THE EXOSAT RESULTS DATABASE
The EXOSAT database will contain the principal results obtained from a standard analysis performed on every observation. The objective is to provide a computer accessible overview of each observation. This will include a summary of the results from each instrument, as well as the option of obtaining the data in a reduced form (e.g. compressed images, background subtracted spectra, lightcurves etc). A data base management system will allow manipulation of the results summary files (e.g. cross- correlating parameters). The database will in many cases circumvent the need for astronomers to analyse the final observation tapes (FOTs). This system will be accessible by remote users via the SPAN network and will include the possibility of file transfer, both of data and programs. A preliminary version of the results summary data base has been generated using the output of the first version of the automatic analysis of the data from the low energy telescope. The reprocessing of all data using second generation software is now well underway and the full database system is expected to be available by the end of 1988.
Over the past twenty years X-ray astronomy has evolved from a field where a few specialists were studying a hundred poorly understood X-ray sources, to the rich field it is today where virtually every known astrophysical object is at some level an X-ray source. This transformation has resulted in non-specialist astronomers using the more recent X-ray astronomy observatories such as Einstein and EXOSAT. One major problem for a newcomer to data obtained from an X-ray observatory like EXOSAT is the complexity of the data acquisition modes used and the rather specialized knowledge required to perform relatively basic tasks such as background subtraction. To undertake the analysis of this data requires a considerable investment in time to write the appropriate software and to gain experience in the nuances of a particular instrument. Such an effort is only worthwhile at institutes expecting to analyse a fairly large fraction of EXOSAT observations. Up to now the non-specialist has been dependent on the EXOSAT interactive analysis system provided by ESA, which has meant traveling to ESTEC/ESOC to access the data. To improve this situation an EXOSAT results database is being setup that will contain the following:
(i) the basic results from each instrument for every observation, so providing an immediate overview of the results from each EXOSAT observation.
(ii) a database management system for cross- correlating those results against each other as well as against other astronomical catalogues and databases.
(iii) the data in a reduced form e.g. background subtracted spectra and images.
(iv) remote access, via SPAN, with the possibility to extract the reduced data and analysis software.
This document describes the structure of the database and the way it is being implemented. While many of the components are now in place the system is relatively flexible and it may be possible to make revisions to include any suggestions arising from this report.
II. The Data Base Management System
There are four instrument summary files which contain the principal results from each instrument (L1, L2, ME and GS) for the entire mission. A fifth slew summary file contains a record of all X-ray sources detected by the ME experiment while the spacecraft was slewing from one target to the next. In addition background subtracted grating spectra will also be available. For the instrument summary files there will be one entry per source, with up to 220 parameters stored for every entry. The summary files are accessed by a database management system (DMS) developed at the EXOSAT Observatory. This software package has been specifically designed to manipulate large amounts of astronomical data. The system, besides providing 0 the basic functions supported by the majority of commercially available packages, also allows a number of statistical tests frequently used in astronomy, The main facilities are:
1. The possibility to display to any terminal the basic results information for every entry. Examples of this are given in Appendix A, B, and C for each experiment.
2. Retrieve information on sources located in a specified area of the sky.
3. Select subsets using any relation of the Boolean algebra.
4. Cross- correlation of the four summary records, or of any subset with catalogues of cosmic sources.
5. Definition of new parameters. These can be virtually any function of the primary parameters i.e. parameters directly stored in the database. Parameters defined in this way are called derived parameters.
6. Possibility to sort the database (or any subset of it) by any primary or derived parameter.
7. Production of ASCII tables of primary or derived parameters.
8. Generation of histograms of any primary or derived parameter.
9. Plotting of parameters against each other.
10. Possibility to fit analytical functions to the data.
11. Statistical analysis e.g. Calculation of mean value, variance, regression analysis, , etc.
Each of the LE records is cross-correlated with well known catalogues of astronomical objects and any coincidences are also written into the record. These catalogues include the Bright Star catalogue, the IRAS catalogue, the revised catalogue of Quasi Stellar Objects, plus several other listings of extragalactic objects. It is also planned to include the catalogues available at the Centre de Donnees Stellaire at the Observatoire de Strasbourg.
