Further informationQuestions regarding this page should be directed to Kentaro Aoki ( ).
Time variation of spectral throughput in several fibers
This section highlights the feature visible in a small fraction of fibers as shown in Figure 1. This image was created by dividing the image of 1-D domeflat spectra from IRS1 in the Low Resolution mode (LR) by another one that was taken at a different time (so each pixel in this image contains the ratio of ADUs between the two domeflat spectra), and applying a 10-pixel binning to the horizontal (i.e. spectral) direction. The color scale is set so that the variation of the ratio between 0.95 and 1.05 is visible most clearly. As the color changes from black, brown, to white, the pixel value varies from large to small. This indicates, although the ratio is close to 1 in most pixels, there are several fibers showing a wave pattern as represented by the spectra pointed by red arrows, implying that the throughput of the fibers changes with time by a few percent and the amount of the variation varies as a function of wavelength like a sinusoidal curve.
Assuming this variation persists in these fibers during observation
(probably it does), it is unlikely due to worse sky subtraction and
calibration that good quality spectra are extracted from these
fibers. More importantly, observers need to be careful not to
observe any stars for tellric absorption and flux calibration using
these fibers (at least they should not use spectra from these fibers
for such purposes), otherwise all the other spectra would be
infected by the wave pattern in the calibration process. Below is
the list of ID numbers of Echidna spines/fibers that show the
variation in the spectra:
As of Aug 2011, the cause of this variation has not been identified, but the wave pattern suggests the presence of some interference effect somewhere. It is also worth noting that it appears only in ∼ 4 % of the 400 fibers. One possibility may be that such an interference is occuring at the fiber-lens coupling in the fiber connectors. At both ends of the Echidna side and spectrograph side, the fiber tip was polished and was glued together with the flat-side of the plano-convex lens using an index matching adhesive, but unfortunately the quality of this gluing is not perfect in some fibers in the sense that there may be small air bubbles left between the fiber tip and lens surface. Looking at the "wavelength" of the wave pattern, it is suggested that interference is happening in a space of which size is roughly ∼ 0.1 μm (recall that the spectral coverage of LR is 0.9-1.8 μm), which might be explained by contamination of very small air bubbles in the adhesive.
Nonlinear time variation of bias component on the IRS2 detector
When a detector is exposed to some illumination, the pixel response should be proportional to the strength of the illumination. As photons come in, the pixel count should be increased monotonically with time and, in particular, the relationship between pixel count and time is linear in the so-called linearity regime of the detector. However, the response of the IRS2 detector to such constant illumination looks quite different: Measuring the counts in a pixel as a function of time in an exposure using non-destructive readouts (NDRs), a steep increase of the pixel count is seen for the first few minutes after an exposure starts and then the slope becomes much shallower as exemplified in Figure 2. Note that this feature is visible only when the detector is exposed to some faint illumination such as a night-sky spectrum in an astronomicaly observation.
We first doubted some non-linear response of the detector, but this was found out to be unlikely: By taking one NDR exposure with giving the detector some faint illumination ("on" frame) and another without such illumination ("off" frame) and subtracting the latter from the former ("on" minus "off" frame), the counts of pixels with illuminated in the "on" exposure turned out to increase linearly with time in the "on" minus "off" frame. This indicates the response of the pixels to illumination is linear. Next, we investigated so-called photon-transfer plot of this detector by measuring the noise level of each pixel as a function of pixel count, and found out that in the regime of low noise level, the pixel count is higher than expected. This implies that there is a contribution to pixel counts that is purely electric and does not contribute to noise level. It is also worth mentioning that the amount of such bias component as suggested in the photon-transfer plot is consistent with the zeropoint offset by the initial steep increase of pixel count in the linearity plot (Figure 2).
Last updated: Jan. 31, 2012
Copyrightę 2000-2011 Subaru Telescope, NAOJ. All rights reserved.