• The Standard Imaging Pipeline
• Current Imaging Status and Performance 

 

The interferometric imaging mode provides correlated visibility data (similar to traditional aperture synthesis radio telescope arrays consisting of antenna elements). Station beam data streams are transferred to the Central Processing (CEP) facility (currently an IBM Blue Gene/P computer is used as the correlator) where they are correlated to produce raw visibility data, which are written in Measurement Set format in the post-processing cluster (known as the CEP2 cluster), also located at CEP.

 

Each "observation" includes not only the LOFAR stations, of which the data streams are cross-correlated to produce "uv-data" visibilities (as in traditional interferometers) but also a number of further processing steps carried out in the "Standard Imaging Pipeline" (see below). Therefore, proposals should also take into account the processing needed to reach their scientific goals.

 

The automatic "Standard Imaging Pipeline" is being developed to provide initial images to the limited specifications defined in each software release. Processing is based on parameters that are defined by the proposers, based on certain well-tested defaults and with consultation and verification by the Science Support Group.

 

Users may wish to further process their data to improve the quality of the initial images or to reach other goals. Users who, in their proposal, demonstrate their expertise in imaging LOFAR data, can request further processing time, either at the CEP cluster or can show availability of processing capability at another processing facility (using the data extracted via the LOFAR Long Term Archive). These "Expert" users can specify pre-application to their data of (parts of) the automated analysis pipeline with specific parameters.

 

A few of the most meritorious proposals from less experienced groups can be earmarked by the PC for (semi)manual further data processing steps and Observatory support staff assistance, on a best-effort basis and limited by the scarcity of processing resources.

 

In LOFAR Version 1.0 available in Cycle 0, one Low Band Antenna (LBA) and one High Band Antenna (HBA) observing mode with associated "recipes" for the imaging pipeline are offered for allocated observations. Expert users can also request modified observing and analysis strategies, which will be considered at the time of the proposal evaluation.

 

Observations with the Low Band Antennae (LBA):

- Single observations that are continuous in time/Hour Angle

- Half of the available bandwidth on the target field (BW<=24 MHz, <=122 subbands) and half on a calibrator (same frequencies as the target, BW<=24 MHz, <=122 subbands).

- Observations in the band of 10-80 MHz (both the 10 MHz and the 30 MHz filters are possible).

- Processing performed with the Standard Imaging Pipeline.

 

Observations with the High Band Antennae (HBA):

- Interleaved short calibrator observations (eg. 2 min) with target field (eg. ~11-30 min), quasi-continuous in HA.

- Up to the full available bandwidth (BW < 48 MHz).

- Observations in two HBA bands: 110-190 MHz, 210-250 MHz

- Processing performed with the Standard Imaging Pipeline

 

As a primary flux calibrator, if at all possible one should use 3C196.

In the HBA, the best flux density calibrators are 3C196 and 3C295, followed by 3C147 and 3C48. For calibrating european baselines, use 3C196, 3C147, or 3C48.

In the LBA, one should not use 3C295, because its spectrum turns over at low frequencies.

Note that for accurate gain calibration at LOFAR-NL or european scales, these sources need resolved models. In Table A we summarize the relevant parameters of the flux calibrators.

 

A. Flux calibrators 

Source	Kind	Band	RA (h m s)	DEC (o ‘ ")
3C196	Seyfert 1 Galaxy	LBA+HBA	08 13 36.07	+48 13 02.58
3C295	Seyfert 2 Galaxy	HBA	14 11 20.52	+52 12 09.86
3C147	Seyfert 1 Galaxy	LBA+HBA	05 42 36.26	+49 51 07.08
3C48	Quasar	LBA+HBA	01 37 41.30	+33 09 35.12

Table A: calibrators to be used in LOFAR observations

 

The list of polarized sources, required for monitoring ionospheric RM changes for accurate polarimetry, is given in Table B:

 

B. Polarized sources

Source	Kind	Band	RA ( h m s)	DEC (o ‘ ")
Western hotspot DA240	Radio galaxy hotspot	HBA	07 49 48.02	+55 54 22.1
PSR B0834+06	Bright pulsar	HBA	08 37 05.64	+06 10 14.56
PSR B1642-03	Bright pulsar	LBA+HBA	16 45 02.04	-03 17 58.32
PSR B1919+21	Bright pulsar	LBA+HBA	19 21 44.82	+21 53 02.25
PSR B1937+21	Bright pulsar	HBA	19 39 38.56	+21 34 59.14
PSR B2217+47	Bright pulsar	LBA+HBA	22 19 48.14	+47 54 53.93

