Reference design simulation tool

From CMB-S4 wiki
Jump to navigationJump to search

October, 2019 - Data management group


This page documents a simulation tool based on the reference design that allows users to explore how various design choices affect CMB-S4 maps. We use limited scope time domain simulations to build an archive of signal, noise and systematic maps that can be combined with appropriate weights to account for

  • observing efficiency
  • survey length
  • detector counts and sensitivity
  • telescope siting
  • levels of systematics


  • Which resolutions to support?
  • Specify the reference design. What are the parameters to vary?


  1. Focalplane geometry 11/15/2019
  2. Instrument noise model 11/08/2019
  3. Observing schedules chosen by 11/15/2019
  4. All inputs for TOD simulation in place by 11/15/2019
  5. Simulated component maps ready by 12/01/2019
  6. Simulation tool written, tested and delivered to the collaboration by 12/15/2019

Instrument model

Reference SAT

Describe frequencies and detector counts on the smallest independent unit (tube?).


Choose some representative geometry (hexagon?) for the independent unit. Full instrument configuration details are here


Instrumental noise is based on ...

It is modulated by elevation dependent factors based on SO V3 LAT noise estimates and fitted by Carlos Sierra.

Reference LAT

Describe frequencies and detector counts on the smallest independent unit (tube?).


Choose some representative geometry (hexagon?) for the independent unit.



White Noise:

The per detector white noise levels for the SAT detectors are calculated from scaled noise maps from BICEP/Keck.

The per detector white noise levels input for the LAT are the output for the baseline design from the SO sensitivity calculator in the v3 goal configuration. We note that the pixel numerology/sizes of the sensitivity calculations were slightly off for the ULF/LF and UHF bands in that the actual pixels are slightly smaller and more numerous. The corrected pixel sizes and numbers are used in the instrument simulations, but the same per pixel noise levels are used, so the sensitivity may be slightly overestimated in these cases.

Noise Elevation Scaling:

The elevation dependence of the noise was simulated for the SO LAT bands/configuration with the SO v3 sensitivity calculator. This elevation dependence of the noise is parameterized with A and C as NET=A/sin(elevation)+C, where the NET is normalized to an elevation of 50 degrees. The S4 sims use the values fit for SO’s LF, MF, and HF bands. The same scaling is assumed between LATs and SATs. The ULF LAT band assumes the same value as the 27 GHz band.

Per-Detector Atmospheric Noise Parameters:

The f_knee, fmin, and alpha values for both the SATs and LATs are the SO LAT values. These were derived from the ACT data release. This data release used gap filling in the spectra, which may make these values pessimistic.

Scan strategy

The simulation requires specifying a scan strategy for each telescope and site considered.

Pole SAT

We use the Pole deep scan strategy from Deep_SAT_from_the_Pole. The boresight scans over RA = [20..60] deg and Dec = [-55..-50] deg in 0.25-degree elevation steps, 30 minutes per step. The 21 steps take 10h 50min to complete with 1min gap between each step. Here is an example 10-day hit map using a dummy 35-degree hexagonal focalplane with 217 pixels:

Hits pole sat.png

For simplicity, we schedule exactly one complete scan for each calendar day.

Pole LAT

The Pole LAT scan strategy is designed to cover the SAT patch. We enlarge the target patch to account for the considerable difference in focal plane sizes. The boresight sweeps over RA = [10..70] deg and Dec = [-65..-40] deg in 0.25-degree elevation steps, 5 minutes per step. The 101 steps take 10h 5min to complete with a 1min gap between each step. Here is an example 10-day hit map using a fake focalplane with 19 pixels:

Hits pole lat.png

Chile SAT

The Chile SAT strategy is based on Deeper_SAT_from_Chile_II. We have circled the Celestial sphere above Atacama with 10x20-degree (RA x Dec) tiles to form an almost continuous chain of low foreground tiles. The tiles are divided into three tiers, each tier having absolute priority over lower tiers when ever they can be targeted. This way the schedule targets two deep patches (North and South) as much as the scheduling constraints allow and embeds them in a wedding cake fashion in a shallower environment. The scheduler considers observing elevations in range of 45-60 degrees with a preference for higher elevations. For the 10-day example we have disabled Sun and Moon avoidance with the understanding that full season observations will lead to similar hit map even with the avoidance enabled.

Hits chile sat.png

Chile LAT

The Chile LAT strategy is based on Modulated_scan_high_cadence_LAT, the experimental scan strategy that modulates the scan rate based on telescope Azimuth. We observe at 40-degree elevation, sweeping at Az = [20..160] degrees or Az = [200..340] degrees. The telescope scan rate is lowest at Az=90 and Az=270 degrees and 2.75 times higher at the start of the turnaround. This allows for a near uniform 65% sky coverage while maintaining daily cadence across the observable sky. As with Chile SAT, we have disabled the Sun and Moon avoidance for the 10-day representative period.

Hits chile lat.png

Sky signal

CMB and foregrounds were simulated with PySM, see github for documentation and map location.

These full sky input maps have been scanned to timelines and then applied a filter-and-bin algorithms to apply the expected transfer function. See details about the mapmaking algorithm below.


Atmospheric noise

Atmospheric signal was simulated using the TOAST atmosphere module. It is a real space simulation of a modified, 3D Kolmogorov spectrum to create realistic distribution of water vapor in the air. Wind effects are simulated by moving a stationary volume at a constant rate in front of the telescope. Detector samples are line-of-sight integrals through the simulated volume.

For efficiency, the volume is simulated in three fields:

  1. Near field (up to 100m distance). 5x5x5m volume elements.
  2. Intermediate field (100m - 1000m distance). 15x15x15m volume elements.
  3. Far field (1000m - 10000m distance). 50x50x50m volume elements.

Wind and surface temperature parameters were drawn for each CES from historical distributions based on NASA MERRA-2 data.

Line-of-sight integrals are truncated at 4km above the telescope. Injection scale was 15m, setting a relatively short correlation length of the fluctuations. Large scale fluctuations would have been filtered out regardless.

Ground pickup

Data reduction

Data reduction must be linear. Each component will be processed separately.

Ground filter

Polynomial filter

Transfer function

Input archive

We'll use NERSC to host the input maps.