The ZEUS Trigger Environment

Chapter 4

The ZEUS Trigger Environment

4.1 Introduction

Triggering is the selection of physics events of interest in conjunction with the rejection of background processes which it is not desired to investigate. The success of any HEP experiment is critically dependent on its ability to achieve a high trigger efficiency.

Identification of interesting physics must be as close to perfect as possible in order to avoid the introduction of unacceptable systematic errors and to maximize the amount of recorded data relating to physics events. A trigger is a complex entity comprising, at HERA, readout electronics, hardwired algorithms and much sophisticated software running on powerful dedicated processors well matched to particular tasks.

At HERA, triggering has assumed even greater importance than in the past partly due to the high rates of background and partly due to the short beam-crossing interval of only 96 ns. At other machines, a longer interval simplifies trigger design so that no pipe-lining of data is required. For example, the Large Electron-Positron collider at CERN has a crossing every 10 μs. Experience gained at HERA will prove invaluable in the design of the yet more complex triggers which will be required at the next generation of colliders, notably the Superconducting Supercollider in Texas which will have a beam crossing interval of 16 ns.

4.1.1 Overview of Dataflow

An overview of the system is shown in figure 4.1.

Figure 4.1: Flow of data through the DAQ system.

Each component feeds data to local pipelines and the global first level trigger (GFLT). If this decides to accept the event, the pipelines are read out to the second level trigger (SLT): raw data from each component remains separate at this stage though the global second level trigger (GSLT) can clearly consider the results from processing in all components. If the GSLT issues an accept, the event builder (EVB) assembles the whole of the information acquired for the event and sends it to the third level trigger (TLT). This, mediated by the Central Data Acquisition (CDAQ) VAX and run control (RC), writes events passing the final stage to tape. There is also some facility for local disk storage.

4.2 Rates and Background

The trigger philosophy has been developed with the characteristics of physics events in mind. These in general have many tracks coming from the origin which will be observed in the tracking detectors. Large depositions of energy, especially at high angles to the beam-line, often result from physics processes. If there is a neutrino in the final state, this will not be seen by any part of ZEUS and so and asymmetry in transverse energy may be found.

It is envisaged that there will be three main sources of background in ZEUS; cosmic rays passing through the detector, losses from the proton beam, and interactions of the beam with residual gas inside the beam-pipe. The latter are known as beam-gas interactions and much effort has been expended to try and devise triggering strategies to prevent them from causing triggers.

Cosmic ray events will on the whole be rejected by the use of timing information from the calorimeters.

Protons not following the nominal beam trajectory hit machine elements thus producing hadronic showers including pions. These can subsequently decay into muons which are very penetrating. There are approximately 2 x1013 protons in the beam. The circumference of the ring is 6,336 m. If it is assumed that the beam will have a lifetime of about ten hours, these interactions may occur at a rate of up to 100 kHz per meter of beam-line. However, structural elements such as collimators, beam scrapers and the veto-wall will substantially reduce the rate of these events causing activity in the detector.

It is estimated that the rate of beam-gas interactions will be up to 2 kHz per meter.[23] Near the interaction region they will fake good events in the tracking trigger. Upstream of it, their modest energy deposition may be misinterpreted by the calorimeter as a high transverse energy deposition representing a large Q2 interaction. This is because upstream tracks which are in fact only leaving the beam-pipe by a shallow angle can arrive a long way away from it once they are intercepted at the interaction region.

Also, upstream beam-gas can have secondary interactions producing tracks which come from the interaction region. This background is potentially the most serious. All of these backgrounds can be reduced by combining together triggers from different detector components. These have different discrimination powers against the various types of background and by combining them in a flexible way the sensitivity to physics can be maintained while minimizing the background. This is discussed in the next section.

4.3 The Trigger

Assuming the design luminosity of 1.5 x 1031 cm2 s-1 leads to the rates shown in table 4.1 for events observed in the acceptance of the ZEUS detector.

Table 4.1: Rates of physics and background.

As mentioned above, the ZEUS trigger will have three levels. In order to allow more sophisticated processing on a more complete subset of component data at successive levels, each level will have a longer period of time with which to make a decision. This is shown in table 4.2.

