The 35-ton prototype

DUNE logoIntroduction

Although there are several single-phase LAr TPCs in operation (ICARUS, MicroBooNE, ArgoNeuT) or under construction (SBND), the current design for the DUNE far detector includes various specific features that have not been field-tested.  The 35-ton prototype was designed to incorporate these features so that they could be tested in a realistic environment before the construction of ProtoDUNE-SP.  The 35-ton working group lists these as "the FR4 field cage construction, the gaps between modules, the wrapped wire planes, the light guide style photon detection system, cold ADC ASIC in the readout electronics, and the development of triggerless DAQ operation."  It was also the first LAr TPC for neutrino physics to use a membrane cryostat—a system designed for bulk shipment of liquid natural gas, in which the support structure is warm and the cryogenic liner is not load-bearing.  Some of these, such as the segmented anode plane assemblies and the wrapped wire planes, are also being implemented in SBND and ProtoDUNE-SP.

Construction and operation of the 35-ton prototype

Schematic of the 35-ton prototypeA schematic diagram of the 35-ton prototype is shown in the figure.  Like the reference DUNE single phase design, but unlike SBND and ProtoDUNE-SP, it has a double-sided anode plane assembly (APA) within the liquid argon volume.  The DUNE design is symmetrical, but this proof-of-principle prototype has the APA close to one end, with the main drift region on one side and a nominal 20 cm drift region on the other.

As described in the introduction to LAr TPCs, 3D track reconstruction is achieved by using the electron drift time for the third coordinate.  The reference time, corresponding to the time at which the particle crossed the detector volume, is determined by reading out the scintillation light from the liquid argon, since the light travel time is negligible compared to the electron drift time.  Liquid argon is a good scintillator, but it emits in the far ultraviolet (~128 nm), which is not optimal for most photodetectors.  Consequently, the photon detection is done using lightguides coated with a wavelength shifter which absorbs the UV light and re-emits in the visible.  This secondary emission is then detected by "silicon photomultipliers" (SiPMs)—a form of avalanche photodiode.  The wavelength shifting lightguides were installed between the wire planes of the APA.

The 35-ton prototype operated in two phases: a test phase from December 2013 to February 2014, which was a test of the cryostat and purification system with no detector installed, and a fully instrumented run in March 2016 (cooldown and filling December 2015 – January 2016; detector commissioning February 2016; data run started March 7 2016).  Unfortunately, the latter phase came to an abrupt end on March 19 when a pipe broke as a result of vibration in the pump to which it was connected, resulting in contamination of the liquid argon with air.  Including data taken during commissioning, this yielded a total of 22 days of good-quality data for analysis.  The data consist of through-going cosmics triggered using the cosmic ray counters surrounding the detector.

The cold camera

Cold camera for the DUNE 35-ton prototype

One of Sheffield's contributions to the 35-ton prototype was a camera system designed to operate in the liquid argon volume and monitor the detector during and after cooldown and filling.  The components of the camera are shown in the photo, both separately and assembled in the housing.  A total of 8 cameras were installed to monitor various parts of the detector.

The principal purpose of the camera system was to observe the interior of the detector during cooldown, filling and HV switch-on, with the aim of identifying problems such as breakdown in the high-voltage system. This was successfully accomplished: the picture is a still from a movie showing the liquid argon filling.  It proved possible to power cycle the cameras in the cold: in particular, after a 9-day power outage they all came back on without problems.

Still from video of LAr filling in 35-tonThe cameras used were not, of course, designed to operate at LAr temperatures, and it was in fact necessary to provide a small amount of local heating, using resistors as labelled in the photo.  This did not cause any problems with detector operation (in particular, worries that they might cause the formation of bubbles of gaseous argon proved unfounded).  The images did tend to degrade with longer time spent in the cold, but this is a relatively minor disadvantage: the main value of the cameras is to help diagnose any problems that become apparent at or shortly after the start of detector operation.

Event reconstruction

Data from the 35-ton prototype turned out to be unexpectedly noisy, with a signal to noise ratio of ~10 on the collection plane (vertical wires, y) and only ~2 on the induction planes (diagonal wires, u and v).  Worse, the detector sometimes entered a "High Noise State" involving coupled oscillation of all detector components.  This obviously adversely affected event reconstruction and data analysis.  Subsequent investigations have identified most of the noise sources and proposed modifications to the electronics that would eliminate them.  There is also a proposed, though as yet not verified, explanation for the High Noise State.

As a result of this work, it is unlikely that the ProtoDUNE and DUNE electronics will suffer from the same pathologies.  However, this does not help those trying to analyse the 35-ton data, in which the noise is unavoidably present. 

Electron shower viewed on an induction plane of the 35-ton

High noise levels cause problems in track reconstruction by making the initial hit-finding procedure more difficult.  Typically, hit-finding algorithms look for a peak in the signal from each wire.  If there is a lot of electronic noise, a low threshold for defining a peak will result in a large number of spurious hits, whereas a high threshold will cut out the noise but will reduce hit-finding efficiency, as some real hits are also likely to be lost.  Either of these outcomes is likely to reduce the efficiency and precision of the track reconstruction.

Sheffield students are among those trying to mitigate the effects of the noise by developing more sophisticated hit-finding algorithms.  One of these, developed by Matthew Thiesse, is RobustHitFinder, which uses a variable threshold optimised for each individual wire.

Another feature of hit and track reconstruction which can be studied in the 35-ton prototype, but not in detectors such as SBND where the APAs are at the edges of the LAr volume, is the behaviour of a track that crosses the APA.  Because of the geometry of the electric field, ionisation electrons produced between the wire planes of the APA and the grounded mesh that separates the two sides of the double-sided APA actually drift back on to the plane that the track has just passed through.  If the resulting drift time is interpreted as being in the usual direction, this results in a hook-like feature in the track, which can be seen in the data (see figure).  Clearly it will be possible to correct for this effect, but only after hits have been assigned to tracks.

Track crossing an APA in the 35-ton

Lifetime measurements

A key goal of the 35-ton prototype was to measure the lifetime of the ionisation electrons, which is a function of the purity of the liquid argon (argon itself being a noble gas with a full outer electron shell, electrons are much more likely to be captured on impurities).  A standard method of measuring this is to reconstruct throughgoing cosmic muon tracks—which, since muons are minimally ionising particles, should lose, on average, a constant amount of energy to ionisation per unit track length—and look at the amount of charge collected as a function of drift distance.  The result should be a negative exponential, q(t) = q0 exp(–t/τ), where q =〈dQ/dx〉is the mean charge deposited per unit track length and the time t is derived from the drift distance.

This measurement is also affected by high noise levels.  The reason for this is as follows:

  • The most probable value of dQ/dx (a more stable proxy for the mean value) is found by fitting a Landau-Gaussian convolution to the distribution of deposited charge values.
  • However, if the threshold for identifying hits has to be raised because of noise, hits with lower dQ/dx are systematically missed, and the most probable value is overestimated.  Because the threshold is not sharp (since hits are Gaussian peaks rather than delta functions) and the rising edge of a Landau-Gaussian convolution is quite steep, this does not produce a noticeably worse fit to the Landau-Gaussian, so it is not obvious that this has happened.
  • This effect becomes more pronounced as the true most probable value of dQ/dx becomes smaller, so the result is that the exponential fall-off appears slower than it really is, and the lifetime is overestimated.

This effect can be simulated by adding real-data noise (from events with no tracks) to simulated hit waveforms with known electron lifetimes.  The results of such studies suggest that the true electron lifetime is ~3.2 ms, which is in agreement with the value of ~3.5 ms implied by direct measurements of the purity of the argon.