1310.7555v2.pdf

Mon. Not. R. Astron. Soc.

000

, 1–11 (2013)

Printed 11 March 2014

(MN L

X style file v2.2)

The RoboPol Pipeline and Control System

O. G. King

, D. Blinov

, A. N. Ramaprakash

, I. Myserlis

, E. Angelakis

M. Balokovi ́c

, R. Feiler

, L. Fuhrmann

, T. Hovatta

, P. Khodade

A. Kougentakis

, N. Kylafis

, A. Kus

, D. Modi

, E. Paleologou

G. Panopoulou

, I. Papadakis

, I. Papamastorakis

, G. Paterakis

V. Pavlidou

, B. Pazderska

, E. Pazderski

, T. J. Pearson

, C. Rajarshi

A. C. S. Readhead

, P. Reig

, A. Steiakaki

, K. Tassis

, J. A. Zensus

Cahill Center for Astronomy and Astrophysics, California Institute of Technology, 1200 E California Blvd, MC 249-17,

Pasadena CA, 91125, USA

Department of Physics and Institute of Theoretical & Computational Physics, University of Crete, PO Box 2208,

GR-710 03, Heraklion, Crete, Greece

Inter-University Centre for Astronomy and Astrophysics, Post Bag 4, Ganeshkhind, Pune - 411 007, India

Max-Planck-Institut f ̈ur Radioastronomie, Auf dem H ̈ugel 69, 53121 Bonn, Germany

Toru ́n Centre for Astronomy, Nicolaus Copernicus University, Faculty of Physics, Astronomy and Informatics,

Grudziadzka 5, 87-100 Toru ́n, Poland

Foundation for Research and Technology - Hellas, IESL, Voutes, 7110 Heraklion, Greece

Astronomical Institute, St. Petersburg State University,Universitetsky pr. 28, Petrodvoretz, 198504 St. Petersburg, Russia

Aalto University Mets ̈ahovi Radio Observatory, Mets ̈ahovintie 114, 02540 Kylm ̈al ̈a, Finland

Accepted XXX. Received YYY; in original form ZZZ

ABSTRACT

We describe the data reduction pipeline and control system for the RoboPol

project. The RoboPol project is monitoring the optical

-band magnitude and

linear polarization of a large sample of active galactic nuclei that is dominated

by blazars. The pipeline calibrates and reduces each exposure frame, producing

a measurement of the magnitude and linear polarization of every source in the

′

×

′

field of view. The control system combines a dynamic scheduler, real-

time data reduction, and telescope automation to allow high-efficiency unassisted

observations.

Key words:

galaxies: active – galaxies: jets – galaxies: nuclei – polarization –

instrumentation: polarimeters – techniques: polarimetric.

1 INTRODUCTION

The RoboPol project

is monitoring the

-band optical

linear polarization and magnitude of a large sample of ac-

tive galactic nuclei (AGN). The statistically well-defined

sample is drawn from gamma-ray loud AGN detected by

Fermi

(Abdo et al. 2010a; Nolan et al. 2012) and is dom-

inated by blazars, as described in V. Pavlidou et al., in

prep. The main science goal of the RoboPol project is to

understand the link in AGN between optical polarization

behavior, particularly that of the electric vector position

angle (EVPA) (e.g., Marscher et al. 2008; Abdo et al.

2010b), and flares in gamma-ray emission.

The RoboPol polarimeter (A.N. Ramaprakesh et al.

in prep.) is an imaging photopolarimeter that measures

the linear polarization and magnitude of all sources in

the 13

′

field of view. It is installed on the 1

3-m

telescope at the Skinakas Observatory

in Crete, Greece.

E-mail:ogk@astro.caltech.edu

http://www.robopol.org/

http://skinakas.physics.uoc.gr/

The large amount of observing time (four nights a week

on average over the Skinakas 9-month observing season)

and long duration of the project (at least three years) will

generate a large amount of data, requiring a fully auto-

mated data reduction pipeline and observing procedure.

While the RoboPol instrument is optimized for operation

in the

-band, it can also observe in the

and

-bands.

All observations described in this paper were made with

a Johnson-Cousins

-band filter.

Blazar emission at optical wavelengths is highly vari-

able and the optical polarization events we aim to char-

acterize can occur very rapidly. This requires a flexible

observing scheme capable of responding to changes in the

optical polarization of a source without human interven-

tion. In this paper we describe the data reduction pipeline

and control system developed to meet these requirements.

