The data is taken in the same manner as August, i.e. we used the thermal emission from aluminum plate on instrument window as a light source. Measurement was done when MOIRCS was detouched from telescope and stayed in the IR stand-by room. No day-time activity was done as it was holiday. K-continuum filter is used. This time we apply three WIPE to the detectors between each exposure set. The result shows much improvement. We take 8 images with various exposures (25 to 350sec with NDUMMYREAD=2) for various count level. The stability of the thermal emission was carefully checked, and confirmed to be fairly stable during the measurement(~1%). Bias was not subtracted because it was proven to be totally negligible in previous measurement. The graphs below are the result of the linearity measurement.
Figure 4a.
Figure 4b
Figure 4a: The result of the linearity measurement. Horizontal axis is the exposure time and vertical axis is the median counts of each detector in ADU units. Fitting results are also shown. Figure 4b: Deviation of the data from the linear fit. The linearity is now shown as extremely good (-0.2% to +0.4% in range) until ~30000 ADU.
The result of linearity measurement shown above is an order of magnitude better than the estimate we did during August. The difference between these dataset is whether we applied WIPE or not. Although we still does not address the origin of the discrepancy yet, we consider that the new data should be more realistic because the WIPE is applied before each exposure in the usual operation. Another support is the value of the estimated Gain (below), which is similar to the laboratory value we tested before.
Next we show the result of the gain estimate. We again apply the standard photon-transfer curve method. This method assumes that the rms noise of the image is purely from the photon-statistics and the read-out noise. We combined all images except the 1st frame for each exposure data under 3-sigma clipping, then the sigma images are simultaneously generated by the IRAF imscombine task. We measure the mid-point from the combined images, as well as their rms noise from the sigma images. The region [700:1300,700:1300] is used for statistics.
We plot the variation(=sigma^2) with the mid-point values. It should be a linar relation. Suppose that the average image count is N (ADU) which is actually shown as g (e-/ADU) * F (e-), and the image rms noise is E (ADU) with the readout noise of R (e-). Then there is a relation as follows.
The result is shown in Figures 5.
Figure 5.
The result of the gain and read-out noise estimate is summarized below.
The values for chip 2 is larger than the estimate by Ichikawa et al.(2007, SPIE) which showed the gain value of 2.86 e-/ADU. The discrepancy may arise from the difference of the way of estimating the image statistics. In Ichikawa et al. they used the data taken by the partial-read mode as well as the whole read mode. Later we recognized that the use of partial-read mode may introduce a bias to the raw count data (see Information page of MOIRCS website). The new data shown here is based on the whole read, and no such uncertainty is introduced.
Readout noise for each detector is also independently estimated using the sigma images by 10 21-sec dark images, assuming that the rms noise in each pixel is dominated by readout noise. The result is 35.6e- and 33.7e- for chip 1 and 2, respectively. The result is larger than the one estimated by photon transfer method. The direct estimate by dark images may be more realistic value, as the estimate of cross section by photon-transfer method tends to be affected by the small errors in fitting.
In the graph the horizontal axis indicates the median counts around the center of each detector. The vertical axis is the offset from the linear fit (assuming the y-cross section of 0). The fitting is performed using the data less than 30000 ADU (across the red arrow). The data by new chip (Chip 1) shows almost the same behavior as Chip 2, indicating that the new detector has enough good linearity performance for scientific use. Note that the counts below 5000ADU shows relatively large deviation. We suspect that it is caused by the persistence and not true.