IRS: Rogue Pixel FAQ

• WHAT IS A "ROGUE PIXEL"?

  A rogue pixel is a pixel with abnormally high dark current and/or photon responsivity (a "hot" pixel) that manifests as pattern noise in an IRS BCD image. In background-subtracted images, a rogue pixel may appear negative. The term "rogue" was used originally to distinguish pixels whose hotness was unpredictable, but rogue pixel masks can include those that are permanently as well as temporarily hot. At current bias levels rogue pixels constitute about 6% of the illuminated portion of the Long High array and 2.5% of the illuminated portion of the Long Low array. Less than 1% of the Short Low and Short High arrays are rogue pixels.

• HOW DO WE IDENTIFY ROGUE PIXELS?

  We identify rogue pixels using the dark current measurements taken for calibration of the IRS arrays (observations of a very dark spot in the sky). The rogues are chosen by thresholding a high spatial frequency "component" of the dark current image. All pixels in the high frequency image with dark current greater than 4-sigma are labeled rogue.

• WHAT IS A ROGUE MASK?

  A rogue mask is a fits file containing a 128 x 128 image with all detected rogue pixels set to 1, and all other pixels set to 0. Rogue masks are available for every IRS module and campaign starting from IRS nominal operations campaign 1 (IRS1) and extending to the present. A new rogue mask will be made available on the web site within two weeks of the end of a campaign.

• WHY IS THERE A ROGUE MASK FOR EACH CAMPAIGN?

  Because the rogue pixels are a time-dependent phenomenon (this is why they are called "rogue"). Observers who need to correct for the effects of rogue pixels on their data should use a rogue mask developed as close in time to the date of observation as possible. For many observers, the campaign-based rogue masks are a good starting point. The Campaign-based Long High rogue masks made from the beginning of Nominal Operations (December 2003) through IRS24 show an average net increase of about 1.5 new rogue pixels per day (see Figure 1 below). After the LH bias change in IRS25 (October 2005), the rate decreased by a factor of 2. The rate for Long Low was 0.7 new rogue pixels per day up until the LL bias and temperature change in IRS45 (October 2007). Given the average time between the beginning of one campaign and that of another (about 4 weeks), the current trend in LH and the trend for most of the lifetime of LL yielded a little over 20 new rogue pixels from one campaign to the next. To keep track of these changes, we provide a new rogue mask each campaign.
  
  Figure 1: Number of rogue pixels detected in Long Low and Long High campaign-based dark data as a function of time. A linear fit to LH data from Campaigns 1 through 24 indicates an average net gain of 1.5 new rogue pixels per day. A separate fit to Campaign 25 and later LH data yields 0.7 rogue pixels per day after the bias change. The trend for LL up until Campaign 45 is very similar to the post bias-change LH behavior. The LL bias and temperature were reduced in IRS45, decreasing the rogue pixel count by a factor of 3. PDF version

• WHAT CAN I DO WITH A ROGUE MASK?

  Use rogue masks with care. Since they are derived from dark data, they may not indicate the exact distribution of hot pixels when the arrays are exposed to photon sources of moderate or high brightness. Furthermore, rogue pixel detection is not an exact science and the rogue masks we distribute are derived from dark measurements taken over as large a period as three weeks. That said, the rogue masks do provide users with a priori knowledge about which pixels are likely to be hot and at least may help explain unusual spikes in extracted spectra. In the extreme case, the mask could be used to eliminate pixels from consideration in extraction (eg in SMART or SPICE). However, we do not recommend doing this, since eliminating pixels in regions of bright emission cause unphysical negative spikes in the extracted spectra. The masks are best used to identify pixels to be replaced or corrected by interpolation and other image cleaning procedures.

• IS THERE A SOFTWARE PACKAGE THAT USES THE ROGUE MASKS TO FIX MY DATA?

  A rogue mask editing and image cleaning tool called IRSCLEAN is available from the SSC. For details, see the web site. This software produces files that are compatible with SPICE.

• SHOULD I ALWAYS RUN IRSCLEAN ON MY DATA?

  No. If your source is bright enough that the rogue pixels do not make a significant contribution to the flux, or if you were lucky enough to get a good match to the rogue pixels in background subtraction, you may choose not to clean any pixels. This is of course the best option, since "cleaning" a pixel in effect throws away the data for that pixel. The tradeoff should always be improvement in the extracted spectrum.

• HOW DO I KNOW IF A ROGUE MASK IS APPROPRIATE FOR MY DATA?

