1. Pixel Non-Uniformity Study
Owen Miller
23/04/2008

2. 1. Pixel Non-Uniformity
Pixels do not seem to behave uniformly with the same threshold settings.
This non-uniformity is visible on threshold scans (shown in the next few slides).
This is probably a contributing factor to the low efficiency demonstrated by the sensor so far.
Hopefully by understanding and correcting this behaviour we can improve the sensor efficiency in future.
Owen Miller
23/04/2008

3. Some Samples of individual Pixels:
This is what a single pixel threshold scan looks like
Owen Miller
23/04/2008

4. Individual Pixel Behaviour by Column
Some examples of pixel non-uniformity in the same column, and in the same row: As you can see there appears to be no correlation between pixel behaviour and either row or column number except that there is a definite difference in behaviour between shapers and samplers (which we will return to).
Owen Miller
23/04/2008

5. Individual Pixel Behaviour by Row
Some examples of pixel non-uniformity in the same column, and in the same row: As you can see there appears to be no correlation between pixel behaviour and either row or column number except that there is a definite difference in behaviour between shapers and samplers (which we will come back to).
Owen Miller
23/04/2008

6. Mean thresholds Across the Whole Sensor
Some examples of pixel non-uniformity in the same column, and in the same row: As you can see there appears to be no correlation between pixel behaviour and either row or column number except that there is a definite difference in behaviour between shapers and samplers (which we will return to).
Owen Miller
23/04/2008

7. Variation Between shapers and samplers
Note the Shapers are centred on zero while the samplers are skewed to the right, centred around 40 or 50
Owen Miller
23/04/2008

8. Overall variation in behaviour
Peak Thresholds
Mean Thresholds
These histograms show how the behaviour of different pixels varies across the whole sensor, demonstrating a range of peak points and means.
Owen Miller
23/04/2008

9. 2. Pixel Trimming
Pixel non-uniformity can (in theory) be dealt with by assigning separate trim values to individual pixels.
A preliminary trim file was made based on the average threshold of the individual pixel threshold scan histograms, here’s how it works:
Pick a target value for the mean.
For each pixel calculate the trim value which should result in that pixel having the target mean.
Sounds simple, but the tricky part is selecting a target mean that minimises the number of pixels that would need impossible trim values (<0 or >15).
Pretty much says it all; the way we pick the target mean is by testing all the means between the maximum mean and the minimum mean and seeing which one works best
Owen Miller
23/04/2008

10. The Impact of Pixel Trimming on Threshold Peaks
Before
After
Hopefully you should see this distribution get narrower when a trim is applied, the narrower the distribution, the better the trim.
Owen Miller
23/04/2008

11. The Impact of Pixel Trimming on Threshold Means
Before
After
Hopefully you should see this distribution get narrower when a trim is applied, the narrower the distribution, the better the trim.
Owen Miller
23/04/2008

12. 3. Improving the Trimming
As you can see the pixel trimming is a long way from perfect.
Some possibilities for improving pixel trimming are:
Calculate trim values separately for shapers and samplers.
Use different statistics in as the basis for calculating the trim values.
Overall pixel behaviour will depend on which values are used to calculate the trim settings for individual pixels.
So what exactly would you like to see?
At present when we try to line up pixel characteristics using trims we are trying to line up all the pixels on the same value, we might have more success trying to line up the shapers on one value and the samplers on another.
We can base trims on mean, peak value, highest threshold with more than X hits, or whatever else you can come up with, pick a statistic and I will try and ensure that it is the same for all pixels. E.g. pick peak and all the threshold scans will peak at the same point, pick zero cut-off and all the threshold scans will cross zero at the same point.
Owen Miller
23/04/2008