Fast and precise map-making for massively multi-detector CMB experiments

作者：Sutton D^*; Zuntz J A; Ferreira P G; Brown M L; Eriksen H K; Johnson B R; Kusaka A; Naess S K; Wehus I K

来源：Monthly Notices of the Royal Astronomical Society, 2010, 407(3): 1387-1402.

DOI：10.1111/j.1365-2966.2010.16954.x

• methods: data analysis
• methods: statistical
• cosmic background radiation

Future cosmic microwave background (CMB) polarization experiments aim to measure an unprecedentedly small signal - the primordial gravity wave component of the polarization field B mode. To achieve this, they will analyse huge data sets, involving years of time-ordered data (TOD) from massively multi-detector focal planes. This creates the need for fast and precise methods to complement the maximum-likelihood (ML) approach in analysis pipelines. In this paper, we investigate fast map-making methods as applied to long duration, massively multi-detector, ground-based experiments, in the context of the search for B modes. We focus on two alternative map-making approaches: destriping and TOD filtering, comparing their performance on simulated multi-detector polarization data. We have written an optimized, parallel destriping code, the DEStriping CARTographer (DESCART), that is generalized for massive focal planes, including the potential effect of cross-correlated TOD 1/f noise. We also determine the scaling of computing time for destriping as applied to a simulated full-season data set for a realistic experiment. We find that destriping can outperform filtering in estimating both the large-scale E-and B-mode angular power spectra. In particular, filtering can produce significant spurious B-mode power via EB mixing. Whilst this can be removed, it contributes to the variance of B-mode bandpower estimates at scales near the primordial B-mode peak. For the experimental configuration we simulate, this has an effect on the possible detection significance for primordial B modes. Destriping is a viable alternative fast method to the full ML approach that does not cause the problems associated with filtering, and is flexible enough to fit into both ML and Monte Carlo pseudo-C(l) pipelines.

• 出版日期2010-9-21