Check Design Tool Sims versus scaled BK15 part II

2021 Mar 23 - Clem Pryke

In 20210303_dt1_vs_bk15 I compared the high \(\ell\) ("white noise") level of the Design Tool (DT) Pole SAT noise sims against what one gets from scaling from BK15 published \(N_\ell\) according to the now time honored recipe. In this post I filter correct the DT sim and have a look at how the low \(1/f\) part compares.

Measuring the suppression factor

In 20200817_DM_AWG_sim Colin looked at the 08/20 round of sims, and plotted the filter suppression factor. Since Reijo and Andrea have supplied the input LCDM map and the corresponding filtered output we can take the ratio of the spectra of these and cancel out sample variance and the beam to just get the filter suppression factor. The plot below shows the ratio of the output over input \(C_\ell\)'s for TT and EE. The color range is red to blue low to high frequencies - we see that there are only 4 distinct beam widths. The numbers go crazy above the \(\ell\) number supported by the beam size at each frequency. The dashed black line is the functional form suggested by Colin the above linked post with \(\ell_\mathrm{min}=10\) and \(\ell_0=50\) - this is clearly not a great fit and is not used below.

Fig 1:

It is not possible to extract a BB filter function without using a power spectrum estimation that maintains very high E/B separation purity. I believe there is nothing currently in the DT sims which would result in different filtering between E and B so I just use the EE filter factor for BB.

Comparing suppression factor corrected BB noise spectra

I next take the spectrum of the DT noise sim maps (single realization provided), and divide by the filter function from above. In the plot below I compare the result to the BK15 scaled expectation as detailed in 20210303_dt1_vs_bk15. At high \(\ell\) we see the same result as Fig 3 of that posting - 30/40/85/95 GHz basically the same, while 145 GHz and above over perform in DT. This time we can compare the \(1/f\) part as well. The DT sims do not reproduce the \(1/f\) as fit to the BK spectra very well. One can see the original BK fits from Ben Racine in 20190220_noiseparams_bk15_forS4.

Fig 2:

Do we expect DT sims and BK-scaled to agree exactly?

• The DT sims and the BK re-scaling above both use the same NET's. The DT sims build in an explicit variation of NET with observing elevation. The BK re-scaling implicitly bakes in any such variation in as much as the distribution of elevation angles is the same - which it pretty much is.
• The BK scaling uses the sky hit pattern from the DT sims.
• Both sets of sims use the same number of detector-years.
• The DT sims include an effective observing efficiency which is taken from BK data - but not exactly the BK15 data - see 20210318_bk15_obseff. However, as noted in that post the "weighting inefficiency" is not included in the DT sims and is significant. The BK re-scaling bakes in all inefficiencies.
• The DT sims attempt to include a penalty from correlated noise between many detectors observing simultaneously. In the current round there are effectively 143 at 30/40, 1764 at 85/95/145/155 and 4296 at 220/270 GHz. The BK-scaled calculation implicitly includes such penalty for the number of simultaneous detectors used in BK15. For 95GHz and 220GHz this is 576 and 1024, while for 150GHz the effective number is \(\approx1000\). The BK15 penalty might be expected to be larger for given \(N_\mathrm{det}\) since the field of view of the BICEP2/Keck generation receivers is half the diameter of the BICEP3/S4 generation.
• The failure of \(1/f\) polarized noise to integrate down with \(N_\mathrm{det}\) in BK data has been investigated empirically in 20201116_noise_int_down. At 150GHz and 2500 detectors the penalty was found to be \(\approx30\)%. An investigation at 95GHz and 2500 detectors is in preparation. Again - this penalty is baked in to the BK15 re-scaling for the number of detectors given above.

Comparing suppression factor corrected TT noise spectra

Since as noted above the number of simulataneous detectors is not that different between the DT sims and the BK15 obs the massive over production of TT \(1/f\) noise seen below seems to argue that the atmospheric model is nowhere near reproducing reality.

Fig 3: