Please note that there are limitations inherent to the Operations Commissioning Phase (shared-risk environment). In the course of the last months we learned a lot more about the instruments and the environment that they are operated in and some prior unknown technical limitations were encountered. For ViSP for example, some of these limitations had an impact on the frame rate and as such the time spent on an individual slit position and map cadences the ViSP can achieve.
In general, summit science operations staff (i.e. resident scientists and science operations specialists) strive to match the requests of any observing proposal the best they can, but there are no guarantees that for example, the lengths of observations, cadences or requested seeing will be 100% consistent with the proposal. If you do have any questions about (summit) science operations and the execution of your observing program(s) you may also contact the DKIST Program Scientist for Operations atritschler@nso.edu or use the DKIST Help Desk.
These caveats cover all the ViSP datasets released so far through OCP 1.5. There have been many updates to the ViSP pipeline code during the OCP. Many of these have been to fix bugs or update the robustness of key routines such as the geometric correction. Some updates were in response to optical issues discovered during OCP while others were planned as part of ongoing polarization systems calibration work.
There is a problem with the ViSP CDELT2 keyword (see also the following issue), which is the keyword giving the pixel plate scale along the slit. In the early data that was taken with ViSP this keyword provides a wrong value. We have not yet had the chance to fix it in the files that we made available to you but that is something that may be done in the future. Christian Beck, the ViSP Instrument scientist, has supplied the following correct values for CDELT2.
arm 1 630 nm: 0.0298"/pixel
arm 2 397 nm: 0.0245"/pixel
arm 3 854 nm: 0.0194"/pixel
The keywords NAXIS1 and NAXIS2 give the lengths of the spatial and spectral axis of the data array, respectively. However, the keywords CDELT1, CDELT1A, CRPIX1, CRPIX1A, CRVAL1, CRVAL1A, CTYPE1, CTYPE1A, CRDATE1, and CRDATE1A all refer to spectral quantities as opposed to spatial quantities. There may be other keywords that are "transposed" as a result of this that issue. This results in the WCS axis 1 referring to pixel axis 2 in the data array and vice versa.
There are issues with the calculated wavelength and dispersion based on the header information in ViSP as illustrated in Figure 1. No attempt is made to correct these at the moment. A wavelength calibration for ViSP data is being investigated for inclusion in a future version of the calibration pipeline.
Figure 1: Observed and Atlas Spectrum for Ca II 8542 A.
The ViSP instrument scientist has supplied a better approximate calibration of the wavelength in the 3 arms, show below in pseudo code.
630nm = 629.495+findgen(1000)*1.285/1000. ; dispersion 1.285 pm /px
397nm = 396.418+findgen(1000)*0.77/1000. ; dispersion 0.77 pm / px
854nm = 853.182+findgen(1000)*1.882/1000. ; dispersion 1.882 pm / px
The ViSP data has incorrect pointing information in the WCS headers. This is not currently correctable.
We note the current known issues with the ViSP data. Th
Data analysis and testing has identified multiple sources of stray light in ViSP. One set of tests done with showed that a significant part of the stray light enters from the sides and top at the end of the camera arms that were open to the environment. Reduction of stray light to about the same level as during a dark current observation with the GOS dark shutter in the beam was only possible when both the camera arms and camera lenses where completely covered. Installation of various baffles was done between May and June. These were successful in reducing the stray light from external sources. Data taken through at least OCP 1.4 will be contain stray light from this source.
A second source of background light is identified to come from within the beam itself and is only seen when the beam is allowed to pass through the slit. It affects all ViSP data currently taken. This second source cannot be mitigated with external baffles or enclosures, and must be mitigated using other means. It has a different signature (spatial and spectral) at different wavelengths and/or ViSP arms. Analysis of this signal shows it to be additive and mostly unpolarized, much like a dark or background signal. This signal is a much more significant contribution in frames with overall low flux (e.g. 396 nm and 854 nm channels, and the dimmer of the two dual beams). In order to mitigate this, the Data Center is currently using an ad hoc algorithm created by the Polarization Scientist Dave Harrington, that uses the PolCal frames taken at a single slit position. We use the assumption that the modulation should be spectrally constant over the ~1nm bandpass covered within a ViSP camera arm. By normalizing each of the raw PolCal intensity spectra to the mean over all spectral pixels, we get a spectrum compensated for the intensity modulation. Variation in these normalized intensities with wavelength is measure of the stray light impact. Spectral invariance of modulation has been confirmed in each camera arm, and also by comparing the intensity modulation curves of both orthogonally polarized beams recorded strictly simultaneously in the dual beams of each camera. The worst behavior has been observed in the 854nm channel, in one of the two beams as seen below:
The algorithm finds a single background unpolarized spectrum, that when subtracted from all the individual modulated spectra, minimizes the spectral variation of the mean-normalized intensities. These background intensities correlate well with a known stray light optical pathway. An example of the stray light background in the 854nm channel from June is below. We note that this background is recorded after, and is not impacted by installation of the external baffles discussed above.
