Data Acquisition Systems

The kind of system for which we investigate the application of object oriented design and implementation can be characterised as a heterogeneous network environment. It comprises a collection of different networks and processing units which have to be coordinated to serve a common purpose. The task performed by such a system is recording, processing and storage of data that originate from particle detector machinery.

Definition 2   A data acquisition system (DAQ) is a set of hardware and software components which are used to read data from detectors, process them according to rules and transfer them to persistent store for subsequent use.

DAQ systems can be found in a variety of environments: Space agencies such as NASA have DAQ systems[AR95] for handling information from earth observation satellites. These data are combined and made human interpretable in the processing phase. The data sets are stored for use in studies of climate or habitat development. Such a system is characterised by large amounts of data, disconnected operation, high availability and long lifetime. A different example for DAQ can be found in experimental fusion reactor experiments[RLN$^$97,LM99]. There, it is vital to collect data from a large number of sensors within a short period of time in order to be able to react quickly to fluctuations in the plasma and take countermeasures. Finally, we would like to mention the area of particle physics experiments[GG99b, pp. 83-85] and their industrial applications, e.g. in medicine[Ama96]. Highly energetic particles are used to modify the structure of materials used for medical treatment or examination. Another branch that is gaining more and more importance is that of treatment with particle beams. In all cases it is highly important to obtain detailed information about the particles for exact dosing of the energy quantity[AWHG96]. Data acquisition for particle physics experiments will be outlined in more detail in this thesis. Highly reliable operation, records of large data sets and small processing latencies characterise these types of systems.

In all cases we experience that performance for DAQ systems boils down to the ability to meet three major requirements:

1. Speed. This property consists of latency and throughput. It is a parameter affected by the transferred data size and the frequency of these operations. Not meeting this requirement means not having an efficient system. After all, it is the main task of a data acquisition system to transmit data. Hence we must aim for performing this task at high speed.

2. Scale covers both the number of units forming the distributed system as well as its physical dimensions. For both issues we must work on solutions that allow the system to work efficiently, regardless of its size. Adding further elements shall therefore not decrease the overall performance.

3. System Management deals with the configuration and operation of the data acquisition system. It has to reflect openness in terms of configuration parameters and adaptability to specific needs. A system must not only be extensible, but also easy to manage. This is important for its static and dynamic structure. A lack of these characteristics can also affect performance as it is not possible to select the configuration which will allow efficient operation. System management is also coupled to the lifetime of the systems, as a long living installation is more likely to be modified frequently.

I do not include dependability[Lor89] as explicit parameter. All DAQ systems shall be highly dependable. I could measure this as a percentage of allowed data loss, but I could not find quantitative evidence for such characteristics in existing data acquisition. Although speaking of high performance systems, the described systems tend to operate in best-effort mode: if the data rate drops below a given limit, operation has to be suspended in order not to harm persons, destroy equipment or record meaningless data. In this respect they differ from hard real-time systems that must meet defined deadlines[LL73] and thus shall never drop below a specified performance level.

This thesis deals with data acquisition for high energy physics experiments. Such a system is composed of a variety of different sensor clusters for identifying elementary particles, e.g. electrons, photons or particles composed of quarks. The observed particles have been produced in collisions inside a particle collider. Such machinery accelerates them to high energies and brings them to collision at an interaction point. Particle physics detectors have to provide very high resolution, both in time and in space. Therefore proprietary readout electronics is provided by the detector designers to achieve response times in the nanosecond range. Good space resolution is achieved by building grids of sensors and grouping them into multiple layers. Tens of millions of data channels lead to so called readout units which are able to buffer a certain amount of recorded data. These are then processed according to the physics algorithms in on-line mode.

Some experiments just need to transform the data into a specific format before they can be stored for later use. This includes data compression as well as indexing. In an experiment with small time resolution and a massive amount of channels it might not be possible to store all the data because of physical or financial limitations. Therefore just those particle collisions are kept which show certain interesting characteristics with a sufficiently high probability. This task is known as event filtering. Furthermore, information about the detector has to be taken into account before the data can be used. This process is called calibration. The correction becomes necessary because the materials that are used in the detectors react differently when exposed to varying temperatures, currents or radiation levels. Some examples of applying calibration values to acquired data are offset calculation of geometry due to tension and heat deformation, and subtraction of background noise that can be found in any kind of electronic sensors. Before I outline the scenario with which we are confronted in our field of study, I would like to shed light on the role of software in DAQ systems. These issues are valid for all DAQ systems. The degree of performance that has to be met however, will vary from case to case.