PLATO PLAnetary Transits & Oscillations of stars Data onboard treatment PPLC study February 2009 on behalf of Reza Samadi for the PLATO data treatment team
Onboard processing modes
Main functions of onboard/onground processing Observation mode: smearing correction weighted mask photometry / aperture photometry kinematic aberration correction jitter correction Configuration mode: measuring/modelling PSF measuring/modelling sky background
Smearing correction CCD registers overscan rows measure smearing thanks to overscan rows subtract from each image Normal telescopes: sampling: 25s integration time: 23s readout time: 2s Fast telescopes: sampling: 2.5s integration time: 2.25s transfer time: 0.25s beforeafter
Weighted mask photometry minimizing the impact of confusion the target star can be polluted by a neighbouring star to avoid confusion : use of a weighted mask weights = integral of PSF over pixel need to know the PSF normal aperture photometry to be used for brighter stars PSF weighted mask ~ 90 % of the flux target star m V = 11 nearby faint star cf CDF study
PLATO: large field of view : 42° pixel size : 12.5” (14.3”) the effect is much more important than for CoRoT star displacements over 1 month : ~ 7 pixels (worst case) will induce an unacceptable decrease of the flux thermoelastic variations of the telescope pointing direction can also induce star displacement time flux - update the mask position frequently - avoid flux loss - introduce periodic perturbation - need to limit impact of this perturbation - update every hr (tbc) - hourly update is entirely predictible - less frequent update for telescope variations Differential kinematic aberration m V =11 5 months
Is jitter correction at all necessary? CoRoT : 0.25'' rms + orbital components PLATO : specified : 0.2'' rms PRNU does not seam to be a problem Depending of the jitter noise level and nature : the perturbation can be important or negligible For bright star the contribution can be important if the jitter is ~ 0.5'' rms or more Aperture photometry results in negligible perturbations with ref to photon noise
Jitter correction Surface for the jitter correction from the knowledge of the PSF, we can predict the perturbations induced by any displacement: This method also corrects for differential aberration The presence of polluting sources can be accounted for in the correction surface Accurate knowledge of the star displacements: x, y is needed Accurate PSF is needed Fialho et al (2007, PASP) PSF mask
PSF determination (configuration mode) Assumptions, for each telescopes : The PSF varies slowly across the field of view We have available N (=1000) reference stars with associated image time series We have a functional form of the PSF as a function of K parameters a i (eg. width , skewness, etc): PSF(x,y) = f a1,a2,…,aK (x-x 0,y-y 0 ) For each star, for each telescope: We constrain the parameters using the image time-series. The fitted parameters a i (j) are then considered as a function of the position [x 0 (j) and y 0 (j)] of the star j. A 2D polynomial interpolation is then performed to derive the values of the parameters at any position across the field of the telescope. However, PSF can depend on the star colour => 3D polynomial interpolation (x,y,colour) ? Procedure to apply at TBD frequency (once a week?) PSF used to calculate mask weights and jitter correction surface
Sky background determination (configuration mode) set 400 background windows per telescope (100 per CCD) collect a long enough time series of background measurements background is modeled using a 2D polynomial fit The sky background level can then be estimated at any position, then for all stars in the FOV.
Onboard processing dimensioning: star samples Sample P1 : mv < ; noise level < 27 ppm/h stars : 50s, 600 s Subset : N = 1000 references stars, mv= , individual light curve Sub-images (imagettes) : n = s sampling Sample P2 : mv < 12 ; noise level < 80 ppm/h s Oversampled : s sampling Sample P3 (P4) : 4.75 < mv < 7.3 noise level < 27 ppm/h 500 (1 000) 50s Subset: 100 stars 2.5 s Sub-images (imagettes) : m = 50 s Sample P5 : mv < 13.5 ; noise level 80 ppm/h ; no centroids measured s Oversampled : s Background windows : 400
Onboard processing architecture 1 DPU per telescope + 1 ICU (+1 redondant) - case 1: perform onboard average - case 2: downlink all individual LC trade-off needed very soon !
Normal telescope DPU processing
Normal telescope data flow and TM volume
Fast telescope data flow and TM volume
Total TM rates case 1: perform onboard averagecase 2: downlink all individual LC
Case 1.vs. Case 2 trade-off Case 1 : only 1000 LCs from Sample P1 are downloaded : 31 Gb/day (with compression) Case 2 : all LCs are downloaded : 71 Gb/day (with compression) Case 1 : jitter correction to be done onboard ! Outlier discarding and LC average to be done on board. Strong constraints on the onboard processing, no replay possible. Case 2 : jitter correction can be done onground ! Outlier discarding and LC average done on ground.
Onboard processing H/W dimensioning CPU for one DPU LEON processor at 100 MHz CPU occupation rate = 40%
Open issues Trade-off between Case 1 and Case 2. Case 2 is preferred, but can we afford to downlink 71 Gb/day of science data ? Pointing performances ? Level and nature of the jitter ? Is jitter correction needed? Exact threshold in magnitude between weighted photometry and aperture photometry ? Model for the PSF ? Resolution required for the jitter correction ? Resolution required for the calculation of the weighted mask ? Photometry of the saturated stars ? Down to which magnitude ? Calculation of the barycenter : thresholding ? simple mask ?