SPIDER 1 CMB Data

This directory contains data products associated with the SPIDER 1 CMB results paper (https://iopscience.iop.org/article/10.3847/1538-4357/ac20df/pdf ).

The types of products are divided into three categories:

General Products:       beam, filter transfer function, bandpower window
                        functions, bandpowers, and covariance matrices

Map Products:           mask, sample simulation map pairs, map pixel weights

Likelihood Products:    r likelihood curve

For all cases, the maps have been cleaned with a 353-100 GHz Planck template map
using XFaster's best-fit alpha coefficients (0.018 for 95 GHz and 0.045 at 150
GHz). Files are given for the independent frequencies, 95 and 150 GHz, as well
as the combined result, which includes the individual frequencies and their
cross-spectra.

Bandpowers

Bandpower files are 3 columns: [ell bin centers, band powers, error bars].
Everything is in uK_CMB^2. Files labeled Cl are C_ells; files labeled Dl are
`ell (ell+1) C_ell / (2 pi)`. Error bars include sample variance.  Additionally,
they include error from template-subtraction. EE, BB, and EB spectra are
provided as separately labeled files.

Covariance Matrices

Covariance matrices are 27 x 27 blocks, where elements are [EE 9 bins, BB 9
bins, EB 9 bins].  Bin centers match those given in the bandpowers files.
Sample variance is included. Error from template subtraction is included only
approximately: the diagonal of the covariance has template-subtraction error
incorporated, matching the square of the error bar vector in the bandpowers
files. Off-diagonal elements do not account for error from template
subtraction. Units are uK_CMB^4.

Bandpower Window Functions

Bandpower window functions are 10 columns of length 408. This corresponds to 408
ell values and the 9 science bins. These are normalized such that for each bin,
summing over the column in the file times a normalization function `N_ell`
results in 1 for each bin. (The bandpower window functions are zeroed below
SPIDER's minimum multipole bin, ell=8.)

The Cl window functions can be used as follows to obtain bandpowers for the nine
BB bins from a theory C_ell spectrum `Cl_th_BB`:

.. code-block:: python

    import numpy as np
    wbl_Cl = np.loadtxt("wbl_Cl_combined_bb.txt", unpack=True)
    ell, wbl_Cl = wbl_Cl[0], wbl_Cl[1:]
    N_ell_Cl = (2.0 * ell + 1.0) / 4.0 / np.pi
    norm = np.sum(N_ell_Cl * wbl_Cl)   # this should be an array of 1's
    Cb = np.sum(N_ell_Cl * wbl_Cl * Cl_th_BB[:408])

The Dl window functions are the same, but with a different normalization
function and using the theory spectrum in D_ell's:

.. code-block:: python

    wbl_Dl = np.loadtxt("wbl_Dl_combined_bb.txt", unpack=True)
    ell, wbl_Dl = wbl_Dl[0], wbl_Dl[1:]
    lfac = ell * (ell + 1.0) / 2.0 / np.pi
    N_ell_Dl = (2.0 * ell + 1.0) / 4.0 / np.pi / lfac
    norm = np.sum(N_ell_Dl * wbl_Dl)   # this should be an array of 1's
    Dl_th_BB = lfac * Cl_th_BB
    Db = np.sum(N_ell_Dl * wbl_Dl * Dl_th_BB[:408])

Beam Products

Beam Products (B_ell, not B_ell^2) are 3 columns of length 350.
[ell, beam value, beam error estimation].
Beams are normalized to 1 at ell=0.

Filter Transfer Functions

Filter transfer function files (F_b) are 3 columns x 16 bins.
[left bin edge, right bin edge, estimated binned transfer function F_b].

Latlon, Pointsource Removed Mask

The latlon, pointsource removed mask is a FITS file of True and False values,
where True-valued pixels are to be included in the analysis. The map is in
equatorial coordinates.  The mask can be loaded using the standard
`healpy.read_map` routine:

.. code-block:: python

    import healpy as hp
    mask = hp.read_map("mask_latlon_pointsource_removed.fits")

Sample Simulation Map Pairs

These represent the result of the end-to-end simulations of the SPIDER 2015
timelines.  They include a small ensemble of signal and noise maps generated
from re-observed LCDM signal realizations and SPIDER's time domain noise model
that can be used for observation planning, forecasting, or for Galactic
astrophysics.  We provide 10 signal and 10 noise maps for each frequency.  All
maps are in equatorial coordinates and have units in uKcmb.  The input spectrum
for the sky model, which is prior to re-observation, is located:
`input_spec_signal_flatBBDl.dat`.  A much larger ensemble is required for
accurate cosmological analyses.  When run on the data, the XFaster
cross-correlation pipeline determines an empirical noise scaling factor between
the data and the noise model.  This factor has been applied to the noise maps,
and is documented in the header.

Map Weights

The SPIDER detector timestreams are weighted by the detector noise for every
10-minute chunk of data.  These detector weights are then binned into maps to
compute a set of map pixel weights, which can be used alongside the signal and
noise simulation maps.  This weighting is provided as weight_95.fits and
weight_150.fits.

.. code-block:: python

    import healpy as hp
    mask = hp.read_map("weight_95.fits")

r Likelihood Curve

1000 points of the density of the r likelihood curve as computed by the XFaster
algorithm are provided in a text file.  The first column is the r value and the
second is the value of the normalized density for that r value.

Notes:

Since the publication of the paper, we have migrated all the analysis code from
python 2 to 3, and re-validated the results.  Additionally, we have implemented
a number of changes in the XFaster implementation with respect to the published
results.

   The calculation of the gcorr factor (compute_gcal.py) uses a simple estimate
   of the variance rather than fitting the distribution.

   Numerical outliers in the simulation ensemble are identified and flagged
   prior to their inclusion in the bandpower estimate.

Taken together, these changes result in numerical differences between the
current code base and the published value of the bandpowers and r-likelihood.
In all cases these differences are of negligible amplitude.