VHE) gamma rays from elestial objets. The

light of the night sky (LONS) is a strong bakground for these telesopes. The

gammaraypulsesbeingvery short,anultra-fastread-outof anACTanminimize

the inuene of the LONS. This allows one to lower the so-alled tail uts of the

showerimageandtheanalysisenergythreshold.Itouldalsohelptosuppressother

unwantedbakgrounds.

Fast 'ash' analog-to-digital onverters (FADCs) withGSamples/s are available

ommerially; they are, however, very expensive and power onsuming. Here we

present anoveltehniqueof Fiber-OptiMultiplexingwhihusesa single2GSam-

ples/sFADCto digitize16 read-out hannels onseutively. Theanalog signalsare

delayedbyusingoptialbers.Themultiplexed(MUX)FADCread-outreduesthe

ost byabout85% omparedto usingone ultra-fast FADCperread-out hannel.

Two prototype multiplexers, eah digitizing data from 16 hannels, were built

and tested. The ultra-fast read-out system will be desribed and the test results

willbereported.Thenewsystemwillbeimplementedfortheread-outof the17 m

diameterMAGICtelesope amera.

Key words: fastdigitization,FADC, multiplexer,analog ber-optilink,

Cherenkovimagingtelesopes, gamma-ray astronomy.

Correspondingauthor.

Email address: hbartkomppmu.mpg.de(H.Bartko).

MAGIC is the world-wide largest Imaging Air Cherenkov Telesope (IACT).

Itaimsatstudyinggammarayemissionfromthehigh energyphenomenaand

the violent physis proesses in the universe, at the lowest energy threshold

among existing IACTs. An overview about the gamma ray astronomy with

IACTs is given in [1℄. MAGIC is a unique detetor in that it will over the

presently unexplored energy range between gammaray satellite missionsand

other ground-based Cherenkov telesopes [2℄.

The amera of the MAGIC Telesope onsists of 576 Photomultiplier tubes

(PMTs), whih deliver about 2 ns FWHM fast pulses to the experimental

ontrol house. The urrently used read-out system [3℄ is relatively slow (300

MSamples/s).Toreordthepulseshapeindetail,anartiialpulsestrething

toabout 6.5ns FWHMis used.This auses more lightof the nightsky to be

integrated, whih atsasadditionalnoise.Thusthe analysis energythreshold

of the telesope is limited, and the seletion eÆieny of the gamma signal

fromdierent bakgrounds is redued.

Forthe fastCherenkov pulses(2nsFWHM), aFADCwith 2GSamples/san

provide at least four sampling points. This permits a reasonable reonstru-

tionofthepulseshapeandouldyieldanimprovedgamma/hadronseparation

basedontiming.Suhanultra-fast read-outan strongly improvethe perfor-

mane of MAGIC. The improved sensitivity and the lower analysis energy

threshold willonsiderably extend the observation range of MAGIC, and al-

low one tosearhfor very weak souresat high redshifts.

A few FADC produts with 2 GSamples/s and a bandwidth 500 MHz

are available ommerially; they are, however, very expensive and power-

onsuming. To redue the ost of an ultra-fast read-out system, a 2 GSam-

ples/s read-out system has been developed at the Max-Plank-Institut fur

Physik in Munih. It uses the novel tehnique of Fiber-Opti Multiplexing,

anapproahpossible beausethe signalduration(fewns) and the triggerfre-

queny(typially1kHz)resultinaverylowdutyyleforthedigitizer.The

new tehnique uses a single FADC of 700 MHz bandwidth and of 2 GSam-

ples/s to digitize 16 read-out hannels onseutively. The analog signals are

delayed by using optial bers. A trigger signal is generated using a fration

ofthelight,whihisbranhed obyber-optilightsplittersbeforethe delay

bers. With the Fiber-Opti Multiplexing a ost redution of about 85% is

ahieved ompared tousing one FADC per read-out hannel.

Thesuggested2GSamples/smultiplexed(MUX)FADCsystemwillhavea10

bit amplitude resolution. For large signals the arrival time of the Cherenkov

pulse an be determined with a resolution better than 200 ps. The system

ommeriallyavailable, while the multiplexer eletronis has been developed

at the MPI in Munih. Two prototype multiplexers, for 32 hannels in total,

were builtandtestedin-situasread-outof theMAGICtelesopeinLaPalma

inAugust 2004.