It is organized as follows. The telescope and instrument

are described in Section 2. The data reduction pipeline

and its performance are described in Section 3, and the

control system is described in Section 4. We conclude in

Section 5.

©

2013 RAS

arXiv:1310.7555v2 [astro-ph.IM] 7 Mar 2014

O. G. King et al.

0°

67.5°

CCD

(inset)

Figure 1.

Diagram showing the basic operation of the

RoboPol instrument. The pupil of the instrument is split in

two, each half incident on a half-wave retarder followed by a

Wollaston prism, labelled

λ/

2 and WP respectively, with dif-

fering fast axis and prism orientations as indicated. The blue

pair split the rays horizontally to produce the spots labelled

2 and 3, while the red pair produce the vertical spots 0 and

1. The linear polarization parameters are then calculated us-

ing Eqn. 1. (inset) The pattern of spots at each position on

the CCD is described by this model. The distance between the

spots ∆

and ∆

, their distance from the intersection point

and

, and their angle with respect to the CCD axes

and

all vary independently across the field.

2 TELESCOPE AND INSTRUMENT

2.1 Telescope

The 1

3-m telescope at the Skinakas Observatory (1750 m,

◦

′

′′

E, 35

◦

′

′′

N) has a modified Ritchey-

Chr ́etien optical system (129 cm primary, 45 cm sec-

ondary,

54). It has an equatorial mount, built by

DFM Engineering

, with an off-axis guiding system. The

telescope is equipped with several other instruments in

addition to the RoboPol polarimeter, including an imag-

ing camera, IR camera, and spectrograph.

Control of the telescope and its subsystems is spread

over several computers. The guiding camera, its focus con-

trol, the RoboPol filter wheel, and the RoboPol CCD cam-

era are connected directly to the main control computer.

The secondary mirror focus control, the dome control, and

the equatorial mount control are connected to the tele-

scope control system (TCS) computer, which interfaces

with the main control computer through a serial link. A

third computer monitors the weather station.

2.2 Instrument

The RoboPol instrument (A.N. Ramaprakesh et al., in

prep) is a 4-channel imaging photopolarimeter designed

with high observing efficiency and automated operation

as prime goals. It has no moving parts other than a filter

wheel. Instead, as shown in Fig. 1, the instrument splits

the pupil in two – each half incident on a half-wave re-

tarder followed by a Wollaston prism (WP). One prism

is oriented such that it splits the rays falling on it in the

horizontal plane (blue prism and rays in Fig. 1), while the

other prism’s orientation splits them in the vertical plane

(red in Fig. 1).

Every point in the sky is thereby projected to four

http://www.dfmengineering.com/

points on the CCD. The fast axis of the half-wave re-

tarder in front of the first prism is rotated by 67

◦

with

respect to the other retarder. In the instrument reference

frame the horizontal channel measures the

U/I

frac-

tional Stokes parameter, while the vertical channel mea-

sures the

Q/I

fractional Stokes parameter, simultane-

ously, with a single exposure. This design eliminates the

need for multiple exposures with different half-wave plate

positions, thereby avoiding systematic and random errors

due to sky changes between measurements and imperfect

alignment of rotating optical elements.

The expressions for the relative Stokes parameters

and their uncertainties are (see A.N. Ramaprakash et al.

(in prep.) for the derivation):

−

, σ

√

)

(

)

−

, σ

√

)

(

)

(1)

where

,...,N

are the intensities of the upper, lower,

right and left spots, as shown in Fig. 1, and

,...,σ

are

their uncertainties. We estimate the uncertainty in a spot

intensity

following the method outlined in Laher et al.

(2012):

√

sky

phot

sky

phot

sky

(2)

where

is the spot intensity,

sky

is the sky in-

tensity (background) in a single pixel,

phot

is the area (in

pixels) of the photometry aperture, and

sky

is the area

of the background estimation annulus (see Section 3.3.2).

The first two terms account for counting statistics of the

source and sky, while the third describes the uncertainty

in the background estimation.

The instrument has a large 13

′

field of view

that enables relative photometry using standard cata-

log sources and the rapid polarimetric mapping of com-

pact sources in large sky areas. While the instrument is

designed to operate in the optical

, and

-bands,

RoboPol monitoring observations are generally made us-

ing a Johnson-Cousins

-band filter. An example of an

image from the instrument is shown in Fig. 2.