If we then subtract the background signal from all the individual spectra, prior to normalization, then we get the following modulation-normalized spectral shape:
There’s still some residual difference between the normalized spectra, but overall, the normalized spectra look much more spectrally constant. The resulting modulation curves similarly agree much better in overall contrast and uniformity. Ultimately the best way to deal with this issue is to remove the stray light with appropriate optical aperture stops and masking. Work on this topic is ongoing. The algorithm presented above is not a perfect solution. It is only intended to get data “good enough” at this point, and further tuning of the algorithm settings is necessary. You may notice that there are some line artifacts visible. If you do have any questions or if you see any issues with line signals, you are encouraged to ask (DKIST Help Desk.)
During calibration testing several issues were noted with both the lamp and solar gain tasks.
Lamp gains were being used directly. The lamp was designed to be bright, (1% to 15% of the full solar on-disk DKIST beam) but this came at the cost of spatial uniformity. Many of the lamp gains have their own strong optical response and thus this response was affecting all downstream data (both PolCal and science).
The Solar Gain calculation attempted to preserve detector variations through some complicated interpolation steps. Not only was this fundamentally incorrect but the repeated interpolation of narrow solar spectral lines (especially present in 630nm data) left very large spectral residuals. These residuals also affected.
To remove the optical variations from the Lamp Gain we apply a High-Pass Filter (HPF). The detector variations are, by their very nature, at the highest frequency possible in the image so a HPF with a very high cutoff frequency can successfully remove all optical variations from a gain image, leaving only detector variations. Tuning the cutoff frequency of the HPF needs to be done on a per wavelength (arm) basis and researching the best frequencies to use is still being investigated.
To identify and characterize the solar spectrum in the Solar Gain images we need to account for variations in the spectral shape along the slit that occur as a result of physical non-uniformities in the actual slit construction. In other words we can’t simply take the median spectrum along the slit because the true solar spectrum varies along the slit. To compute the “characteristic spectra” we run a moving 1D Gaussian average along the slit. The width of this Gaussian window is an important tunable parameter.
The core gain algorithms are simple: use a filtered Lamp Gain to remove detector variations, and use a Solar Gain (with solar spectrum removed) to remove several other optical variations. There are some important parameters to tune and further work is necessary to determine the best values for each wavelength region. In some/many cases we may be unable to perfectly remove all optical variations.
Further improvements have been made to the geometric calculations: The algorithm for rotation between beams during the dual-beam merge was improved to fix some failures with some datasets. The algorithm to compute XY shifts between individual modulation states was updated.
An efficiency issue with beam 2 / arm3 (currently 854nm) have been noted.
854nm Beam2 (Arm3) has anomalously low modulation efficiency (35% vs 50%). 397nm Beam2 (Arm2) also has reduced efficiency. It is suspected that this is due to low polarization beam splitter contrast. Optical mitigation likely will be required, and modeling is ongoing. We note that subtraction of the intrinsic stray light (above) improved the modulation efficiency, but it remains at / below 38%.
Spatial scale for demodulation sampling is yet to-be-finalized. The initial “checkerboard” interpolation pattern, seen in earlier releases of ViSP data has now been fixed. However, several other issues create background signals. Further assessment is necessary to find the appropriate trade-off between errors and smoothness. The QUV continuum levels are at variable levels around 1%. We are currently investigating different observing techniques to also improve the polarization zero point performance.
Cross-talk Possibilities
Please check the quality report for your data set to note any failures in PolCal fitting outputs. We have seen data sets where certain variables (transmission of polarization calibration optics) are far away from metrology expectations. For example, in some cases the variable representing the polarizer transmission is fitted to be near 100% transmission when the optic is known to be 91.5% transmission +/-0.3%. This fitting error can create cross-talk of all types through the correlation between fitted variables. The settings for polarization calibration are currently being investigated, and we expect improvements in the demodulation matrix accuracy as the algorithms are tuned.
The following issues have been found / are being worked on with VBI datasets.
A wrong value is stored in the CRPIX[N] fits header keywords for proposals carried out in OCP 1.2 / 1.3 (February 2022 through April 2022). This should be corrected for runs from April 2022 onward.
We are working on understanding the response of the DKIST Wavefront Correction System (WFC) to varying atmospheric seeing conditions. The WFC system performance is still in the process of being optimized. The Fried parameter keyword AO___001 within your data set headers provides an estimate of the prevailing seeing condition. Due to technical limitations in the way the estimate is generated by the WFC system, the value provided can be misleading. Therefore, the reconstruction process occasionally fails even if a good Fried parameter is estimated. Furthermore, unrealistic Fried parameters (in the meter range) are estimated whenever the WFC system encounters conditions that are too severe for operation. In that case, a complete image reconstruction is not attempted.
For a movie demonstrating these issues please visit: [LINK]
We have encountered unexpected technical issues with the VBI cameras leading to a variety of noise artifacts in the data. This includes an overall dynamic noise pattern in the images which is amplified by the reconstruction process, a vertical stripes pattern in the images, and increased noise at strong gradients in the images as seen in particular in high contrast images (see Ca II K image at [LINK]).
We have developed a variety of algorithms to improve the data quality. These issues continue to be approached from multiple angles to provide further improvements and solutions.
In a few rare cases you might notice an overexposure in the G-Band and the Blue Continuum images which worsens during the observing sequence (as the sun rises). If there is a second observe sequence on the observing day, the exposure time will have been adjusted for this second run.
Please don't hesitate to contact us if you have questions.
Alisdair Davey
DKIST Data Center Scientist
adavey@nso.edu
Alexandra Tritschler
DKIST Program Scientist for Operations
atritschler@nso.edu