The HERMES Spectrometer - math.uwaterloo.cajemerson/papers/... · 0W.K.Kellogg Radiation Lab,...

Nuclear Instruments and Methods in Physics Research A 417 (1998) 230—265

The HERMES Spectrometer

The HERMES Collaboration

K. Ackerstaff!, A. Airapetian", N. Akopov", M. Amarian", V. Andreev#, E.C. Aschenauer$,%,R. Avakian", H. Avakian&, A. Avetissian", B. Bains', S. Barrow), W. Beckhusen!,

M. Beckmann*, St. Belostotski#, E. Belz+, Th. Benisch,, S. Bernreuther,, N. Bianchi&,J. Blouw%, H. Bottcher-, A. Borissov-,., J. Brack+, B. Braun,,/, B. Bray0, S. Brons-,1,W. Bruckner., A. Brull., H.J. Bulten2, G.P. Capitani&, P. Carter0, P. Chumney3,

E. Cisbani4, S. Clark+, S. Colilli4, H. Coombes5, G.R. Court6, P. Delheij1, E. Devitsin7,C.W. de Jager%, E. De Sanctis&, D. De Schepper8,9, P.K.A. de Witt Huberts%, P. Di Nezza&,

M. Doets%, M. Duren,, A. Dvoredsky0, G. Elbakian", J. Emerson:,1, A. Fantoni&,A. Fechtchenko;, M. Ferstl,, D. Fick!!, K. Fiedler,, B.W. Filippone0, H. Fischer*,

H.T. Fortune), J. Franz*, S. Frullani4, M.-A. Funk!, N.D. Gagunashvili;, P. Galumian5,H. Gao', Y. Garber-, F. Garibaldi4, G. Gavrilov#, P. Geiger., V. Gharibyan", V. Giordjian&,F. Giuliani4, A. Golendoukhin!!, B. Grabowski!, G. Graw/, O. Grebeniouk#, P. Green1,5,

G. Greeniaus5,1, M. Gricia4, C. Grosshauser,, A. Gute,, J.P. Haas3, K. Hakelberg!,W. Haeberli2, J.-O. Hansen8, D. Hasch-, O. Hausser:,1,1, R. Henderson1, Th. Henkes%,

R. Hertenberger/, Y. Holler!, R.J. Holt', H. Ihssen!, A. Izotov#, M. Iodice4, H.E. Jackson8,A. Jgoun#, C. Jones8, R. Kaiser:,1, J. Kelsey9, E. Kinney+,*, M. Kirsch,, A. Kisselev#,P. Kitching5, H. Kobayashi!", E. Kok%, K. Konigsmann*, M. Kolstein%, H. Kolster/,

W. Korsch0, S. Kozlov#, V. Kozlov7, R. Kowalczyk8, L. Kramer9, B. Krause-,A. Krivchitch#, V.G. Krivokhijine;, M. Kueckes1, P. Kutt), G. Kyle3, W. Lachnit,,R. Langstaff1, W. Lorenzon), M. Lucentini4, A. Lung0, N. Makins8, V. Maleev#,

S.I. Manaenkov#, K. Martens5, A. Mateos9, K. McIlhany0, R.D. McKeown0, F. Mei{ner-,F. Menden1, D. Mercer+, A. Metz/, N. Meyners!, O. Mikloukho#, C.A. Miller5,1,

M.A. Miller', R. Milner9, V. Mitsyn;, G. Modrak-, J. Morton6, A. Most', R. Mozzetti&,V. Muccifora&, A. Nagaitsev;, Y. Naryshkin#, A.M. Nathan', F. Neunreither,,

M. Niczyporuk9, W.-D. Nowak-, M. Nupieri&, P. Oelwein., H. Ogami!", T.G. O’Neill8,R. Openshaw1, V. Papavassiliou3, S.F. Pate9,3, S. Patrichev#, M. Pitt0, H.J. Plett,,H.R. Poolman%, S. Potashov7, D. Potterveld8, B. Povh., V. Prahl!, G. Rakness+,

*Corresponding author. Tel.: #49 40 89984578. e-mail: [email protected].

0168-9002/98/$19.00 ( 1998 Elsevier Science B.V. All rights reserved.PII: S 0 1 6 8 - 9 0 0 2 ( 9 8 ) 0 0 7 6 9 - 4

V. Razmyslovich#, R. Redwine9, A.R. Reolon&, R. Ristinen+, K. Rith,, H.O. Roloff-,G. Roper!, P. Rossi&, S. Rudnitsky), H. Russo,, D. Ryckbosch$, Y. Sakemi!",

F. Santavenere4, I. Savin;, F. Schmidt,, H. Schmitt*, G. Schnell3, K.P. Schuler!, A. Schwind-,T.-A. Shibata!", T. Shin9, B. Siebels5, A. Simon*,3, K. Sinram!, W.R. Smythe+, J. Sowinski.,M. Spengos), K. Sperber!, E. Steffens,, J. Stenger,, J. Stewart6, F. Stock,,., U. Sto{lein-,

M. Sutter9, H. Tallini6, S. Taroian", A. Terkulov7, D. Thiessen:,1, B. Tipton9,V. Trofimov#, A. Trudel1, M. Tytgat$, G.M. Urciuoli4, R. Van de Vyver$,

J.F.J. van den Brand%,!#, G. van der Steenhoven%, J.J. van Hunen%, D. van Westrum+,A. Vassiliev#, M.C. Vetterli1,:, M.G. Vincter1, E. Volk., W. Wander,, T.P. Welch!$,

S.E. Williamson', T. Wise2, G. Wobke!, K. Woller!, S. Yoneyama!",K. Zapfe-Duren!, T. Zeuli8, H. Zohrabian"

! DESY, Deutsches Elektronen Synchrotron, 22603 Hamburg, Germany" Yerevan Physics Institute, 375036, Yerevan, Armenia

# Petersburg Nuclear Physics Institute, St.Petersburg, 188350 Russia$ Department of Subatomic and Radiation Physics, University of Gent, 9000 Gent, Belgium

% Nationaal Instituut voor Kernfysica en Hoge-Energiefysica (NIKHEF), 1009 DB Amsterdam, Netherlands& Istituto Nazionale di Fisica Nucleare, Laboratori Nazionali di Frascati, I-00044 Frascati, Italy

' Department of Physics, University of Illinois, Urbana, IL 61801, USA) Department of Physics, University of Pennsylvania, Philadelphia PA 19104-6396, USA

* Fakulta( t fu( r Physik, Universita( t Freiburg, 79104 Freiburg, Germany+ Nuclear Physics Laboratory, University of Colorado, Boulder, CO 80309-0446, USA, Physikalisches Institut, Universita( t Erlangen-Nu( rnberg, 91058 Erlangen, Germany

- DESY, 15738 Zeuthen, Germany. Max-Planck-Institut fu( r Kernphysik, 69029 Heidelberg, Germany/ Sektion Physik, Universita( t Mu( nchen, 85748 Garching, Germany

0 W.K.Kellogg Radiation Lab, California Institute of Technology, Pasadena, CA, 91125, USA1 TRIUMF, Vancouver, British Columbia V6T 2A3, Canada

2 Department of Physics, University of Wisconsin-Madison, Madison, WI 53706, USA3 Department of Physics, New Mexico State University, Las Cruces, NM 88003, USA

4 Istituto Nazionale di Fisica Nucleare, Sezione Sanita% , 00161 Roma, Italy5 Department of Physics, University of Alberta, Edmonton, Alberta T6G 2N2, Canada6 Physics Department, University of Liverpool, Liverpool L69 7ZE, United Kingdom

7 Lebedev Physical Institute, 117924 Moscow, Russia8 Physics Division, Argonne National Laboratory, Argonne, IL 60439, USA

9 Laboratory for Nuclear Science, Massachusetts Institute of Technology, Cambridge, MA 02139, USA: Department of Physics, Simon Fraser University, Burnaby, British Columbia V5A 1S6 Canada

; Joint Institute for Nuclear Research, 141980 Dubna, Russia!! Physikalisches Institut, Philipps-Universita( t Marburg, 35037 Marburg, Germany

!" Department of Physics, Tokyo Institute of Technology, Tokyo 152, Japan!# Department of Physics and Astronomy, Vrije Universiteit, 1081 HV Amsterdam, The Netherlands

!$ Department of Physics, University of Oregon, Eugene, OR 97403 USA

Received 15 June 1998

Abstract

The HERMES experiment is collecting data on inclusive and semi-inclusive deep inelastic scattering of polarisedpositrons from polarised targets of H, D, and 3He. These data give information on the spin structure of the nucleon. Thispaper describes the forward angle spectrometer built for this purpose. The spectrometer includes numerous trackingchambers (micro-strip gas chambers, drift and proportional chambers) in front of and behind a 1.3T.m magnetic field, as

231K. Ackerstaff et al. /Nucl. Instr. and Meth. in Phys. Res. A 417 (1998) 230—265

2HERA is an e—p collider but only the e beam is used byHERMES.

well as an extensive set of detectors for particle identification (a lead-glass calorimeter, a pre-shower detector, a transitionradiation detector, and a threshold Cherenkov detector). Two of the main features of the spectrometer are its goodacceptance and identification of both positrons and hadrons, in particular pions. These characteristics, together with thepurity of the targets, are allowing HERMES to make unique contributions to the understanding of how the spins of thequarks contribute to the spin of the nucleon. ( 1998 Elsevier Science B.V. All rights reserved.

Keywords: Magnetic Spectrometer; Deep inelastic lepton scattering; Semi-inclusive scattering; (e,e’h)

1. Introduction

The HERMES experiment (HERA MEasure-ment of Spin) is a second-generation polarised deepinelastic scattering (DIS) experiment to study thespin structure of the nucleon. It is being run at theHERA storage ring at DESY.2

Several experiments over the last decadehave provided accurate data on the polarisationasymmetry of the cross-section for inclusive scatter-ing where only the scattered lepton is detected.These experiments have been interpreted as show-ing that at most 30% of the nucleon spin comesfrom the spins of the quarks. Further knowledgeof the origin of the nucleon’s spin can be gainedby studying semi-inclusive processes involvingthe detection of hadrons in coincidence with thescattered lepton. These data offer a means of ‘fla-vour-tagging’ the struck quark to help isolate thecontributions to the nucleon spin of the individualquark flavours, including the sea quarks. The inter-pretation of semi-inclusive data is made clearer ifthe type of hadron is identified. This is a centraltheme of the HERMES experiment, which canidentify pions and will add kaon identification for1998.

The physics program for HERMES is verybroad. The experiment contributes inclusive datawith qualitatively different systematic uncertaintiesto improve the world data set for the x dependenceand the integral of the spin structure function g

1(x)

for both the proton and the neutron. Most impor-tantly, HERMES is providing new precise data onsemi-inclusive processes by virtue of the good

acceptance of the spectrometer combined with had-ron identification and the purity of the targets.

The HERA storage ring can be filled with eitherelectrons or positrons, which are acceleratedto 27.5GeV. Positrons have been used since 1995because longer beam lifetimes are then possible.Since with few exceptions such as the luminositymeasurement, the physics processes are the samefor positrons and electrons, the term ‘positron’ willbe used for both in this paper.

In addition to studying polarised DIS using Ho ,Do ,and 3Ho e targets, data are collected with un-polarised gases (H

2,D

2,3He,N

2). This provides in

a relatively short period of time (2—3 weeks) highstatistics data sets that are used to study importantproperties of the nucleon not related to spin, suchas the flavour asymmetry of the sea, as well ashadronisation in nuclei.

2. General description

The HERMES experiment is located in the Easthall of the HERA storage ring complex at DESY.The spectrometer is a forward angle instrument ofconventional design. It is symmetric about a cen-tral, horizontal shielding plate in the magnet. Dueto this symmetry, the description of the detectorscontained in this paper will apply typically to onlyone half of the spectrometer, in particular withrespect to the number of detectors quoted and theirdimensions. A diagram of the spectrometer isshown in Fig. 1. The coordinate system used byHERMES has the z-axis along the beam mo-mentum, the y-axis vertical upwards, and the x-axishorizontal, pointing towards the outside of the ring.The polar (h) and azimuthal (/) scattering angles aswell as the initial trajectory for the determination of

232 K. Ackerstaff et al. /Nucl. Instr. and Meth. in Phys. Res. A 417 (1998) 230—265

Fig. 1. Schematic side view of the HERMES spectrometer. See the text for the meaning of the labels.

the particle’s momentum are measured by the fronttracking system, which consists of microstrip gaschambers (MSGC, referred to as the vertex cham-bers (VC)) and drift chambers (DVC, FC1/2). Themomentum measurement is completed by two setsof drift chambers behind the magnet (BC1/2 andBC3/4). In addition, there are three proportionalchambers in the magnet (MC1/3) to help matchfront and back tracks as well as to track low mo-mentum particles that do not reach the rear sectionof the spectrometer.

Particle identification (PID) is provided bya lead-glass calorimeter, a pre-shower detector (H2)consisting of two radiation lengths of lead followedby a plastic scintillator hodoscope, a transitionradiation detector (TRD) consisting of six identicalmodules, and a threshold Cherenkov detector (C[ ).The particle identification system was designed toprovide a hadron rejection factor of 104 to yielda very clean DIS positron sample, and exceeds thisin practice. Pion identification is provided bya threshold Cherenkov detector. An upgrade toa ring imaging Cherenkov detector that will ident-

ify both pions and kaons is being prepared foroperation in 1998. The calorimeter and pre-showerdetector are included in the trigger along with a sec-ond hodoscope (H1) placed in front of the TRD. Anadditional trigger hodoscope (H0) was included in1996 in front of FC1 to reduce the trigger rate frombackground caused by the proton beam.

The acceptance is limited at small angles by aniron plate in the beam plane, which shields thepositron and proton beams from the magnetic fieldof the spectrometer magnet. Both beams gothrough the spectrometer, separated by 72 cm. Par-ticles with scattering angles within $170mrad inthe horizontal direction and between #(!)40and #(!)140mrad in the vertical direction areaccepted. Therefore, the range of scattering anglesis 40—220mrad. Spin structure functions depend onx (x"Q2/2Ml where Q is the four-momentumtransfer in the DIS reaction, l is the energy transfer,and M is the mass of the nucleon). The variablex can be interpreted as the fraction of the nucleon’smomentum carried by the struck quark. Thex range covered by the HERMES experiment


Fig. 2. Schematic of the target region. See the text for moreexplanation.

is 0.02—1.0, although there is little count rate forx50.8 when W'2GeV (W is the photon—nuc-leon invariant mass).

The experiment is mounted on a large platformthat can move on rails together with an attachedtrailer (ET) containing the electronics and the gassystems. A fixed but removable concrete wall be-tween the platform and the ET shields the mainpart of the hall from radiation and hence allowsaccess to the electronics while the accelerators arerunning. Cables and gas pipes between the de-tectors and the ET are routed beneath the platformto a large cable tray passing under the shieldingwall. The experiment was assembled in the Easthall outside the shielding during the 1994 HERArun. After the shielding wall had been dismantled,the platform/trailer was moved into place in thering. The shielding wall was re-erected between theplatform and the electronics trailer in such a waythat the experiment could be moved far enoughwithin the shielding to allow the HERA mainten-ance tram access to the tunnel on 24 h notice.

3. The target region

The HERMES target consists of gas froma polarised source fed into a storage cell internal tothe HERA positron ring. In 1995, an optically pum-ped polarised 3Ho e cell was used to supply gas to thestorage cell [1], while an atomic beam source (ABS)was used in 1996—97 to produce a Ho target [2]. TheABS will be modified for operation with Do in1998—99. It is also possible to inject unpolarisedgases into the storage cell. The following were usedin the first three years of operation: H

2,D

2,3He,N

2.

The target region is shown schematically inFig. 2. The gas enters an open-ended (i.e. win-dowless) T-shaped tube that confines the gas atomsin a region around the positron beam. The storagecell increases the areal target density by about twoorders of magnitude compared to a free atomicbeam. The gas atoms leak out the open ends of thetarget cell and are pumped away by a high speeddifferential pumping system. In this way, the num-ber of atoms seen at the target location by the beamis maximised while minimizing the effect onthe stored beam. The storage cell is an elliptical

tube, 9.8mm high by 29.0mm wide and 400mm inlength. It was made of 125lm thick ultra-purealuminium in 1995 and 75lm aluminium in1996—97. The transition from the cell to the beampipe was made smooth to avoid the generation ofwake fields that could cause heating and increasethe emittance of the beam. This was accomplishedusing thin perforated tubes called ‘wake field sup-pressors’.

The amount of synchrotron radiation createdupstream of the HERMES area is considerable,even though it is reduced by the addition of twoweak dipoles downstream of the last bending mag-net in the arc of the accelerator. If it were allowed tohit the target cell, it would damage the cell coating,and on the order of 1013 — 1014 soft synchrotronphotons per second would Compton scatter intothe spectrometer acceptance. This is prevented bya system of two collimators installed near the tar-get. C1 is placed 2 m upstream of the target andprotects the next collimator from direct radiation.It actually consists of two moveable collimators,C1H in the horizontal direction and C1V in thevertical, separated by 0.5m. These collimators areopened during beam injection and operated ata narrow but safe setting (15p#0.5mm) duringHERMES operation. During the first three years ofoperation, typical beam parameters werepx"0.31mm and p

y"0.07mm. C2 has a fixed

aperture that is slightly larger than C1, and islocated right in front of the storage cell to protect itfrom photons scattered in C1. It also protects the


cell from direct radiation during beam injection.Note that only C2 is shown in Fig. 2.

Particles scattered into the spectrometer accept-ance exit the target chamber through a thin(0.3mm) stainless steel foil. Background fromshowers (initiated mostly by c’s from p0 decay) isreduced by also making the beam pipe just down-stream of the target chamber of thin stainless steel.

4. The magnet

The HERMES spectrometer magnet is of theH-type with field clamps in front as well as behindin order to reduce the fringe fields at the positionof the drift chambers FC2 and BC1. A massiveiron plate in the symmetry plane shields the posit-ron and proton beams as they pass through themagnet.

The most important features of the magnet are:

f It is capable of providing a deflecting power of:Bdl"1.5Tm, although it is operated at1.3 Tm to reduce power consumption. Thesmaller field integral does not significantly im-pair the performance of the spectrometer. Thevariation of the deflecting power within the ac-ceptance is less than 10%.

f The gap between the pole faces encloses thegeometrical acceptance of $(40—140)mrad inthe vertical direction. In the horizontal direction$170mrad plus another $100 mrad startinghalf-way through the magnet is provided. Thepole faces are tilted parallel to the limits of thevertical angular acceptance. Due to the limitedspace of 8.5m for the HERMES spectrometerbetween the centre of the target and the firstquadrupole magnet of the electron machine, only2.2m between the drift chambers FC2 on thefront and BC1 on the rear side are utilised for themagnet.

f The fringe field at the positions of the adjacentdrift chambers does not exceed 0.1T.

f An effective magnetic shielding substantially re-duces the effect of the magnet on the proton andpositron beams. In particular, the sextupole mo-ment of the magnetic field in the beam tubes isminimised.

f A correction coil with a deflecting power of0.08Tm is accommodated inside the shielding ofthe positron beam pipe. This coil is used tocorrect for fringe fields and the imperfections ofthe magnetic shielding in this section of the ironplate. It is also intended to compensate the trans-verse holding field of the target when operatingwith transverse polarisation. It serves as an ele-ment of a closed orbit bump with no net globaleffect on the positron beam.

Magnetic model calculations were done with the3D-codes MAFIA [3] and TOSCA [4]. The calcu-lations agree with subsequent magnetic measure-ments within a few percent. The field measurementswere done with a 3D-Hall probe on an automated3D-mapping machine. The results exhibit an over-all reproducibility of about 10~3, and a field pat-tern as expected. No homogeneous regions arefound due to the high gap to length ratio of themagnet, and a pronounced step in the field is ob-served that reflects the structure of the iron shield-ing plate. The detailed field map was integratedinto the track reconstruction algorithm, as de-scribed later.

5. The tracking system

The tracking system serves several functions,which vary somewhat in their performance require-ments:

f Determine the event vertex in the target cell toensure that reconstructed events come only fromthe sub-mm beam envelope in the target gas,with the expected z-distribution. Also, the analy-sis of events with multiple tracks requires eithera common vertex, or possibly an additional dis-placed vertex for a reconstructed particle decay.

f Measure the scattering angles for kinematic re-construction.

f Measure the particle momentum from the trackdeflection in the spectrometer dipole magnet.

f Identify the hits in the PID detectors associatedwith each track.

When all tracking detectors are fully operational,the momentum resolution for positrons is limited


Table 1Tracking chamber properties

CHAMBER Vertex Drift vertex Front Magnet Back

Detector name VC1 VC2 DVC FC1 FC2 MC1 MC2 MC3 BC1/2 BC3/4mm from target 731 965 1100 1530 1650 2725 3047 3369 4055 5800

Active areaHorizontal (mm) 323 393 474 660 660 996 1210 1424 1880 2890Vertical (mm) 137 137 290 180 180 263 306 347 520 710

Cell design Microstrip gas Horizontaldrift

Horizontal drift MWPC Horizontal drift

Cell width (mm) 0.193 6 7 2 15A—C plane gap (mm) 3 3 4 4 8Anode (A) material 200lm glass(Al) W(Au) W(Au) W(Au) W(Au)Anode wire diameter 7lm 30lm 20lm 25lm 25 lmPotential wire mat@l Al strip Be—Cu(Au) Al(Au) Be—Cu(Au)Potential wire dia. 85lm 50lm 50lm 127lmCathode (C) material Al on glass Al on Mylar Al on Mylar Be—Cu wires C on KaptonCathode thickness 200lm 34lm 6.4lm 90lm @ 0.5 mm pitch 25.4lmGas composition: DME/Ne Ar/CO

2/CF

4Ar/CO

2/CF

4Ar/CO

2/CF

4Ar/CO

2/CF

4(%) 50/50 90/5/5 90/5/5 65/30/5 90/5/5U, V stereo angle #5°, !90° $30° $30° $30° $30°Resolution/plane (p) 65lm 220lm 225lm 700lm 275lm 300lmWires in X plane 1674 2046 80 96 96 496 608 720 128 192Wires in º,» plane 2170 2170 96 96 96 512 608 720 128 192Module config. VUX XVU XX@UU@VV@ UU@XX@VV@ UXV UU@XX@VV@Rad. length/module 0.8% 0.25% 0.075% 0.29% 0.26%Number of modules 1 1 1 1 1 1 1 1 2 2(upper or lower)Channels/module 6014 6386 544 576 576 1520 1824 2160 768 1152

Total channels 24 800 1088 2304 11 008 7680

by Bremsstrahlung in the material of the target cellwalls, the 0.3mm thick stainless-steel vacuum win-dow, as well as the first tracking detectors. Even theresolution in scattering angle is normally limited bymultiple scattering in this material. An exceptionwas 1995, the first year of operation, when the lowefficiency of most planes of the Vertex Chambers(improved for 1996 for the top detector) made itmore attractive to reconstruct the tracks using onlythe drift chambers (DC), as will be explained laterin more detail. This mode of operation resulted inthe resolution of reconstructed kinematic quantit-ies being much more sensitive to the DC spatialresolution.

The locations of the tracking detectors are shownin Fig. 1 and listed in Table 1, which also sum-marises their properties.

The microstrip gas counters VC1/2, the drift ver-tex chambers DVCs, and the front drift chambersFC1/2 provide both vertex reconstruction to thetarget, and definition of the scattering angle. Inconjunction with the front tracking, the back driftchambers BC1/2 and BC3/4 measure the magneticdeflection and hence the momentum. The BCs alsoidentify the cells in the PID detectors associatedwith each track. The proportional chambers insidethe magnet (MC1/3) were originally intended toensure that multi-track ambiguities could be re-solved. As it happens, chamber occupancies are lowenough that this can be accomplished using thedrift chambers alone. However, the MCs are foundto be very useful for momentum analysis of lowenergy decay products that are deflected too muchto reach the downstream tracking detectors.


Fig. 3. Schematic view of a drift cell in a MicroStrip GasCounter, showing the field lines. This figure has been taken fromRef. [6].

The horizontal length of the BCs precluded theuse of long horizontal wires for y measurements.Hence, all planes are one of three types withwires oriented either in the vertical (for X measure-ment), or tilted 30° right or left (º and » planes). Inorder to allow the use of fast reconstructionalgorithms, all tracking detectors are restrictedto this geometry. Like the rest of the spectrometer,the tracking system is symmetric about the beamplane.

5.1. Vertex chambers

The purpose of the vertex chambers is to providehigh-precision measurements of the scatteringangle and the vertex position over the full accept-ance of the experiment, in the presence of a signifi-cant background flux. The proximity of the VC tothe target implies that at the minimum scatteringangle, the tracks from the downstream end of thetarget cell pass the VC active area 20mm from thebeam axis, or only 5 mm from the beam tube. Incombination with the need to minimise high-Z ma-terials near the beam tube that could convert thenumerous photons from p0 decay into high multi-plicity showers, this geometry presented severeproblems in mechanical design that would havebeen very difficult to solve in the context of wirechamber frames. All these requirements are metthrough the use of a relatively new technology,a microstrip gas chamber (MSGC). In this tech-nique [5], metal strips are etched on a thin, semi-rigid substrate of high resistivity, allowing forhigher drift fields and much smaller cell pitch thanin a conventional drift chamber. Each of the upperand lower VC assemblies contains six MSGCplanes, grouped into two ‘modules’ (VUX andXUV for VC1 and VC2 respectively), which aresufficiently separated to provide a precise deter-mination of the front track by the VC alone. Thetotal radiation length of one module is 0.8%X

0and the thickness of the target chamber vacuum

window is 1.7% X0, so that for momenta below

about 7 GeV the angular accuracy becomes limitedby multiple scattering. An inherent cost associatedwith the MSGC choice is the large channel count— 24 800 channels in total. This requires ahighly integrated readout system, including analog

electronic signal storage on the planes and multi-plexed digitisation at the detector.

The basic features of the MSGC drift cell areshown in Fig. 3. Ionisation electrons created bya traversing particle in a 3 mm thick gas volumemove in a drift field of about 5 kV/cm directly awayfrom a planar cathode and towards a 200lm thickglass substrate that carries a pattern of fine etchedaluminum strips. The field is produced by a 1800Vpotential on the cathode planes. A voltage differ-ence of 580 V is applied between the 7 lmwide anode strips and the 85lm wide cathodestrips forming the pattern on the substrates. Asa result, gas multiplication occurs in the vicinity ofthe anode, a process comparable to that in a wirechamber. High gas amplification at 580 V is ob-tained by using a dimethyl-ether(DME)/Ne mix-ture (50/50%). The gas gain is set at a level suchthat the signals are about three times that of silicondetectors of the same material thickness, leading toa better signal to noise ratio. Even with only singlebit per cell digital read-out, a spatial resolutionmuch better than the cell pitch is achieved, due tocharge sharing between cells.

The metallic strip pattern that creates the electri-cal field configuration is carried on substrates, thematerial of which is a critical choice. Low Z is


preferred to minimise the effects of multiple scatter-ing and Bremsstrahlung. Surface properties are im-portant as electrons, ions, and gas pollutants maystick and cause field modifications, discharges, age-ing, and radiation damage. Since part of the ava-lanche falls on the substrate instead of on theconductive strip, the substrate should have a cer-tain degree of conductivity in order to remove thecollected surface charge. D-263 glass was chosen asthe substrate material [7]. It consists of SiO

x(64.1%), Al

2O

3(4.5%), K

2O (6.6%), TiO (4.2%),

ZnO (6.1%), Na (6.1%), with the remainder (8.4%)unspecified. It has a bulk resistivity of 1015) cm,equivalent to a surface resistivity of 1017 )/h,caused by the migration of electrons and Na` ions.The thickness of the substrates is 0.2mm, the min-imum required for safe handling and to have suffi-cient rigidity to resist electrostatic distortions.

The active area of the º or » planes is as large as484]137mm2. The largest substrate area thatcould be processed is 150]200mm2. Therefore,five substrates are combined to form one 30° plane.The substrates are cut along the cathode strip, andglued on a carrier frame, leading to a gap of onaverage 461lm between the anode wires of con-secutive substrates. No loss in efficiency at theseboundaries has been observed.

The PC carrier frame to which the substrates areglued has sufficient mechanical stiffness to be self-supporting. The carrier frames are mounted ina fixture attached to the inside of the lid of the boxthat acts as the gas enclosure. The fixture providesa means of precisely aligning the planes with re-spect to fiducials on the outside of the lid. The wallof the box, which is shaped to curve around thebeam tube, is only 0.5mm thick and made of low-Zmaterial. The thin windows for the detected par-ticles have two layers enclosing non-flammableflush gas to ensure no leaks of flammable gas.

In order to minimise external data flow, digitiz-ation of the signals is done in an electronic assem-bly mounted on the detector. The high degree ofdigitiser multiplexing required for economy is madepossible by continuously storing the charge bucketscollected by each anode strip in a switched-capaci-tor analog pipeline. There are 64 preamplifiers andpipelines, each 32 steps deep, integrated on eachAnalog Pipeline Chip (APC). The pipelines are

clocked with the HERA bunch frequency (10.4MHz). If an event trigger occurs any time within3.1ls (32]96ns) after an event, the pipelines arestalled before the data are shifted out and lost. Thepipelines are advanced until the buckets of interestin the vicinity of the bunch causing the trigger areabout to appear at the pipeline output stage. Thenthe pipelines are shifted much more slowly as thebuckets of interest are presented to the ‘trackingdiscriminators’, which were designed to compen-sate the rising baseline of the APC chip. A sampleand hold circuit is used to keep the threshold ata fixed value above the baseline offset, given bythe readout value of the previous channel. If a hitis detected, the baseline restore operation issuspended, until the following channel no longerexceeds the discriminator level. For that reason, thesystem is limited to handle a maximum cluster sizeof up to four neighboring anode strips. The firsttwo channels of each chip are not connected to ananode strip, and serve as a reference for the baselinerestoration. The data from each discriminator (62channels) are stored in its shift register. As soon asall digital data have been stored, the inputs of theAPC chips are enabled and a new trigger can beaccepted. The dead time during readout of the APCchips and digitisation into the shift register is 56ls.Simultaneously, serial readout of the shift registersis started. Addresses of signals above threshold areloaded into memory stacks. If during this operationa new trigger arrives, the new event waits in theAPC pipeline until the data of the previous eventare transferred into the memory stacks. Sub-sequently, the memory stacks are read out seriallyvia a data driver into a Struck ECL300 FASTBUSmodule.

The efficiency of the vertex detector is measuredin two ways. In the internal tracking approach,tracks are identified using VC data alone, requiringhits in five of six planes. The track is reconstructedfrom four planes, and a hit is accepted in the fifthplane if it is within 200lm of the track. Only tracksare considered that point to the target and arereconstructable by the rest of the tracking system.The efficiency of the fifth plane is then defined asthe fraction of events with a hit close to the track. Inthe external tracking approach, the efficiency isdetermined with tracks reconstructed using the


Fig. 4. Vertex reconstruction in the transverse coordinates us-ing the VC alone. This plot was made using two-track eventsonly to get a well-defined vertex. The outline of the target cell isshown as the ellipse.

drift chamber system, extrapolated to the vertexchamber. In both cases, efficiencies in the range60—90% were found for the 1995 data, wherea prototype version of the APC chip was used thatexhibited a strongly non-linear baseline. Thiscaused high noise. Due to the fragility of some ofthe anode resistors that were etched on the substra-tes, the high voltage could not be raised sufficientlyto get a high enough signal-to-noise ratio. For 1996running, the substrates and APC chips of the upperchamber were replaced, resulting in much im-proved efficiencies of around 95%. The lower halfwas similarly upgraded for 1997 running. The res-olution has been determined by comparing thedistance between the track, formed with five planes,and the hit in the sixth plane, and also by compar-ing the horizontal angle of the track calculatedfrom the X planes with that from the º and» planes. Both methods yield a resolution of about65lm (p) per plane.

The vertex reconstruction capability of the VC isillustrated in Fig. 4, which is a histogram of thetransverse coordinates of the point of closest ap-proach of two tracks in an event. A well-definedbeam envelope is seen. Furthermore, the figureshows that the vertices are well separated from thetarget cell wall represented by the ellipse.

Over three years of operation, there has been noobserved degradation of performance that wouldsuggest ageing effects.

5.2. The drift chambers

The drift chambers DVC, FC1/2, BC1/2 andBC3/4 are of the conventional horizontal-drift type.Each layer of drift cells consists of a plane of alter-nating anode and cathode wires between a pair ofcathode foils. The cathode wires and foils are atnegative high voltage with the anode sense wires atground potential. The chambers are assembled asmodules consisting of six such drift cell layers inthree coordinate doublets (ºº@, XX@ and »»@). Thewires are vertical for the X planes and at an angle of$30° to the vertical for the º and » planes. TheX@, º@ and »@ planes are staggered with respect totheir partners by half the cell size in order to helpresolve left-right ambiguities. Each sense plane con-sists of a fiberglass-epoxy frame laminated toa printed circuit board (PCB) that has solder padsfor the wires. Except in the case of the DVCs, thePCBs carry the traces leading past O-ring gas sealsto external connectors. The opposite ends of thecathode wires connect to internal distributionbuses for the negative high voltage. Each module isenclosed between metallised Mylar gas windowsmounted on metal frames. Identical modules aresituated above and below the beam pipe.

The frames of all the drift chambers except theDVCs have sufficient length to allow both ends ofall wires in the º and » planes to terminate on thelong edges of the frames. Hence, all º and » wiresand their PCB traces have the same length andsimilar capacitance. The onboard electronics aremounted along only the one long edge opposite tothe beam pipes. This arrangement does not sacrificemechanical stiffness of the frame. Since the insideaperture of each frame is only as large as the activearea in common with all the frames, each stressedframe member is as short as possible. A shallowrelief is milled in the inactive end regions of theº and » frames to allow the wires to stand off fromthe frame material. Proximity to the beam of theactive areas of all chambers required externalnotches to be machined in the frames to accom-modate the beam pipes. As these notches are short,


they do not seriously compromise the mechanicalstiffness of the frames.

The choice of gas mixture for the drift chamberswas governed by the serious inconvenience of con-trolling the hazards of a flammable gas in a tunnelenvironment. As well, there is an advantage inlimiting the drift cell occupation time and hence thenumber of extraneous hits that must be accommod-ated in track finding. The mixture Ar(90%)/CO

2(5%)/CF

4(5%) is both fast and non-flammable

[8]. Its drift velocity has been measured as a func-tion of field — e.g. '7 cm/ls at E"800V/cm [9].FC2 and BC1 operate in a dipole fringe field thatreaches 0.1 T. Mixtures containing CF

4have been

shown to result in compromised resolution due toelectron attachment [10], and large Lorentz driftangles in magnetic fields. However, they have theadvantages of short occupation time and longchamber lifetime [11].

The gas pressure and flow through all drift cham-bers is regulated by a system of valves and massflow controllers, all under the supervision of anindustrial programmed controller. A large fractionof the flow (80—90%) is recirculated through purifi-ers that remove oxygen and water vapour, but notnitrogen. In 1995, nitrogen from gas leaks in thedrift chambers and from diffusion in the windowfoils accumulated in the loop to an equilibriumconcentration of 1.1%. This influenced the driftvelocity in the gas mixture and resulted in a gainreduction as confirmed by independent studies[12]. Work on the chamber gas leaks as well asdecreasing the re-cycled fraction from 90% in 1995to 80% in 1996 (20% exhaust), reduced the nitro-gen contamination to a stable level of 0.4% withnegligible influence on both drift velocity and gain.

The DC readout system consists of Ampli-fier/Shaper/Discriminator (ASD) cards mountedonboard the drift chambers, driving ECL signals on30 m long flat cables to LeCroy 1877 MultihitFastBus TDCs in the external electronics trailer.The design goal (especially in the case of the FCsand DVCs) was chamber operation at low gas gain(&104) to maximise chamber lifetime and reduceparticle flux dependence of the gas gain due tospace charge effects. To combine this goal withgood spatial resolution requires operation at lowthreshold — well below 105 electrons from the wire.

This in turn requires both low loise performance ofthe amplifiers as well as great care in the radio-frequency design of the electronics enclosure on thechambers, to maintain electronic stability. Allground elements including the ASD cards themsel-ves should be considered as radio-frequency res-onators, and the lowest frequency of any resonantmode should be kept well above the pass-band ofthe amplifiers. For example, this dictates that theground plane of each 16-channel ASD card con-tinuously contact the electronics enclosure alongboth sides of the card via special high-conductivitycard guides.

During the first year of operation in 1995, alldrift chambers performed satisfactorily and aloneprovided the tracking data that were used in thephysics analysis. As discussed later, performancewas improved in 1996—97 by the addition of theDVCs.

5.2.1. The front drift chambersThe front drift chambers FC1 and FC2 provide

good spatial resolution immediately in front of thespectrometer magnet. Each chamber above or be-low the beam consists of one module of six senseplanes; thus the total front drift chamber packagepresents 12 sense planes to a charged particle. De-sign parameters are listed in Table 1.

A drift cell size of only 7mm (3.5mm maximumdrift length) was chosen to provide fine spatialgranularity to maximise tolerance to both corre-lated and uncorrelated background. It emergedthat uncorrelated background is mostly from col-limator showers, at the average level of only 0.1tracks per event. For a minor fraction of the events,the FC occupation from correlated tracks can besubstantial, but is always acceptable.

The ASD design is similar to that of the LeCroy2735DC card, except with a low-noise front endand pulse shaping added. With the input transistor(MMBR941LT1) used in common emitter modewith feedback, channel to channel gain variation isabout $10%. In order to optimise pulse pairresolution and hence rate capability, pulse shapingbetween stages is also added. A long tail due topositive ions from the avalanche drifting to thecathodes is cancelled by shaping with two stages ofpole-zero-cancelled differentiation. To facilitate


Fig. 5. FC resolution as function of drift distance.

Fig. 6. FC single-plane efficiency as function of drift distance.

testing of the chamber system, a linear monitorpoint is provided for each channel.

The threshold can be varied from approximately0.07—1.25lA at the input. At the lowest value, it willtrigger efficiently on an impulse charge of approx-imately 12 000 electrons, resulting in 8 ns widepulses at both the linear monitor and ECL logicoutputs. An impulse charge of 1600 electrons pro-duces a linear pulse of about the same magnitude asthe RMS noise. In the 1995 running period, the FCthreshold was dictated by ambient noise radiatedby the target r.f. system. In 1996, the target r.f. noisewas reduced and the electronics enclosure was im-proved, allowing a threshold of 140 nA from thewire, corresponding to a charge impulse of about20 000 electrons. The power consumption perboard is 0.3A at #5V and 0.55A at !5.2V, or270mW per channel. Hence moderate forced aircooling is provided.

Values of the resolution and efficiency of thechambers in 1996 are plotted as a function of trackposition across the drift cell in Figs. 5 and 6. Thevalues shown for the resolution are the result ofa fitting technique that removes the contribution ofthe tracking resolution to the track residuals. Thesevalues therefore better represent the resolution ofthe chambers.

5.2.2. The back chambersThe back drift chambers BC1/2 and BC3/4 are in

the size regime where special attention is requiredin the mechanical design to maintain the rigidityand stability necessary for uniform wire tensionand precise alignment. These detectors are de-scribed in detail in Ref. [13]. The active areas of themodules were chosen according to their z-positionsand the acceptance of the spectrometer. The geo-metrical parameters are listed in Table 1. Eachmodule contains 14 fiberglass/epoxy (GFK) framesof 8mm thickness, including six wire frames. Theyare clamped between two 38mm thick non-mag-netic stainless steel frames. The total thicknessof a module is 190mm. Because of their greatlength, the frames were prestressed to maintain thewire and foil tensions. The wires are located pre-cisely by contact with accurate pins made of poly-oxymethylene with 5lm tolerance, inserted in holesdrilled in the GFK frames with a precision of better

than 50lm. The holes are located along a curvedconvex locus that compensates the expected stressdeflection of the frames, in order to prevent dis-placement of the º and » wires. A mean deviationof 14lm for the position of the holes was measured.Pins precisely located in holes can be producedmuch more easily than combs and allow conve-nient installation of the wires. The gas is injecteddirectly into the wire gaps and exhausted via thewindow gaps. Work on the chamber gas leaks after


Fig. 7. Width of the BC1/2 and BC3/4 residual distributionas function of the drift distance for tracks with an incidentangle (1°.

Fig. 8. BC efficiency as a function of the drift distance for theupper BC modules under normal running conditions.

the 1995 running period reduced the leak rate of theBCs to the level of diffusion through the windowfoils. The cathode wires and foils are operated atthe same negative high voltage of typically 1770 V;the anode wires are at ground potential. Moredetails on the BC mechanical construction can befound in Refs. [14,15].

The chamber BC1 is located in the fringe field ofthe magnet. It has been verified by simulationsusing the program GARFIELD [16] that a verticalcomponent of 0.1 T contributes to a deviation inthe measured coordinate of (60lm, which can beneglected in the analysis.

The BC performance was studied extensively intest beam measurements illuminating essentiallyonly one drift cell per plane. Under these condi-tions, the HV/threshold setting 1750 V/50mV wasfound to be the optimum working point, yieldingp"175lm resolution (p"150lm in the centralregion of the BC drift cell) and 96.3% efficiency at6% cross talk. The efficiency was determined fromthe occurrence of chamber signals in a rather nar-row $4p corridor around the reference track,which was defined by a silicon micro strip detectortelescope [17,18].

The BC performance was evaluated using pro-duction data from normal running. The widths ofGaussian fits to the central portions of the residualdistribution for all tracks crossing the plane with an

angle (1° is given in Fig. 7 as function of the driftdistance for both BC1/2 and BC3/4. This plot wasobtained by combining the data from all cells of allplanes. Tracks close to the sense wire (i.e. small driftdistances) are excluded because this data point isstrongly affected by the difficulties in resolving theleft/right ambiguity. However, a smooth behaviourof the residual can be assumed. As expected, thebest value is measured in the central region of thedrift cell. The results are approximately 210lm forBC1/2 and 250lm for BC3/4.

