source: trunk/admin/requirements1.tex

\documentclass[11pt]{article}
\usepackage{a4}
\usepackage{calc}
\usepackage{smartref}
\usepackage[dvips]{graphicx}

% Adjust the page size
\addtolength{\oddsidemargin}{-0.4in}
\addtolength{\evensidemargin}{+0.4in}
\addtolength{\textwidth}{+0.8in}

\setlength{\parindent}{0mm}
\setlength{\parskip}{1ex}


\title{ATNF Single Dish Spectral Line Software Requirements }
\author{Chris Phillips \& Malte Marquarding}

\newcounter{requirement}
\addtoreflist{requirement}
\newcommand{\reqref}[1]{R\ref{#1}-\requirementref{#1}}

\newcommand{\makenote}[1]{{\bf \tt \em#1}}

\newcommand{\anitem}[2]{\smallskip \parbox[t]{2cm}{#1}\parbox[t]{\textwidth-2cm}{#2}}

\newcommand{\showreqcounter}{{\stepcounter{requirement}
    \bf R\arabic{section}.\arabic{subsection}-\arabic{requirement}}}

\newcommand{\requirement}[2]{
  \hspace*{2mm}\begin{minipage}{\textwidth-2mm}
    \setlength{\parindent}{-2mm}
    \showreqcounter\ #1 \\ 
    \hspace*{1cm} {\em Priority #2}
  \end{minipage}
}

\newcommand{\extendedrequirement}[2]{
  \hspace*{2mm}\begin{minipage}{\textwidth-2mm}
    \setlength{\parindent}{-2mm}
    \showreqcounter\ #1 
    \hspace*{1cm} {\em Priority #2}
  \end{minipage}
}

\newcommand{\reqeqn}[1]{\\\hspace*{1cm} $#1$}

\let\oldsection\section
\renewcommand{\section}[1]{\setcounter{requirement}{0}\oldsection{#1}}

\let\oldsubsection\subsection
\renewcommand{\subsection}[1]{\setcounter{requirement}{0}\oldsubsection{#1}}

\begin{document}

\maketitle

\section{Introduction}

The spectral line single-dish software {\tt spc} at the ATNF has
become increasingly difficult to maintain. There also have been many
requests for features which are not possible to implement without
major change to this package. The interface is based on outdated
operating systems and programming languages. The decision was made to
replace {\tt spc} with a new package which supports existing and
future ATNF single-dish instrumentation. A survey of users was taken
into account creating this requirements document.

\section{Scope}

The software should be able to process all spectral line single-dish
observations from ATNF telescopes (Parkes, Mopra \& Tidbinbilla). This
includes reading the data produced by the telescope, calibration and
reduction of the data and basic analysis of the data such as fitting
line profiles.

It has been assumed that the following processing is out of the scope
of the new software.
\begin{itemize}
\item Raster or ``on-the-fly'' mapping (This is handled by
``livedata'' and gridzilla).
\item Very complex or specific data processing. (A route into
  Class\footnote{Part of the GLIDAS software package, produced by
  Institut de Radio Astronomie Millime\'trique http://www.iram.fr}
  should be available for advanced processing).
%%TODO%% give example
\item Continuum data.
\item Pulsar timing observations.
\end{itemize}

\section{Priorities}

Requirements have been given a value of 0 to 3. The ``0'' priority
functions will be implemented in a demonstrator version of the
software (with no GUI interface). The other requirements will be
implemented mainly depending on priority, with ``1'' the
highest. Priority 3 and some priority 2 requirements will probably not
get implemented in the first released version of the software.

\section{User Interface}

The user interface (UI) is the most important part of a single dish
processing package, but probably the most difficult to get right. The
UI for this software will consist of three parts.
\begin{itemize}
  \item A graphical user interface (GUI).
  \item An interactive command line interface (CLI).
  \item A scriptable interface for batch processing.
\end{itemize}

The CLI and scriptable interface may essentially be the same.

The software does not {\em need} to be able to run solely from a
``vt100'' style terminal. It can be assumed that the user is running
the software from within a windowed (i.e. X11) environment. This will
mean it will not necessarily be possible to run the software remotely
over a slow network connection (e.g. internationally or from home).
Where possible, operations on the data should be possible from all
three aspects of the user interface. 

The user interface needs to be implemented so that the user can easily
and transparently work on spectra either one at a time or by
processing multiple spectra in parallel. This means there must be an
easy way to select specific or multiple spectra to display or process.

\subsection{Graphical User Interface}

The GUI is intended to be the main interactive interface with the
software. 

