For finding up the crosstalk source a series of resistors were added to each of the biases, finding that the main source was the common load resistors connected to a common VDDA, as was already explained. The 10K loads were moved from the fanout board to the video cards where they were biased by separate buffers. This change the crosstalk from positive to negative, and reduced it from about one part in 1000 to one part in 15,000, at which point it is no longer visible in the noise of a single frame, at least not until the banding effects are controlled.
The mask consists in a one hole per quadrant. In the following images, the brighter spot corresponds to the actual hole, while the other are pure crosstalk from the other quadrants
figure 1.1 When a 30 ohms series resistor was added to VDDA
Crosstalk ratio: 1/100
figure 1.2: Normal images, 2 ohms wire resistance on VDDA:
Crosstalk ratio: 1/1000
figure 1.3: After the load resistors were moved to the
video card (and separate buffers were used for each quadrant):
Crosstalk ratio: 1/15000
Measure electronic noise: ~0.8 adu (~15 uV ==> ~ 2 - 5 e-)
(See here the noise considerations for
the preamplifier)
Banding effect? ... ....it seems be dissapear ... what it was?
fan noise
we found that the fan produces a magnetically induced noise mainly
over the most adjacent channel to the fan We confirmed that it is
not an electric noise since we saw the same effect when using an external
power supply for the fan. This noise was found to be about 6 adu peak to
peak (figure 2.1, lower right quadrant)
60 Hz
Very strong 60Hz noise was found. This noise was fixed in the lab by
connecting all the grounds together (controller ground to chasis, power
supply ground to chasis, dewar to controller ground). See figure 2.2
figure 2.1: Fan noise (lower right quadrant) (just controller):
fig 2.2: 60 Hz noise (enginering array)
figure 2.3: science grade array in the lab, after all the noise was removed. The only visible effects are the bias tilt . This dark was taken with 0 secs idle time.
figure 3.1:
- initial black bands: can this be a transient in the biases?. There are two interesting clues: in the first case (reset-read line to line) the transient appears independent on the exposure time; this means that it is present when there was and when there was not an idle time. This may discard the idea that it is caused by a "transient" in the biases, since even when the biases where not relaxed (no idle time) it is there. A reconfirmation on this would be that in the second case (reset-read frame to frame) this transient is not present, even when you have an idle time diferent from 0 (which implies a relaxing in the biases)
- tilt: As we can see, in the second case the effect it is highly minimized. What is the diference?. In the first case, we see that the effect lasts less than 3 seconds since the reset clock stoped (when there is no idle time, it takes about 700 hundreds lines to dissapear, which is about 2.5 seconds read time, and if we have 2.5 seconds of idle time the effect does not appears ...). In the second case, when you start reading the exposed frame, even when there was no idle time, already passed 3 seconds since the reset clock stoped, and this may explain why it does not appears so strongly. Is this suggesting that this is some reset-related issue?
4.1 Reset-Read line by line waveforms
fig 4.1.1: 0 secs idle time
figure 4.1.2: 0 secs idle time
fig 4.1.3: 3 secs idle time
4.2 Reset - Read frame by frame waveforms
fig 4.2.1: 0 secs idle time
fig 4.2.2: 0 secs idle time
A 10 K load resistor to a 10V p-p square wave was also connected to
determine the sensitivity of the signal to a 200 mV modulation of the bias
voltage.
| 200 ohm in series with... | DC Current
(mA) |
sensitivity
(ADU/200mV) |
noise bands
(ADU p-p) |
Crosstalk ch2 to ch1
(rejection ratio) |
signal memory
(rejection ratio) |
| Ground = DSUB = CellDrain = Drain | |||||
| VDDA:load resistor [1] | 0.77[2] | 3700 | 3.6 | 118 | |
| VDDA: substrate bias | |||||
| BIASPOWER | 0.02 | 15060 | 1 | 840 | 1600 |
| BIASGATE | 0 | [4] | 1-0.6 | 860 | 1520 |
| VRESET | 1.6 [3] | 370 | 1 | 1020 | 1400 |
| VDDD | 0.01 | 157 | 0.7 | 760 | 1400 |
Notes
[1] When VDDA crosstalk (etc) was first measured, the 10K output load resistors were connected to a common VDDA. The crosstalk was thus due to the common series resistance. The 10K loads were moved from the fanout board to the video cards where they were biased by separate buffers. This change the crosstalk from positive to negative, and reduced it from about one part in 1000 to one part in 15,000, at which point it is no longer visible in the noise of a single frame.
[2] Expected total video load current = 4 * (VDDA - video_DC)/10Kohm = 4*(5-3.5)/10K = 0.6 mA so true VDDA current to mux ~0.17 mA.
[3] Why does VRESET source any current other than photocurrent ? Is there some leakage poath in parallel wiht the detector ?
[4] This measurement was invalid since we failed to notice that the reset frame was driving into the lower rail. The sensitivity to BIASGATE fluctuations should be similar to BIASPOWER.
Video gain = 10
Video ADC: 10V unipolar 16 bit ... 152 uV/ADU
More measurements need to be performed to complete the previous table
T: ~ 79 K
G: ~5.47 e-/adu
read method: CDS, using one non-destructive read
ranging time measured: from 100 to 1000 secs
Dark current measured: ~0.048 adu/s ==> ~0.26 e-/s
Values measured on the telescope:
gain = 5.467 e-/adu.
CDS noise: 13.76 e-
Out the telescope the shown (figure 8.1) curve was measured. Counts varied from about 1000 to 20000 (about same range in which linearity was measured). The coeficients obtained for the diferent channels were between 5.39 and 5.5 (e-/adu), which corresponds with the value measured on the telescope.
figure 8.1: example mean/variance curve obtained for channel
3 (slope adjusted by legendre, order 2).
System Non-linearity was measured to be less than 1% (!!?). The curve
shown below was done considering a gain of 5.4 e-/adu. ADU ranges varied
then from about 1000 to 18000 counts.
The electronics (controller) non-linearity (without detector) by the
other hand, was measured to be about 0.025%
fig 9.1: System Non-linearity
Detector:
Finish the crosstalk investigation:
rewire BiasPower for decreasing the cable resistance
Understand exposure zero point
paradox.
Finish measuring sensitivity of
each bias (and clock?) to noise.
Measure noise on each bias and
improve where necessary.
Measure noise vs bandwidth.
Determine 1/f corner frequency.
Investigate more preciselly what
leads to negative dark current (bias drift?)
Determine cause of "image lag";
fix if possible.
Determine cause of "image tilt";
fix if possible
Test amplifier glow control strategy
(play with DSP waveforms)
Measure & mitigate glow from
row and column select switches.
Measure the departure from root
N noise reduction for multiple sampling. The above tests should give
us a clue to the cause.
Measure performance versus temperature.