The primary scientific goal of the project is to mon-

itor the linear polarization of blazars, which appear as

point sources at optical wavelengths, so we optimized the

instrument sensitivity for a source at the centre of the

field by using a mask in the telescope focal plane. The fo-

cal plane mask has a cross-shaped aperture in the center

where the target source is placed. The focal plane area

immediately surrounding this aperture is blocked by the

mask. This prevents unwanted photons from the nearby

sky and sources from overlapping with the central tar-

get spots on the CCD, increasing the sensitivity of the

instrument for the central source. The sky background

level surrounding the central target spots is reduced by a

factor of 4 compared to the field sources. The focal plane

mask and its supports obscure part of the field, reducing

the effective field of view.

The polarimeter is attached to an Andor DW436

CCD camera which has an array of 2048

2048 pixels

and can be cooled to

−

°C

where it has negligible dark

noise (

001 e

−

pixel

−

©

2013 RAS, MNRAS

000

, 1–11

RoboPol Pipeline and Control System

Figure 2.

An example of a RoboPol image. Each point in

the sky has been mapped to four spots on the CCD. A focal

plane mask, held in place by four support legs, reduces the sky

background level for the central target.

2.2.1 Model of the instrument

Inspection of Fig. 2 reveals that the pattern of four spots

on the CCD corresponding to a source is dependent on the

location of the source in the field. This is expected, and is

due to optical distortions in the instrument. In addition

to this geometric spot-pattern effect, there are systematic

errors that affect the intensity in each spot. For an un-

polarized source, the number of photons falling on each

spot should be equal. However, unavoidable imperfections

in the optics result in deviations of the ratios

and

from 1, and

In our model (detailed in Appendix A), the measured

intensities (

,...,N

) are dependent on the location of

the source on the CCD (

x,y

), and are related to the true

intensities (

∗

,...,N

∗

) by:

=[1

−

(

x,y

)]

(

)

(

)

∗

=[1 +

(

x,y

)]

(

)

(

)

∗

=[1

−

(

x,y

)]

(

)

(

)

∗

=[1 +

(

x,y

)]

(

)

(

)

∗

(3)

Here

(

x,y

) and

(

x,y

) are functions that describe

the instrumental polarization errors – they are the only

terms that remain in the calculation of

and

, Eqn. 1.

The functions

(

), and

(

) describe the in-

strumental photometry errors: the position and prism de-

pendent optical transmission of the instrument. The form

of the error functions – and their dependence on either

, or both – were determined by inspection of data

from unpolarized standard stars. The residuals between

the data and the instrument model are uniformly dis-

tributed across the field, indicating that the model ade-

quately describes the spatial dependence and scale of the

action of the instrument.

The model also predicts the pattern that the spots

make on the CCD. As shown in Fig. 1, we model the

distance between the spots ∆

and ∆

, the angle between

the spots and the CCD axes

and

, and the distance

of the spots from the intersection of their joining lines

and

. This is modelled at every point in the field, and

is used by the pipeline to identify which spots correspond

to which astronomical source. To produce the model we

take multiple exposures of an unpolarized standard star

at many locations in the field and map the variation in the

non-ideal behavior. We fit the model to the measured spot

pattern and intensities and save the model coefficients to

disk for use by the data reduction pipeline. We re-fit the

model parameters each time the instrument is removed

and replaced.

3 PIPELINE

3.1 Overview

The RoboPol pipeline measures the magnitude and linear

polarization of every unobscured source in the field, i.e.

every source that is not obscured by the focal plane mask

and its supports. A flow-diagram of the pipeline is shown

in Fig. 3. The pipeline is written in Python, with some

subroutines written in Cython

to improve the processing

time. The operation of the pipeline can be described in

five basic steps:

(i)

Source identification

, Section 3.2: Find all the

spots on the CCD, match them up to sources in the sky,

solve for the world coordinate system (WCS) that de-

scribes the image, and calculate the source coordinates

from the spot coordinates.

(ii)

Photometry

, Section 3.3: Perform aperture pho-

tometry on each of the spots.

(iii)

Calibration

, Section 3.4: Use the instrument

model to correct the measured spot intensities.

(iv)

Polarimetry

, Section 3.5: Measure the linear po-

larization of every source in the field.