In setion 2the MAGIC experiment is briey desribed inthe ontextof the

data aquisition(DAQ)system using ultra-fast FADCs.The speiationsof

the ultra-fast read-out are desribed in setion 3, followed by the measured

performanefortheMUX-FADCprototypeinlaboratorytests(setion4)and

asread-outoftheMAGICtelesope(setion5).Finally,setion6isdediated

todisussions and onlusions.

2 Priniple and Signal Proessing of the MAGIC Telesope

Sine the details of the MAGIC telesope are desribed elsewhere [4℄, only

items relevant to the FADC system are presented in this setion. Figure 1

shows the working priniple of an air Cherenkov telesope. A high energy

gammarayentering theearth'satmosphereinitiatesashowerasade of ele-

trons and positrons. These radiateCherenkov light, whih is olleted by the

mirrorandfoussedontothePMTameraoftheMAGICtelesope.Themain

bakgroundoriginatesfrommuhmore frequentshowers indued by isotropi

hadroni osmi rays.

189mm

°

0.6 48

55 62 69 76 83 90 96 103 110 117 124 131 138 145 152 159

S [au]

Signal (raw) Mean 103.7 RMS 30.17

189mm°

0.6 48

55 62 69 76 83 90 96 103 110 117 124 131 138 145 152 159

S [au]

Signal (raw) Mean 103.7 RMS 30.17

189mm°

0.6 48

55 62 69 76 83 90 96 103 110 117 124 131 138 145 152 159

S [au]

Signal (raw) Mean 103.7 RMS 30.17

189mm°

0.6 48

55 62 69 76 83 90 96 103 110 117 124 131 138 145 152 159

S [au]

189mm°

0.6 48

55 62 69 76 83 90 96 103 110 117 124 131 138 145 152 159

S [au]

θ

Top of atmosphere

under characteristic angle Cherenkov light emission

Primary particle

~ 20 m

~10 km a.s.l.

~ 10 km Camera (cleaned event)

Cherenkov Telescope

atmosphere at about 20 km height

Fig. 1.IACT priniple: A osmi high energy gamma ray penetrates in the earth's

atmosphereand initiatesa shower asade of eletrons andpositrons, whih radiate

Cherenkov light. This light is olleted and foussed onto the amera, providing an

image of the air shower. Pituretaken from [5℄.

gamma and hadron indued shower images as well as for images of single

muons. The timing information is therefore expeted to improve the separa-

tion of gamma events from the bakground events [6℄. Figure 2 shows the

mean amplitude (a,) and time (b, d) proles for gamma (, d) and hadron

(a, b) indued air showers images on the amera plane of the MAGIC tele-

sope.The impat parameter isxed to120 m and the initialgammaenergy

isset to100 GeV,while the protonenergy isset to200 GeV.The prolesare

obtained by averaging over many simulated showers [7℄. Although the total

shower duration of gamma and hadron indued air showers is omparable,

the photon arrival time varies smoothlyover the gamma shower image while

hadronshower imagesare struturedintime.Thetime strutureofthe image

an alsoprovide essential informationabout head and tailof the shower.

2

photons / 9 mm

10

10

1

x [mm]

-400 -200 0 200 400

y [mm]

0 200 400 600 800 1000

a)

time [ns]

0 0.5 1 1.5 2 2.5 3

x [mm]

-400 -200 0 200 400

y [mm]

0 200 400 600 800 1000

b)

2

photons / 9 mm

10

10

1

x [mm]

-400 -200 0 200 400

y [mm]

0 200 400 600 800 1000

c)

time [ns]

0 0.5 1 1.5 2 2.5 3

x [mm]

-400 -200 0 200 400

y [mm]

0 200 400 600 800 1000

d)

Fig.2.Mean amplitude (a,) andtime (b, d)prolesfor gamma(, d) andhadron

(a, b) indued air showers images on the MAGIC amera plane from MC simula-

tions.Theimpatparameterisxedto120mandtheinitialenergy ofthegammais

setto100GeV, whiletheprotonenergy issetto200GeV. Theproles areobtained

by averaging over many simulated showers [7℄. The timing struture of the image

an provide viable information about the headand the tail of the shower as well as

help to disriminate between gamma and hadron indued showers.

The MAGIC read-out hain, inluding the PMT amera, the analog-optial

link, the majority trigger logi and FADCs, is shematially shown in gure

3.Theresponseof thePMTs tosub-ns inputlightshows apulseof FWHMof

1.0 - 1.2ns and rise and fall times of 600 and 700 ps orrespondingly [8℄. By

modulatingvertialavitysurfaeemittinglaser(VCSEL)diodesinamplitude

the ultra-fastanalogue signalsfromthe PMTs are transferred via 162m long,

multimode graded index 50/125 m diameter optial bers to the ounting

house[9℄.Aftertransformingthe lightbaktoaneletrialsignal,theoriginal

VCSEL diode receiver FADC

central control

~162 m optical fiber

data storage DAQ

PMT PC

majority trigger logic

Fig. 3. Current MAGIC read-out sheme: the analog PMT signals are transferred

via an analog optial link tothe ounting house where after the trigger deisionthe

signalsaredigitized byusinga 300MHzFADCssystemandwrittentothehard disk

of a DAQPC.

In order to sample this pulse shape with the present 300 MSamples/s FADC

system,thepulseisstrethedtoaFWHMof>6ns(theoriginalpulseisfolded

with a strething funtion of about 6ns). This implies a longer integration

of LONS and thus the performane of the telesope on the analysis level is

degraded.

BeausetheurrentMAGICFADCshavearesolutionof8bitonly,thesignals

are split intotwo branhes with a fator of 10 dierene ingain. One branh

isdelayed by 55nsandthen bothbranhesare multiplexedand onseutively

read out by one FADC. The FADC system an be read out with a maximum

sustained rateof1kHz. A512kbytesFIFOmemoryallows short-timetrigger

rates of up to 50kHz.

3 The Ultra-fast Fiber-Opti MUX-FADC Data Aquisition Sys-

tem

TheMAGICollaborationisgoingtoimprovetheperformaneofitstelesope

by installinga fast (2GSamples/s) FADC system, whih fully exploits the

intrinsi time strutures of the Cherenkov light pulses. The requirements for

suha system are the following:

10bitresolution at a2 GSamples/s samplingrate

500 MHzbandwidth of the eletronishain inludingthe FADC

up to1 kHz sustained event triggerrate

deadtime 5%.

It is interesting to note that in experiments where FADCs are used to read

out a multihannel detetor in the ommon event trigger mode, only a tiny

fration of the FADC memory depth is oupied by the signal while the rest

iseetively \empty"[11,12℄. One antry toorretthis\ineÆieny ofuse"

by \paking" the signals of many hannels sequentially in time into a single

FADC hannel,i.e. by multiplexing.

Following this simple idea, a multiplexing system with ber-opti delays has

beendevelopedfortheMAGICtelesope.Theblokdiagramisshowningure

4.The ultrafastber-opti multiplexeronsists of three main omponents:

ber-optidelays and splitters

multiplexereletronis: fast swithes and ontrollers

ultra-fastFADCs.

fastF ADC

PMT PMT PMT

receiver receiver receiver transmitter

transmitter transmitter

receiver receiver

MUXcontroler +switches triggerlogic

receiver optical

splitters

optical delay lines

optical fibers optical

fibers

shown are three out of 16 channels optical delays:

0..15 * 40 ns

Fig. 4. Shemati diagram of the multiplexed ber-opti ultra-fast FADC read-out.

Part of the analog signal that arrives via the ber-optilink from the PMTamera

isbranhed o and fed into amajority trigger logi. Theother partof the signalsis

onseutively delayed in optial bers. Channels are onneted onseutively one by

one to the ultra-fast FADC, using fast swithes. Thereby the noise from the other

hannels iseÆiently bloked.

AftertheanalogoptiallinkbetweentheMAGICPMTameraandtheount-

inghousetheoptialsignalsaresplitintotwoparts.Onepartofthesplitsignal

isused asaninputto thetriggerlogi.The otherpart isusedfor FADC mea-

surements after passing through a ber-opti delay line of a hannel-spei

length.

Themultiplexereletronisoperatesinthefollowingway:Theommontrigger

theanalogsignaltopassthroughand bedigitizedbythe FADC.Allthe other

swithesarelosedduringthistime.Whenthedigitizationwindowfortherst

hannelis overthe orresponding swith islosed.The losed swith strongly

attenuates the signal transmission by more than 60 dB for the fast MAGIC

signals. Then the swith number two is opened suh that the aordingly

delayed analog signal from the seond hannel is digitized and so forth, one

hannel at a time until the last one is measured. In this way one \paks"

signalsfromdierenthannels inatime sequene whih an bedigitizedby a

single FADC hannel.

Beause of the nite rise and fall times of the gate signals for the swithes

and beause of some pik-up noise fromthe swith one has toallow for some

swithing time between the digitizationoftwoonseutive hannels. Thegat-

ingtimefor eahhannel wasset to40ns, ofwhihthe rst and last5nsare

aeted by the swithing proess.

For the use in MAGIC a 16! 1 multiplexing ratio was hosen. 16 hannels

are read out by a single ultra-fast FADC hannel. The hosen multiplexing

ratio isa ompromise between

Deadtime:the digitizationofoneevent takes16*40ns=640ns. Duringthis

timenoothereventan be reorded.Compared tothe maximumsustained

triggerrate of up to 1 kHz this dead time isnegligible.

Noise due to ross-talk through the losed swithes: The attenuated noise

ofthe otherhannelsould inuene the ativesignal hannel.

Cost of the FADCs

Mehanial onstraints,e.g. board size, length of wires and bers.

3.2 Optial Delaysand Splitters

Optial bers were hosen for the analog signal transmission between the

PMT ameraand the ounting house beause they are lightweight,ompat,

immunetoeletro-magnetipik-upnoiseandprovidenopulsedispersionand

attenuation[9℄.Thesignalattenuationat1kmberlengthisabout2.3dBfor

thehosen850nm wavelengthofthe VCSELs.Theanalogsignaltransmission

oersa dynami range largerthan 60 dB.

Using ber-opti delays ultra-fast analog signals an be delayed by several

hundreds of ns. Thus a large number of suessively delayed signals an be

multiplexed and read out by a single hannel FADC. Part of the analog sig-

nal has to be split o before the delay lines for trigger. Therefore ber-opti

splittersof type 1!2 are used.

length,orrespondingtoadelayof710nsand750ns.Figure6showsamodule

offourgradedindex(GRIN)-typeber-optisplitterswith50:50splittingratio

(for a tehnial desription see [13℄). The modules have standardized outer

dimensions and an be assembled in 3U hight 19" rates. The splitters and

optialdelaylinesare ommeriallyavailablefromthe ompany Sahsenkabel

[14℄.

Fig.5.Twohannelber-optidelaymoduleof 142and150mlength,orresponding

to a delay of 710 and 750 ns, respetively. Mehanial dimensions: 235 mm * 130

mm(3U) * 35 mm (7HP).

Fig.6.Fourhannelber-optisplittermodule,GRINtehnology and50:50splitting

ratio. Theouter dimensions are:235 mm * 130 mm (3U) * 35mm (7HP).

Themultiplexereletronisonsistoffourstages.Therststageisaberopti

reeiver, where the signals from the optial delay lines are onverted bak to

eletrial pulses using PIN diodes. In a seond stage, part of the eletrial

signal is branhed o and transferred to a monitor output. The third stage

onsists of ultra-fast swithes whih are ativated one at a time. In the last

stage all 16 hannels are summed to one output. The multiplexed signals

are then transferred via 50 oaxial ables to the FADC hannels. Table 1

summarizes the speiations of the multiplexer eletronis.

Table 1

Speiations of the eletronis for analogsignal multiplexing.