III. The Reduced Data Files
The reduced data files created by the automatic analysis are used to produce the instrument summary files by operating on them with a standard set of analysis routines e.g. source detection and spectral fitting programs. These files are saved on an off-line medium (currently magnetic tapes, but eventually optical disks) and include such things as background subtracted spectra, images, lightcurves; etc. These files will be given in a format that will be compatible with a suite of interactive programs to provide further analysis of these data. These programs will include:
1. An image processing package to detect, or set upper limits to sources in the field of view.
2. A spectral fitting program.
3. Fast Fourier transform, folding and general plotting facility (the latter may be system dependent).
4. A program to convert the files to FITs format.
These programs and the data will be made available to be run on Vax systems.
IV. The LE Automatic Analysis
The revised LE automatic analysis software performs a detailed standard
processing of LE data, taking into account the final calibration of the
telescopes and detectors. For every observation a compressed image file is
created for each filter used. A compressed image is a file containing all the
information necessary to create a standard EXOSAT X-ray image and, because each
images is a sparse array, is written in way that enables a factor 10 saving in
disk space with respect to more common methods of storing images. The details of
how this is done are given elsewhere in this volume. Compressed images also allow
rapid transmission down data links. A program will be provided to convert these
images to FITs format after the image has been received. The images are searched
for sources and for every one detected its position, count rate, significance of
the detection and several other parameters of scientific interest are calculated
and written to the database. In addition for each source a file containing all
the relevant timing information is created and analysed by the timing analysis
software. The compressed images (for each filter used) and the time series files
for all detected sources plus the associated background. The main advantages of
the new LE automatic analysis are the following:
(i) Smaller uncertainties in source coordinates. The use of
new and improved star tracker calibration and of more precise misalignment values
allow 6 arc sec 90% confidence positions to be obtained in the center of the
field of view.
(ii) The broadening of the point spread function with increasing off-axis angles is properly taken into account (see accompanying article in this volume).
(iii) Non-uniformities in the CMA background are included.
(iv) Many sources that in the old automatic analysis were missed because of their proximity to other sources are now detected.
(v) A proper approximation to the boron point spread function is used.
V. The ME Automatic Analysis
The ME auto analysis output that observers received some ten weeks after their EXOSAT observation was designed before launch and, although extensively modified, could not be modified to incorporate the latest thinking on the background subtraction, spectral fitting, timing analysis etc. In order to produce a database of 'final products' such as spectra and lightcurves, the ME interactive analysis software has been modified to run automatically.
The new ME automatic analysis system has been designed to produce, in an automatic fashion, as many high quality spectra and lightcurves as possible. In many cases the quality is such that users need not access the original FOT. These files are operated on by a standard set of analysis programs and the results written to the summary file.
Great care has been taken to design a system that can cope with the myriad of OBC modes, observing configurations and types of source observed by the ME. One of the most difficult areas of ME analysis is the choice of background subtraction technique (discussed below). It is envisaged that a second pass through some 20-30% of the observations will be required to 'fix' problems due to poor background subtraction. The automatic processing software has now been throughly tested on many types of source and will, in principle, give results that are as good as the ME interactive analysis e.g. QPO from GX5-1 were easily detected. The limiting sensitivity for spectra is strongly dependent on the method of background subtraction and the stability of the background. For many observations the automatic analysis gives acceptable spectra for sources as faint as 1 UFU. With similar constraints, the limiting sensitivity for detection is ~0.25 UFU.
(i) Description of the Processing
The new auto analysis utilises programs from the ME interactive analysis. This has the advantage that the whole range of spectral, display and timing programs developed for the interactive analysis can access the new auto analysis files. The new auto analysis proceeds as follows:
1. Selection of the method of background subtraction.
2. Creation of spectral and rates files.
3. Spectral Analysis.
4. Timing Analysis.
5. Slew Analysis.
The selection of the most appropriate method of background subtraction for a particular observation is an important step in obtaining the best possible results. The software works in a hierarchical manner selecting, in order of merit, the following background subtraction methods:
a. Array swaps if complete set of swaps present.
b. Array swaps if incomplete set present.
c. Slew background available.
d. No suitable background available.