(v)

Relative photometry

, Section 3.6: Measure the

-band magnitude of every source in the field by perform-

ing relative photometry using field sources.

3.2 Source identification

Every point in the sky or focal plane is mapped to four

points on the CCD by the RoboPol instrument. The first

step in the pipeline is to identify which spots on the CCD

correspond to which source in the sky, i.e. to reverse the

7→

4 mapping of the instrument.

We use SExtractor (Bertin & Arnouts 1996) to find

the pixel coordinates of the centre of every spot on the

CCD. After finding the location of the mask, and dis-

carding spots whose photometry aperture is obscured by

it, we find all sets of spots on the CCD that originate from

the same astronomical source (described in Section 3.2.1

below).

We use the central point, defined as the intersec-

tion of the line that joins the vertical spots and the line

that joins the horizontal spots, for each set of four spots

to determine the WCS that describes the image using

the Astrometry.net (Lang et al. 2010) software. We then

use this WCS to transform the central pixel coordinate

for each set of four spots to a J2000 coordinate. The

Astrometry.net software bases its astrometry on index

files that are calculated from either the USNO-B catalog

(Monet et al. 2003) or the 2MASS catalog (Skrutskie et al.

2006); the RoboPol pipeline uses the 2MASS-derived in-

dex files

. We have found that the astrometry solutions

for the target source are within 3 arcseconds of the catalog

position 90% of the time, independent of seeing conditions

or position on the sky.

http://cython.org/

Downloaded from

http://data.astrometry.net/4200/

©

2013 RAS, MNRAS

000

, 1–11

O. G. King et al.

Figure 3.

Flow chart representation of the operation of the pipeline. The FITS image from the CCD is first processed in a source

identification step (Section 3.2), in which we match spots to sources in the sky and calculate their astronomical coordinates. We

then perform photometry on all the identified sources (Section 3.3). The measured spot counts are corrected for instrumental errors

in the calibration step (Section 3.4) before we measure the linear polarization (Section 3.5) and relative photometry (Section 3.6).

3.2.1 Spot matching method

A typical RoboPol exposure, such as the example shown

in Fig. 2, contains a large number of sources in the field.

Each source forms four corresponding spots on the CCD.

We describe here a method for determining which spots

on the CCD correspond to which source in the sky, i.e.

a method for finding sets of four spots automatically. We

use our knowledge of the expected spot pattern from the

instrument model to do this.

Suppose we have found

spots on the CCD, with

pixel coordinates (

)

,...,

(

). For each spot, we

can then use the instrument model (Section 2.2.1 and

Appendix A) to predict the location of the intersection of

the line joining the vertical spot pair and the line joining

the horizontal spot pair, i.e. the central point. However,

this requires us to know what type of spot each spot is,

i.e. 0

2, or 3. Since we do not know this a priori, we

calculate where the central point would be in each of these

four cases, producing four potential central points for each

spot arranged above, below, left, and right of the spot on

the CCD.

We then have a set of 4

predicted central points.

The four spots which correspond to a particular source

will have the same predicted central point. We search the

set of predicted central points for groups of four points

that lie within a threshold of

∼

3 pixels of each other, to

account for centroid and model errors.

3.3 Photometry

After the spots have been detected and matched to

sources in the sky, we measure the intensity of each

spot using aperture photometry. We calculate the mean

FWHM across the field using 10 spots that are bright,

unblended, and unsaturated. We fit both a Gaussian and

a Moffat profile (Moffat 1969), and use the FWHM esti-

mate from the best-fit profile.

3.3.1 Mask detection

We must find the exact location of the mask in order to

perform aperture photometry on the central target, and

to identify and reject sources whose photometric aperture

Figure 4.

An image of the central area of the RoboPol field.

The four low background areas due to the focal plane mask,

containing the four spots of the central science target, are lo-

cated in the centre of the field. The edges of the mask pattern

are indicated in green. The yellow squares are the background

estimation boxes for the central science target, the red diamond

is the mask centre, and the blue cross is the pointing centre.

Optical distortions result in the mask centre and pointing cen-

tre being slightly different: the pointing centre offset from the

mask centre was determined empirically.

intersects with the mask. We find the position of the mask

by fitting the known mask pattern to the image. This is

done by finding the mask pattern position that maximises

the difference in background level between stripes of pixels

on either side of the mask pattern edge. Pixels contam-

inated by bright sources located near the mask pattern

edge are excluded from the procedure.