\smallskip
\requirement{It should be simple, intuitive and
uncluttered. Specifically, use of many windows simultaneously should
be discouraged, as should hiding functionality behind layers of dialog
boxes.}{1}

\requirement{The plotting window should be a major component of the GUI
control, not a separate isolated window.}{1}

\requirement{The interface should use minimal ``always visible''
  controls, with use of pull down menus and maybe a toolbar for
  frequency used functions. }{1}

\requirement{Keyboard shortcuts should be available.}{2}

\requirement{Most user preferences (i.e. keywords in the CLI) should
be presented in a popup, tabbed, dialog box.}{2}

\requirement{When performing line profile fitting, a spreadsheet type
window should be viewable which shows the current parameter values
(amplitude, velocity etc) for each line fitted and allow the user to
change these parameters or set the current value as fixed. This gui
should stay synchronised with any CLI changes to these values.}{2}

\subsection{Command Line Interface}

\requirement{While the GUI should be the main interface for new users
and for basic manipulation, some tasks can be more efficiently
performed using a CLI. A virtual CLI could be integrated as part of
the GUI.}{1}

\requirement{The CLI should have a keyword/argument form and never
prompt the user for specific values (the user should be able to change
values which are retained until they wants to change them again).}{1}

\requirement{The CLI should be case insensitive and accept minimum
matching and short forms of keywords.}{2}

\requirement{The user must be able to quickly and easily see from the
command line the available routines and keywords which affect it, so
they can see which parameters may need changing.}{1}

\subsection{Scripting}

\requirement{It must be possible to run the software in a scripting
mode. This would be to process large amounts of data in a routine
manner and also to automatically reproduce specific plots etc (So the
scripting must have full control of the plotter). Preferably the
scripting ``language'' and the CLI would be the same. }{1}

\requirement{It would be worthwhile having a method to auto-generate
scripts (for reduction or plotting) from current spectra history, or
some similar method.}{3}

\section{Plotter}

The plotter should be fully interactive and be an integral part of the
GUI and software interface.

\requirement{It must be able to produce plots of publishable
quality. This includes being able to specify line thickness, character
size, colours, position of axis ticks, axis titles etc and producing
hard copies in postscript and .png format.}{0}

\requirement{It must be possible to flexibly select the data to plot
  (e.g. Tsys vs time etc as well as plots such as amplitude vs channel
  number or velocity). Preferably any of the header values for a
  selection of scans could be plotted on a scatter plot (e.g. Tsys vs
  elevation)}{2}

\requirement{It must be possible to overlay multiple spectra on a
  single plot using different colours and/or different line
  styles. (Including multiple stokes data and multiple IFs).}{1}

\requirement{It must be possible to plot either the individual
  integrations (in either a stacked fashion, or using a new subplot
  per integration) or some type of average of all the integrations in
  a scan.}{2}

\requirement{Multi-panelling of spectra in an n$\times$m size grid. If
  more spectra than can fit on the plot matrix are to be plotted, then
  it must be possible to step back and forth between the viewable
  spectra (i.e. ``multi-page'' plots). It must be possible to quickly
  and easily change the number of plots per page, and define the ``n''
  and ``m'' values.}{1}

\requirement{When using multi-panelling, it should be possible to
  easily change the display to flip between a single spectra display
  and multi-panels display (i.e. zoom into a specific sub-window).}{3}

\requirement{It must be possible to interactively zoom the plot
  (channel range selection and amplitude of the spectra etc.)  This
  includes both GUI control of the zooming as well as command line
  control of either the zoom factor or directly specifying the zoom
  bounds. }{1}

\requirement{On a single plot, it should be possible to plot the full
  spectrum and a zoomed copy of the data (using a different lie style)
  to see weak features. The user must be able to specify the zoom
  factor.}{3}

\requirement{Optionally when stacking multiple spectral plots in one
  subwindow, a (user definable) offset in the ``y'' direction should
  be added to each subsequent spectra.}{2}

\requirement{The plotter should automatically update to reflect user
processing, either from the CLI or GUI. The user should have to option
to turn this feature off if they so wish.}{2}

\requirement{It should be possible to plot individual integrations
(possibly from multiple scans) in a ``waterfall'' plot. This is an
image based display, where spectral channel is along the x-axis of the
plot, time (or integration number) along the y-axis and greyscale or
colour represent the amplitude of spectra. Interactive zooming and
panning of this image should be supported. }{3}

\requirement{When plotting ``waterfall'' plots, it should be possible
to interactively select regions or points and mark them as invalid
(i.e. to remove RFI affected data). The plotter should also show the
time/velocity of the pixel beneath the cursor.}{3}

\requirement{It should be possible to export the ``waterfall'' plot
images as a FITs file, for user specific analysis.}{3}

\requirement{Line markers overlays, read from a catalogue should be
  optionally available. This would include the full Lovas catalogue,
  the JPL line catalogue, radio recombination lines and a simple user
  definable catalogue. The lines would be Doppler corrected to the
  line velocity. The user must be able to plot just a sub-section of
  the lines in any specific catalogue (to avoid clutter).}{2}

\requirement{Optionally plot fitted functions (e.g line profiles or
  baseline fit). If multiple components (e.g. Gaussians) have been
  fit, it should be possible to show the individual functions or the
  sum of the components}{1}

