The "bars" are narrow stripes that cross the quadrants of an array, and occur identically in all 4 quadrants at the same rows/columns in each. It is thought that these bars are caused by pick-up of an amplifier signal on one of the row/column address lines, causing a momentary change in the bias for that pixel. The bars became more common in data taken after a software patch to turn the readout amplifiers on during autoflush in late August of 1997, but were also seen before that time.
They typically run the length of a quadrant (128 pix), and are 3 pixels wide - the first pixel is lower than the mean, the second is at the mean level and the third is higher than the mean, giving the impression of an undersampled sinusoidal spike with an amplitude of up to ~10 DN peak-to-peak. If a bar appears in the 0th readout, it will be subtracted from all the other readouts as part of the normal calibration process, and will appear to be the negative of the above description. The bars are almost always broken in at least one place, with a shift of 2-10 pixels in the narrow direction. These sometimes correspond loosely with the breaks in the "bias jump" bands in the last readout of a MULTIACCUM sequence, indicating they may be related to the same electronic phenomenon.
The bars run parallel to the slow readout direction, which is vertical in NIC1, and horizontal in NIC2 and NIC3. Normally there is only one bar per quadrant, but occasionally there are more, always reflected in all 4 quadrants. Sometimes the bars will be "in sync" and appear at the same place in successive or every other readout, or may appear to march across the frame over the course of a MULTIACCUM sequence.
See memo for fix. The bars problem can be identified and corrected in a non-interactive fashion. We have developed and included a bars correction step in the version of CALNICA released in May 1999. The new algorithm searches for and flags pixels affected by bars in each readout. When the reads are combined in the "CRIDCALC" step of CALNICA, the flagged pixels are rejected from the fitting process.
Since Cycle 11, we have implemented a modified readout sequence for the three NICMOS cameras which reduces the probability that a detector is being reset while another is read. This procedure is completely transparent to users and should significantly reduce the electronic bands problem.