Knowing the geometry of the mask, we can then iden-

tify the location of the low background areas, squares

′′

in size in which the central target should be

located. An image of the central area from a RoboPol

image, with the mask pattern and low background areas

©

2013 RAS, MNRAS

000

, 1–11

RoboPol Pipeline and Control System

RoboPol intensity [ADU]

−

Relative difference, APT [%]

Figure 5.

Comparison of the source intensity measured by the

RoboPol pipeline with that measured by Aperture Photome-

try Tool (APT). The APT used the same aperture settings as

used by the RoboPol pipeline. The median difference is 0.04%.

The red points are the 1

uncertainties in the RoboPol intensi-

ties, calculated using Poisson statistics. The outlier points are

sources close to the focal plane mask: the RoboPol pipeline

removes pixels in the focal plane mask from the background

estimation annulus, while APT does not.

outlined, is shown in Fig. 4. The mask detection is also

used in the target acquisition procedure outlined in Sec-

tion 4.3.

3.3.2 Aperture photometry

For the field sources we use circular apertures centred on

each spot to measure the intensity and an outer annulus

to estimate the background level. We use the SExtrac-

tor positions of the spot centres, and flag blended spots

(these are currently not analysed by the pipeline). The fo-

cal plane mask restricts the area around the central target

that can be used to estimate the background level, so we

use a square aperture, as indicated by the yellow boxes

in Fig. 4, for the central target spots to maximize the

number of pixels used in the background estimation. The

location of the square apertures is set by the location of

the mask, regardless of the location of the source spots.

We estimate the background level using a method

outlined in Da Costa (1992). The background level is the

mode of the smoothed distribution of all pixels that have

an intensity within 3

of the median level in the aperture.

We evaluated the performance of the RoboPol aper-

ture photometry code by comparing its output to the

output of the Aperture Photometry Tool (APT) software

(Laher et al. 2012). We ran APT in batch mode on a set of

RoboPol images. We fixed the photometry apertures used

by APT to match those used by the RoboPol pipeline.

The results are shown in Fig. 5, where we plot the rel-

ative difference between the RoboPol intensity and that

obtained by APT. The APT and RoboPol results have

excellent agreement, with a median relative difference of

0.04%. The outlier points are due to errors in the back-

ground level estimation due to proximity of the source to

the focal plane mask.

3.4 Calibration

We correct the measured spot photometry for known in-

strumental measurement errors before we calculate the

linear polarization and relative photometry. We use the

instrument model corrections (Section 2.2.1) to correct

−

0

8

−

0

6

−

0

4

−

RBP

−

CAT

,[%]

Jun 2013

Jul 2013

Aug 2013

Sep 2013

Oct 2013

−

2

0

2

4

6

8

10

χ

RBP

−

χ

CAT

◦

]

HD 215806

HD 236633

Hiltner 960

VI Cyg #12

BD+64

◦

106

HD 155197

HD 204827

Figure 6.

Light curves for a selection of polarization standard

stars. We plot the difference between the RoboPol measured

linear polarization percentage (

RBP

) or angle (

RBP

), and the

catalog value (

CAT

,χ

CAT

). The mean values for each stan-

dard are listed in Table 1, including references for the catalog

values.

the measured spot counts

...

and obtain the corrected

spot counts

...

−

(

)]

(

)

(

)

[1 +

(

)]

(

)

(

)

−

(

)]

(

)

(

)

[1 +

(

)]

(

)

(

)

(4)

where (

) is the intersection of the lines joining the

vertical spot pair and the horizontal spot pair on the

CCD, the central point.

3.5 Polarimetry

The relative linear Stokes parameters

and

are cal-

culated with Eqn. 1 using the corrected spot counts from

Eqn. 4. The linear polarization fraction

and electric vec-

tor position angle (EVPA)

are then calculated using:

√

, σ

√

(5)

tan

−

(

)

, σ

√

(

)

(6)

If the polarization of the source is low, i.e. (1 +

)

(1 +

)