Table 4.2: Processing time allowed per event by level of trigger.

4.3.1 The First Level Trigger

It is impossible to decide whether or not to accept an event within the 96 ns between beam crossings. In the first level trigger (FLT),[24], [25], [26] this forces the storage of data in pipelines which must be able to hold data relating to 5 μs.

Processing takes place both at the level of individual components and in the Global First Level Trigger Box or GFLTB.[27], [28], [29], [30], [31], [32] Because of these constraints the sophistication of processing that may be done by components at this level is restricted.

The output rate from the FLT will be 1 kHz, after the fast clear (section The components must write the data relating to the event to their internal pipeline.

Each of the components have 26 beam crossings to perform calculations on their data. They must then send the results of these calculations to the GFLTB. If the GFLTB decides to accept the event, it will send an accept bit to each component exactly 20 beam crossings later. The GFLTB must therefore complete its calculations within this 20 beam crossing period. The components then read out the relevant data to the component second level trigger.

The tracking detector FLT is is central to the work presented in this thesis. Its discussion is therefore postponed to the following chapter. Calorimeter FLT

The calorimeter first level trigger (CALFLT)[32], [33], [34] is designed to detect isolated electrons and muons and to measure momenta and energy deposition. It is essential to use angular information in this trigger. The distinction between transverse energy and momentum in the CALFLT is an important one. Momentum is a vector quantity whereas energy is a scalar. The difference between energy and momentum is expressed in the statement that momentum is signed so that tracks in a opposite hemispheres might sum to give zero transverse momentum whereas energy would always add.

Transverse energy is a calculated value in which depositions at high angles are accorded more weight. This quantity is a good measure of activity in the detector characteristic of desired physics events. Look-up tables (LUTs) are used to consider transverse energy deposition in order to recognize patterns associated with good events.

The original intention to measure longitudinal momentum will not now be fulfilled due to financial reasons. The calorimeter is mostly non-projective: only the electromagnetic section of the barrel has cells aligned parallel to lines radiating from the interaction point. For this reason, the subdivision of the calorimeter into regions for trigger purposes is different to its physical division. Entities known as ‘trigger towers’ are formed from calorimeter cells such that a straight line from the interaction point will be fully contained within them.

Most towers contain two electromagnetic calorimeter (EMC) cells representing approximately 25 radiation lengths as was shown in table 2.1. Beyond that are the two pairs of HACs (hadronic cells) which map most closely on to the EMCs. In a small number of towers at the edges of the FCAL and of the RCAL, the BCAL is between the first cell in the tower and the interactions region. In this case, the tower contains only HACs (see section 2.2.1). The makeup and number of towers in the calorimeter is shown in table 4.3.

Table 4.3: Calorimeter tower numbers and makeup by location.

It can be seen from the table that there are a total of 1,360 towers: these all provide a HAC sum. Of this total, 974 also provide an EMC sum. The non-projective cells must be grouped into projective towers. This is done by using the EMC sections to define tower geometry and then assigning HACs behind them to form a tower with the best possible match. It transpires that 896 projective towers with a sensible geometric division emerge from this process. The calorimeter is now divided into sixteen trigger regions: four for each of the RCAL and FCAL and eight for the barrel. This is shown in figure 4.2. Each region contains 7 x 8 towers.

Figure 4.2: Trigger regions in the calorimeter.

Each calorimeter cell is read out by two photomultiplier tubes. EMC and HAC energy depositions are summed within a tower by on-board cards known as trigger sum cards (TSCs). These sums are sent to trigger encoder cards (TECs) in the rucksack: each TEC covers four towers. So there are 14 TECs in a crate to cover the 56 towers in each of the sixteen trigger regions.

For each tower, EMC and HAC energy deposition is measured on two digitization scales by flash analogue-to-digital converters (FADC): high gain (12.5 GeV on an 8 bit scale) and low gain (400 GeV over 8 bits in the FCAL and 100 GeV over 8 bits in the RCAL and BCAL). If the deposition exceeds a scale an overflow bit is set. If neither the HAC nor the EMC in a tower set off the high-gain channel, the TEC ceases to perform energy sum calculations and begins testing for electrons and minimum ionizing particles as described later.