Cases (a) and (b) will generally give the best results because background data is obtained from the same detectors at a different time while they were pointing away from the target. This introduces another uncertainty since offsetting the detectors alters their background counting rate by a small amount - called the difference spectra. For (a) the software generates its own set of difference spectra using the + and - offset data (see EXOSAT Express No. 16, p.21). This generally gives the best results. If an incomplete set of array swaps were not obtained (case b) the program uses standard difference spectra. If no suitable array swap data are available then background spectra obtained during the slew on or off (or both) the source are used for background (case c). If no suitable slew data are available then standard background spectra are used (case d). This generally only gives satisfactory results for bright sources.
After selecting the method of subtraction the software creates a single spectrum (integrated over the entire observation) and a number of rates (counts against time) files. Rates files are created at the highest possible time resolution for channels corresponding to energies of 1-3, 3-6 and 1-6 keV for the source and 1-6 keV for the background data. If the source intensity in >4 cts/s/4det then rates files of any available high time resolution data will be produced. If the source count rate is >0.25 cts/sec/4det then an automatic spectral fitting procedure is run as described in §VIII.
The next procedure is to run the timing analysis software described in §VII to search for coherent periodicities, QPO, bursts and dips and to quantify the amounts of variability observed on various timescales and energies.
Any slew data associated with the observation is processed to produce rates files for the aligned and offset data as well as a file containing pointing positions against time during each slew. If data is available with detector or quadrant ID then rates files are produced separately for each offset quadrant since they point in different directions.
(ii) The Instrument Summary Records
The summary records includes information from the spectral
and timing analyses as well as recording the way in which the analysis proceeded.
Each record contains count rates, the results of a periodicity search, pointing
position, observation class (HLX etc.), rms variability, best fit spectral
parameters etc. This makes for a very flexible way of searching for particular ME
observations of interest. For example, it is possible to select all HLX sources
that have best fitting power-law models or plot softness ratio against hardness
ratio for all AGN's etc. The summary record also contains information on the
location of associated hardcopies and archived reduced data files making it easy
to use the interactive analysis to further analyse the auto analysis results.
Until a quality control check is made, for some of the observations the background subtraction may not be as good as can be obtained by careful use of the interactive analysis. In particular, for background selection cases (a) and (b) the offset data is not screened for background variations. In background selection case (c) the slew data is examined for the presence of point sources only. Any long term variations in background counting rate will probably be missed and produce an anomalous subtraction. Also, a slightly different type of limitation is that only one spectrum is produced per observation. This means that for burst sources, any bursts are included in the spectrum etc. It is anticipated that up to 30% of the observations will require some sort of manual intervention to correct problems. Checking the quality of the ME database will be a time consuming and tedious task and it is anticipated that this will take at least one year to complete after the automatic run has finished.
(iv) Reduced data files.
The following reduced data files will be archived:
1. Time series files with the highest PHA time resolution available at energies of 1-3, 3-6 and 1-6 keV.
2. Time series files for 1-6 keV with a time resolution of 30 sec for both source and background.
3. Image files from the FFT analysis.
4. Slew time series files
5. The history of the spacecraft pointing during slews.
6. Spectra from the offset quadrants if there is evidence for background contamination.
7. Spectra for each observation plus a summary file with results from the automatic spectral analysis
Only 2, 4 and 7 will be kept permanently available. The others are either too large or not of sufficient interest to keep on disk, but they can be be restored on request.