Table B: polarized sources for LOFAR observations


 

The "Standard Imaging Pipeline"


Processing of the raw uv data, which consists of calibration and imaging, is handled offline via a series of automated pipelines (see "Software Pipelines"). Calibration is an iterative process consisting in obtaining the best estimates of instrumental and environmental effects such as electronic station gains, clock and ionospheric delays

 



Figure 1: Layout of The Imaging Pipeline. A short overview of the pipeline is given by Heald et al (2010)[PDF].

 

The Standard Imaging Pipeline is under development, and its first version deployed in LOFAR Version 1.0 is graphically illustrated in Figure 1.

The first standard data processing steps are encapsulated within a sub-pipeline called the Pre-processing Pipeline, which consists of two steps:

 

(a) The Calibrator Pre-Processing Pipeline, which flags the data in time and frequency, and optionally averages them in time, frequency, or both (the software that performs this step is labeled NDPPP - New Default Pre-Processing Pipeline). This stage of the processing also includes a subtraction of the contributions of the brightest sources in the sky (the so called "A-team": Cygnus A, Cassiopeia A, Virgo A, etc...) from the visibilities through the 'demixing' algorithm (B. van der Tol, PhD thesis). Eventually, the solutions for the calibrator are computed, using the BlackBoard Selfcal (BBS) system - specifically developed for LOFAR.

 

(b) The Target Pre-processing pipeline performs RFI excision and removes the contributions of the A-team from the visibilities. This step is followed by the BBS calibration, which applies the externally generated solutions for the calibrator to the target field. In the future, the calibration of the complex station gains will be achieved using a local sky model (LSM). This will be generated from the Global Sky Model (GSM), which will soon provided by the MSSS survey.


Following the pre-processing stage, the calibrated data are further processed in the Imaging Pipeline, which begins with an imaging step that uses the AWImager. AWImager, still under development and testing, is able to perform both W-projection and A-projection, a scheme that can potentially take into account all direction-dependent effects in the deconvolution step. Source finding software is used to identify the sources detected in the image, and generate an updated local sky model. In future versions of the pipeline one or more `major cycle' loops of calibration (with BBS), flagging, imaging, and LSM updates will then be performed. At the end of the process, the final LSM will be used to update the GSM, and final image products will be produced.

 

The entire end-to-end process, from performing the observation through obtaining the final images, is overseen by the SCHEDULER, software specially developed for LOFAR, which manages both the observational and the computational resources available.




Final Imaging Data Products

 

The final data products include the calibrated uv data, optionally averaged in time and frequency, and corresponding images/image cubes. Visibility averaging is performed to a level that reduces the data volume to a manageable level, while minimizing the effects of time and bandwidth smearing. The averaging parameters are determined from the request of the users upon consultation with the Radio Observatory's Science Support Group.

These final products will be stored at the LOFAR Long Term Archive where, in the future, significant computing facilities may become available for further re-processing. It is possible to routinely export datasets from the LTA to investigators for reduction and analysis using their own resources or through the use of suitable resources on the GRID.

This mode requires medium to long-term storage of un-calibrated or partially calibrated data at the central processing facility, until processing is complete, following detailed inspection of results by the user. The resulting storage and processing requirements will impose limits on the amount of such customized reprocessing, which may be conducted in the early years of LOFAR operation.

 

 

 

Current Imaging Status and Performance

 

The imaging performance of LOFAR has been assessed using the products of both data reduction strategies performed by experienced commissioners and the Standard Imaging Pipeline. The results of the two approaches are reported below.

 




Manual Reduction Strategies

Expert users have adopted standard LOFAR software to produce images, exploring various analysis strategies, in some cases involving also self-calibration. Their results are reported in Table 1 for both HBA and LBA observations.