1, then the expression for the EVPA uncertainty

can be written as

SNR

(7)

i.e. the uncertainty in the measurement of the EVPA is

determined by the signal to noise ratio (SNR) of the po-

larization fraction

measurement SNR

We tested the performance of the RoboPol pipeline

by observing a number of polarized standard stars with

©

2013 RAS, MNRAS

000

, 1–11

O. G. King et al.

known polarization properties in the Johnson-Cousins

band. The polarization standards we observed are listed

in Table 1. In Fig. 6 we plot light curves of the differ-

ence between the polarization fraction measured by the

RoboPol pipeline and the catalog value, and the difference

between the RoboPol polarization angle and the catalog

value. No de-biasing has been applied, as the SNR of each

measurement is large (

20:1). There is no systematic dif-

ference between the RoboPol polarization percentage and

the catalog value: the mean difference in polarization per-

centage is (3

−

. The polarization angles measured

by the RoboPol instrument are on average 2

◦

larger than the catalog angle. This is due to a rotation of

the telescope polarization reference frame with respect to

the sky.

3.6 Relative photometry

We measure the brightness of the objects in the RoboPol

field relative to a set of non-variable reference sources

in the field. This requires a reliable reference photomet-

ric catalog. There are two catalogs that have significant

overlap with our sources, the PTF (Palomar Transient

Factory)

-band catalogue (Ofek et al. 2012b) and the

USNO-B1.0 catalog (Monet et al. 2003). However, we

have found the USNO-B1.0 magnitudes to be unsuitable

for use as photometric standards due to their marginal

photometric quality.

The PTF

-band catalogue magnitudes are of a high

quality, with very low systematic errors of

∼

02 mag,

but the data were taken using a Mould

filter and the

resultant catalog magnitudes are in the PTF photometric

system (Ofek et al. 2012a). We transform the

PTF

mag-

nitude to the Johnson-Cousins system using the transfor-

mation provided in Ofek et al. (2012a), Equation 6. Since

we do not know the color of each object in the field a-

priori, we use the median color term

c,R

= 0

214 for all

sources in the PTF catalog to obtain the transformation:

PTF

+ 0

086

(

−

) + 0

124

(8)

We then used SDSS data

to study the colors of the PTF

reference objects. We found 13,091 PTF reference sources

with corresponding SDSS

−

colors, with the mean of

the color distribution being 0.18 and the width (standard

deviation) 0.19. We therefore ignore the negligible color-

dependent part of the transformation and use the rela-

tionship:

PTF

−

124

(9)

This approximate photometric transformation will be-

come redundant when we complete our catalog of

Johnson-Cousins reference magnitudes, as discussed in

Section 5.

We identify all reference sources in the RoboPol

frame that are uncontaminated, i.e. that are not blended

sources and that do not have any sources in their back-

ground estimation annulus. We find their catalog magni-

tude and convert it to a flux using the zero point for the

Johnson-Cousins photometric system. We find the best-

fit line to the total source intensity (sum of the four spot

intensities

∑

) vs flux for the reference sources, and

http://cas.sdss.org/astro/en/tools/crossid/upload.

asp

−

Magnitude difference:

−

RBP

Number of sources

Expected

Data

Figure 7.

Distribution of the difference between the RoboPol-

measured

-band magnitude

RBP

and the PTF

-band mag-

nitude (corrected using Eqn. 9)

. The red curve shows the

distribution of the expected magnitude uncertainty calculated

RBP

(where

is the source intensity and

is its

uncertainty), mirrored around 0. The difference in the magni-

tudes is consistent with the level expected from photon count-

ing statistics; no systematic difference is evident.

use this relationship to convert the total intensity for all

the sources in the frame to an

-band magnitude. We

measure the standard deviation of the difference between

the RoboPol magnitude and the catalog magnitude for

the reference sources, and call this the “standards” uncer-

tainty. This systematic uncertainty is the same for every

source in the field. The uncertainty in the magnitude of

a source is then the quadrature sum of the statistical un-

certainty for that source (SNR in intensity measurement)

and the “standards” uncertainty.

In Fig. 7 we show the distribution of the difference

between the RoboPol

-band magnitude and the PTF

catalog

-band magnitude (corrected using Eqn. 9) for

a set of RoboPol field sources. The magnitudes are very

similar, and the scatter in the difference is consistent with

that expected from the SNR in the spot photometry mea-

surement.

4 CONTROL SYSTEM

4.1 Overview

The RoboPol control system is designed with high ob-

serving efficiency and dynamic scheduling as prime goals.

High efficiency is achieved by full automation of the ob-

serving process, and dynamic adjustment of the exposure

time for a target to reach a specified SNR goal.