Mehanialsize 370 mm (9 U)* 220 mm* 30 mm (6 HP)

Numberof hannels 16

Analog input via 50/125 mgradedindexber, E2000onnetor

Gain 25, inludingthe VCSEL transmitter

Dynamirange max output amplitude:1V

Power supplies +12 V,5V

Power dissipation 20 W

Trigger input LVDS

One multiplexer module onsists of one 6 layer motherboard and 16 double

layer swithboards, whih are plugged into the motherboard via multiple pin

onnetors. Figure 7 shows a photo of the printed iruit MUX motherboard

with 16mounted daughter swithboards.

The motherboard inludes the following omponents:

16opto-eletrialonverters

16monitoroutputs

the DigitalSwith Controliruit (DSC)

the triggerinput toativate the DSC

16ultra-fast swithes on16swithboards

the 16hannel summingstage.

One opto-eletrial onverter onsists of a reeptale, a PIN photo diode,

paked in the E2000-onnetor. The photodiode is biased by 12 V to redue

its intrinsi apaity for speed and noise optimization. The urrent signal

of the PIN photo diode is onverted into an equivalent voltage signal by a

transimpedane-amplier. Its amplier-IC has a gain-bandwidth produt of

about 1.5 GHz and a very high slew rate of about 4000 V/s. The trans-

impedaneis1000.Amonitoroutputonsistsofanultra-wideband(UWB)-

driver-amplier,whihtransmitsthe signalfrom thetransimpedane-stage to

ofa triggerinput,16opto-eletri onverters, 16monitor signaloutputs,theDigital

Swith Control iruit (DSC), 16 daughter swith boards and two summing stages.

Theoverall size is 370 mm(9 U) * 220 mm * 30mm (6 HP).

a50 -SMA-onnetor.

Fig. 8. Ciruit diagram of the Digital Swith Control iruit (DSC): The trigger

initiates a sequeneof 16 PECL high levelsof 40ns duration applied onseutively

tothe swith boards.

Figure8showsthe iruitdiagramoftheDigitalSwithControliruit,DSC.

It onsists of the following parts:

50MHz to800 MHz and works inPECL mode. It is rystal stabilizedand

set to800 MHz.

A digital delay line (DDL) that an be set from 2 ns to 10 ns with 11 bit

auray.It an beused toadjustthe trigger timesbetween dierentMUX

motherboards.

Adigitallok-in-iruit(DLC)synhronizes the MUX-sequenetothetrig-

ger signal.The lok-in jitteris 1.25 ns(=1=[800MHz℄).

16dierentialPECL-driversthattransmittheMUX-sequenesignalstothe

orrespondingswithboards.

Eahswithboard inludestwo ultra-wideband (UWB)-amplieriruits, fol-

lowed bytwons-swithingMOSFETs operatedinseriesandoneUWB-driver-

amplier-iruit. MOSFET swithes were hosen due to their fast swithing

propertiesand averyfast stabilizationof the signalbaselineafterthe swith-

ing. The smallross-talk through the losed swith is further redued by the

serial operation of two swithes. An on-board PECL to CMOS onverter

distributes the digital swith-ontrol-iruit (DSC)-signal to the MOSFET-

swithes in parallel. Figure 9 shows a photo of the swith board, while its

iruit diagramisshown ingure 10.

Fig. 9. The printed iruit board for fast swithing. The swith board ontains two

ns-MOSFET-swithes operated in series. Its mehanial dimensions are 80 mm *

20mm *5 mm.

In a passive summation, the swith parasiti apaitanes would add up and

an signiantlywiden the signalpulse. Toavoid this,a two-step ativesum-

mationwashosen:In therst step,the outputsoffourhannelsaresummed

together. In the seondstep, the fourresulting outputsare summedintoone.

ForthesummingUWB-ampliersareused.Thetwo-stepsetupkeepsthehan-

nelwiresshort, andpermitstouse the ampliersinthe faster invertingmode

whilekeepingthe signalpolaritynon-inverting.Finally,anUWB-driversends

Control iruitopens both of the ns-MOSFETswithes operated in series.

the multiplexed signals over a 50 -SMA-oaxial onnetion to the FADC.

The iruit diagramof the summation stage is shown ingure 11.

Fig. 11. Shematis of the two-step summing stage of the multiplexed signals. The

two-step setup keepsthe hannel-wires short and allows touse the UWB-ampliers

in the faster inverting mode while keepingthe signal polarity non-inverting.