COMMENTARY ON TIANLAI JANUARY 2018 DISH OBSERVATIONS OF THE NORTH - PowerPoint PPT Presentation

COMMENTARY ON TIANLAI JANUARY 2018 DISH OBSERVATIONS OF THE NORTH CELESTIAL POLE longest continuous run to date ~10 days using pipeline calibrated data / no RFI removal Tianlai Collaboration Meeting 2018 Albert Stebbins Pingtang County, Guizhou Province, China Fermilab Theoretical Astrophysics 19 September 2018

TIANLAI DISH “POLARSCOPE” TIANLAI HIGHLY CONFIGURABLE! BY POINTING DISH ARRAY TOWARD POLE WILL INTEGRATE DOWN TO LOW MAP NOISE TEMPERATURE VERY RAPIDLY SINCE ONE IS ALWAYS POINTING AT SAME SPOT ON SKY. dish array in week for ( 100 Mpc ) 3 voxel 500 raw δ T voxel (µ k ) 200 100 Δ z = 0.01 50 Δν = 4 MHz - 90 - 60 - 30 0 30 60 90 pointing declination (°)

January 2018 Dish Data • This is the longest run to date ~10 days. • Analysis starts with Timestream data calibrated by Cas A observation at the beginning of run generated by John Marriner. • No RFI removal was applied. • There were a few iterations of rawTimestream → Timestream as we learned about how to use the pipeline. • Further analysis was done using Mathematica code. • There are apparently some bugs still left (possibly in Mathematica code).

example of one “good” visibility V 3x31 . this is raw correlator data 11 sidereal days top to bottom 2018-01-02 02:44:15 to 2018-01-11 20:44:12 brightness gives |V 3x31 | color gives complex phase frequency ⇒ bright swath are fringes of the Sun fringes vary because Earth rotates bow occurs when Sun passes “above” baseline direction Sun dominant source even 110° off axis sidereal time ⇒ dimmer fringes still visible at night

nighttime only view of V 3x31 nighttime mean subtracted for each frequency image intensified to show nighttime fringes visibilities repeat with sidereal day period signal from the sky fringe pattern more complicated than bow indicating multiple bright sources in 2° beam centered on North Celestial Pole frequency ⇒ bright and dark patches constructive/destructive interference between sources good S/N in 1min x1MHz pixels rapid frequency fringe rate N-S baseline: large ν -dependent phase difference variable temporal fringe rate caused by Earth rotation - sources at different declinations sidereal time ⇒

Good and Bad Baselines John Marriner w/ others worked 1min × 244kHz pixels past few weeks on understanding how to use python pipeline - applying it to this run. Apparently there are still some worst misunderstandings leading to large mis-calibrations. best

WORST frequency ⇒ frequency ⇒ sidereal time ⇒ sidereal time ⇒

ribbon “Good” Median Nightly Averages candy have good measure of visibilities from foreground sources

looking for non-smooth spectrum components Hi- k || Analysis Cone of Silence To search for non-smooth spectrum emission one might initially look as far away from the foreground wedge as possible, Foreground Wedge e.g . large k || for individual visibilities. This requires no knowledge of the beam , only that | k || | ≫ | k ⟂ |. Due to chromaticity of beams non-smooth angular structure At present 21cm intensity mapping of sources will leak into frequency structure (mode mixing) filling the foreground wedge . Ideally the cone of silence , experiments su ff er from non-detection of the complement of the wedge, is not contaminated by 21cm emission. “Finding it” should be a smooth spectrum sources, however there is no well defined high priority . We can do so by looking for boundary so one must determine the leakage of smooth spectrum sources into all parts of k-space. non-smooth spectrum emission away from the wedge. Discover the Hilbert subspace of the “space of beams” with little contamination.

Power Law Decomposition One can expand a visibility covering a Numerical Tests on Sinusoids+ given band into Legendre polynomials (or any other polynomial expansion) N a n P n [ 2( ν − ν mid ) ∑ V [ ν ] = ] ν max − ν min -80db numerical floor -75db numerical floor n =0 the larger n is the less smooth the frequency dependence. A carefully designed discrete analog of this works better than a discrete Fourier transform. The integer index n is an analog of k || . For small bandwidth -70db numerical floor -70db numerical floor 2 π k || ≈ ( n + 1) Δ R co much spill-back, rapid fall-o ff , little spill-forward

Are Good Tianlai Visibilities Smooth Spectrum? 1 nights 3 nights 11 nights rms = 5.44 K , 4.64 K , 4.34 K 1 nights 3 nights 11 nights rms = 5.29 K , 4.55 K , 4.1 K 1 nights 3 nights 11 nights rms = 6.55 K , 5.75 K , 5.53 K ( Mpc - 1 ) ( Mpc - 1 ) ( Mpc - 1 ) pseudo k ∥ pseudo k ∥ pseudo k ∥ 0.01 0.1 1 0.01 0.1 1 0.01 0.1 1 2 60sec × 244kHz pixels 60sec × 244kHz pixels 60sec × 244kHz pixels NORTH NORTH NORTH 3 12 3 31 5 25 mode amplitude ( Kelvin [ nominal ]) 1 mode amplitude ( Kelvin [ nominal ]) mode amplitude ( Kelvin [ nominal ]) 1 WEST WEST WEST EAST EAST EAST 1 0.5 SOUTH SOUTH SOUTH 0.5 0.5 noise floor noise floor noise floor 0.2 0.2 0.2 0.1 0.1 0.1 noise floor 1 5 10 50 100 500 1 5 10 50 100 500 1 5 10 50 100 500 1 + polynomial order 1 + polynomial order 1 + polynomial order 1 nights 3 nights 11 nights rms = 5.59 K , 4.93 K , 4.55 K 1 nights 3 nights 11 nights rms = 5.65 K , 4.88 K , 4.54 K 1 nights 3 nights 11 nights rms = 6.34 K , 5.72 K , 5.31 K ( Mpc - 1 ) ( Mpc - 1 ) ( Mpc - 1 ) pseudo k ∥ pseudo k ∥ pseudo k ∥ 0.01 0.1 1 0.01 0.1 1 0.01 0.1 1 2 2 60sec × 244kHz pixels 60sec × 244kHz pixels 60sec × 244kHz pixels NORTH NORTH NORTH 7 29 9 17 9 22 mode amplitude ( Kelvin [ nominal ]) mode amplitude ( Kelvin [ nominal ]) mode amplitude ( Kelvin [ nominal ]) WEST WEST WEST EAST 1 EAST EAST 1 1 SOUTH SOUTH SOUTH 0.5 0.5 0.5 noise floor noise floor noise floor 0.2 0.2 0.1 0.1 0.1 1 5 10 50 100 500 1 5 10 50 100 500 1 5 10 50 100 500 1 + polynomial order 1 + polynomial order 1 + polynomial order

Power Distribution 03x31 ← pseudo-m → n=0 n → m=+471 m=-470 m=0 log of power in 1min × 244kHz pixels pseudo-m because not full sidereal day n=472

Noteworthy Conclusions • for n >50 the visibilities do indeed seem to decrease in amplitude like the inverse of the square root of integration time - and can be thought of as system noise. • in the n - m representation of the data space nearly all of the area appears to be uniform white noise - totally untainted by the foregrounds - this includes a lot of regions where 21cm lives. • If n -expansion works as well for removing Sun as for polar foregrounds one could use both nighttime and daytime data.

Things to do • find remaining bugs in implementation of pipeline and reprocess January data • this may bring many more baselines into the “good” category • estimate e ff ective noise temperature from hi- k || white noise region (!) • check whether hi- k || analysis e ff ective in removing the Sun • if so obtain full day maps of white noise region and proceed with true m -mode analysis. • fit to NVSS catalog and see if it make any sense • process/calibrate older NCP dish runs and combine all together • this could more than double the total integration time • predict how much more NCP integration time expected to reach 21cm floor • request additional NCP runs as needed • see if we find a di ff erent floor (systematics? other contamination?) en route to 21cm floor