The typical IR array read sequence begins by reading a frame with nearly zero exposure, in the case of the Ohio State ICIMACS systems [1,2] with Hawaii arrays, by resetting a full line, reading that line, and advancing to the next line. This "pre-read" is stored digitally. The exposure timer is started at the beginning of this pre-read process. When the desired exposure time is complete the process is repeated, with the exception of using dummy delays instead of the line resets, to form the post-read frame. The pre-read frame is then subtracted from the post-read frame. Structure in the pre-read frame is dominated by the offsets from the individual preamplifiers associated with each pixel, and these offsets subtract out. There is also, inevitably, a small amount of coherent noise at the PLF and harmonics. The image in Fig. 1, from the Ohio State MOASIC/TIFKAM , was obtained by subtracting two frames taken as outlined above.
The diagonal banding is the result of the random phase of the line frequency disturbance with respect to the read process. The bands point toward the center because this type of device reads out in quadrants arranged so that the read process proceeds from the corners toward the center. Figure 2 was taken in an identical fashion with the exception that the start of both pre-reads and post reads are synchronized to the PLF phase. The RMS in Fig. 1 is 2.74 data numbers while the RMS in Fig. 2 is 2.67 data numbers. Thus the improvement in noise level is largely cosmetic.
Figure 3 shows the circuit used to generate the synchronizing signal. The output of a small 12 volt transformer is clipped with d1 and d2. A low pass filter formed form R1 and C1 attenuates any spikes. Comparator U1 then detects the sign of the resulting signal. This square wave is applied to the RI (ring input) pin of a UART, though any available digital input that can be poled could be used. A hardware interrupt could also be used. In our case the RI input is sampled at ~50 kHz and the read process is started when a zero is detected on sample n and a one on sample n+1.
Note that the scheme fails if the line frequency is changing rapidly enough that the total number of power line cycles in a pre-read and a post-read are different. The effect would be good cancellation of the PLF disturbances in the pixels read first, where the phases match due to the active synchronization, and decreasing cancellation as the read process continues. If the pre-read PLF were dramatically different for the pre and post reads there would be bands of good and bad cancellation as the pre and post PLF beat.
Was this article helpful?