The geometric position of each tower in terms of θ and φ is known to the TEC. It uses these and the finest resolution energy scale available (depending on whether the high or low gain channel has been used) to find transverse energy depositions. Total and transverse energy sums for the four towers covered by each TEC are sent to a trigger adder card (TAC). There are two of these in a crate.

The TEC’s run test procedures may result in three bits being set for each tower. An E-bit is set if the depositions found are characteristic of an isolated electron: these will predominantly deposit their energy in the EMC part of a tower. The design aims to find all electrons with energy greater than 5 GeV.

The EMC threshold is set at 2.5 GeV however since an electron may deposit its energy in adjacent cells. Since there is a small likelihood that an electron with energy between 2.5 GeV and 5.0 GeV will ‘punch through’ the EMCs to reach the HAC layer, only 0.1 GeV is allowed in the HAC layer. If the EMC deposition is greater than 5.0 GeV, then the ratio EEMC to EHAC must be greater than 10. A slightly different requirement is implemented in the more active FCAL but clearly the requirement for this bit to be set is also based on substantial symmetry between the two types of cell.

The rate at which charged particles passing through matter lose energy by ionization depends on their energy. In fact, the rate decreases to a minimum and then increases to a plateau at high energy. Particles above the minimum are called minimum ionizing particles or MIPs. The energy deposited by a particle at the minimum in a tower is shown in table 4.4.

Table 4.4: Total HAC and EMC energy deposited by a MIP by location of tower.

If a situation not unlike the reverse of what is necessary to set the E-bit occurs, then an M-bit (M is for minimum ionization particle) is set. It is required that the deposition E fulfills the condition 0.2 EMIP ≤ E ≤ 2 EMIP. It is generally likely that a muon is the cause. Muons are comparatively penetrating and so do not deposit most of their energy in the EMCs. Genuine hadrons will usually have energies which are much too large to set the M-bit.

Towers in the active region around the beam-pipe are not permitted to set E or M bits. If insufficient energy is deposited to set either of these bits, LUTs are used to find if the tower is ‘low-activity’ for the Q-bit. ‘Q’ stands for ‘quiet’. In fact, the requirement to set the Q-bit is that the pulse height be less than 20% of the pulse height required to set the M-bit.

In the TACs, pattern logic searches for groups of up to four E or M-bits set and surrounded by Q-bits in each of the sixteen regions. NC events have a high-energy isolated electron and this pattern logic forms an excellent trigger on these events. On the other hand, isolated muons are characteristic of many interesting physics processes including heavy quark production.

The exact thresholds for these bits vary depending on the location of the tower being processed. The thresholds for the E, Q and M-bits must be matched to each other because otherwise a legitimate electron may fail its isolation requirement. Therefore a quiet tower is defined by having less than the minimum EMC energy for an E-bit and the minimum HAC energy for an M-bit. For example, in the FCAL a quiet tower must have EEMC < 2.5 GeV and EHAC < 2.268 GeV. These bits are sent to the CALFLT processor.

The CALFLT processor receives the energy sums for the sixteen regions and also on a finer sub-region scale. This finer scale is designed to have better resolution around the beam-pipe and to prevent loss of the flexibility to examine data relating to areas covered by more than one trigger crate. The CALFLT processor will be able to examine in this way deposition in the FCAL and the RCAL in annular regions at different radii from the beam-pipe. This is useful because beam-gas events are more likely to have high deposition around the beam-pipe region than physics events for Q2 values of interest to ZEUS.

Sums are made of the number of towers in each region which have energy sufficient to set the bits. This enables the processor to search for jets which will appear as clusters of towers with bits set.

The processor sends data to the GFLTB relating to the whole calorimeter and to the 16 sub-regions. The global data is: EEMC, EEMC + EHAC, Ex, EMC + Exm HAC , Ey, EMC + Ey, HAC, Ex, EMC, Ey, EMC , missing energy, cluster data and the total number of E and M-bits set.