 

1. Commissioning fields - HBA & LBA




Commisioner(s)	Iacobelli	Pizzo	Orru	Broderik
Band	HBA	HBA	HBA	HBA
Target Field	Fan Region	Abell 2255	B1835+62	B0329+54
Total observing time	12 hrs	6 hrs	7hrs	6hrs
Bandwidth	27.7	36.8	33.5	47.8
Resolution $ (arcsec)	90x70	169x178	24 x 21	147x122
Imaged FOV (deg)	16.7	11	5	11.4
Final RMS Noise (mJy/beam)	0.3	1.5	1.2	1.2
Equivalent noise over 2 MHz bandwidth and 6 hours (mJy/beam)	2.2	6.5	5.7	5.9
Noise /thermal noise ratio	17	15	12	14

 

Commisioner(s)	Bonafede	Van Weeren	MSSS
Band	LBA	LBA	LBA
Target Field	MAC 0717+35	Abell 2256	L070+69
Total observing time	10 hrs	10 hrs	1.65 hrs
Bandwidth	5.3	5.3	15.6
Resolution $ (arcsec)	38x12	25x25	310x210
Imaged FOV (deg)	7	5	15
Final RMS Noise (mJy/beam)	13	10 *	60
Equivalent noise over 2 MHz bandwidth and 6 hours (mJy/beam)	33.2	28.2	43.9
Noise /thermal noise ratio	7	4	10.5

^$ The resolution depends on the stations used for the imaging

^* Analysis used self-calibration

Table 1: Examples of sensitivities reached in commissioning observations in HBA and LBA

 

in the LBA, expert users have demonstrated that, using a simple analysis strategy that does not  employ position dependent calibration or self-calibration, total intensity sensitivities of ~10 times the theoretical thermal noise in relatively long observations (6-10 hours) can be reached. Using more involved calibration techniques (as self-calibration or direction dependent calibration), sensitivities of 4-5 times the theoretical thermal noise have already been achieved.

For the HBA, experienced users using simple analysis strategies reached sensitivities of the order of 12-17 times the theoretical thermal noise in images at a resolution of 20"-160". One group, after investing considerable post-processing effort on a 6-hour HBA dataset, has achieved a thermal noise limited image with a dynamic range of half a million (Labropoulos et al. in prep).

 

Imaging Pipeline Performance

The Standard Imaging Pipeline is under development and is based on the experience gained by analyzing commissioning observations with different strategies. In its first implementation in LOFAR Version 1.0, it produces calibrated data and initial images using only one imaging step. Development of extensions and refinements including "Major Cycle" calibration is under way. Moreover, since it was initially developed to process MSSS data, the Standard Imaging Pipeline is currently optimized to analyze LBA observations; the optimal reduction strategy for HBA observations is still under investigation.

 

To assess the performance of the Imaging pipeline, HBA and LBA observations were performed and reduced using the standard strategies mentioned above. Specifically, a typical field with no dominant sources was selected from the MSSS catalogue: L227+69, centered at RA = 15h07m49s and DEC = +69d14m24s. To explore any possible dependence by observing frequency, the data were analyzed in datasets of ~2 MHz bandwidth separated by ~ 2 MHz from each other. For the HBA, the final bands were distributed between 115 and 185 MHz, for the LBA between 54 and 75 MHz. The imaging was performed for the full FOV and using various physical baseline selections, specifically 3, 6, 12, 24, 48, and 79 km. Uniform weighting was used to make the maps and deconvolution was applied using 2500 cleaning cycles. The results for 4 bands spread uniformly over the full observing bandwidth are reported in table 2A for the HBA band and 2B for the LBA band. The detailed overviews of the behaviour of the noise, resolution, and imaging time as a function of baseline cut are available here for the HBA and here for the LBA. Note that the thermal noise has been computed though the LOFAR Image Noise Calculator, which uses the relevant relations reported here. The adopted factors assume an ideal performace of the instrument.

 

 

2A. L227+69 - HBA

Central Frequency HBA band	116 MHz	141 MHz	160 MHz	180 MHz
Resolution $ (arcsec)	15 x 6	14 x 6	13 x 6	13 x 6
Field of View Imaged (deg)	6	6	6	6
Achieved RMS noise over 2MHz bandwidth and 6 hours (mJy/beam)	3.5	2.2	1.5	1.4
Noise/thermal noise ratio	16	13	9	7

 

2B. L227+69 - LBA

 

Central Frequency LBA band	57 MHz	65 MHz	71 MHz	75 MHz
Resolution $ (arcsec)	28 x 13	26 x 11	23 x 10	26 x 9
Field of View Imaged (deg)	3	3	3	3
Achieved RMS noise over 2 MHz bandwidth and 6 hours (mJy/beam)	18	22	34	63
Noise /thermal noise ratio	5	6	9	15

Table 2A & 2B : characteristics of the images of the MSSS field L227+69 in HBA (top) and LBA (bottom). The field does not host dominant sources (their typical peak flux is < 15 Jy).

 

When analyzing these results it is important to realize that significant improvement can be made in the final images by adopting more complex reduction strategies. The low frequency radio sky is very bright and the FOV seen by dipole instruments such as LOFAR is very wide. The rapid density fluctuations of the ionosphere induce artifacts associated with strong sources located within the FOV; this severely limits the dynamic range of the final images made by adopting simplified reduction strategies such as those currently offered by the Standard Imaging Pipeline. To improve the map quality it is crucial to apply direction-dependent calibration towards the offending sources. This technique is known to be compute-time intensive, although recent developments made it more efficient and applicable to large low-frequency datasets. The noise estimates reported here will significantly improve once this technique will be part of the Standard Imaging Pipeline.

 

To assess the limitations induced by off-axis errors, an observation of an empty field (Elais, RA = 16h14m00s and DEC = +54d29m59s) was performed in HBA. The data were processed by Vibor Jelic through automatic scripts performing similar steps as the Standard Imaging Pipeline. The results for 4 bands uniformly distributed over the full observing bandwidth are reported in Table 3. The detailed overview of the noise dependence with frequency is reported here.

 

3. Elais field - HBA

 

Central Frequency HBA band	138 MHz	141 MHz	160 MHz	180 MHz
Resolution $ (arcsec)	15 x 6	14 x 6	13 x 6	13 x 6
Imaged FOV (deg)	6	6	6	6
Achieved RMS noise over 2 MHz bandwidth and 6 hours (mJy/beam)	1.2	1.1	0.9	0.9
Noise /thermal noise ratio	5.8	5.7	4.5	5

Table 3: image characteristics for the Elais field

 

When comparing these results with those obtained for L227+69 in HBA, it is clear that direction-dependent effects are one of the major effects that limit the quality of the images of low-frequency fields hosting strong off-axis sources. Direction-dependent calibration will be included in the Standard Imaging Pipeline in the near future, but will not be available for the Cycle 0 observing proposals.

 

A few quality metrics were investigated in the final maps of L227+69 and Elais, specifically the positional and flux accuracy of the sources within the primary beam. The positions of the sources in the LOFAR maps are correct to within 10" with respect to the catalogue positions. Since gain transfer from a calibrator was adopted during the data reduction, the flux accuracy of the sources in the target field depends on the accuracy of the calibrator fluxes, which is about 5-10%. The accuracy of the primary beam correction was also investigated during the commissioning work and it showed that primary beam corrected fluxes are within 10-20% from the interpolated VLSS-WENSS values up to 3 degrees from the field center.

 

Preliminary polarization calibration and full-Stokes imaging has already been accomplished during the commissioning phase, but the performance of this application is currently not yet well characterized. While work is underway to offer polarization imaging as a standard mode, it currently requires a carefully argued technical case by groups with sufficient expertise.

 

 

Forthcoming Improvements in the instrument and in the Standard Imaging Pipeline

 

Over the next few months, the performance of the instrument will improve due to several factors that are currently under development:

A. Hardware improvements. It is known that some of the current stations have significantly lower sensitivity due to differential delays in the processing boards. The rms noise in the current images is dominated by the affected stations. This problem will be solved with the installation of a new clock distribution board to be completed in September 2012. The new static delay calibration tables that will be created then, will also improve the focus of the station beams, which will in turn reduce the effect of off-axis sources in the final images. The improvement in the image noise level will likely be a factor of 2 - 2.5 in the HBA images. In the LBA, the improvement may be smaller as ionospheric effects will become more dominant.