The average BC plane efficiency for electron andpositron tracks is found to be well above 99% for1996 data. The situation is somewhat worse for alltracks, which are mainly hadrons, due the smallerenergy deposited in the chambers by hadrons. Theplane efficiency drops to 97% when all tracks areconsidered. The efficiency is calculated by using thereconstructed track as a reference and adopting thesame corridor width as used in the reconstructionprogram to find the hits belonging to a track, i.e.about $3p, where p is the plane resolution. Thedependence of the efficiency on the drift distanceaveraged over all drift cells is shown in Fig. 8.


Fig. 9. Averaged BC plane efficiency as a function of the atmo-spheric pressure for the 1995 data sample (open circles) anda 1996 data sample after introducing a gain stabilisation scheme(filled circles). For details see the text.

Investigations of the long-term stability of the1995 data showed the expected correlation betweenchamber performance and atmospheric pressure,which led to varying working conditions.

A system to compensate for atmospheric pres-sure variations by dynamically adjusting the highvoltage was introduced during the 1996 runningperiod. A ‘nominal’ high-voltage setting of 1770Vwas chosen for an atmospheric pressure of1013mbar. The actual operating voltage was com-puted using a parametrisation of the gain versusatmospheric pressure given in Ref. [19]. The highvoltage setting is corrected in steps of 1V overa range of $20V. This scheme significantly im-proves the stability of the chambers as can be seenin Fig. 9. The dependence of the average planeefficiency on the atmospheric pressure is presentedfor data before and after introduction of the gainstabilisation scheme in the middle of 1996 running.The overall improvement in efficiency using thissystem is 1—2%. Threshold settings of 65 mV for theBC1/2 and 80 mV for the BC3/4 have been used.The operation of the eight BC modules was re-markably stable and reliable in the first 2.5 yearswith low dark currents.

5.3. The drift vertex chambers

Notwithstanding the adequate performance ofthe tracking system during the 1995 commissioningyear, redundancy in the front region was insuffi-cient. Loss of a single FC plane could have hadserious consequences. There was also foreseen theneed for front tracking elements with larger accept-ance for a future upgrade to provide muon detec-tion beyond the standard acceptance. Hence anadditional Drift Vertex Chamber (DVC) pair wasconstructed, and installed immediately followingthe VC’s for use starting in the 1997 running period.The DVC drift cell design is similar to but slightlysmaller than that of the FCs. However, the DVCacceptance extends vertically from $35 to$270mrad, and covers $200mrad horizontally.

An acute design problem arising from the DVClocation near the target chamber exit window is theclose proximity of the active area to the beam pipe.This is solved in part by separating the functions ofgas enclosure and chamber frames. The gas enclos-

ure or box design is similar to that of the VCs, witha very thin wall less than 1mm from the beam tube.Each of six wire planes in a box is clamped betweenits own pair of cathode foil planes. All planes can beinstalled together in the box while they are matedwith the signal feed-throughs in the lid. However,the planes are thereafter precisely aligned by twopins equipped with gas seals, externally insertedthrough the box and all the planes. The ends ofthese pins provide external alignment fiducials.A gas injector is also externally inserted throughthe box wall into each anode plane after its installa-tion in the box. The gas spills out of the active gapsinto the box, and hence to a common exit port.Both the signal feed-throughs and the gas injectorsare designed for mechanical accommodation suffi-cient to avoid constraining the alignment of theplanes with respect to the box.

The readout system of the DVCs is identical tothat of the FC’s, resulting in similar performancein 1997 except for one of the 12 DVC planes,which suffered a broken wire immediately afterinstallation.

5.4. The magnet chambers

The proportional wire chambers MC1 throughMC3 located in the gap of the magnet were originally


intended to help resolve multiple tracks in case ofhigh multiplicity events. Since low backgroundshave made this unnecessary, their primary functionis now the momentum analysis of relatively lowenergy particles, from the decay of K’s for example.Since the MCs operate in a strong magnetic fieldand resolution of $1 mm is sufficient, multi-wireproportional chambers with digital single bit-per-wire readout were chosen. The magnet hasa tapered pole configuration that implies a differentsize for each pair of chambers. The active areas ofthe three chambers given in Table 1 are dictated bythe spectrometer acceptance together with mag-netic dispersion at a nominal trigger thresholdpositron energy of 3.5GeV.

Each chamber consists of three submodules º,Xand », laminated into one module with commongas volume. Each submodule consists of an anodeplane at ground potential and two cathode planeswith a common negative HV connection unique tothis submodule, typically at 2850V. The distancebetween anode and cathode planes is 4$0.03mmand the centers of neighbouring submodules areseparated by 21mm. The gas mixture has the sameconstituents as the drift chamber gas, but withproportions (Ar—CO

2—CF

465 : 30 : 5) optimized for

MWPC operation (see Table 1).The most difficult aspect of the MC design was

the limited space inside the magnet for the framesand onboard electronics. The electronics are moun-ted on the front and back faces of the chambers inthe barely adequate space between the magnet poleface and the spectrometer acceptance. Thin flexibleprinted circuit foils connect the wire frames to theelectronics. The readout is the LeCroy PCOS IVsystem, including an on-chamber card design con-figured specifically for this application [20]. In ad-dition to signal amplification, these cards providediscrimination and delay as well as latching inresponse to the event trigger. The high-speed serialreadout (up to 20 Mbit/s) greatly reduces thecabling in the very confined space of the magnetgap. The most severe constraint on MC operationis the inaccessibility of the detectors and the on-board electronics for the entire running period. It isnot possible to work on the detectors during regu-lar monthly accesses because they are buried insidethe magnet. Unfortunately, a faulty batch of

discriminator/latch chips together with a less-than-optimal design of the output motherboardsused in 1995 created a reliability problem. Thefailure of a single chip that reads out only eightwires disabled as many as 16 cards, correspondingto 256 wires. This process in combination witha low voltage distribution malfunction eventuallyresulted in the failure of a large fraction of the MCsystem, with overall tracking efficiency of only45%. A redesigned motherboard eliminated thisproblem for 1996 operation. The MC system thenperformed reliably with typical efficiency per planeof 98—99%. Electronics failures in the 1996—97 run-ning period resulted in the loss of only 0.5% of thechannels (3 of 688 cards).

The very restricted air flow in combination withthe power dissipation of 250mW per channelmakes water cooling a necessity. Flexible metallicfingers contact the tops of the SMD chips andconduct the heat to water-cooled plates betweenthe cards. The water system operates at a pressureless than atmospheric to eliminate the danger ofleaks.

The contribution to the spectrometer momentumresolution for 10GeV positrons by Coulomb mul-tiple scattering in the magnet chambers and airbetween FC2 and BC1/2 is h

3.4"0.15mrad. (If

the air were replaced by helium, it would be de-creased to 0.11mrad.) Thus the ratio of h

3.4to the

magnet deflection angle is equal to 0.0024, indepen-dent of energy. This is small compared to othercontributions.

5.5. Alignment

Errors in the relative alignment of the variouselements of the tracking system are an importantcontribution to the resolution in reconstructedkinematic quantities. The initial alignment dur-ing detector installation was done using conven-tional optical techniques, but difficulties withoptical access to the components limited the ex-pected accuracy of this process to a few hundredmicrometers in x and y, and of order 1mm in z. Itwas always expected that the relative detectoralignment would have to be improved in the dataanalysis through the study of track residuals withthe spectrometer magnet turned off. This procedure


Fig. 10. Fresnel zone plate made of 13lm thick stainless steelwith a diameter of 24.5mm.

was refined over the first three years of operationto the point that drift chamber alignment errorsare no longer a major contribution to trackingresolution [21].

5.5.1. Laser alignment systemIt was considered essential to have a means to

continuously monitor the relative alignment of thetracking detectors for two reasons. First, the de-tectors are mounted on a complex support systemthat is not simply interconnected so that its re-sponse to temperature changes and gradients isdifficult to predict. Also, the front detectors aresupported ultimately via the spectrometer magnetfield clamp, which might suffer distortion frommagnetic forces. Such an effect was detected withthis monitoring system, and eliminated by stabilis-ing the clamp mount. Secondly, if a detector mustbe replaced with a spare after the laborious align-ment analysis has been done, it is desirable to beable to avoid having to repeat this process immedi-ately with new field-off data.

A solution to both of these problems is to haveon each detector module optical targets that arecontinuously accessible to a remotely operated im-age recording system [22]. The type of system thatwas chosen employs two well-collimated parallellaser beams passing through the entire spectrom-eter in the beam plane, on either side of and parallelto the beam axis. The 2 cm diameter beams arecreated by one He—Ne laser and a semi-transparentmirror. Two optical target assemblies are mountedprecisely and reproducibly on each detector ele-ment. Each assembly includes a remotely control-led actuator that can move an optical target intothe path of a laser beam, arriving at a precise andreproducible stop. Only one target is in a laserbeam at a time. The type of target used — theFresnel zone plate — is similar to that used in theSLAC accelerator alignment monitoring system. Itis made by etching a precise pattern through a thinmetal foil. This avoids any deflection of the beamfrom imperfections in the planarity of transparentmaterial. A Fresnel zone plate is shown in Fig. 10.The large unperforated central regions were left forthe sake of mechanical stability. They have thedisadvantage that they produce secondary fringesin the images.

The zone plates on each detector are specifiedwith a focal length equal to the optical path lengthfrom the target to the image recording component,which is a lensless CCD camera with a pixel size of11lm ]11lm. Hence, any shift in the detector ortarget is reproduced with unity magnification asa shift in the focal pattern at the camera. Thecamera image frames are digitised and recorded viaa VME-based system.

Since there are two targets on each detector atx"$45 cm, all transverse detector coordinatescan be monitored. No absolute alignment is attem-pted with this system. In fact, long-term stability ofthe laser beams is not assumed, since the opticalpaths contain several mirrors that are subject toinstability. The beams must remain stable only forthe 10min period required to cycle through alltargets in each beam path. Only the relative xand y positions of the images recorded within thatcycle are interpreted as being significant. Thermal


Fig. 11. Typical Fresnel pattern image detected by the CCDcamera. A one-dimensional projection is shown at the top andon the right. Nine peaks are visible at the centre. The position ofthe central peak is fitted.

wavering of the beams is reduced by insulating thebeam path through the warm magnet shieldingplate using a foam-cell tube. Several camera framesare averaged to minimise the effects of thewavering.

A Fresnel pattern image is shown in Fig. 11. TheFWHM of the central peak is about 300lm. Theprecision of the laser alignment system is 30—50lm.

5.6. Track reconstruction

The HERMES reconstruction program (HRC)[23] is very efficient because it makes use of twounusual methods: the tree-search algorithm for fasttrack finding and a look-up table for fast mo-mentum determination of the tracks. The full re-construction of 30 Monte Carlo events on an SGIR4400 processor takes only 1 s.

5.6.1. The fast pattern recognition algorithmThe main task of the reconstruction program is

to identify particle tracks using the hits in thetracking chambers. Each detector plane gives spa-tial information in one coordinate, and only bycombining the information of many detectors is itpossible to reconstruct the tracks uniquely in space.There are several track-finding algorithms in com-mon use. HRC uses the tree-search algorithm,

which turns out to be very fast. This algorithm isimplemented in the following way.

As a first step, ‘partial’ tracks have to be found inprojections, separately in the region in front of andbehind the spectrometer magnet. The track projec-tions are approximately straight lines in those re-gions, except for small curvatures caused by themagnetic fringe fields, and kinks coming from sec-ondary interactions and straggling.

The basic idea of pattern recognition using thetree-search algorithm is to look at the whole hitpattern of the detectors with variable (increasing)resolution as illustrated in Fig. 12. At each step ofthe iteration the entire partial track is seen. How-ever, only at the end is the full detector resolutionreached. The detector resolution is roughly 250 lmand the active width of a chamber is of the order ofa few meters. Therefore, after about 14 steps in thebinary tree, the resolution of the detector of&1 : 214 would be reached. For the purpose oftrack finding (not fitting) a resolution of 1 : 211 issufficient, which reduces the maximum number ofiterations in the HRC tree-search to about 11.

In each step of the iteration, the algorithm checksif the pattern (at the given resolution) containsa sub-pattern that is consistent with an allowedtrack. Fig. 13 shows examples of allowed and for-bidden patterns. All allowed patterns are generatedand stored in a database at the initialisation phaseof the program. The comparison is very fast as onlylook-up tables are used; no calculations have to bedone during event processing.

The number of allowed track patterns at the fullresolution is of the order of 108. Without optimisa-tion, this large number would make the algorithmuseless for two reasons: memory space problems tostore all the patterns, and CPU time problems tocompare with each of them. The following methodsavoid these problems:

f If one compares the allowed patterns fromone iteration to those from the following one(called parent and child in the following) it be-comes obvious that only a very limited numberof children exist for each parent. At initialisationtime all links between the allowed patterns ofeach generation are calculated, i.e. all childrenfor each parent are stored in the database. The


Fig. 12. The tree search algorithm looks at the hits of thetracking detectors with artificially reduced resolution. In everytree search level the resolution is doubled until a resolution isreached that is optimal for track finding.

Fig. 13. At each tree search level, the allowed and forbiddenpatterns are well defined. The program compares the measuredpattern with the database and determines if the measured pat-tern is consistent with an allowed track.

number of comparisons now becomes small: ifa pattern is recognised at one generation, only itschildren have to be checked in the next genera-tion. The number of children for each parent istypically 4—8. This solves the CPU problem.

f Symmetry considerations can be used to reducethe number of patterns that have to be stored. Iftwo patterns are mirror-symmetric, or if they areidentical except for a transverse shift, they arestored as one pattern. Most important is that ifa child is identical to a parent, then the child islinked to the parent and all grand children be-come identical to the children. This simplifica-tion is based on the fact that we are looking onlyfor approximately straight tracks. This reducesthe number of branches in the tree-search signifi-cantly. The number of patterns in the database isof the order of 50 000, which is small comparedto the original number of 108.