Further, the result of a beam-gas likelihood algorithm is sent. [This uses the regional energy sums and also the sum of energy in the beam-pipe region (because beam-gas events cause much activity here). Also the presence of towers showing kinematically disallowed energies is a useful sign. Because they have upstream vertices, beam-gas events can fake larger transverse energy than would be possible for any real event.] On the sub-region scale, the M and Q-bitmaps are sent to the GFLTB along with Etot , Etrans, Eemc , Ex and Ey. Fast Clear

To ensure that the accept rate to the second level trigger is no greater than 1 kHz an element of parallel processing of calorimeter data has been introduced[35]. The fast clear (FC) will consider data simultaneously from the FCAL, RCAL and BCAL relating to events which have had an FLT issued. Each accept is accompanied by an indication of whether the GFLTB will permit the FC to abort the event, based on the strength of its acceptance by components other than the calorimeter.

The FC works by searching for clusters[36] and finding their angle and energy. Cuts are made to discriminate against beam-gas which can be quite stringent compared to those in the FLT because the FC will be permitted to abort a trigger only if the other components show a weak accept decision.

An important quantity in the FC is shown in equation 4.1.

In CC events, hadron jet do not often have trajectories which take them through the RCAL. On the other hand, about 80% of particles in beam-gas interactions are hadronic. So physics events have high-angle clusters with large Ef. It has been shown that a cut based on this ratio for the highest energy cluster in the RCAL yields a rejection factor of 400[37]. This clearly indicates that efficient recognition of electron or hadron jets is possible.

If an abort does occur, the GFLTB stops component readout to the second level. The FC operates in around 10 to 30 μs. This is longer than the 5 μs available to the GFLTB because the FC does not have to consider every event. In this way, more detailed considerations of clusters are possible thus enhancing efficiency while the design goal of 1 kHz input to the GSLT is not compromised. In fact, the exact amount of time available depends on the FLT rate but flexibility has been inbuilt here by simply declaring that the FC will cease incomplete operations on an event and allow it to proceed through the readout chain as soon as the next FLT decision is issued. Other FLT Components

Forward muon detectors

A muon trigger[38] will be formed taking account of direction and momentum by requiring a strip-to-strip coincidence between first and last planes of streamer tubes (LT1 and LT5, see section 2.3.1). The susceptibility of this method to background is reduced by additionally requiring signals in corresponding φ-sectors of all five planes LT1 to LT5, as shown in figure 4.3.

Figure 4.3: Forward muon detector first level trigger.
Figure 4.4: Barrel muon detector first level trigger.

The time-of-flight plane will assist the association of a triggered muon with its correct beam-crossing. The FMUFLT will have three subdivisions in terms of polar angle as shown in table 4.5.

Table 4.5: FMUFLT polar angle subdivision.

Correlation matrices select tracks consistent with having originated in the interaction region. This is done by logically dividing the readout channels into θ and φ windows as shown in figure 4.4 for the BMUO. The RMUO is covered by four sections.

Veto-wall signals indicating the passage of a muon from the beam halo inhibit triggers in the RMUO while CTD timing data reduce the cosmic background in the barrel to a manageable level. Coarse scale muon multiplicities are sent to the GFLTB: these give the number of muons found in left or right barrel and rear regions.

Veto-wall By virtue of its presence, this device (section 2.4.1) will reduce rates from beam-gas and beam-halo[39]. Apart from the veto signal to the RMUO described above, it is instrumented to set three flags. These will indicate to the GFLTB the presence of signal in the inner and outer scintillator planes and sum up such activity to produce a multiplicity.

Luminosity monitor The LUMI[40] continuously scans the energies of photons registered in its photon detector and of electrons in its other sub-component. The energies measured are sent to the GFLTB. The LUMIFLT raises a flag if the arrival times and the sum of the two energies are consistent with a bremsstrahlung event in the interaction region: Ee + Eγ = Ebeam. A photo-production flag indicates Ee in a proper window and photon energy below a threshold (in practice no deposit).

Leading proton spectrometer Horizontal and vertical position measurements will be made for the FLT in the last three stations (section 2.4.3). Coordinates in the three planes are linearly related for straight tracks from the interaction region (figure 4.5). Selection logic searches for valid spatial coincidences. Beam halo events however are expected to produce a rate of 3 kHz so this will not be a standalone trigger. By combination with an independent trigger the GFLTB will use the hit pattern from the LPS to obtain an additional background rejection factor.