  You should always check the match between a given rogue mask and your data. In particular the pixels in the rogue mask should be significantly brighter than their surroundings (NOTE: for background-subtracted data, rogue pixels can also be much fainter than surrounding pixels). Sometimes this is obvious from visual inspection. You can use irsclean_mask to overlay a rogue mask on your data, examine the brightness of pixels, and add or remove pixels from the mask: IDL> irsclean_mask,[data file name],inRmask=[rogue mask file]

• HOW CAN I USE A ROGUE MASK TO FATALLY MASK PIXELS IN SPICE?

  This is not recommended, since eliminating pixels in regions of bright emission cause unphysical negative spikes in the extracted spectra. Instead of masking out rogue pixels, we recommend using irsclean_mask to "clean" your data before running SPICE. This produces more stable results.
  The binary rogue mask needs to be converted to a 16-bit unsigned integer mask to be used in the SPICE extraction pipeline. Since the fatal bmask value is currently 28800, the rogue mask should be multiplied by 28800 and saved to a file which is then used as the mask input to SPICE. The rogue mask can also be combined with the pipeline bmask using the boolean "OR" operation. For example, in IDL one might type:
  
  IDL> rmask = READFITS('b3_rmask_IRS9.fits')
  IDL> rmask16bit = rmask * 28800
  IDL> bmask = READFITS('bmask.fits',bmask_header)
  IDL> b_r_mask = rmask16bit OR bmask
  IDL> b_r_mask_header = bmask_header
  IDL> SXADDHIST,'Combination of BMASK and Campaign rogue mask',b_r_mask_header
  IDL> WRITEFITS,'b_r_mask.fits',b_r_mask,b_r_mask_header
  

• WHAT DO THE FILENAMES MEAN?

  The mask file names are of the format b?_rmask_IRS?[?].fits, where the first ? is the module number (SL=0, SH=1, LL=2, LH=3) and the second ?[?] is the campaign number.

• HOW DO I KNOW WHAT CAMPAIGN MY DATA BELONG TO?

  The file campaigns.txt, included with this release, provides the translation between IRS campaign number (starting at 1) and the SSC campaign label given in the FITS header keyword CAMPAIGN in your data (starting at IRSX002500).

• WHY ARE THERE FOUR MASKS FOR EACH MODULE IN IRS21 AND TWO MASKS IN IRS23 AND IRS51?

  Campaign IRS21 was split into four mini-campaigns due to intervening short IRAC and MIPS campaigns for Target of Opportunity observations. The SSC aign label (fits header keyword CAMPAIGN) is different for each of the mini-campaigns, so we created four identical rogue masks corresponding to IRS21.1, IRS21.2, IRS21.3, and IRS21.4. We derived a single rogue mask per module using dark calibration data for the entire campaign, and then copied the masks into four files per module. IRS23 was split into two campaigns, IRS23.1 and IRS23.2, due to an intervening short MIPS campaign. IRS51 was also split into two campaigns, IRS51.1 and IRS51.2, to observe a gamma ray burst afterglow with IRAC. All sub-campaign rogue masks are identical, but are associated with a different campaign label (see campaigns.txt).

• WHY DID THE NUMBER OF LONG HIGH ROGUE PIXELS SUDDENLY DROP IN CAMPAIGN 25?

  The bias voltage on the Long High array was reduced from 2.0 to 1.6 volts in IRS25, reducing the dark current and photon sensitivity of the array. The main reason for this change was to reduce the strength and number of rogue pixels and improve the signal to noise in observations of faint sources. For many purposes, the data and rogue pixel masks before and after the bias change should be treated as coming from two different instruments.

• WHY DID THE NUMBER OF LONG LOW ROGUE PIXELS SUDDENLY DROP IN CAMPAIGN 45?

  For the same reason we changed the bias in Long High, in IRS45 the bias voltage on the Long Low array was reduced from 1.8 to 1.6 volts and the array temperature was reduced from 4.4 to 4.1 K. As for LH, this reduced the dark current and photon sensitivity of the array. In the case of LL, the number of detected rogue pixels decreased to 1/3 of its previous value. This combination of effects increases the overall signal-to-noise ratio in the LL array by up to 60%, depending on wavelength.

• WHY DID THE CAMPAIGN 25-27 LONG HIGH ROGUE MASKS CHANGE ON FEBRUARY 8, 2006?

  All IRS data are processed such that certain "permanently bad" pixels are automatically excluded from consideration and set to the value NaN (not a number). Prior to February 8, the Long High dark measurements for all campaigns were processed using an old "permanently bad" mask, or "pmask", derived at the old Long High bias setting. Thus, many pixels that were actually well-behaved at the new bias were excluded from consideration. Using a new pmask, we were able to open up many pixels that had previously been designated NaN. Unfortunately, we also opened up a set of pixels that are rogues at the new bias, about 200 of the excluded pixels.

Go back to Rogue Pixel page

Go back to IRS page