After applying the tree search algorithm to theº and » planes, the partial tracks in these projec-tions are defined. They are called tree-lines. Bytesting all combinations of tree-lines and mergingthem with hits in the x coordinate, the partialtracks in space are found. The tree-search algo-rithm is not applied to the x-coordinate directly as


portions of the X-planes of the VC chambers aretilted and thus do not fulfil the symmetry condi-tions. However, the x projections are used for trackfinding in the back region.

The procedure described until now is appliedindependently to the front and back regions, result-ing in a set of front partial tracks and another set ofback partial tracks. All combinations of front andback partial tracks are tested to see if they matchspatially within a specified tolerance at the x—yplane in the center of the magnet. Those combina-tions that match are combined to form a full track,after refitting the track geometry to the chamberhits, subject to the match condition. The matchingcondition is refined by various small correctionsdepending on partial track geometry and deflectionangle from the first iteration.

5.6.2. The fast momentum look-upAs a second feature in HRC, a very fast method

has been developed to determine the momentum ofa track, which is given by the deflection in theinhomogeneous field of the spectrometer magnet.Tracking through a magnetic field is very CPU-intensive. The new method makes obsolete thetracking through the magnetic field on a track bytrack basis. Instead, a large look-up table is gener-ated only once during initilisation. It contains themomentum of a given track as a function of thetrack parameters in front of and behind the magnet.The relevant track parameters are the position andthe slope of the track in front of the magnet and thehorizontal slope behind the magnet. The resolutionof the table has been chosen such that, using inter-polation methods, the contribution by HRC to theprecision of the track momentum determination isbetter than *p/p"0.5%. The look-up table con-tains 520 000 numbers. This method of momentumdetermination is extremely efficient.

The vertical slope and position of the track be-hind the magnet are not kinematically relevantdegrees of freedom and are used only to determinethe track quality and reduce the number of ghosttracks. They are not used in the determination ofthe kinematic parameters, as the resolution andalignment of the VC-FC system is superior to theresolution of the back chambers.

5.6.3. Tracking in 1995 without VCs and DVCsAs mentioned earlier, the VC performance in

1995 was not optimal due to difficulties in produc-tion of the APC readout chips. The performance ofthe substitute prototype chips resulted in marginalefficiencies (60—90%) and many ‘hot’ channels. TheDVCs were conceived later and became availableonly for 1997. Hence, an alternative track recon-struction method was developed using only thedrift chamber data. The back partial tracks aregenerated as usual and the front partial tracks aregenerated using only the FC data. The front—backmatching search is done as usual, except with some-what larger tolerances. Then, for all matching pairsof partial tracks, the match point at the centre ofthe magnet defined by the back partial track is usedto refine the front partial track by pivoting it abouta conserved space point midway between FC1 andFC2. Thus, the front partial track is forced to agreeat the magnet midpoint with the presumably high-er-quality information from the back partial track.This process was used successfully for the 1995/96physics analysis, with approximately a factor of twoloss of resolution in kinematic quantities relative towhat could be expected if the VCs were fully opera-tional.

5.7. Tracking system performance

The performance of the tracking system has beenstudied using the detailed HERMES Monte Carlo(HMC) simulation of the experiment, which isbased on the GEANT [24] software package. Res-olutions for several quantities were estimated byperforming a Gaussian fit to spectra of the differ-ence between the generated and reconstructedquantities. Results for the measured quantitiesp (momentum) and h (scattering angle) are shown inFigs. 14 and 15. The momentum resolution is0.7—1.25% over the kinematic range of the experi-ment, while the uncertainty in the scattering angleis below 0.6mrad everywhere. Bremsstrahlung inmaterials in the positron path cause the momentumresolution function to have a standard deviationconsiderably larger than that of the fitted Gaussian.The resolutions for x and the square of themomentum transfer are shown in Figs. 16 and 17.The x resolution varies from 4 to 8% while


Fig. 14. Momentum resolution in the HERMES spectrometer,deduced from Monte Carlo studies.

Fig. 15. Resolution in scattering angle deduced from MonteCarlo studies.

Fig. 16. Resolution for the Bjorken x variable, deduced fromMonte Carlo studies.

Fig. 17. Resolution for the square of the four-momentum trans-fer, Q2, deduced from Monte Carlo studies.

the Q2 resolution is better than 2% over most of thekinematic range.

Finally, the absolute calibration of the spectrom-eter can be checked using the decay of K

Smesons

into two pions. The invariant two-pion mass isplotted in Fig. 18. The reconstructed K

Smass

agrees to one part per mil with the value of theParticle Data Group [25] (497.4 vs. 497.7MeV).The width of the peak (p"0.0057, or +1.1%)agrees with the resolution determined by MonteCarlo methods.

5.8. Kinematic variables

Numerous measured and calculated quantitieshave been compared to the Monte Carlo simula-tion of the experiment (HMC). Fig. 19 shows someof these comparisons for the momentum p of thescattered particles, the variable x defined earlier,the four-momentum transfer squared (Q2), and thesquare of the photon—nucleon invariant mass (¼2).The normalisations of the histograms are arbitrary.The points represent data while the shaded histo-grams are the Monte Carlo simulation. The agree-ment is good over most of the range of thevariables.


Fig. 18. Invariant mass of p`—p~ pairs. The reconstructedK

Smass agrees very well with the PDG value.

Fig. 19. Kinematic variables p, x,Q2, and ¼2. The points aredata while the shaded histogram is a Monte Carlo simulation.The normalisations of the histograms are arbitrary.

Fig. 20. Kinematic plane for the HERMES experiment. See textfor details.

The kinematic plane for the experiment (Q2 vs. l)is shown in Fig. 20. The typical DIS acceptance isdefined by the following software cuts shown as thebold lines on the plot:

f Scattering angle: 40mrad (h(220mrad;f Fractional energy transfer: y"l/E(0.85;f ¼2'4 GeV2;f Q2'1GeV2.

6. The particle identification detectors

6.1. General description

The HERMES particle identification system dis-criminates between positrons, pions, and otherhadrons. It provides a factor of at least ten inhadron suppression at the DIS trigger level to keepdata acquisition rates reasonable. The rate of DISpositrons is much exceeded by that of hadrons fromphotoproduction by a factor as high as 400 : 1 incertain kinematic regions. The system providesa hadron rejection factor (HRF) of at least 104 inoffline analysis to keep the contamination of thepositron sample by hadrons below 1% forthe whole kinematic range. The HRF is defined asthe total number of hadrons in the spectrometer


Table 2Properties of F101 glass

Chemical composition (weight %)

Pb3O

451.2

SiO2

41.5K

2O 7.0

Ce 0.2

Radiation length (cm) 2.78Density (g/cm3) 3.86Critical energy (MeV) 17.97Moliere radius (cm) 3.28Refractive index 1.65Thermal expansion coefficient (C~1) 8.5]10~6

acceptance divided by the number of hadrons mis-identified as positrons. It depends on the selectedefficiency of positron identification (see Table 3later for typical efficiencies and contaminations).Since most of the hadrons are pions, we will oftenrefer to a pion rejection factor (PRF). The PIDsystem also discriminates pions from other hadronsfor the important semi-inclusive measurements thatwill allow the contribution of the valence and seaquarks to the nucleon spin to be isolated.

The PID system consists of four sub-systems:a lead-glass calorimeter, two plastic scintillatorhodoscopes, one of which is preceded by two radi-ation lengths of lead and which acts as a pre-shower detector, a transition radiation detector,and a threshold Cherenkov detector (see Fig. 1).The calorimeter and the hodoscopes are used in thefirst-level trigger to select DIS events. Beam tests atCERN have shown that this combination givesa HRF of several thousand for an electron efficien-cy of 95% in off-line analysis. The rejection factor isestimated to be approximately 10 in the trigger.The TRD consists of 6 modules in each half andprovides an additional HRF of over 100 for 90%e` efficiency (several hundred if a probability basedanalysis is used). The main function of the thresholdCherenkov detector is to distinguish pions fromother hadrons for the semi-inclusive measurements.In future, kaons will also be identified using a ringimaging Cherenkov detector (RICH). An upgradeof the current threshold detector to a RICH is beingprepared for 1998. Each sub-system is described indetail in the following sections.

6.2. The calorimeter

The function of the calorimeter is to providea first-level trigger for scattered positrons, based onenergy deposition in a localized spatial region(53.5GeV in 1995 and the beginning of 1996 and51.4GeV in late 1996 and 1997); to suppresspions by a factor of 510 at the first-level triggerand 5100 in the off-line analysis; to measure theenergy of positrons and also of photons from radi-ative processes or from p0 and g decays.

The solution chosen to meet these requirementsconsists of radiation-resistant F101 lead-glassblocks [26] arranged in two walls of 420 blocks

each above and below the beam. The properties ofF101 glass are listed in Table 2.

Each block is viewed from the rear by a photo-multiplier tube (PMT). The blocks have an area of9]9 cm2, a length of 50 cm (about 18 radiationlengths), and are stacked in a 42]10 array. Theblocks were polished, wrapped with 0.051mm thickaluminized mylar foil, and covered with a 0.127mmthick tedlar foil to provide light isolation. Eachblock is coupled to a 7.62 cm Philips XP3461 PMTby a silicone glue (SILGARD 184) with refractiveindex 1.41. A l-metal magnetic shield surroundsthe PMT. A surrounding aluminum tube housesthe l-metal and provides the light seal. It is fixed toa flange that is glued to the surface of the lead glass.This flange is made of titanium, matching the ther-mal expansion coefficient of F101.

To prevent radiation damage of the lead glass,both calorimeter walls are moved away verticallyfrom the beam pipe by 50 cm for beam injection.The monitoring of gain and ageing is achievedusing a dye laser at 500 nm, which sends light pulsesof various intensities through glass fibres to everyPMT of the calorimeter, and additionally to a refer-ence counter photodiode. The various intensitiesare produced by a rotating wheel with several at-tenuation plates. The light is split in several stagesbefore being fed into the glass fibres. As the gain ofthe photodiode is stable, the ratio of the PMTamplitude to that of the photodiode signal can beused to monitor relative gain changes in the photo-multipliers. Over three years of operation, there has


Fig. 21. E/p distribution measured for all counters (one entryeach) for the runs collected during the first year data takingperiod.

Fig. 22. p0 mass reconstructed from 2c cluster events.

been no observed degradation of performance thatwould suggest ageing effects.

Radiation damage to the lead-glass is alsomonitored indirectly using TF1 [27] blocks placedbehind the calorimeter. This material is much moresensitive to radiation damage than F101. There-fore, a degradation of the response would be seenmuch sooner in these monitor detectors if therewere a large radiation dose incident on the back ofthe calorimeter caused by showers produced bybeam loss in the proton machine. So far no vari-ation has been observed in their response, indicat-ing that the effect of radiation damage is negligible.

Measurements with 1—30GeV electron beamswere performed at CERN and DESY with a 3]3array of counters. Each lead-glass block was alsocalibrated within +1% at DESY in a 3GeV elec-tron beam incident at the center of the block. Theperformance of a 3]3 array of counters showed[28]: (i) an energy response to electrons linear within1%, over the energy range 1—30 GeV; (ii) anenergy resolution that can be parameterised asp(E)/E [%]"(5.1$1.1)/JE [GeV]#(1.5$0.5):this is similar to that obtained for other large lead-glass calorimeters; (iii) a spatial resolution of theimpact point of about 0.7 cm (p); (iv) a PRF of+(2.5$1.2)]103 integrated over all energies incombination with the preshower detector, fora 95% electron detection efficiency.

The central values of E/p distributions for posi-trons, measured for tracks incident on each blockduring the 1995 data taking period are plotted inFig. 21. E and p are respectively the energy ofpositrons measured by the calorimeter and the mo-mentum determined by the spectrometer. The ratiohas a central value of 1.00 with a width (p) of 0.01,demonstrating that the response of the counters isuniform to + 1%. The long term stability of thedetector response of 1% is determined from themean value of the E/p distribution, measured foreach run for the counters nearest to the beam. Thisvalue also includes the contribution of the radi-ation damage produced in one year of operation,indicating the effect of the radiation damage isnegligible.

The overall calibration of the calorimeter can bedetermined by the reconstruction of p0 decays.Fig. 22 shows the p0 invariant mass for events with

two c clusters. The centroid of the peak for the 1995data is (134.9$0.2) MeV with a p"(12.5$0.2)MeV in good agreement with the particle datagroup value [25]. The calorimeter is described inmore detail in a separate paper [29].

6.3. The hodoscopes

A scintillator hodoscope (H1) and Pb-scintillatorpreshower counter (H2) provide trigger signals andparticle identification information. Both counters


Fig. 23. Response of the pre-shower detector to hadrons andpositrons.

are composed of vertical scintillator modules (42each in the upper and lower detectors), which are1 cm thick and 9.3 cm ] 91 cm in area. The mater-ial for the modules is BC-412 from Bicron Co.,a fast scintillator with large attenuation length(300—400 cm). The scintillation light is detected by5.2 cm diameter Thorn EMI 9954 photomultipliertubes, one coupled via a light guide to the outsideend of each scintillator (away from the beam plane).The modules are staggered to provide maximumefficiency with 2—3mm of overlap between eachunit.

In addition to providing a fast signal that iscombined with the calorimeter and H1 to form thefirst-level trigger (H0 was implemented in 1996; seebelow), the H2 counter provides important discrim-ination between positrons and hadrons. This isaccomplished with a passive radiator that initiateselectromagnetic showers that deposit typicallymuch more energy in the scintillator than minimumionizing particles. The passive radiator consists of11mm (2 radiations lengths) of Pb, sandwichedbetween two 1.3mm stainless-steel sheets. The re-sponse of the H2 counter to positrons and hadronsis shown in Fig. 23. While pions deposit only about2MeV, positrons produce a broad distribution of

deposited energies with a mean of 20—40MeV thatdepends weakly on the energy of the incident posit-ron. A PRF of &10 is possible with 95% efficiencyfor positron detection.