VI. The GSPC Automatic analysis
This will primarily consist of a background subtracted
spectrum for each observation where the ME count rate was greater than 5
ct/s/half. For weaker sources the signal to noise is not sufficiently high to
The analysis creates a rates file for both the slew and the source in the energy ranges 2-7 keV and 8-15 keV. Any sources in the slew are removed and the average count rate used to re-normalize the standard background. This standard background is then subtracted from the source spectrum and the resulting spectrum analysed in the same way as the ME spectra. The structure of the GSPC database is similar to that from the ME with for each observation one record written with interesting parameters such as count rates, the results of the spectral fitting etc. The spectral and time series files are archived on magnetic tape.
VII. The Timing Analysis
The files containing the source and background light curves for the ME and the LE are processed by a timing analysis system and part of the results stored into the database summary file for the relevant instrument. Many tasks of the timing analysis are common to both instruments. These are:
1. Test source and background constancy.The variance of the light curve is compared to that expected from a constant source (when instrumental and computer dead time effects are taken into account). The results are evaluated in terms of the rms variability of the source (or the background). The advantage of the rms variability is that its value depends only on the source (or background) variability properties over the range of timescales explored, while is independent of the effective area of the instruments, source count rate etc. If the source (or the background) is consistent with being constant a 90% confidence upper limit to the rms variability is calculated.
2. Search for bursts and dips. This search is carried out both to detect bursts and dips from the source and to reveal background activity or on board computer problems. Bursts and dips are defined by a departure larger than 5 times the standard deviation of the light curve points. A maximum number of 48 bursts and dips is allowed for each light curve analysed. The time of occurrence, duration and maximum or minimum count rate is calculated for each burst or dip.
3. Power spectrum analysis.
The power spectrum is computed for each lightcurve using an FFT to search for coherent periodicities,
quasi-periodic oscillations and other features in the source variability (e.g.
red noise, shot noise etc.). In the cases in which the light curve consists of
more than one interval, the power spectra from individual intervals are summed.
The results are presented in graphic form and the maximum value of the power (and
the corresponding frequency) is output over two different ranges of
frequencies (0.0 - 0.1 and 0.1 - 1.0 times the Nyquist frequency).
For all ME observations the analysis first uses the original binning time of the light Curves (usually 10s) and then is repeated for a binning time ten times longer. For the sources > 4 ct/s/half the high time resolution ME data (with binning times of a few milliseconds or less) are analysed by a dedicated timing analysis program which computes the source power spectrum up to very high frequencies, for each 4096 data points. The power spectra obtained are then summed to give an average power spectrum which is then searched for coherent and quasi periodic oscillations and displayed as in (c). In addition the program produces a "Time Image File" which is used to display the source light curve together with a colour-coded representation of each 4096 points power spectrum. This type of power spectrum display has proven very useful in searches and studies of fast (1-50 Hz) quasi-periodic oscillations in the X- ray flux bright galactic X-ray sources.
For the LE the time resolution is optimized for each observation such that the average number of counts per bin is ~0.3. Again the analysis is repeated using a binning time that is ten times larger.
The results from each run are stored in Timing Summary Files which are archived. An example of the contents of one of these files is given in Appendix D. The most important results are also put into the data base summary files.
The automatic spectral analysis program fits a variety of standard spectral models to the ME and GSPC spectra generated by the automatic analysis. The following models are fit to the data and the best fitting parameters and stored for each model : thermal, powerlaw, blackbody and Compton. If an acceptable is not obtained then the same models are fitted with the addition of an iron line. If this also fails to give a satisfactory fit then multi- component models are also tried: a power law with a high energy cut-off, Compton plus blackbody and Compton plus a powerlaw tail. Such automated fitting programs are far from infallible and in many cases an inappropriate model will give the best fit. During the quality control check these faults will be corrected by reading into the database a fit made using an interactive fitting program.
In addition to this it is recognized that many spectra will be generated by users of the interactive system. Many of these will have been processed with extreme care to obtain the best possible background subtraction. These spectra will be stored into an ME spectral file and will also be available. A flag will be put into the ME summary file to alert users that a better spectrum exists. This will for example allow for a canonical EXOSAT spectrum of such objects as Cas A or Perseus cluster to be simply obtained by an outside user. It will be possible to generate and extract via remote file transfer the appropriate detector response. The spectra obtained from the objective grating spectrometer will also be available in a similar manner.
