The behavior of noise in the pixel space as a function of number of receivers coadded has been investigated in 20201027_noise_scaling_withrx for Keck 2012/13 data. The behavior of \(N_\ell\) with number of coadded receivers for Keck 2013,2017 and 2018 has been investigated in 20201116_noise_int_down and 20201116_noise_int_down_K17_18, where we observe that the average \(N_\ell\) for \(10 \lt l \lt 100\) (bandpowers 2 and 3) shows deviation from simple scaling as a function of number of detectors. Also, the \(TT\) spectrum for the Keck 2013 and B2 has been investigated in 20210330_noise_int_down_t. This posting is a follow-up to the same in order to investigate the scaling of \(BB\) and \(TT\) \(N_\ell\) with number of detectors in BICEP 3 using per-pair coadded maps for BICEP 3 2017 data.
BICEP3 has got the same number of pairs as K2013- about 1200, but at 95 GHz rather than 150 GHz.We want to see if the excess noise effect is band dependent. We accumulate 5021 tags for 2017 and create the per-pair coadded map (jack02) for the analysis. The data-products used for this analysis are are stored in maps/5901 and aps/5901.
We use the jack02 per phase coadded map, equally divide the pairs into 8 groups of 155 pairs each (after shuffling), and accumulate over them so as to cover the entire focal plane each time with an increasing density of detectors.




We then simulate 100 realizations of \(N_\ell\) for each of these 8 combinations of accumulated tiles. We plot those in Fig.2a (Note: \(N_\ell\) has been divided by a factor of \( \ell(\ell+1)/2\pi\) to plot the 'raw \(N_\ell\)' ). We then take the mean of those 100 realizations and plot in Fig 2.b.
The high \(\ell\) component integrated faster than the low \(\ell\) component as a function of number of accumulated receivers. To illustrate it better we plot the mean \(N_\ell\) in bandpowers 2 and 3 (in blue) and in the range \(200 \lt \ell \lt 400\) (in red).

The mean \(N_\ell\) integrates with \(N_{\det}\) closely in accordance with expectations (for \(\ell \gt 10\)). Subdividing into low \(\ell\) (bandpowers 2 and 3) and high \(\ell\) range, the mean \(N_\ell\) in the low \(\ell\) range scales almost similarly compared to the high \(\ell\) range. Although, the mean \(N_\ell\) for the low \(\ell\) range has a deviation from the expectation (around 15 percent), the deviation is not as high as seen in case of Keck data. The reason these B3 \(N_\ell\) spectra turn down at lower \(\ell\), as compared to Keck data, is because of the poly and scansync filtering. This filtering also suppresses the low \(\ell\) power for 150/220GHz but there the \(1/f\) noise is strong enough that it still kicks up to lower \(\ell\) even after the filtering.
The individual pol weight maps from the 8 combinations are here.

make_noise_sims.m is for creating the xxx6 jack02 realizations. make_aps_jack02.m is for creating the aps for the noise realizations. plot_aps_B32017_jack02.m and plot_aps_B32017_TT_jack02.m for creating all the plots in the posting.