Figure 4.5: LPS input to FLT: proton search. Global First Level Trigger Box

The GFLTB collates data from all participating components and performs calculations to make the final decision at this level on whether or not to accept an event. It also has test and calibration functions. It will send data to RC enabling online investigation of dead-time, luminosity etc.

To make an event decision, it performs logical operations of great complexity. These have been designed to a high level of sophistication in advance of data taking at ZEUS, but flexibility exists to make adjustments because it is certain that reality will differ to some extent from simulation.

The information from the components comes on 16-bit cables. Fifty-one 16-bit words of data arrive for every beam-crossing. This information is fed into a set of sub-triggers. The hardware allows for 64 such sub-triggers to be defined, all of which must eventually be combined into a single decision. The sub-triggers are grouped into several functional classes dealing with similar data as shown in figure 4.6.

Figure 4.6: Schematic of logic in the GFLTB.

An example of the kind of cross-matching possible in the global box may be seen from the diagram: isolated muons found in the calorimeter are correlated with tracks from the tracking detectors which may plausibly have been produced there by the same muon. Further, transverse energy from the calorimeter is multiplied by the track count from the tracking detectors: this quantity should be large for good events.

4.3.2 The Second Level Trigger

The SLT has access to a more complete and precise set of data than the FLT by virtue of the longer timescale on which it operates. It is currently envisaged[41] that the Global Second Level Trigger (GSLT) box[42], [43] will make an event decision available to components around 7 ms after the beam crossing. Unlike the FLT, the SLT is asynchronous: different parts of the system are at any given moment analyzing data which was not all acquired at the same time. Tracking Detector SLT

The algorithm for the CTDSLT[44], [45], [46], [47], [48] proceeds in two stages: segment finding[49] and track finding[50]. Segment finding is the grouping of hits in an eight wire cell to produce small portions of tracks: these are then combined to form a complete track. The pulse heights from the DSPs (section will enable electron tracks to be identified when the events are fully reconstructed because their characteristic dE/dx differs from that of other charged particles.

Drift times are the input to the CTDSLT which resides on a network of transputers. These are microprocessors with four bidirectional communication channels which mean that a wide range of topologies are available. They have their own language (Occam[51]) which is designed to fully exploit the inherent parallelism of the networks.

For applications in the CTDSLT, factors of four improvements in time requirements have been measured using Occam[52] as compared to more conventional languages.

In axial SLs only, hits in each cell are examined to find track segments. Each cell is considered in turn, and the ‘single cell mask’ is stepped around the whole chamber.

‘Roads’ are defined so that the drift time at the nest wire is predicted from the previous hit on a segment. The gradient, intercept, variance and the mean z and r coordinates are passed on to the track finding stage.

The track finding sorts segments in overlapping octants making use of their angular values to consider groups likely to be on the same track. Three-dimensional tracks are formed from z-by-timing data associated with r segments via a straight line fit in rz. The CTDSLT will send two tables of results to the GSLT. Exit point and direction and pt will be available with error estimates for each track that has been found.

Also the charge and origin will be known. The vertex for the event as a whole is calculated, as is the total number of tracks found together with an estimate of how many tracks were missed (from the number of unused segments).

The present design of the FTDSLT envisages a tree search method which will be implemented in online memory. It will identify coordinate outputs from the chamber corresponding to straight tracks from the interaction region. It will require one cell hit in each layer: this corresponds to a polar angle requirement of 7° < theta < 30°. The FTDSLT should find all such tracks with momentum over 1 GeV/c coming from within 20 cm of the vertex. Calorimeter SLT

As is common in the SLT as a whole, transputer networks are used for readout and triggering[53].