6.4. Forward trigger scintillators

In 1995, a large number of showers generated bythe proton beam were able to satisfy the triggerrequirements for DIS positrons. There were typi-cally between 20 and 100 triggers per second of thistype, compared to about 35 positron triggers and35 pion-induced triggers per second (with the stan-dard 3He target operation and about 30 mA ofpositron current).

A forward trigger scintillator (H0) placed directlyupstream of the front drift chambers was introduc-ed in 1996 to eliminate these triggers by distin-guishing forward and backward going particlesusing the time of flight between the front and rearscintillators.

The total time required for a light-speed particleto traverse all trigger detectors is on the order of18 ns, and hence a backward-going particle pro-duces a pulse in the front counter that is displacedin time by about 36ns from the normal triggercondition. Such a large time displacement allowsfor easy elimination of these background events inthe trigger.

The front trigger detectors consist of a singlesheet of standard plastic scintillator, 3.2mm thick(0.7% of a radiation length). The occupancy of thescintillators is such that segmentation of the de-tector is not necessary. The rate in each paddle isabout 106 per second. The scintillation light iscollected along the edge farthest from the beam axisinto two sets of lucite strips and then into two5.08 cm Thorn EMI 9954SB phototubes.

6.5. The Cherenkov detector

Pion identification is provided by a pair ofsingle-gas radiator threshold Cherenkov counters,one each above and below the beam. The units arelocated between the back drift chamber groups,BC1/2 and BC3/4. The radiator is a gas mixture atatmospheric pressure of nitrogen and per-fluorobutane, C

4F10

, the composition of which can


Fig. 24. Response of the Cherenkov detector from 1995 data;note that events below 0.15 photoelectrons have been sup-pressed.

be varied to control the momentum interval overwhich pions can be distinguished unambiguouslyfrom other hadrons. The depth of the radiatorvolume is 1.17m. During the first year of operation,the radiator was pure nitrogen for which theCherenkov momentum thresholds for pions, kaons,and protons are 5.6, 19.8, and 37.6GeV, respective-ly. In 1996—97, a mixture of 70% nitrogen and 30%perfluorobutane was used, which gives correspond-ing thresholds of 3.8, 13.6, and 25.8GeV.

The body of each unit is constructed of alumi-num. The entrance and exit windows have areas0.46m ] 1.88m and 0.59 m ] 2.57m, respectively.Each window consists of a composite foil of 100 lmof mylar and 30lm of tedlar separated by a 1 cmgas gap from a second identical composite foil. Drynitrogen gas continuously flowing through this gapin each window provides a barrier against the diffu-sion of atmospheric gas through the windows intothe radiator gas volume. An array of 20 sphericalmirrors (radius of curvature: 156 cm) mounted atthe rear of the counter on Rohacell forms focusesthe Cherenkov light emitted along particle trajecto-ries onto corresponding members of an array ofphototubes. The mirrors are coated with aluminumand magnesium fluoride and have a measured re-flectivity of 90% at 400 nm. The mirror array ismounted with individual adjustment on a plane of3.0 cm Rohacell foam reinforced with fibre glassepoxy films. The average effective particle pathlength in the gas viewed by the mirrors is 0.95m.

The phototubes (Burle 8854) have 12.7 cm dia-meter photocathodes, and are fitted with Hinter-berger—Winston light cones to maximise the lightcollection. The funnels have an entrance diameterof 21.7 cm and can accept light at angles of up to26° from the optical axis. They are coated withaluminium to give a reflectivity of 85% at 400nm.Magnetic shielding for the phototubes is providedby three concentric l-metal shields and a soft ironhousing. The photocathode faces are coated witha layer of 200lg/cm2 of p-terphenyl and a protec-tive coating of magnesium fluoride. Thiswavelength shifter increases the photoelectron yield[30]. The mean number of photoelectrons fora b"1 particle with a pure nitrogen radiator wasmeasured to be slightly less than 3. LEDs mountedon each Winston cone provide an on-line gain

monitoring system. Calibration spectra are accu-mulated during normal running with the LEDamplitudes set sufficiently low so that strong single-photoelectron peaks are resolved cleanly. Thesepeaks provide an accurate measure of the gain perphotoelectron for each channel.

The typical response of the Cherenkov detectoris shown in Fig. 24. The single photoelectron peakis prominent and reasonably well separated fromthe pedestal.

6.6. The transition radiation detector

The purpose of the transition radiation detector(TRD) is to provide a pion rejection factor of atleast 100 for 90% positron efficiency at 5GeV andabove. Only positrons produce transition radiationin the HERMES energy regime and the detection ofone or more TR X-rays in coincidence with thecharged particle can be used to discriminate be-tween positrons and hadrons. The X-ray is difficultto distinguish because it is emitted at an angle of 1/cto the momentum of the charged particle and istherefore not separable in the detector from thepositron.

Because the space restrictions imposed on theTRD were not severe, a relatively simple and econ-omical design was chosen. Rather than optimiseeach module of the TRD for maximum performance,


the overall performance is assured by the additionof extra modules. The design of the TRD was opti-mised using a Monte Carlo program that trackscharged particles and photons down to an energyof 1 keV. Care was taken to model properly theproduction of d-rays in the radiator and thedetector gas as well as hadronic showers inthe radiator. Elements of the TRD simulationwere later integrated into the general spectrometersimulation [31].

The HERMES TRD consists of six modulesabove and below the beam. Each module containsa radiator and a Xe/CH

4filled proportional cham-

ber. In addition, there are two flush gaps on eitherside of each detector through which CO

2is flowed

to reduce the diffusion of oxygen and nitrogen intothe chamber. Such contaminants would affect theresponse significantly. CO

2is used in the flush gaps

because it is easily removed from the chamber gasduring re-circulation.

The use of polyethylene foils as a radiator is notpractical for HERMES because of the large size ofthe modules (active area: 325 ] 75 cm2). It is notpossible to maintain a uniform separation of thefoils for such a large radiator. A good approximationto a foil radiator is provided by a pseudo-randombut predominantly two-dimensional matrix offibers [32]. Several radiators were compared ina test beam at CERN: foils, fiber matrices withdifferent fiber diameters, and polyethylene foams.Fiber radiators were shown to produce a responseonly slightly less than that from a foil radiator. Thefinal design uses fibers of 17—20lm diameter ina material with a density of 0.059 g/cm3. This den-sity is significantly less than would be found to beoptimal if hadronic showers and delta rays from theradiator were ignored. The radiators are 6.35 cmthick, which corresponds to an average of 267 di-electric layers.

The detectors are proportional wire chambers ofconventional design, each of which consists of 256vertical wires of 75lm gold-coated Be—Cu separ-ated by 1.27 cm. The wires are not soldered butrather held in crimped pins. This makes it muchsimpler to replace a broken wire in situ. The wiresare positioned with 25lm accuracy using preciselymachined holes for the crimp pins in plastic strips(Ultem) that are glued to the aluminium chamber

frames. The chambers are 2.54 cm thick. Xe/CH

4(90 : 10) is used as the detector gas because of

its efficient X-ray absorption. The electric field inthe chamber is produced by a #3100V potentialapplied to the anode wires; the cathode foils are atground potential. The large voltage on the anodewires is needed to produce an electric field in thecritical region between the wires of sufficientstrength to collect the ionisation electrons withinthe 1ls ADC gate used in the readout electronics.The wire diameter was chosen to be unusually largeto allow operation at this high voltage while limit-ing the gas gain to about 104. The drift cell proper-ties were optimised using the computer programGARFIELD [16].

The active gas volume of the detector is definedby the cathodes which are 50lm thick mylar foilsaluminised on both sides. The foils are stretched toan unusually high tension (1.75kN/m) to make themmore stable against pressure differences between thedetector and the flush gaps. This is important be-cause a 10 lm deviation in the position of thecathode causes a 1% gain shift in the detector. Allframes supporting the anode wires and the cathodefoils were pre-stressed during construction.

The tension of each wire was measured usingresonant vibration induced by the combination ofa current flowing through the wire and an externalmagnetic field. Furthermore, each wire was scan-ned with an X-ray source every 2.54 cm duringoperational tests after assembly. Any gain vari-ations of more than 10% caused by imperfectionsin the wire resulted in the wire being removed andre-strung.

The signals from the wires are amplified by elec-tronics mounted along the long edge of the detectorfurthest from the beam. Balanced differential sig-nals are transmitted over 35m long twist/flat cablesto the electronics trailer. The use of differentialsignals reduces common mode noise, which wouldseverely degrade TRD performance. Receiver mod-ules filter, amplify, and delay the signals by 256ns,while a first-level trigger decision is made, at whichtime they are digitised by LeCroy 1881M FastbusADCs. Test pulses are injected via an antenna atthe other end of the wires to allow for a quick checkof the detector and readout electronics withoutbeam.


Fig. 25. Response of a single TRD module to positrons andhadrons, integrated over all momenta. Data from 1995 wereused for this plot.

Since Xe is expensive, the chamber gas must berecirculated and purified. Impurities in the gas arekept at acceptable levels by molecular sievesand activated copper (N

2+1%, CO

2+0.25%,

O2+ppm). As a further precaution to keep impu-

rities low, all detectors were He-leak-checked dur-ing construction. This was particularly importantto identify and plug leaks in the crimped pins usedto fix the wires. A sophisticated pressure controlsystem based on an industrial programmed con-troller adjusts the flows into the detector and thegaps in such a way that the differential pressurebetween these volumes is less than a few lbar [33].A pressure stability of 1—2 lbar corresponds toa gain stability of a few percent. The chamberstrack atmospheric pressure in order to avoid usingvery thick foils to contain the gas. However, experi-ence has shown that even during periods of largeatmospheric pressure variations, the differentialpressure between the detector volume and the flushgaps is maintained below 1—2lbar by the controlsystem. The gas system is described in more detailin Ref. [33].

Both positrons and hadrons deposit energy inthe detector due to the ionisation of the chambergas (dE/dx). The most probable energy depositedby a 5GeV pion is about 11 keV. Positrons depositon average approximately twice this amount ofenergy due to transition radiation and the relativis-tic rise in dE/dx. The response to positrons andhadrons in one module is shown in Fig. 25. Purepositron and hadron samples are selected by plac-ing stringent, inefficient conditions on the otherPID detector responses. The distributions are verybroad and it is clear that information from severalmodules must be combined for high quality hadronrejection. In particular the long, high-energy tail inthe hadron distribution overlaps significantly withthe positron distribution. A simple but effectivemethod of analysis is the truncated mean method.In this technique, the largest signal from the sixmodules is discarded and the average of the re-maining five modules is used. This procedure re-duces the mean of the positron distribution but hasa much more significant effect on the hadron tailthat is due to rare events — production of energeticknock-on electrons. Fig. 26 shows the truncatedmean distributions for the HERMES TRD derived

from 1995 data. It is clear that the separation of thehadrons and positrons is greatly enhanced. Thepion rejection factor varies with energy becauseboth the energy loss by charged particles and theproduction of transition radiation depend on theenergy of the incident particle. It is customary whendescribing the performance of a TRD to place a cuton the response of the detector such that 90% ofthe positrons are above the cut. The cut is set forhigher efficiency for physics analysis. Using thetruncated mean method, the energy-averaged PRFis about 150 for 90% positron efficiency. If the dataare analysed as a function of momentum, a PRF of130 is deduced at 5GeV, exceeding the design goal.The PRF is somewhat smaller for lower momenta(80 at 4GeV) and larger for higher energies (up to150).

The PRF can be improved using a probabilitybased analysis similar to the one described in thesection on PID system performance below. A prob-ability function can be defined as

CTRD

"log10

[P%`

/P)],

where P%`

and P)

are the probabilities that a par-ticle was a positron or a hadron respectively. Theresult of such an analysis for 1996 data is shown inFig. 27. A cut for 90% positron efficiency results ina PRF of 1460$130. Even at 95% positron effi-ciency, the PRF remains very good (489$25).


Fig. 26. Truncated mean for six modules for positrons andhadrons, integrated over all momenta. Data from 1995 wereused for this plot.

Fig. 27. The logarithmic likelihood CTRD

, integrated overall momenta, for hadrons (positrons) is plotted as the solidline (dashed line). Data from 1996 were used for this plot.The vertical line indicates the location of a 90% efficiencypositron cut.

6.7. PID system performance

The responses of the four PID detectors arecombined to provide good hadron rejection. Thishas been done in several ways. The simplest tech-nique is to consider the responses individually, de-termine a cut that separates positrons and hadronsin each detector, and define a positron/hadron asa particle identified as such in all detectors. A very

clean sample of a given particle type can be produc-ed in this way but at the cost of efficiency. Typically,this technique is used only for detector studies. Theefficiency of the system can be improved whileretaining good particle type separation by usinga probability based analysis. The responses of thecalorimeter, the pre-shower detector, andthe C[ herenkov detector are combined to producethe quantity PID3 defined as

PID3"log10

[(P%C!-

P%P3%

P%C%3

)/(P)C!-

P)P3%

P)C%3

)],

where Pijis the probability that a particle of type

i produced a given response in detector j. The Pijs

are determined by comparing the detector re-sponses for each track to typical response functionsfor each detector called parent distributions. Thelatter are either derived from the data or modelledusing Monte Carlo techniques. The use of mo-mentum-dependent parent distributions is requiredto account for the varying responses of the de-tectors. PID2 is defined in a similar way to PID3using only the calorimeter and the pre-shower de-tector. PID4 including the TRD is also defined butwill be discussed later.

For the 1995 data, PID3 was used in combina-tion with the TRD truncated mean to make a two-dimensional cut identifying positrons and hadrons.This is shown in Fig. 28 where it is clear thata clean positron sample can be isolated at positivePID3 and large TRD truncated mean (TRD Sig-nal). The advantage of using this technique ratherthan combining all four detectors is that it allowsdetailed systematic studies to be done on the prob-ability analysis [34]. For example, stringent cuts(low efficiency, but high purity) can be applied tothe TRD to define clean samples of positrons andhadrons that can be used to study the PID3 re-sponse and vice versa.

One can incorporate the TRD into the probabil-ity analysis to improve performance even further.The resulting function is labelled PID4. However,one then loses the straightforward systematiccross-checks described above and Monte Carlosimulation must be used to estimate the contamina-tion of each particle sample as well as the detectionefficiency. A graphical comparison of PID2, PID3,and PID4 is shown in Fig. 29. In order to keepas many events as possible for the final analysis,


Fig. 28. Plot of the PID3-TRD plane. Positrons appear atpositive PID3 and large values of ‘TRD Signal’.

Fig. 29. Comparison between the PID parameters PID2, PID3and PID4 for the 1995 data set. Positrons have positive values ofPID2, PID3, and PID4, while hadrons have negative values forthese quantities.

Table 3Positron efficiency and hadron contamination in % for thePID3-TRD method (1995)

x-bin PID3-TRD

e` eff. h` cont.

0.023—0.04 97.77 1.180.04—0.055 98.38 0.810.055—0.075 98.78 0.550.075—0.1 99.21 0.370.1—0.14 99.44 0.220.14—0.2 99.64 0.160.2—0.3 99.71 0.160.3—0.4 99.72 0.160.4—0.6 99.72 0.11

a so-called PID downshifting scheme was imple-mented. If all PID detectors provide valid data,a cut on a two-dimensional PID3-TRD plot isused, which is equivalent to a straight cut on PID4.However, if information from either the TRD or theCherenkov detector is not available a more limitedscheme is used: if the TRD data are not availablePID3 is used, while if the Cherenkov data are notuseful a two-dimensional cut on the PID2-TRDplane is considered. For the analysis of the inclusivedata in 1995, the full PID3-TRD scheme was usedfor over 97% of the useful events.

The hadron contamination of the positronsample was determined in several ways. In the morelimited PID schemes, a value can be derived fromclean particle samples obtained using the detectornot considered in that particular method. However,for the full PID scheme, the hadron contaminationcan only be estimated due to the fact that all de-tectors are used in the analysis and there is nocontrol detector to define clean samples. The differ-ent analysis techniques give consistent results. Themost straightforward of these methods is a simplefit to the total probability distributions using anassumed shape for the tail for each particle sample

in the overlap region, for example a Gaussian or anexponential. The values of the hadron contamina-tion for a given positron efficiency are listed inTable 3 for the x bins used in the analysis of the


Fig. 30. Time of flight vs. momentum for hodoscope H1.

Table 4Properties of NaBi(WO

4)2

crystals

Density q (g/cm3) 7.57Radiation length X

0(cm) 1.03

Moliere radius RM

(cm) 2.38Critical energy E

C(MeV) 9.75

Index of refraction n 2.15Optical transparency (nm) 5380Radiation hardness (Gy) 7]105

1995 data. More details on the particle identifica-tion system can be found in Ref. [34].

Time of flight information from the hodoscopescan be used to separate pions, kaons, protons, anddeuterons at low momentum (up to +2GeV). Thetime of flight is referenced to the accelerator bunchsignal. The TDC values must be corrected for verti-cal position in the scintillators to take into accountthe transmission time of the light. The time resolu-tion is 300 ps. Fig. 30 shows the time of flight asa function of momentum. Protons, deuterons andperhaps some kaons are visible.

Another very effective way to do particle identi-fication is to reconstruct the mass of particles withdecay products in the spectrometer. This has beendone for p0, q, g, K

S, /, x, J/W, and K. More de-

tails on decaying particles can be found in Ref. [35].

7. The luminosity monitor

The luminosity measurement is based on theelastic scattering of beam positrons from target gaselectrons eè~Peè~ (Bhabha scattering) and theannihilation into photon pairs eè~Pcc. Thecross sections are known precisely, including radi-ative corrections, e.g. eè~Peè~c, eè~Pccc[36]. In the future with an electron beam in HERA,the process e~e~Pe~e~ (Moller scattering) will beused [37]. The scattered particles exit the beam

pipe at z"7.2m and are detected in coincidence bytwo small calorimeters with a horizontal accept-ance of 4.6—8.9mrad, which is limited by the size ofthe beam aperture in the magnet shielding plate.For a beam energy of 27.5GeV the symmetric scat-tering angle is 6.1mrad and both scattered particleshave half of the beam energy.

Due to the high radiation background in theregion very near to the beam, the calorimeter con-sists of Cherenkov crystals of NaBi(WO

4)2

[38—40], which have a very high radiation hardnesson the order of 7]105Gy. To further minimiseradiation damage, the calorimeters are movedaway from the beam pipe exit window by &20 cmin the horizontal direction for beam injection anddumping. The properties of NaBi(WO

4)2

are sum-marised in Table 4. The small radiation length andthe small Moliere radius allow a very compactcalorimeter design. Each calorimeter consists of 12crystals of size 22]22]200mm3 in a 3]4 array.The crystals are wrapped in aluminised mylar foiland coupled with optical grease to 1.9 cmHamamatsu R4125Q photomultipliers with a radi-ation hard synthetic silica window and a bialkaliphotocathode. For monitoring the gain of thecounters, each crystal is fed with light pulses froman optical fibre of the HERMES gain monitoringsystem described in Section 6.2. An energy resolu-

tion of p(E)/EK(9.3$0.1)%/J(E) (E in GeV) hasbeen determined for a 3]3 matrix from test beammeasurements with 1—6GeV electrons.

Fig. 31 shows the hit distribution in the calori-meters. Most of the scattered particles hit the innercrystals near the beam pipe. The deposited energyis reduced by lateral shower leakage, especially forhits near the inner crystal edge. A scatter plot of the


Fig. 31. Hit distribution in the calorimeters of the luminositymonitor. The size of the boxes is proportional to the number ofhits per channel. The shaded area shows the beam pipe accept-ance.

Fig. 32. Scatter plot of the deposited energy in the left luminos-ity detector versus the deposited energy in the right detector.The dashed line shows the trigger threshold of 5GeV in bothdetectors for Bhabha events.

energy in the left detector versus the energy inthe right detector is shown in Fig. 32. Most of thebackground events have a high energy depositionin only one of the detectors while Bhabha eventshave a high energy deposition in both detectors.They are separated from background by triggeringon a coincident signal with energy above 5GeV inboth the left and right calorimeter. A Bhabha coin-cidence rate of 132 Hz was measured with this trig-ger scheme for a beam current of 20 mA and a 3Heareal target density of 1]1015nucleons/cm2. Thisprovides a statistical accuracy of the luminositymeasurement of 1%, within about 100 s.

8. The trigger

8.1. Trigger requirements

The function of the trigger system is to distin-guish interesting events from background with highefficiency, and initiate digitisation and readout ofthe detector signals. HERMES requires physicstriggers corresponding to deep-inelastic positronscattering, photoproduction processes (where nopositron is detected) and additional triggers fordetector monitoring and calibration. The triggerhierarchy is potentially capable of four levels. How-ever, for running in 1995—97 only the first-leveltrigger was required.

The first-level trigger decision is made withinabout 400 ns of the event using prompt signals ofthe scintillator hodoscopes, the calorimeter, anda few wire chambers. This can be divided into thetime necessary for signal formation in the detectorsand transportation to the electronics trailer(+250ns including the transit time of the particlefrom the target), and the time needed for a decisionby the trigger electronics (+150ns). The first-leveltrigger initiates digitisation by the readout elec-tronics. The second-level trigger could use any in-formation available within the 50ls during whichtime a hardware clear of the Fastbus readout elec-tronics is possible. The third-level decision could bemade on a timescale of a few hundred microsecondsby existing digital signal processors in Fastbus,which would first read out only that part of thedetector information required for the decision. Infourth level, the full data stream could be filtered bya farm of Intel Pentium Pro processors that canmake a decision on a timescale of 41ms. Only thefirst-level trigger is described below.

The DIS trigger selects positron events by requir-ing hits in the three scintillator hodoscopes, H0,H1, and H2 (only H1 and H2 in 1995) together withsufficient energy deposited in two adjacent columnsof the calorimeter, in coincidence with the acceler-ator bunch signal (HERA Clock). The forward H0scintillator was installed in 1996 because back-ward-going particles from the HERA proton beamare not adequately suppressed by the H1—H2 tim-ing alone. With H0 included, the primary sources ofbackground are hadrons either produced from


positron beam interactions in the upstream col-limators, or photoproduced in the target. The re-quirement of hits in the H0 and H1 hodoscopessuppresses neutral particle background. The pri-mary charged background discrimination is doneusing a calorimeter threshold set above the min-imum ionising energy deposition of 0.8GeV. Thecalorimeter has high efficiency for electromagneticshowers, but relatively low efficiency for hadronicshowers. The hodoscope thresholds were set belowminimum ionizing. A calorimeter threshold settingof 3.5GeV (used for the 1995 running and corre-sponding to y40.85) suppresses the charged had-ronic background rate by a factor of 510 in thetrigger. The typical trigger rate at a luminosity of5]1032nucleons/cm2/s, was about 50Hz in 1995.The threshold was lowered to 1.4GeV during 1996to increase the y range to y40.95 and improve theacceptance for semi-inclusive particles.

The photoproduction trigger detects hadronssuch as K, q, D0, J/w and K0 that are produced atlow Q2 and decay to two or more charged particles.Typically, the scattered positron angle is too smallfor detection. The trigger requires charged particletracks in both the top and bottom detectors, asidentified by the three hodoscopes and the BC1chamber as well as the HERA Clock. The backchamber requirement eliminates those showers ori-ginating in the upstream collimators, which areconfined near the beampipe and hit the tips of thehodoscopes but not the wire chambers which arewell shielded by the magnet steel.

8.2. First-level trigger architecture

The signals needed for the first-level triggers arethe hodoscopes (H0, H1, H2), a multiplicity of twoor more hits in the back wire chambers (BC1), anda cluster of calorimeter blocks with total energyabove threshold. Hits in the top AND bottomhalves of the spectrometer are required for thephotoproduction trigger.

Each hodoscope photomultiplier (PMT) signal ispassively split with one output going to a LeCroy1881M analog to digital convertor (ADC) and theother going to a LeCroy 3420 Constant FractionDisciminator (CFD). The individual CFD outputsare fed to LeCroy 1875A time to digital convertors

(TDC). The high impedance OR outputs of theCFDs for each hodoscope wall are connected to-gether in a chain to form the hodoscope logicsignals.

Each calorimeter wall consists of 42 columns of10 lead-glass blocks each. PMT signals from eachblock in a given 10-block column are split to pro-vide ADC inputs, and added linearly to form aColumn Sum using custom-built NIM modules.Segment Sums of 20 blocks are formed by linearaddition of pairs of adjacent Column Sums, inLeCroy 428F modules. At least one Segment Sumcontains more than 95% of the energy of eachelectromagnetic shower. All combinations of adjac-ent columns are used to fully cover a calorimeterwall. The analog Segment Sum signals are dis-criminated by CFDs to form TDC and logic signals.

The back chamber signals are used only for thephotoproduction trigger. ECL signals are utilisedfrom each of the 256 wires of the top and bottomX-planes of BC1. LeCroy 4564 modules are used toproduce logical ORs of adjacent wires in groups of16. The multiplicity of 16-wire groups that fired isdetermined in a LeCroy 4532 module.

The resulting ECL logic signals are reshaped bya LeCroy 4415A differential discriminator and col-lected into a signal bus. Relative delay adjustmentscan be made with a LeCroy 4518 programmableCAMAC logic delay module. The trigger logic isestablished in LeCroy 4508 programmable logicunits (PLU), which allow changes to be made eas-ily. The signal bus provides the inputs to the PLU,which is strobed with the HERA clock. The PLUECL outputs are converted to NIM, prescaled ap-propriately with CERN prescalers, and broughttogether in the Master Event OR. The MasterEvent is retimed by the HERA Clock and gated bythe data acquisition NOT BUSY signal to producethe Master Trigger that initiates digitisation andreadout of the event.

8.3. Trigger performance

The purity of the deep inelastic scattering (DIS)trigger is good for a high calorimeter thres-hold (3.5GeV) and acceptable for a low threshold(1.4GeV). For a 3.5GeV calorimeter threshold,two-thirds of the triggers had tracks, 95% of


reconstructed tracks came from the target, and onethird had accompanying positrons. For a 1.4GeVcalorimeter threshold, the physics trigger rate in-creased by a factor of about 6 while the fraction ofDIS positrons only went up by about 10%. Theseruns had significant contamination coming fromthe collimator just upstream of the target. Abouttwo-thirds of the events had tracks, but only 70%of the tracks came from the target, with most of theremaining coming from the collimator.

Approximately, one-third of the photoproduc-tion triggers have reconstructable tracks. Of these,over 3/4 have two or more tracks, due to the re-quirement of hits in the lower and upper part of thedetector simultaneously. Only 6—7% have posit-rons, which is not surprising given that the aim ofthe trigger is to detect events at low Q2 where thepositron goes down the beampipe. However, mostof the tracks come from the target region.

During normal polarised running with the highcalorimeter threshold, 15% of the photo-produc-tion triggers were also DIS triggers. Roughly, 2.5photo-production triggers were collected for eachDIS trigger. During polarised running with the lowcalorimeter threshold the overlap increased to 40%due in part to the increased rate in the high-y DIStriggers. The number of photo-production triggersremained the same, since the calorimeter is not inthis trigger.