The control system operates the Skinakas 1

3-m tele-

scope robotically during RoboPol observing sessions, and

allows full manual control of the telescope the rest of the

time. As described in Section 2.1, the control of the tele-

scope subsystems is spread over several computers run-

ning a variety of operating systems. The RoboPol con-

trol system is written in Python and consists of a num-

ber of independent processes running on these computers,

communicating with each other over ethernet using TCP

sockets.

A simplified flow chart of the main observing loop

in the master control process is shown in Fig. 8. Some of

the other independent processes are shown in purple. The

control system processes are:

Master

: Control the observing process.

©

2013 RAS, MNRAS

000

, 1–11

RoboPol Pipeline and Control System

Table 1.

Comparison of the RoboPol pipeline results for a set of polarized standard stars observed in the

-band (linear polarization

percentage

RBP

,σ

RBP

and position angle

RBP

,σ

χ,

RBP

) and their catalog values (subscript CAT). The values listed here are the

unweighted means of the measurements shown in Fig. 6. Catalog values are from:

, Schmidt, Elston & Lupie (1992),

, Whittet

et al. (1992).

Source

RBP

[%]

RBP

[

◦

]

χ,

RBP

CAT

[%]

CAT

[

◦

]

χ,

CAT

VI Cyg #12

7.78

0.05

119.2

0.2

7.893

0.037

116.23

0.14

HD 236633

5.22

0.23

95.4

1.3

5.376

0.028

93.04

0.15

Hiltner 960

5.45

0.08

56.4

0.4

5.210

0.029

54.54

0.16

BD+64

◦

106

5.19

0.10

98.0

0.6

5.150

0.098

96.74

0.54

HD 204827

5.29

0.06

61.6

0.3

4.893

0.029

59.10

0.17

HD 155197

3.92

0.09

104.5

0.7

4.274

0.027

102.88

0.18

HD 215806

1.96

0.09

69.6

1.3

1.830

0.040

66.00

1.00

Figure 8.

Flow chart representation of the main observing loop. The observing loop gets the next object from the scheduler process

(Section 4.2) and instructs the telescope control process to slew to the target. A target acquisition loop (Section 4.3) then ensures

that the science target is centred in the mask. The source is then observed until the SNR goal is reached (Section 4.4). The loop

then acquires the next object from the observing queue.

Scheduler

: Provide the next object to observe (Sec-

tion 4.2).

Pipeline queue

: Analyse the FITS images from the in-

strument and provide the science target magnitude and

linear polarization to the master and scheduler processes.

Gamma-ray data pipeline

(not shown): Process the

gamma-ray data provided by the

Fermi

LAT telescope

offline and provide the latest data to the scheduler pro-

cess.

Telescope control

: Interface with the mount, dome,

and focus control through the TCS computer, control of

the RoboPol filter wheel and CCD.

GUI

(not shown): A graphical interface to the control

system to provide the telescope operator with feedback

and allow manual intervention if necessary.

Weather

(not shown): Monitor a weather station to

provide information to the watchdog processes and for

logging.

Watchdogs

(not shown): Monitor and maintain the

stability of the control system.

In addition to the fully-automated main observing

loop, the control system runs an automated focus routine

(Section 4.5) several times during the night, automati-

cally acquires flat-field exposures (Section 4.6) to moni-

tor dust contamination of the optics, and has a target-of-

opportunity mode that can interrupt the main observing

loop to observe, for instance, gamma-ray burst optical af-

terglows.

All exposures made by the control system are stored

on disk at the telescope and transferred once a day to

servers at the University of Crete. From there the data are

distributed over the internet to the partner institutions

for redundant backup. A database of light curves for all

the sources in every RoboPol field is maintained at the

University of Crete.

4.2 Dynamic scheduling

The RoboPol control system is designed to allow dynamic

scheduling. At the start of each night the scheduler pro-

cess produces a nominal schedule of the sources from the

RoboPol catalog that are due to be observed. As each

source is observed its measured magnitude and linear po-

larization are passed to the scheduler process to allow

©

2013 RAS, MNRAS

000

, 1–11

O. G. King et al.

100

200

300

400

500

600

700

Initial pointing offset [arcseconds]

000

001

002

003

004

005

006

007

008

Fraction of observations

Number of pointing corrections

Figure 9.