Timing of energy deposition in the calorimeter is very precisely measured at the second level. Because the distance from the interaction region is not the same for the FCAL and the RCAL there will be a 2 ns difference in arrival times for good physics events. More importantly, most beam-gas events originate from upstream of the interaction point at negative z-coordinates. These are expected to produce a difference in arrival times of 12 ns[54]. This permits discrimination between physics and background. Prior to this enhancement of capability, the design called for those calorimeter towers around the active beam-pipe region to be disbarred from setting isolated electron bits because of the intolerable leakage rate that would result. With this timing information however it appears that this restriction may be relaxed thus improving efficiency. In addition, events with unphysical longitudinal momentum will be vetoed. Other SLT Components

Other components are in communication with the GSLTB. It is clearly to be expected that the quality and quantity of information available at the second level will in general be superior to that at the FLT.

GFLTB The GFLTB sends the results of its calculations to the GSLTB along with component data and the FC information.

BAC Eight-bit 10 MHz FADCs sum charges over two successive beam crossings. Two networks of transputers will be used: one will be in communication with the GSLT and the other with the EVB[55]. If an energy threshold is met, cluster data will be sent to the GSLT. Also, a muon trigger is formed from coincidence logic in the bottom yoke where there are no muon chambers. The data should in general be more precise than that from the BACFLT.

BMUON Coordinates of found muons should be available.

FMUON The FMUSLT will make an estimation of momentum from the sagitta of the particle found at the first level using the LT planes.

LPS A bit will be sent to the GSTLB to confirm or negate the LPSFLT. Further, a measurement of the proton momentum is made and is expressed as a fraction of the beam momentum. Horizontal and vertical projections of the transverse momentum of the proton are supplied.

LUMI The measurements made at the first level remain available. Further, the location of electromagnetic shower centers is measured and also photon shower centers if the bremsstrahlung flag is up.

VETO The number of hits and their positions should be available.

4.3.3 The Third Level Trigger

The composition of the code to be run at this level[56] has proved to be quite volatile. This is due to two factors. Firstly, there is great uncertainty as to the form of events, both background and physics, which will survive the first two levels of trigger. Because of the high efficiency with which these reject beam-gas, the entire Monte Carlo production which has taken place so far has resulted in only around 350 events (from a generated sample of 750,000). This is clearly not a large enough sample to base definitive conclusions on.

Secondly, the compute power available within resources at this level has fluctuated. The system is now based on Silicon Graphics (SGI) processors, the specification of which have quite considerably improved over time. Initially, it was thought that code would need to be written specially for this application: later it seemed that it would be possible to run the full offline reconstruction code online! At present, there has been some retreat from this bold position so that now it is envisaged that there will be some form of vertex rejection to eliminate remaining beam-gas before running the ‘offline’ code. Possibilities for this include pattern recognition or a helix fit to find tracks and look for upstream vertices. Use of CTD stereo layers will permit a better z-resolution so as to improve on the SLT. Also, the VXD data becomes available here for the first time. It is likely that most of the remaining beam-gas will have a vertex very close to the interaction region so improvement of the resolution here is the critical factor.

The TLT[57], [58], [59] is intimately linked to the Event builder (EVB). The EVB resides on transputer-based standard ‘VME’ boards similar to those of the GSLT and CALSLT.

The EVB assembles the events and makes them available in a Triple Port Memory. The TLT then reads events into SGI workstation memory. It is hoped that the SGI workstations will provide an online event display. These will also read out the EVB to IBM and VAX computers. The IBM will write events to tape with a rate of 0.5 Mb s-1. The speed with which this is possible is the final constraint on the whole of the trigger. Over 100 Gb will be accumulated during a year of operation.

The amount of processing time is a function of the input rate and the number of TLT nodes. It is currently envisaged that there will be 32 4D/35 processor nodes and 6 4D/25 machines handling communications. The time allowed per event is given by dividing the number of nodes by the number of events which must be handled every second: 32/100 = 0.32 s per event.

The VAX is the main Central Data Acquisition computer (CDAQ) and represents the interface between the experiment and operators in the control room. Many interlinked processes will run on the CDAQ VAX. Run control will start TLT processes and setup runs without human intervention. For monitoring purposes, RC will be connected to components down optical transputer links running from the rucksack to the control room. These will carry ‘analyse’, ‘reset’ and ‘error’ signals. Slow control will monitor parameters not varying on the timescales of beam crossings, such as temperatures in the racks housing readout electronics, cooling fan status and gas flow rates.

Next Chapter: