Map-making in small field modulated CMB polarisation experiments: approximating the maximum-likelihood method

TL;DR

This paper introduces a destriping map-making method for CMB polarisation experiments that approximates maximum-likelihood results, significantly reducing computation time while maintaining accuracy in detecting B-modes.

Contribution

The paper presents Descart, a destriping algorithm that efficiently produces near-optimal CMB maps, offering a practical alternative to maximum-likelihood methods for large datasets.

Findings

Descart achieves 5-22 times faster computation than maximum-likelihood methods.

The destriping method produces maps with negligible difference from optimal, without biasing E or B-mode spectra.

Proper destriping of atmospheric 1/f noise is essential for B-mode detection in single detector maps.

Abstract

Map-making presents a significant computational challenge to the next generation of kilopixel CMB polarisation experiments. Years worth of time ordered data (TOD) from thousands of detectors will need to be compressed into maps of the T, Q and U Stokes parameters. Fundamental to the science goal of these experiments, the observation of B-modes, is the ability to control noise and systematics. In this paper, we consider an alternative to the maximum-likelihood method, called destriping, where the noise is modelled as a set of discrete offset functions and then subtracted from the time-stream. We compare our destriping code (Descart: the DEStriping CARTographer) to a full maximum-likelihood map-maker, applying them to 200 Monte-Carlo simulations of time-ordered data from a ground based, partial-sky polarisation modulation experiment. In these simulations, the noise is dominated by either detector or atmospheric 1/f noise. Using prior information of the power spectrum of this noise, we produce destriped maps of T, Q and U which are negligibly different from optimal. The method does not filter the signal or bias the E or B-mode power spectra. Depending on the length of the destriping baseline, the method delivers between 5 and 22 times improvement in computation time over the maximum-likelihood algorithm. We find that, for the specific case of single detector maps, it is essential to destripe the atmospheric 1/f in order to detect B-modes, even though the Q and U signals are modulated by a half-wave plate spinning at 5-Hz.