\requirement{It should be possible to plot the residual data with or
  without subtraction of fit functions. This includes plotting the
  spectra with or without baseline removal and the residual after
  subtracting Gaussian fits. The default should be to plot the data
  with baseline subtracted but profile fits not subtracted.}{2}

\requirement{Basic header data (source name, molecule, observation
  time etc) should be optionally shown, either on the plot or next to
  it. This may either consist of a set of values, or only one or two
  values the user specifically wants to see (source name and molecule,
  for example). Preferably the user would be able to define where on
  the plot the header info would appear.}{2}

\requirement{Optionally, relevant data such as the current mouse
  position should be displayed (maybe with a mode to display an
  extended cross, horizontal or vertical line at the current cursor
  position).}{2}

\requirement{The user should be able to define simple
  annotations. This would include text overlay and probably simple
  graphics (lines, arrows etc).}{3}

The user must be able to use the plotter window to interactively set
initial values and ranges used for fitting functions etc. The use of
keyboard ``shortcuts'' or other similar ``power user'' features should
be available to save the time of experienced users. 

The plotter should be used to set the following values:

\requirement{Range of spectral points needed for specific tasks (See
 requirement \reqref{ref:chansel})}{1}

\requirement{Initial Gaussian parameters (velocity, width, amplitude)
for profile fitting.}{1}

\requirement{Change the parameter values of existing line profile
fits, or channel ranges used for baseline fits.}{3}

\section{Functionality}

\subsection{Import/export}

The software needs a set of import/export functions to deal with a
variety of data formats and to be able to exchange data with other
popular packages. These functions should be flexible enough to allow
the user to perform analysis functions in an different package and
re-import the data (or vice versa). The import function must be
modular enough to easily add new file formats when the need arises.
To properly import data, extra information may have to be read from
secondary calibration files (such as GTP, Gated Total Power, for 3~mm
wavelength data taken with Mopra). The import functions should be
flexible enough to gracefully handle data files with missing headers
etc. They should also be able to make telescope and date specific
corrections to the data (for ATNF observatories).

The software must be able to read (import) the following file formats.

\requirement{The rpfits file format produced by all current ATNF
  correlators.}{0}

\requirement{SDFITS (currently written by {\tt SPC}).}{1}

\requirement{Simple ``image'' FITS (used by CLASS}{2}

\requirement{Historic ATNF single dish formats (Spectra, SPC,
SLAP). Possibly a set of routines to translate these formats to SDFITs
would suffice.}{3}

\requirement{PSRFIT for pulsar spectroscopy.}

\requirement{For online analysis, the software should be able to read
an rpfits file which is is still currently open for writing by the
telescope backend processor.}{1}

\requirement{Data which has been observed in either a fixed frequency
or Doppler tracked fashion needs to be handled transparently.}{1}

The software should be able to export the data in the following formats.

\requirement{Single Dish FITS.}{1}

\requirement{Simple ``image'' FITS (used by CLASS. It must be possible
to to export multiple spectra simultaneously, using default file name
and postfix.}{2}

\requirement{In a format which can be imported by other popular
    packages such as Class. }{2}

\requirement{Simple ascii format, suitable for use with programs such
    as Perl, Python, SuperMongo etc.}{2}

\requirement{The exported data should retain as much header data as
possible. It should also be possible to request specific data be
written in the desired form (B1950 coordinates, optical velocity
definition etc).}{2}

\requirement{The import function should apply relevant corrections
(especially those which are time dependent) to specific
telescopes. See $\S$\ref{sec:issues} for a list of currently known
issues.}{1}

\subsection{Sky subtraction}
\label{sec:skysubtraction}
To remove the effects of the passband filter shape and atmospheric
fluctuations across the band, sky subtraction must be performed on the
data. The software must be able to do sky subtraction using both
position switching (quotient spectra) and frequency switching
techniques.

\requirement{\label{ref:skysub} Position switched sky subtraction
should be implemented using the algorithm \medskip\reqeqn{T_{ref}
\times \frac{S}{R} - T_{sig}} -- removes continuum\bigskip
\reqeqn{T_{ref} \times \frac{S}{R} - T_{ref}} -- preserves
continuum\medskip}{0}

\requirement{The user should be able to specify an arbitrarily complex
reference/source order (which repeats), which can then be used to make
perform multiple sky subtractions in parallel.}{3}

\requirement{Frequency switched sky subtraction should be
        supported. (Ref. Liszt, 1997, A\&AS, 124, 183) }{2}

%\requirement{For wideband multibit sampled data it may be desirable or
%even required to assume Tsys has a frequency dependency. Appropriate
%sky subtraction algorithms will need to be investigated.}{3}

\requirement{For pulsar binned data, the (user specified) off pulse
bins can be used as the reference spectra. Due to potentially rapid
amplitude fluctuations, sky subtractions may need to be done on a per
integration basis.}{3}

Multibeam systems can observe in a nodding fashion (called MX mode at
Parkes), where the telescope position is nodded between scans so that
the source is observed in turn by two beams and a reference spectra
for one beam is obtained while the other is observing the target source.

\requirement{For multibeam systems, it must be possible to perform sky
subtraction with the source and reference in an alternate pair of
beams}{2}

\subsection{Baseline removal}

Baseline removal is needed to correct for imperfections in sky
subtraction. Depending on the stability of the system, the residual
spectral baseline errors can be small or quite large. Baseline removal
is usually done by fitting a function to the (user specified) line
free channels.

\requirement{The software must be able to do baseline removal by
  fitting a n'th order polynomials to the line free channels using a
  least squares method.}{0}

%\requirement{The baseline fitting should be reversible, i.e. the fit
%parameters must be retained.}{2}

\requirement{Removal of standing wave ripples should be done by
fitting a Sine function to the line free channels.}{2}

\requirement{``Robust'' fitting functions should be available, which are more
  tolerant to RFI.}{1}

\requirement{Automatic techniques for baselining should be
  investigated.}{3}

\subsection{Line Profile Fitting}

The user will want to fit multicomponent line profiles to the data in
a simple manner and be able to manipulate the exact fitting
parameters.

\requirement{The software must be able to do multi-component Gaussian
  fitting of the spectra. The initial amplitude, width and velocity of
  each component should be able to be set by the user and specific
  values to be fit should be easily set.}{0}

\requirement{The reduce Chi square (or similar statistic) of the fit
should given to the user, so that they can easily see if adding extra
components give a statistically significant improvement to the fit.}{1}

%\requirement{The fit parameters should be stored with the data so that
%the user can work on multiple data sets simultaneously and experiment
%with different fitting values. These values should be saved to disk
%along with the data.}{1}

\requirement{For multiple polarisation data, the individual stokes
parameters or polarisation products should be fit independently.}{1}

\requirement{There should be an easy way of exporting the fit
parameter from multiple spectra, e.g. as an ASCII table.}{2}

\requirement{It should be also possible to do constrained fitting of
multiple hyperfine components, e.g. the NH$_3$ hyperfine
components. (The constraints may be either the frequency separation of
the individual components or the amplitude ratio etc.)}{3}

\requirement{It must be possible to alter the line profile fit
parameter values by hand at any stage.}{2}

\requirement{It must be possible to ``fix'' particular values of the
  line profile parameters, so that only subset of lines or (say) the
  width of a line is fit.}{1}

\requirement{The software should allow hooks for line profile shapes
other than Gaussian to be added in the future, possible user
specified.}{2}

%\makenote{Should it be possible to attach multiple sets of fits to the
%data (similar to CL tables in classic AIPS), so the user can
%experiment with different ways of fitting the data?}

%\makenote{Should calculations of rotational temperatures etc be
%handled when fitting hyperfine components, or should the user be doing
%this themselves?}

\subsection{Calibration}

The software should handle all basic system temperature (Tsys) and
gain calibration as well as opacity corrections where relevant. The
Tsys value should be contained in the rpfits files. The actual
application of the T$_{\mbox{sys}}$ factor will be applied as part of
the sky subtraction ($\S$\ref{sec:skysubtraction}). The units of Tsys
recorded in the data may be either in Jy or Kelvin, which will affect
how the data is calibrated. The rpfits file does {\em not} distinguish
if the flux units are Kelvin or Janskys. 

\requirement{Gain elevation corrections should be implemented using a
elevation dependent polynomial. The polynomial coefficients will be
telescope and frequency dependent. They will also have a (long term)
time dependence.}{1}

\requirement{The user may wish to supply their own gain polynomial.}{2}

\requirement{When required by the user, the spectral units must be
converted from Kelvin to Jansky. At higher (3mm) frequencies this
conversion is often not applied. The conversion factor is\medskip
\reqeqn{\mbox{Flux (Jy)} = \frac{T \times 2 k_b \times
10^{26}}{\eta A}},\medskip\\where $k_b$ is Boltzmann's constant, A is
the illuminated area of the telescope and $\eta$ is the efficiency of
the telescope (frequency, telescope and time dependent)}{1}
%%TODO%% Use planks formula here?

\requirement{In some cases the recorded Tsys values will be
wrong. There needs to be a mechanism to scale the Tsys value and the
spectrum if the Tsys value has already been applied (i.e. a simple and
consistent rescaling factor).}{2}

\requirement{The data may need to be corrected for opacity effects,
particularly at frequencies of 20~GHz and higher. The opacity factor
to apply is given by\medskip\reqeqn{C_o = e^{\tau/cos(z)}}\medskip\\
where $\tau$ is the opacity and z is the zenith angle (90-El). These
corrections will generally be derived from periodic ``skydip''
measurements. These values will not be contained in the rpfits files,
so there should be a simple way of the software obtaining them and
interpolating in time (the user should not {\em have} to type them in,
but may want to). Reading in an ascii file which contains the skydip
data along with a time-stamp would be one possibility.}{1}

\requirement{For wideband, multibit observations, the software should
have the option to handle Tsys which varies across the band. The exact
implementation will have to be decided once experience is gained with
the new Mopra digital filterbank. This will affect the sky subtraction
algorithms (requirement \reqref{ref:skysub}).}{2}

%\makenote{Is the dependence of gain on frequency weak enough for one
%set of coefficients for each receiver, or is a full frequency dependent
%set of values needed?}

%\makenote{Should it be possible to read ``correct'' Tsys values from
%an external ascii file?} 

\subsection{Editing \& RFI robustness}

In a data set with many observations, individual spectra may be
corrupted or the data may be affected by RFI and ``birdies''. The user
needs to be able to easily flag individual spectra or channels. This
may affect other routines such as sky-subtraction, as this will
disrupt the reference/source sequence.

\requirement{The user must be able to set an entire spectra or part
thereof (individual polarisation, IF etc) as being invalid. The
effected channels should either be blanked or interpolated depending
on the user wishes. When blanked data is plotted, the plotting routine
should also either interpolate the data on the fly or show a blank in
the spectrum, depending on the users preferences.}{1}

\requirement{The user must be able to indicate an individual spectral
point or range of spectral points are invalid. This should be applied
to an individual spectra, or set of spectra.}{1}

\requirement{The user should be able to plot the average spectral flux
across the band, or part of the band, as a function of time and
interactively select sections of data which should be marked as
invalid (similar to IBLED in classic aips).}{3}

\requirement{Where relevant, fitting routines etc should have the
option of selecting RFI tolerant (``robust'') algorithms. This will
require investigating alternate fitting routines other than the
least-squares approach.}{3}

\requirement{A routine to automatically find birdies or RFI corrupted
data and indicate the data as invalid would be useful.}{3}

\requirement{Other routines must be able to cope with portions of data
which are marked as invalid.}{1}

\subsection{Spectra mathematics and manipulation}

A flexible suite of mathematical operations on the spectra should be
possible. This should include options such as adding, subtracting,
averaging and scaling the data. For common operations such as
averaging and smoothing, it must be simple for the user to invoke the
function (i.e. not to have to start up a complex spectral
calculator). Where it makes sense, it should be possible to manipulate
multiple spectra simultaneously.

The spectral manipulations which should be available are:

\requirement{Add or subtract multiple spectra.}{1}

\requirement{Averaging multiple spectra, with optional weighting based
on Tsys, integration or rms. If the velocity of the spectra to be
averaged is different, the data should be automatically aligned in
velocity.}{0}

\requirement{Various robust averaging possibilities (e.g. median
averaging, clipped means etc) should be possible.}{2}

\requirement{Re-sampling or re-binning of the data to a lower (or
higher) spectral resolution (i.e. change the number of spectral
points). The re-sampling factor may not necessarily be an integer.}{2}

\requirement{It must be possible to shift the data in
  ``frequency/velocity''. This should include channel, frequency and
  velocity shifts of an arbitrary amount.}{1}

\requirement{Spectral smoothing of the data. Hanning, Tukey, boxcar
and Gaussian smoothing of variable widths should be possible.}{1}

\requirement{Scaling of the spectra.}{1}

\requirement{Calculate basic statistical values (maximum, minimum,
rms, mean) on a range of spectral points. The range may not be
contiguous. The calculated rms value should be retained with the
spectra so it can be optionally used for weighted averaging of
spectra.}{1}

\requirement{It must be possible to calculate the flux integral over a
range of channels. The units should be Jy.km/s (or Kelvin.km/s). The
channel range for the calculation should be specific via the GUI or
CLI.}{2}

\requirement{It must be possible to calculate the numerical ``width''
of a line (full width at half maximum type measurement). This should
be calculated by specifying a channel range and finding the maximum
value in this range and then finding the interpolated crossing point
of the data as a user defined fraction of the maximum (default
50\%). The profile width and velocity mid-point should then be
computed. If the profile shape is complex (e.g. double arch) with
multiple crossing points of the fraction value, the minimum and
maximum width values should be calculated. There should be the option
of using a user specified ``maximum value''.}{2}

\requirement{The user must be able to easily change the rest-frequency
to which the velocity is referenced.}{1}

\requirement{FFT filtering for high- and lowpass filtering and
tapering.}{3}

\requirement{It should be possible to FFT the data to and from power
  spectra to the autocorrelation function.}{2}

\requirement{The user may wish to compute the cross correlation
function of two spectra. The result should be a standard ``spectra'',
which can be displayed and analysed using other functions (max, rms
etc).}{3}

\requirement{Complex experiment specific processing can often be done
using a series of the simple of basic functions. A spectral calculator
options should be added to the CLI to perform a series of
manipulations on a set of spectra.}{3}

The user may want to perform specific analysis on the data using the
functionality above, but wish to do the manipulation between two
polarisations or IFs. Allowing the functions to also, optionally,
specify specific polarisations or IF would be an implementation and
interface nightmare. The simplest solution is the allow the data to be
``split'' into separate spectra.

\requirement{It must be possible to take multi IF, multibeam or
polarisation data and split out the individual spectral portions to
form self contained spectra.}{2}

\subsection{Polarimetry}

The software must fully support polarmetric analysis. This includes
calibration and basic conversions. Observations may be made with
linear or circular feed and the backend may or may not compute the
cross polarisation products. As such the software must cope with a
variety of conversions. The software should be able to calculate
stokes parameters with or without solving for leakage terms.

%\makenote{It is debatable whether stokes I is the sum or average or
%two dual polarisation measurements.}

\requirement{All functions on the data (calibration, sky subtraction
spectral mathematics) must support arbitrary, multiple, polarisation
(linear, circular \& stokes and single, dual \& cross
polarisations.}{1}

\requirement{It must be possible to calculate stokes I from single or
dual polarisation observations.}{1}

\requirement{Average a mixture of dual polarisation and single
polarisation data and form average stokes I (e.g. for a long
observation of a source, in which one polarisation is missing for some
time.}{3}

\requirement{Full stokes parameters should be obtained from dual pol
(linear or circular) observations where the cross polarisation
products have been calculated.}{1}

%\requirement{If the observations used linear polarisations and the
%cross polarisations were not computed, the source needs to be
%observed with the feeds set at least 3 different parallactic angles
%(note that if dual linear feeds are available, 2 orthogonal
%parallactic angles are obtained at once). The Stokes parameters can be
%solved using a least squares fit to the equation:
%\reqeqn{Iu/2 + Ip * cos^2 (PA + p)},\\ 
%where PA is the linear feed position angle, p is the polarisation
%angle, Iu and Ip and the unpolarised and linearly polarised
%intensity. {\em Stolen from SPC. I need to write this in more useful
%language. Is this technique likely to be used anymore?.}}{3}

\requirement{If dual circular polarisation measurements are taken,
without computing the cross products, the software should still be
able to compute stokes I and V.}{2}

\requirement{The software should be able to calculate leakage terms
from a calibrator source and correct the data either before or after
conversion to Stokes. (ref. Johnston, 2002, PASA, 19, 277)}{3}

\requirement{The software should be able to determine absolute
position angle from a calibrator source and correct the data either
before or after conversion to Stokes.}{3}

\requirement{Zeeman splitting factors should be derived from
(previous) profile fitting and the left and right circular
polarisations. The velocity shift varies linearly with the magnetic
field, but the scaling factor depends on the molecule and
transition. Scaling factor for common transitions should be known by
the software and the user able to enter factors for less common
transitions. Correctly identifying Zeeman pairs is crucial in getting
the correct result. The software should attempt to make an initial
guess of pairs (based on component velocity and width) but make the
user confirm and override the pairing if required.}{3}

\subsection{Data Selection}
While the software is running the user will usually have loaded
multiple (possibly many) spectra each of which may have multiple IFs,
data from multiple beams and multiple polarisations. The user will
want to be able to quickly flip from considering one spectra to
another and, where relevant, want to perform parallel processing on
multiple spectra at once (e.g. baselining a sequence of on/off
observations of the same source which will later be averaged
together).

\requirement{The software needs an easy-to-use mechanism to select
either individual or multiple spectra for viewing, parallel processing
etc.}{1}

\requirement{An easy-to-use mechanism to select individual IFs, beams
or polarisations is needed.}{1}

\requirement{\label{ref:chansel} The range of spectral points to use
for baseline removal, statistical calculations, RFI editing, analysis
etc must be easily set by the user from both the CLI and GUI. From the
CLI there must be the option of setting the range using a variety of
units (channel number, velocity, frequency). The selection range will
probably not be a contiguous set of channels, but many sets of
disjoint channel ranges. For some tasks (such as baseline subtraction
and statistical values), the channel range should be retained and be
available as a plot overlay.}{1}

\requirement{When performing baseline subtraction on many spectra
simultaneously, the software should have a mechanism for identifying
``on'' and ``off'' spectra and automatically selecting the signal and
quotient spectra. The algorithm needs to cope with on/off/on/off
sequences as well as off/on/on/off. If an individual quotient spectra
has been marked as invalid, an alternative should be found.}{3}

\requirement{The software should be able to select sets of sources
based on simple regular expression type filtering (wild cards) on a
range of header values. Examples include G309$*$ or G309$*$w to select
on source name, or NH3$*$ to select on molecule name.}{3}

\subsection{Plugins}

\requirement{It would be desirable to support ``plugins'', user
definable functions for specific processing. The plugin code must have
full access (read/write) to the spectra data and headers. It should
also be possible to create new spectra which the software treats the
same as the original data. Preferably some (limited) access to the
plotter would also be possible and a number of the standard functions
accessible as ``library'' routines.}{3}

\subsection{Pipelining}

\requirement{Some sort of pipelining mode is required. This would
involve doing a quotient spectra, applying appropriate calibration and
possibly fitting a Gaussian to any lines present.}{3}

\subsection{Methanol Multibeam Survey}

The software may need to support reduction of data from the methanol
multibeam project. If so the pipelining will need to be flexible and
powerful enough to support this.

\subsection{Miscellaneous functionality}

\requirement{The software should be able to take a simple ``grid'' of
observations (normally a set of observations in a cross pattern on the
sky) and, for a subset of channels, fit the position of the
emission. The fit positions should be either plotted on the screen or
exported in a simple ascii form.}{3}

\requirement{The kinematic distance of a source should be calculated
using basic Galactic rotation models. Multiple Galactic rotation
models must be supported and a mechanism for easily adding more.}{3}

\requirement{For 1420 MHz observations of HI, the rms (Tsys) values
vary significantly across the band. The software should be able to
compute the rms as a function of frequency across the spectra from the
off-pulse data and then be able to plot n-sigma error bars on the
spectra.}{3}

\requirement{It should be possible to take a selection of calibrated
spectra which are then passed to the ``Gridzilla'' program to produce
an image cube. Analysis of this cube would be done using external
programs (e.g. Miriad, aips++)}{3}

\section{Help}

\requirement{There should be built-in web-based documentation, which
can be easily kept up-to-date}{1}

\requirement{A short and simple end-to-end cookbook for basic data
analysis should be available.}{1}

\section{Data and meta-data}

\requirement{The software must be capable of handling multi-IF
  (potentially dozens of IFs) and multi-beam data with arbitrary
  polarisation (e.g. single pol, dual pol, full stokes etc).}{1}

\requirement{The software should handle pulsar binned data for pulsar
absorption experiments.}{3}

\subsection{History}

\requirement{A user viewable history of data processing steps should
be kept as part of the data. Where possible this should be retained
when data is imported from other packages.}{1}

\requirement{It should be possible to use the history information to
create template pipeline scripts for batch processing.}{2}

\subsection{Multiple IFs}

\requirement{If multiple IFs are present (currently Tidbinbilla can
produce two IFs and the new wideband spectrometer for Mopra may have
dozens of IFs), the software should handle the data
transparently. Potentially each IF may have a significantly different
sky frequency and be observing a different molecule or transition with
a different rest frequency. From the users point of view,
simultaneously obtained IFs should be kept within the same
``container'' (not split into a myriad of separate ``container'').}{1}

%\makenote{Does the case of multiple lines (so multiple
%rest frequencies) within a single IF need special attention?}

\subsection{Multibeam}

\requirement{Basic handling of multibeam data should be possible (ie
in general each beam will be treated as a separate observation, but
all within the same container. The user should be able to view or
process either individual beams or all beams in parallel.}{2}

\requirement{The use of a single beam observing a source and the
rest of the beams as reference beams for sky-subtraction should be
investigated.}{2}

\subsection{Robust fitting}

\requirement{If robust fitting using median filtering is used, then
the individual integrations from the observations should {\em not} be
averaged when the data is imported, but retained within a single
container. Inspection of this data should be optionally of the averaged
or individual data.}{2}

\subsection{Fit parameters}

\requirement{The fitting parameters for functions which have been fit
to the data (e.g. for baseline removal or Gaussian fits) should be
retained as an integral part of the data and stored permantely on
disk.}{1}

\requirement{It must be possible to export fitting values in an
appropriate form. (i.e. ascii text format)}{1}

\requirement{It should be possible to ``undo'' functions which have
been subtracted from the data (e.g. baseline polynomials).}{3}

\subsection{Coordinate frames and units}

\requirement{Coordinate frames and unit selection and handling needs
to be flexible and relatively transparent to the user (i.e. if the
users preference is for LSRK velocities, they do not need to worry
about the reference frame in which the data was observed).}{1}

\requirement{At a minimum the following reference frames and
conventions should be handled: \setlength{\parindent}{0pt}

\smallskip
\anitem{Position}{(RA,Dec) in J2000 \& B1950 (as well as other
  arbitrary epochs), Galactic, (Az,El).}

\anitem{Frequency}{Velocity (Topocentric, Geocentric, Barycentric,
    Heliocentric, kinematical LSR, dynamical LSR, Rest), Frequency
    (MHz, GHz), channel number.}

\anitem{Velocity}{ Optical, Radio, Relativistic.}

\anitem{Flux}{ Jansky, Kelvin (mJy etc).}}{1}

\requirement{All data should be internally labelled with the
  appropriate coordinate frame and units. If this information is
  ambiguous for some reason, it should be set when the data is imported
  and the user should not have to worry about it again.}{1}

\requirement{It should be clear to the user what coordinate frame
(velocity, position etc) the data is being presented as.}{1}

\subsection{Meta-data}

A comprehensive set of header data should be read from the input data
files. In general all meta-data available in the rpfits file should be
retained. The user may wish to enter some specific values by hand.

\requirement{All header data should be viewable and editable by the
user. This includes changes such as scaling the given Tsys values.}{2}

\requirement{Missing header data should be handled gracefully,
i.e. the software should fill the values with ``blanks'' and be able
to continue to process the data if possible.}{1}

\requirement{The user must be able to add missing header data, which
is not present in the RPFITs file. It must be possible to add the same
header data to multiple scans simultaneously.}{2}

\extendedrequirement{
The following  header data would be required per scan:
\begin{itemize}
  \item Source name
  \item Scan type (signal or reference)
  \item Integration time
  \item Scan length (actual time of observation, $\ge$ integration time)
  \item Telescope
  \item UT time and date of observation
  \item Telescope elevation of observation
  \item Parallactic angle
  \item Beam size
  \item Scan ID
  \item Observer
  \item Project
  \item Polarisation
  \item Receiver
  \item Telescope coordinates
  \item Weather info (temperature, pressure, humidity)
  \item LO chain setup
  \item User axis display preference (LSR velocity, frequency etc).
\end{itemize}
}{1}

\extendedrequirement{
\label{req:if}
The following header data is required for each IF, beam etc:
\begin{itemize}
\item Source coordinates and coordinate frame
\item Frequency/velocity axis definition and type
\item System Temperature
\item Antenna gain (if Tsys measured in Kelvin)
\item Beam number
\item Molecule rest frequency$^\dagger$
\item Molecular name$^\dagger$
\item Molecular transition$^\dagger$
\item Molecular formula$^\dagger$
\end{itemize}
}{1}

\requirement{The molecular formula could be stored with embedded
superscripted and subscripted symbols for ``pretty'' printing on the
plotted, but printed in plain text on the CLI or in ascii output}{3}

Some molecular line rest-frequencies are close enough that two or more
molecules or transitions may be observed in a single IF. Typical
examples include the 1665/1667~MHz OH maser pair, NH$_3$ transitions,
and many observations in the 3~mm band.

\vspace{\parskip}
\requirement{The software should optionally support multiple lines per
IF, by storing a set of rest frequencies per IF, rather than a single
value. The header values in requirement \reqref{req:if} marked with a
$\dagger$ would all have to be stored as an array of values rather
than a scalar. A simple mechanism must be posible to change the
currently ``active'' rest-frequency.}{3}

\section{Installation}

\requirement{It must be possible for astronomers to install the
software at their own institute with either a moderate amount of OS
experience or some help from the local system administrators. This
includes installation on a central ``NFS'' server as well as local
desk-tops.}{1}

\requirement{The software must run on Solaris and all major flavours
of Linux (Fedora/Redhat, Debian, etc).
 }{1}

\requirement{It must be possible for users to
install the software on their (unix) laptops and run with no network
connection. }{1}

\requirement{It should be relatively easy to upgrade to the lastest
version of the software.}{2}

\requirement{The software should run on MacOS/X}{3}

\requirement{It would be desirable for the software to run on
Windows.}{3}

\section{Known Issues}
\label{sec:issues}
The following issue are known problems with the data from ATNF
telescopes, which probably should be automatically corrected for if at
all possible. The best place to do this is while loading the data.

\subsection{General}

\begin{itemize}
  \item All polarisations in the RPFITS files are labelled as
  XX/YY. These need to be relabelled as LL/RR when appropriate.
\end{itemize}

\subsection{Mopra}

\begin{itemize}
\item Data obtained in 2002 \& 2003 (and probably before) have an
  error in the frequency headers (this may be corrected by an external
  program). \makenote{Nedd Ladd}

\item The (RA,Dec) positions in the data file are in date coordinates
  not J2000. This causes problems for packages like Class when
  averaging the data. \makenote{Maria Hunt}

\item It is possible Tsys calibration is inconsistent currently.
  \makenote{Cormac Purcell??}

\end{itemize}

\subsection{Parkes}

\begin{itemize}
\item For pulsar data the automatic gain control is disabled. This
means the nominal Tsys measurement does not change and Tsys per
integration is encoded in a non-standard way. \makenote{Simon
Johnston}
\end{itemize}

\subsection{Tidbinbilla}

\begin{itemize}
\item All 20-GHz data is calibrated in flux units of Kelvin.
\end{itemize}

\section{Acknowledgements}

We would like to thank the following people for input directly or from
previous documents: Maria Hunt, Paul Jones, Jim Caswell \& Dave
McConnell, Frank Briggs, Juergen Ott, Daniel Pisano, Jim Lovell, Ray
Norris, WIm Brouw, Maxim Voronkov, Vincent McIntyre, Jessica Chapman,
Simon Johnston, Nedd Ladd, Lister Staveley-Smith, Cormac Purcell \&
Rachel Deacon.

\end{document}