B. Improvements in the analysis steps. Several analysis steps are planned to be included in the Standard Imaging Pipeline as part of the development towards LOFAR Version 2.0. Some of these are major new software elements while others are simple additions to existing procedures:

• Direction-dependent calibration has been shown to improve significantly the quality of the LBA images. An improvement by a factor of 3-5 in the image noise has been obtained when applied to MSSS LBA observations (see B. Adebahr's report at: http://www.lofar.org/operations/lib/exe/fetch.php?media=msss:adebahr-wee...). A similar improvement is expected in HBA images. A script to automatically derive the sources to be peeled has been written as part of the MSSS analysis. This method should be developed further and included in the Standard Imaging Pipeline. Furthermore, investigations of the quality and the computational efficiency of different algorithms and implementations (SAGECal, BBS) should be continued.


• Closing the Major Loop: self-calibration using a model created from the LOFAR image is another step planned to be inserted in the Standard Imaging Pipeline in the next few months. However, at the moment, tests of self-calibration have not demonstrated a significant improvement in the image sensitivity. This is primarily due to the fact that the image currently produced in the simple imaging step of the pipeline in LOFAR Version 1.0 is not significantly better than the starting model. Thus the ability to make improvements through self-calibration will be better when some of the factors discussed above have been solved and there is a better starting point from the first image. Commissioning work has demonstrated that a large number of self-calibration iterations on just a bright source in the center of the field, helps to improve the image quality, as well as reduce the noise in the image by 10% to 50% (see for example http://www.lofar.org/operations/lib/exe/fetch.php?media=commissioning:au...). In general self-calibration has not been investigated in detail on LOFAR data, this is an area of ongoing research.

 

 

Computational Requirements for the Standard Imaging Pipeline

 

The computational requirements of the imaging mode can be substantial and depend both on observing parameters and image characteristics.

In the following, we present practical estimates for the Processing over Observing time ratio (P/O ratio) separately for the pre-processing and the imaging steps. Note that when considering the computational requirements for the observing proposals, users should account for BOTH of these factors.

 

a) Pre-processing Time

Each of the software elements in the pre-processing pipelines has a varied and complex dependence to observation parameters such as the number of stations, and hence a scaling relation is difficult to determine.

In order to acquire more insight in the processing time required, typical 6h observations were made and processed for various combinations of stations. The run times of the pre-processing pipelines are summarized in the Table below:

 

Type	Nr Stations	Pre-processing (hrs)	Pre-processing/ observation ratio
HBA Superterp	12	1.8	0.3
HBA CS	46	8.4	1.4
HBA CS+RS	51	18	3
LBA Superterp	6	0.6	0.1
LBA CS	21	1.2	0.2
LBA CS+RS	31	9	1.5

Table 4: Pre-processing performance for 6h observations involving different numbers of stations

 

 

b) Imaging Time

 

Imaging with AWimager is a computationally intensive task and, given the finite computing resources of the post-processing cluster, special consideration is needed in deriving the proposed parameters of the final images, particularly the resolution and the imaged field of view (FoV).

The ratio of imaging processing time over observing time (P/O) has been estimated for the L227+69 observations described above. The following considerations can be made:

 

a) For the LBA imaging of the full field of view (6 degrees in LBA outer) the P/O ratio varies from ~0.06 for uvranges restricted to 3 km (core) to ~1 for uv-ranges restricted to 24 km (core + inner remote stations). When including all remote stations, the imaging process becomes too intensive. In particular, swap memory is then used on the locus machines, which both put the locus nodes in danger and does not allow the imaging to make any progress. For this reason, when including more remote stations in imaging, it is necessary to restrict the field of view to ~ 2 degrees.

 

b) For the HBA the P/O ratio ranges from ~0.06 for uvranges up to 3 km, to ~0.4 for uv-ranges up to 6 km. For uv-ranges of 12 km and imaging of the full FoV the P/O ratio reaches 2 or above. When including baselines longer than 12 km, the imaging process becomes too intensive and starts using swap memory on CEP2. In this case, successful imaging runs needs FOV restricted to 2 degrees.

 

It becomes evident that with the current available processing power, in order to image a dataset, the Field of View has to be limited, in order to include long baselines and produce high resolution images.

Note that (sub)arcsec resolution imaging, using the international stations, is also possible, but scarcity of computational resources in CEP limit it practically to only a limited (1-2 arcmin) area of the full station field of view. While work is under way to offer this as a standard mode, this application currently requires appropriate "VLBI" expertise and dedicated processing resources (see "Long Baseline Imaging with LOFAR").