9. Data acquisition and readout electronics

The backbone of the data aquisition system isconstructed in Fastbus. It consists of 10 front-endcrates, the event collector crate, and the event re-ceiver crate connected to the online workstationcluster via 2 SCSI interfaces. CERN Host Interfa-ces (CHI) act as Fastbus masters. To enhance theirreadout performance they are equipped in mostplaces with Struck Fastbus Readout Engines(FRE), featuring one or two Motorola 96002 DSPs.The event collector in the electronics trailer is con-nected to the event receiver crate in a computerroom via a Fibre Optical Link (STR330/FOL)plugged into the respective CHIs. Pairs of Fastbuscrates, read via only one CHI/FRE, are formed bycluster interconnects (CI), where overall readout

timing considerations allow for it. The drift cham-ber signals are digitised using LeCroy multi-hitmulti-event 16 bit 1877 sparsifying multiblock 96channel TDCs with a resolution of 0.5 ns/channel.Charge from the various photomultiplier tubes andthe TRD is digitized by LeCroy multi-event 64channel 1881M multiblock ADCs. The resolutionis 50 fC/channel and the full-scale range is 13 bitsabove pedestal. During a multiblock transfer from1881M ADCs or 1877 TDCs, the CHI/FRE combi-nation results in a readout time of 127 ns/word(CHI without FRE: 270 ns/word), where the slavetiming is contributing 90 ns/word and the FRE37ns/word. The multiblock passing time (time fortwo neighbouring modules to pass the token) isfound to be 640ns. Reading beyond a cluster inter-connect adds a penalty of about 50 ns/word. Thisresults in a block read speed of more than 30 MB/sin crates having a master and more than 20MB/s incrates beyond a CI. As soon as first data are col-lected into the FRE’s input FIFO, the active DSPstarts processing the incoming data, writing theoutput to its output FIFO. The LeCroy multi-event 10-bit 64 channel 1875A was choosen fordigitizing the time of flight signals. The resolution isset to 25 or 50ps/channel depending on the de-tector. As only six 1875As are read out in total, thelack of multiblock capability on this unit is onlya minor disadvantage. The magnet chamberreadout is instrumented with the LeCroy VMEbased PCOS4 system, consisting of one 2749 andtwelve 2748 modules. The readout time (includingthe transfer to the event collector memory) is450 ns/hit wire. Vertex chamber data arrive fromthe detector as a 16-bit ECL STR330/ECL datastream, one 16-bit word containing the plane andstrip number of one hit strip. Incoming data arepassed in direct mode from the Struck ECL inter-face to the associated DSPs, where hot channels aresuppressed using a linear 16-bit lookup table. Thetrigger pattern is available to all front ends, toallow for selective readout. Event collection is donefrom each FRE output FIFO via a cable segment,connecting directly to the FRE cable port. Doublebuffering is implemented in all essential places toprovide a low front-end deadtime. Event collectionis done in parallel with the next event readout withincoming events processed alternately between two


DSPs in each of the FREs. A typical overall reduc-tion in event size from +18 to 10KB/event isachieved at the DSP level.

In addition to the standard detector readout,there is implemented a variety of asynchronousindependent events, capable of trigger rates exceed-ing 5kHz, including the luminosity monitor andvarious high-speed calibration and monitoringequipment. Depending on the application, the datafrom these events are either collected by the eventcollector during idle time or by dedicated addi-tional CHIs and FREs. They can be defined eitheras so-called user events, or as a set of histogramsread directly from CHI RAM or FRE DSP BRAMwhere they have been accumulated.

One VME branch with four crates and threeCAMAC branches with a total of nine crates areconnected to the event builder crate to handlespecial data acquisition tasks, such as obtaining theelectron bunch number, and slow control. TheVME interface is a CHI/VSB2 module and VDB.The VME crates are connected via a STR723 incombination with a STR725. The CAMAC cratesare controlled by standard CCA2 type crate con-trollers. Scaler information is obtained fromSTR200 32 channel 32-bit Fastbus scalers.

The code for both the 68030 and the 96002 iswritten in assembly language. An online library ofStandard Fastbus, CAMAC and VME proceduresfor jobs that are not time-critical can be called fromFORTRAN or C on the workstation cluster. Theseprocedures are executed in the respective front-endCHIs. All event data and the slow control informa-tion are available for on-line monitoring and analy-sis in the on-line data stream, and are easilyavailable to code in high-level languages.

Data are written to 9 GB staging disks over thecourse of a fill of the storage rings, typically lasting8—12h. The data are copied between fills to storagesilos on the DESY main site via a FDDI link andassociated hardware. In parallel, they are stored onlocal Exabyte tape drives for redundancy. The localsystem was switched to DLT tapes late in 1996.A total of 2 TB was recorded in 1995 and 4.1TB in1996. The deadtime during standard running istypically well below 10% and the downtime due tothe data aquisition system is estimated to be below1%. The possible throughput of 1.5MB/s was de-

termined by the CPU and I/O bandwidth of theevent distributing work station (Alpha 3000X),which corresponds to a 150 Hz event rate at theaverage event size. By replacing the 175MHz 3000Xwith a 266MHz 5/266, this was doubled for 1997.

10. Beam diagnostics and tuning

Since HERMES is a fixed target experiment, thescattered particles are mostly forward and the de-tectors are necessarily in close proximity to thebeamline. This makes the experiment sensitive tobeam-induced background. The two main sourcesof background associated with the positron beamare synchrotron radiation and showers from haloparticles striking the collimators just upstream ofthe target. Several closed orbit corrections (i.e. thataffect the beam only locally) are available to helpminimise background. These are of two types: sym-metric which affect the position of the beam at theinteraction point (IP), and asymmetric which varythe angle at the IP. It was determined to be crucialto tune the beam in the straight section upstream ofthe experiment to minimise synchrotron radiation.In particular, background from this source is ac-ceptable only when a special asymmetric orbit cor-rection well upstream of the IP is optimised. Thedetectors most susceptable to backgrounds are thefront tracking chambers and the TRD, the formerbecause they must operate in front of the magnetand hence in the presence of a large flux of lowenergy particles, the latter because it is particularlysensitive to photons by design. The TRD was madeless vulnerable to synchrotron radiation by addinga thin sheet of lead (;1mm) in front of the firstchamber to absorb some of the low energy incidentphoton flux.

To help optimize background conditions beforethe detectors are turned on, dedicated scintillatorswere installed around the beam pipe. This system iscalled the Tuning Scintillation Telescope (TST).The TST consists of 2 sets of scintillators (8 each)that are located upstream of the front chambers atz"#1.4m and behind the luminosity detector atz"#8.3m. Each set has two layers of 4 scintil-lators surrounding the beampipe to view the beamhalo. The first layer is unshielded and therefore is


sensitive to synchrotron radiation as well ascharged particles. The second layer is shielded with3mm of lead, so the synchrotron radiation is most-ly suppressed. Charged particle halo and synchro-tron radiation can be identified using coincidencesand anticoincidences between the first and secondlayers of scintillators. Beam tuning is then a matterof minimising the counting rates and making themsymmetric. The TST is used primarily during initialbeam tuning when the accelerators are first turnedon at the start of a running period, and at thebeginning of a fill of the storage rings.

Beam position monitors (BPM) on either side ofthe IP allow the beam to be adjusted to a pre-determined position and angle in a short time. TheBPMs are particularly useful at the beginning ofa fill of the storage rings.

The experiment is also exposed to particles thatcome from showers originating from the protonbeam. A set of scintillators was installed around theproton beam pipe just behind the calorimeter tomonitor this background. Background from thissource varies significantly from fill to fill and evenduring a fill. However, essentially none of it iscorrelated with the events of interest so its maineffect is to increase the trigger rate and decrease theproportion of good events going to tape. Asoutlined previously, the addition of the fronttrigger hodoscope (H0) has significantly improvedthe situation.

11. Summary

This paper describes the spectrometer built tomeasure the products of deep inelastic scattering ofpolarised positrons from polarised internal gas tar-gets at HERA. After a relatively short commission-ing phase, the experiment started recording data in1995. All essential detectors have performed wellwith only a few component failures over the firsttwo and a half years of running. The acceptance ofthe spectrometer allows the detection of hadrons incoincidence with the scattered positron. These im-portant semi-inclusive measurements with the iden-tification of pions and kaons are unique toHERMES and will allow the determination of thespin-dependent quark distribution functions.

Acknowledgements

The HERMES collaboration would like tothank all those who participated in the design andconstruction of the spectrometer. We are parti-cularly indebted to the MEA staff at DESY, manymore of whom contributed than are mentioned inthe author list. We also gratefully acknowledge theDESY management for its support.

This project has been supported by the FWO-Flanders, Belgium; the Natural Sciences andEngineering Council of Canada; the Dutch Foun-dation for Fundamenteel Onderzoek der Materie(FOM); the German Bundesministerium fur Bi-ldung, Wissenschaft, Forschung und Technologie;the German Academic Exchange Service (DAAD);the Italian Istituto Nazionale di Fisica Nucleare(INFN); Monbusho International ScientificResearch Program, JSPS, and Toray Science Foun-dation of Japan; the United Kingdom ParticlePhysics and Astronomy Research Council; the U.S.Department of Energy and National Science Foun-dation; and INTAS, HCM, and TMR contribu-tions from the European Community.

References

[1] D. De Schepper et al., Nucl. Instr. and Meth., in press.[2] A. Golendoukhin et al.; B. Braun et al., in: C.W. de Jager et

al. (Eds.), Contributions to the 12th InternationalSymposium on High Energy Spin Physics, Amsterdam,September 96, World Scientific, pp. 241, 331; J. Stewart, B.Braun, Contributions to the 7th International Workshopon Polarised Gas Targets and Polarised Beams,Urbana-Champaign, August 97, AIP Conf. Proc., to bepublished.

[3] MAFIA Userguide, Version 3.2, Computer SimulationTechnology, Lauteschlogerst. 38, 64289 Darmstadt, Ger-many; Tel.: #49 6151 717057; M. Bartsch et al., Comput.Phys. Commun. 72 (1992) 22.

[4] TOSCA Computer Program. Vector Fields Inc., 24 Bank-side, Kidlington, Oxford OX5 1JE, England, Tel.: (0865)370151.

[5] M. Geijsberts et al., Nucl. Instr. and Meth. A 313 (1992)377.

[6] J. van der Marel, Ph.D. Thesis, Delft University, 1997.[7] D-263 Glass is produced by Desag, Germany.[8] J.M. Butler et al., Nucl. Instr. and Meth. A 290 (1990) 122.[9] A. Trudel, private communication; U. Becker et al., private

communication.[10] S.F. Biagi, Nucl. Instr. and Meth. A 310 (1991) 133.


3The theses mentioned in these references are available asHERMES internal notes, and can be accessed on the HERMESweb pages: http://www-hermes.desy.de/notes/

[11] R. Openshaw, R. Henderson, W. Faszer, M. Salomon,Nucl. Instr. and Meth. A 307 (1991) 298.

[12] F. Schmidt, Diploma Thesis, Friedrich Alexander Univer-sitat Erlangen-Nurnberg, May 1996; HERMES internalnote 96-043.3

[13] S. Bernreuther et al., DESY 98-011, hep-ex/9803005, Nucl.Instr. and Meth., submitted.

[14] S. Bernreuther, Diploma Thesis, Friedrich AlexanderUniversitat Erlangen-Nurnberg, 1995; HERMES internalnote 95-068.3

[15] F. Neunreither, Ph.D. Thesis, Friedrich Alexander Univer-sitat Erlangen-Nurnberg, 1997.

[16] R. Veenhof et al., Program GARFIELD, a drift chambersimulation program; CERN Program Library, LongWriteup, W5050, 1991.

[17] S. Bernreuther et al., Nucl. Instr. and Meth. A 367 (1995)96.

[18] D. Hasch, Diploma Thesis, Humboldt Universitat zu Be-rlin, 1995; HERMES internal note 95-028.3

[19] J.C. Armitage et al., Nucl. Instr. and Meth. A 271 (1988)588.

[20] LeCroy Research Instrumentation Catalog, 1996.[21] E. Garber, Diploma Thesis, Humboldt Universitat zu Be-

rlin, November 1996; HERMES internal note 96-062.3[22] T.-A. Shibata et al., Nucl. Instr. and Meth. A 411 (1998)

79—84.[23] W. Wander, Ph.D. Thesis, Friedrich Alexander Universitat

Erlangen-Nurnberg, April 1996 (English translation:HERMES internal note 97-0313).

[24] GEANT Detector Description and Simulation Tool,CERN Program Library, Long Writeup, W5013, 1994.

[25] Particle Data Group, Phys. Rev. D 54 (1996) 1.[26] The F101 glass blocks were provided by the Lytcarino

factory, 140061 Lytcarino, Russia.

[27] M. Kobayashi et al., KEK Internal Report 93-178, 1993.[28] H. Avakian et al., Nucl. Instr. and Meth. A 378 (1996) 13;

A. Fantoni, Ph.D. Thesis, Universita della Calabria, Feb-ruary 1996; HERMES internal note 96-015.3

[29] H. Avakian et al., Nucl. Instr. and Meth., in press.[30] P. Baillon et al., Nucl. Instr. and Meth. 126 (1975) 13.[31] J.V. Emerson, M.Sc. Thesis, Simon Fraser University, July

1995; HERMES internal note 96-038.3[32] The fibers were provided by Carl Freudenberg Faservlies-

stoffe, P.O. Box 100363, Weinheim 69465. The commercialdesignation of the material is VILEDON C 1900/034.

[33] D.M. Thiessen, M.Sc. Thesis, Simon Fraser University,March 1996; HERMES internal note 96-028.3

[34] R.B. Kaiser, Ph.D. Thesis, Simon Fraser University,April 1997; HERMES internal note 97-0173; R.B. Kaiser(for the HERMES PID Group), HERMES internal note97-025.3

[35] K. Ackerstaff, in: G. D’Agostini, A. Nigro (Eds.), Contribu-tion to DIS96, Rome, World Scientific, p. 526; Cross-sections for vector meson production at HERMES, April1996, in preparation.

[36] F.A. Berends, A. Bohm, in: A. Ali, P. Soding (Eds.), HighEnergy Electron—Positron Physics, World Scientific, Sin-gapore, 1988, p. 27.

[37] Th. Benisch, Diploma Thesis, Friedrich Alexander Univer-sitat Erlangen-Nurnberg, March 1994; HERMES internalnote 94-030;3 Ph.D. Thesis, Friedrich Alexander Univer-sitat Erlangen-Nurnberg, March 1998.

[38] G.I. Britvich et al., Nucl. Instr. and Meth. A 321 (1992) 64.[39] A.V. Antipov et al., Nucl. Instr. and Meth. A 327 (1993)

346.[40] V. Samsonov, in: P. Lecoq et al. (Eds.), Proc. Crystal 2000

International Workshop, Chamonix, France, 1992, Edi-tions Frontieres, 1993, p. 383.


The HERMES Spectrometer - math.uwaterloo.cajemerson/papers/... · 0W.K.Kellogg Radiation Lab,...

Documents

Transcript of The HERMES Spectrometer - math.uwaterloo.cajemerson/papers/... · 0W.K.Kellogg Radiation Lab,...