CCD Tests

Test of electrical levels

Once the circuit was assembled, the levels of voltage in the KAF sensor socket were verified according this table:

CCD sensor tests

The tests were carried out by Rodrigo Badinez. Prior install the sensor, it was taken an offset image without ccd, with the Pisco software. The average was about 460 counts, with a standard deviation of 10 units.

The next step was install the CCD sensor, and take a bias image (0 seconds exposition, 1x1 binning mode). The first obtained images were fully saturated, even in complete darkness. This leads to analyze the video signal with an oscilloscope, and we notice than the clamp circuit was not working because the U7 amplifier was saturated. A resistance of 10K ohms (named R100 in the schematics) inserted before the OpAmp solve the problem.

This way we took another bias image that wasn't saturated. The average of this image was about 6300 counts, with a standard deviation about 800 units.

First Light

For this test the camera was sealed and the test lens was installed. We try to get an image of a coaxial cable connector, located approximately 1.2 meters away from the camera. The connector was illuminated only by a small torch. Hereafter is the first light image.

The vertical lines are due the absence of the shutter, in the process of reading still photons are converted to electrons. Making the shutter effect "manually", and focusing more exactly, we got another image.

Noise reduction

As the first images has a very high noise,  we try several  ways to diminish it. The first obvious one was to connect the Peltier to decrease the temperature of the CCD. The effect was notorious (specially in the number of hot pixels) but most of the noise remains.

Typically the power supply is the main cause of noise in electronics systems. By this we test another power source (of better regulation) to see if the noise was reduced, but didn't happen. Then we built new shorter and ground shielded power cables, and test it. The noise now was reduced significantly.

On the offset image, the standard deviation was reduced from 10 to 4.3 units. On the bias, the standard deviation was reduced from 800 to 325 units.

As the noise is low, the horizontal lines stand out more. This lines doesn't appear to be related with the image, and are located randomly. That's suggested us that were reading errors, due to synchronism loss on the clocks generated by the computer.
    To test this hypothesis, we close all the other running programs an process on the computer, take another image and check if the amount of line had diminish. And exactly that happened. Then we increase the priority of the PISCO process to real-time, and finally the horizontal lines disappears.

Finally I take some new pictures to see how much the image quality has improved.

Noise analysis with MATLAB

In order to have an indicator of the amount of noise, using the PISCO software we calculate the subtraction of two bias image at the same temperature (same value of resistance on the PT-100). The following image was obtained (at 101.5 ohms): the mean value was 0.09 counts, the median was 3 counts, and the standard deviation was 35.8 units.

As the image is the result of a subtraction of two, the readout noise is 35.8/sqrt(2)=25.3 ADU (Analog to Digital Units; in Audine 1 ADU equals a signal of 2 electrons). This value is a little higher than the expected, so Rodrigo Badinez made a frequency noise analysis in MATLAB. The method used was the following one, using the fftnoise2.m command.

• Read the FITS image and load it into the MATLAB Workplace as a matrix.
• Prompt the pixel clock, and the line frequency of the image.
• Calculate the one-dimentional Fourier transform to all the rows of the image.
• Generate a new image with the transformed rows, and export it to FITS format.
• Average all rows, and plot the result. Also plot the center line of the transformed image to compare it.
• Repeats the 3 previous steps, but now for the columns.

Hereafter are the plots of the transforms obtained for the bias image (temperature at 101.5 ohms). The frequency range is a theoretical reference, it's not verified.

There isn't a specific frequency peak on the average of FFTs (except of the central), therefore we conclude that the noise remaining is white noise. The most probable reason for this noise (on this camera) is the thermal Johnson-Nyquist noise from the resistances.

Lens used for the initial tests

The lens used for the first tests, was the Nikon Nikkor 50mm/f-1.4, with a special adapter manufactured here in Paranal.

• Lens Type: Lens
• Maximum Aperture Range: F/1.4
• Minimum focal length: 45 millimeters
• Minimum Focal Range: 17.7 inches
• Focus Type: manual-focus
• Real Angle Of View: 46
• Mounting Type: Nikon F
• Weight: 8.1 Ounces
• Item Display Diameter: 2.6 inches
• Length: 2.5 inches