IX. Slew Survey
When EXOSAT was slewing across the sky between its scheduled targets the ME
detectors were, for most of the mission, left on. The principal intention of this
was to monitor the particle background but it also gave a useful all- sky survey.
Part of the routine analysis of EXOSAT ME data is a search for sources using
10-second integrations of channels 6-30 (1-7 keV). Channels 40-60 are used to
monitor the particle background. It is expected that several hundred sources both previously catalogued and new
will be found with a sensitivity limit of a few millicrabs. The position of a
source can be determined to an accuracy of at best a few tens of arc seconds
along the direction of the slew. In the perpendicular direction it is the width
of the collimator (45 arc min FWHM) which defines the error region.
The DMS will allow for the input of an RA and Dec on the sky and will return the number of times EXOSAT slewed over that area of sky and the closest detected source. This will form the basis for a catalogue of new EXOSAT sources, as well as detections of previously known ones.
X. Remote Access
The DMS will reside on both the observatory micro-vax (known as Vivean) and the HP1000 computers. These will be available for use by the community at ESTEC. The five summary files and the DMS will be kept on line. Because of disk space limitations it is not currently possible to keep all the reduced data files permanently available on disk for instant access. With the advent of optical disks this may become possible in future.
The micro-vax can be accessed via SPAN and this will form the main route for the community to interact with the DMS. A separate article in this volume describes the procedure for remote logon. Because of the differences in network protocols not all members of the community have the same capabilities, e.g. IBM machines will only be able to transfer files back and forth to Vivean (mail access), while VAXes and most other small systems may have the capability for remote interactive access, depending on their network links. In the oncoming years it is expected that the richness of these links will increase greatly so that almost anyone on an academic VAX in Europe, North America or elsewhere should have interactive access. Those users could, in principle, be allowed access equivalent to members of the Observatory on site, although graphics will be slow (image displays in particular would take a long time) unless a high speed link is available.
Access to Vivean will occur on four levels as follows:
(i) Mail access - The request account.
A special account (EXOSAT::REQUEST) has been setup that acts as a general mailbox. This account is checked dally to process incoming mail and would be similar to the Observatory BITNET account on Profs (MAILEX@HNOESA10). The following items would be handled by this account:
1. Requests for information or archive requests.
2. Requests for restoration of reduced data files.
3. Batch processing requests. This would allow mail-only users to submit jobs to access the DMS. The output would then be returned by computer mail (or normal mail).
(ii) Direct Access
Any user on another SPAN node (which essentially means WORLD
access) may access the DMS and any activity allowed to DECNET can be performed.
The prime purpose of this is to allow free use of the DMS to interrogate the
summary records. In this way astronomers will be able to obtain a concise summary
of each observation as well as being able to manipulate the various
(iii) Visitor accounts
In those cases where a more detailed analysis is required it will be possible to grant individual user IDs to observers, as occurs when observers physically visit the observatory. Then the full interactive facility available on Vivean could be used to further analyse the reduced data files. This will include a more detailed spectral analysis and specialized timing and plotting programs. Because some of the reduced data files will not be permanently mounted this will require at least one days notice for the relevant files to be restored. Eventually it is envisioned that the most frequently used reduced data files will be written to permanently mounted optical disks.
(iv) File Transfer
Users with SPAN access will also be allowed to transfer to their local computer
some of the information in the database. This will include images, lightcurves,
spectral buffers (and the associated detector response matrix) plus other
assorted useful items. In this way it will be possible to fit more complex models
to the data that may not be available on Vivean, or directly compare the
EXOSAT data with those from other observatories. Also the executables of the
interactive software required to operate on these files will also be made
available for use on other Vaxes. The transfer out of data for extended analysis
will be encouraged since the capacity of the micro-vax is limited.
The implementation of this plan is now well underway. Below are given
some key dates for both past and future activity along with the personnel responsible for
the various activities outlined.
October 1986 ME autoanalysis reprocessing begins: