CfAO FAll Retreat Strehl Meeting In attendance: Julian Christou (UCSC) Russell Makidon (StSCI) Francois Rigaut (Gemini) Remi Soummer (StSCI) Don Gavel (UCSC) Kieth Knox (AMOS) Marcos van Dam (LLNL) Marshall Perrin (UCB) Anand Sivaramakrishnan (StSCI) Lisa Poyneer (LLNL) Mitchell Troy (JPL) Rich Dekany (CalTech) The discussion started with a presentation by Julian Christou on the nature of the Strehl Campaign. Julian noted the purpose of the campaign was to take a relatively simple synthetic dataset of controlled sources of uncertainty, etc. and distribute them to the group to see how different methods of measuring Strehl ratio performed under controlled circumstances. Christou and Makidon discussed the creation of the recent dataset and the results of the Strehl measurements from the group. As a reminder... The initial set of StrehlCampaign images were created assuming a set of five independent 512x512 Kolmogorov-spectrum phase screens with D/r_0 = 12.0 across a Palomar geometry pupil of 256 pixels across. PSFs were created by co-adding the five PSFs 'imaged' through the individual phase screens at lambda/8D oversampling. The file "TelescopePupil.fits", the exact telescope pupil used for these simulations, was also provided to enable direct comparison or computation of the perfect PSF. `Parabolic' AO correction was assumed for these simulations, as described by Eq. 9 of Sivaramakrishnan et al. 2001, ApJ, 552, 397. The `Parabolic AO' model was determined via an empirical examination of Palomar AO data, and the authors thus felt this was a valid model for these simulations. Variations in Strehl ratio between images were achieved by varying the number of AO actuators across the telescope pupil. The group then started a discussion centering on how each person determines Strehl ratio and is summarized as follows:. Marshall Perrin's code assumes a bias in oversampling. According to Marshall, "cranking up the oversampling" results in on over-optimistic Strehl calculation. Marshall uses the image shift routine with a bi-cubic spline to shift his analytically-determined perfect PSF to match the centroid of the data and calculates Strehl based on direct comparision. Julian noted that the details of Lewis Robert's Strehl ratio algorithm were as yet unknown (NB - They have recently been uploaded to the web page - J.C.). Julian takes the given pupil function (provided with the distributed PSFs) and calculates a perfect PSF on a lambda/8D oversampled grid. He then iterates to find the optimal shift between the centroid of the perfect PSF and the centroid of the aberrated PSF. He then measures the peak pixel after a shift of the perfect PSF in Fourier space - shift is measured by calculating the first-order moment truncated to a 3 x 3 or a 5 x 5 box. Francois Rigaut and Remi Soummer suggested people should consider binning their perfect PSFs by dividing by a sinc function in the Fourier domain, and zero-padding to compare centroids, etc. Essentially comparing MTFs instead of PSFs? Lisa Poyneer gave a description of Don Gavel's code, which she's using for her analysis. It uses a pixellated pupil to create a reference PSF centroided to match the data. The perfect PSF is generated at the same sampling as the data - there may be an issue with respect to sampling the perfect PSF v. binning the perfect PSF, which may be a reason Lisa's Strehl ratios are of order 10% lower than 'truth'. Marcos van Dam calculates the numerical PSF and compares that with the data. He FFT interpolates and up-samples the data to match the perfect PSF. it appears the difference between sampling and binning may be a reason Marcos's results also trend of order 10% below 'truth'. Mitchell Troy first finds the centroid of the data and compares that with the perfect PSF (note: I don't believe we learned how Mitchell actually generates the perfect PSF). He then interpolates to find the peak values in the data and in the perfect PSF; Mitch compares systematics by input of a PSF of 50% Strehl ratio as part of his quality assurance test (very useful for real data). Frank Marchis uses ESO's ECLIPSE code. At the time, there was no concensus as to how ECLIPSE actually works. A question was presented by Julian Christou: what did everyone use for normalization? It's likely everyone used a different area for their PSF normalizations. Both Julian and Lisa used boxes of 256 x 256 on the Nyquist and 2*Nyquist datasets. A short discussion on the results from the campaign using Bruce's data. The results were very similar to those using this dataset, even though Bruce's data was slightly more complicated than these data, and included PSF centroid variations, etc. The question was put forth: where do we go from here? One goal of this project would be to write a paper describing our results. Will we converge on a "best" algorithm to calculate Strehl ratio? Do we have a "roadmap to Strehl (calculations)?" What about the suggestion by Rigaut and Soummer, working in the Fourier domain instead of oversampling? Anand noted the work he and I have done for the JWST NIRCam simulations, using actual detector data as the basis for the simulations. Should that be our next step? Marshall's suggestion about what makes a good Strehl code: an "oversampler" or centroid finder; a "peak finder"; a "perfect PSF generator"; and a "baackground subtractor". Marshall questioned how well we really need to know our Strehl ratios anyway? What is the Strehl ratio really good for? Is a 5% difference relative to "truth" sufficient? What about the question of variability in the measured Strehl ratios, due to noise differences, etc.? Rick Dekany proposes (and the group agreed) that the Strehl Campaign be used to define a standard method for determining Strehl ratio. Can we define a (normalization) window size standard? The group ended the discussion by agreeing on the next dataset: same PSFs at the same Strehl ratios with the inclusion of simple noise sources (photon noise, Gaussian read noise, etc.) and background subraction.