The initial pointing offset of the field centre from

the commanded position. (inset) About 90% of sources require

two or fewer pointing corrections to be properly centred.

changes to the schedule to be made, if necessary. This

dynamic response mode is not being used in the first

observing season while we gather the data necessary to

characterize the behavior of our sources and develop the

algorithms to reliably identify interesting behavior. De-

tails of the dynamical scheduler will be reported in future

papers.

4.3 Target acquisition

The pointing requirements for the RoboPol instrument

are very stringent: we require the science target to be

within 2

′′

of the pointing centre of the mask. We cannot

achieve this precision with a blind slew to a source, so

the control system contains a target acquisition loop to

centre the source in the mask before taking the science

exposures.

After the initial telescope slew, the control system

takes a short exposure of the field. This is processed using

the pipeline to find the mask location and to calculate the

WCS that describes the frame. A pointing correction that

would place the target coordinates at the mask pointing

centre is calculated. This correction is applied and another

short exposure is taken. This loop is repeated until the

target source is properly located. The performance of the

target acquisition system is shown in Fig. 9. Most initial

slews are within

∼

′

of the commanded position, and

∼

90% of sources require two or fewer pointing corrections

to be properly placed in the mask.

It is not necessary for the central target source to be

visible in a single exposure for this procedure to work. As

long as there are enough stars in the field for the pipeline

to calculate the WCS for the frame, the location of the

source in the field can be calculated and the appropriate

pointing correction applied.

4.4 Dynamic exposure time

Both the polarization and magnitude of blazars are highly

variable at optical wavelengths. For greater observing effi-

ciency we expose only long enough to reach a target SNR

of 10:1 in

, which equates to an uncertainty in the EVPA

∼

◦

(see Eqn. 7). Because the blazar emission can

change significantly from night to night (and even within a

night), we calculate the necessary exposure time to reach

our SNR goal from the data as we gather it.

We use the final target acquisition exposure to pro-

vide an initial guess for the required exposure time.

We calculate the amount of time needed to collect

250,000 photons in total from the source, which we have

found gives an SNR in

∼

10:1 in a

∼

3% polarized

source under average observing conditions at Skinakas.

We then take a number of science exposures; as the sci-

ence exposures are accumulated we run the pipeline on the

stacked image and update the estimate of the required

observing time. We stop observing once the SNR goal

is reached, or when the total exposure time has reached

40 minutes.

4.5 Autofocus

The RoboPol instrument is optimized to measure the lin-

ear polarization of point sources. The control system con-

tains an autofocus mode that takes a series of exposures

at different focus positions. It then finds the focus position

that produces the lowest median FWHM across the field.

While the FWHM does vary across the field, the mini-

mum in the median FWHM corresponds to the same fo-

cal position as the minimum in the FWHM of the central

target, and the curve of median FWHM vs focus position

has lower noise than the curve for a single source. This

procedure is run at the beginning and mid-way through

each night.

4.6 Autoflats

The control system automatically takes flat-field expo-

sures at dawn or dusk, which are used to track the pres-

ence of dust in the telescope optics and its effect on the

performance of the instrument.

We select an observing location for the flat-field expo-

sures by requiring that the distance of the target flat-field

sky area from the Moon be more than 50

◦

and that the

distance from the horizon be more than 40

◦

, thereby lim-

iting the gradient of the background to

1% across our

field (Chromey & Hasselbacher 1996). The control system

selects as the target sky area the point on the line of dec-

lination

= +32

◦

that meets these criteria and has the

greatest summed distance from the Moon and the hori-

zon.

According to Tyson & Gal (1993) the logarithm of

the brightness of the sky changes linearly with time, with

possible deviations due to atmospheric dust. We have

found that the sky brightness light curve is better de-

scribed by a 2

order polynomial. We take a series of

short exposures of the sky every 120 s to characterize the

median sky brightness light curve. Once the changing sky

brightness is adequately characterized, we calculate the

optimum time to start taking the flat-field exposures such

that we get a median background count of

∼

10000 ADU

per pixel (

∼

3 of the non-linear point for this CCD)

in the first flat-field exposure. We then take a series of

−

10 s exposures while varying the pointing location of

the telescope, which is used to calculate the master flat-

field image.

5 CONCLUSIONS

We have described the data reduction pipeline and con-

trol system developed for the RoboPol project. We have

©

2013 RAS, MNRAS

000

, 1–11