Previous Article Next Article Table of Contents
Reading Out CCDs: Where Does the Time Go? (1Jun92) (from KPNO, NOAO Newsletter No. 30, 1 June 1992) When CCDs had only a few hundred thousand pixels to be read, and readout times were half minute or less, observers exhibited only mild irritation, and their behavior was relatively rational, at least compared to their usual behavior. However, with the advent of larger CCDs, observers found millions of pixels waiting to be read at the end of each exposure and readout times of 5 minutes, and began to experience correspondingly more discomfort. There have been some changes made since the last Newsletter to try to reduce that level of discomfort by reducing the readout times of the devices that are in service. The goal was to introduce these changes in a manner transparent to the user. There were three major changes. First, the preclear of the array prior to integration was shortened by summing 10 rows and then clearing the horizontal register. This reduced the preclear time to about 11% of the previous time required. Actual time for each CCD is shown in the following table. Second, for those arrays with more than a million pixels, the hardware integrator components were changed to permit shorter pixel integration times and yield approximately the same gain, noise, and dynamic range. This saves 14 microsec per pixel, thus trimming 1 minute off the readout time of the 2048 x 2048 devices. The third change is beneficial when reading out only a sub-array of some region of interest (ROI). Since there are almost an infinite number of combinations of ROI, information is contained in the table and the first figure to permit optimization of readout speed for specific observing programs. Fast # Row Full Pre Preskip Pixels Pixel Read Frame CCD Clear 4 Rows Per Time Time Readout (sec) (ms) Row (us) (ms) (sec) TIx < 1 18.5 892 49 43.7 35 F3Kx < 2 18.5 3134 35 109.7 115 T2Kx < 7 41.0 2110 38 80.2 165 T1Kx < 2 21.0 1116 35 39.1 40 T5Hx < 1 21.0 604 49 29.6 16 S2Kx < 5 21.0 2110 37 78.1 160 S1Kx < 2 21.0 1085 35 37.9 39 [figure not included] What can be done to get even greater speed performance? Faster analog to digital converters, multiple readouts of single CCDs, and perhaps trimming a microsecond or two from some of the hardware overhead requirements. All these things are currently being pursued. Just where does the time go? The second figure demonstrates a typical pixel timing wave form. This time per pixel represents the bulk of the time required to read out a CCD. This waveform is for a single pixel with no binning. Vertical binning divides readout time by the binning factor (2, 3, 4) with a very small overhead for each vertical shift. Horizontal binning adds an HCLK time (see the second figure) for each binning factor (2, 3, 4) then divides that sum by the binning factor. Confused yet? Things will get even more confusing with multiple ROIs, readout amplifier selection/multiple readouts per CCD, and the MOSAIC. [figure not included] Rich Reed
Previous Article Next Article Table of Contents