Photon Transfer Curve

Hi Jens,

Have you made a linearity curve of the sensor, for example by
making different exposure times with the same illumination?
The amount of readout noise could also be very high, but this
should be seperateable from the photon shot noise in the graph?

I have, and there is some residual nonlinearity at low exposure levels,
but nothing too great.
The readout noise doesn't seem to be too high.

My ideas are shrinking, so what else can cause the fact that
the shot noise doesnt increase so much as it should (what a
lower slope means)?

I don't know either!?!? I'm going back to basics here .... evaluating
every step of the acquisition process to see if I went wrong somewhere.

Best regards,
Patrick

Hi Patrick,

Have you made a linearity curve of the sensor, for example by
making different exposure times with the same illumination?
The amount of readout noise could also be very high, but this
should be seperateable from the photon shot noise in the graph?

I have, and there is some residual nonlinearity at low exposure levels,
but nothing too great.
The readout noise doesn't seem to be too high.

But this seems to be important, is the acquisition board designed
for imagers, what method is used, CDS (correlated double sampling)?
What is the exact sensor name?
I think in the time i have all informations needed to get the
reason, i can design and built a complete new camera ;-).

Is the project an astro camera?


Best regards,
Jens

Hi Jens,

Have you made a linearity curve of the sensor, for example by
making different exposure times with the same illumination?
The amount of readout noise could also be very high, but this
should be seperateable from the photon shot noise in the graph?

I have, and there is some residual nonlinearity at low exposure levels,
but nothing too great.
The readout noise doesn't seem to be too high.

But this seems to be important, is the acquisition board designed
for imagers, what method is used, CDS (correlated double sampling)?
What is the exact sensor name?
I think in the time i have all informations needed to get the
reason, i can design and built a complete new camera ;-).

Is the project an astro camera?

The acquisition board is a National Instruments LabView ADC. So, It's
more of a general purpose low noise ADC. It's a proprietary CMOS
sensor, so legally, I can't speak freely about the readout method. I'm
sorry.

It is not an astro camera, but I do come from the astro community so
I'm familiar with some of the specifics of astronomical
instrumentation.

Best regards,
Patrick

Hi Jens,

Ok. I think I'm on to something here ...

The read noise is high: about 280 e rms. At saturation, the noise is
750 e rms. This is only about 2.7 times the read noise. So, I surmise
I'm not actually ever reaching the shot-noise limited region!

What threw me was that I was able to fit (with low residual error) a
line to a major portion of the curve, which I assumed to be the
shot-noise limited region. I no longer believe this is the case.

However, I still have high dynamic range: saturation = 500ke. So
dynamic range is 20*log(500000/280) = 65dB. Is it strange that I can
have both high dynamic range and high read noise? I know the two are
somewhat decoupled, because the full well is quite high, but something
seems a little off here.

Best regards,
Patrick

Hi Patrick,

Ok. I think I'm on to something here ...

The read noise is high: about 280 e rms.

Oh, this explains something.
Sensors normally has to be read out by CDS or dual slope integration
(weighted dsi is the best), and some sensors can be read without
destruction of the charges, so noise can be lowered by multiple
readouts.
Then there are sensors with charge multiplying processes, but this
increases the shot noise.
Timing and correct filtering are important for a low noise readout.


However, I still have high dynamic range: saturation = 500ke. So
dynamic range is 20*log(500000/280) = 65dB. Is it strange that I can
have both high dynamic range and high read noise? I know the two are
somewhat decoupled, because the full well is quite high, but something
seems a little off here.

Far off, if you ask me. High dynamic range doesnt increase the readout
noise in this amount. Sensors with such a high fullwell have sometimes
two different readout modes, one with a higher v/charge output and
decreased fullwell - but a slightly lower read-noise, and one with
the maximum fullwell and somewhat higher readout-noise.
But the values for good CCDs are somewhere about 5 and 8e_rms,
dependend on the speed and method of readout, higher than 20e should be
avoided.

I think you should use a acquisition board, designed or handmade for
the sensor.

Best regards,
Jens

i forgot:
I think you should use a acquisition board, designed or handmade for
the sensor.

And with a 16 Bit converter minimum, especially if the output of the
sensor is linear.

Best regards,
Jens

Hi Jens,

Thank you again for the excellent insight.

We are not doing strict CDS, because of the nature of the CMOS device.
The reset and video levels are sampled and stored at different times
on the device (as in most CMOS imagers). Also, there is only one
output on the device.

As far as acquisition board goes, I think you are exactly right, the
acquisition board may be what is limiting me. I will be moving to a
more sophistiated setup soon.

But does what I see so far at least advance the theory that I'm never
in a shot-noise limited region because at maximum signal, my noise is
only 2.7 times the read noise?

Best regards,
Patrick

Jens Dierks wrote:
Hi Patrick,

Ok. I think I'm on to something here ...

The read noise is high: about 280 e rms.

Oh, this explains something.
Sensors normally has to be read out by CDS or dual slope integration
(weighted dsi is the best), and some sensors can be read without
destruction of the charges, so noise can be lowered by multiple
readouts.
Then there are sensors with charge multiplying processes, but this
increases the shot noise.
Timing and correct filtering are important for a low noise readout.


However, I still have high dynamic range: saturation = 500ke. So
dynamic range is 20*log(500000/280) = 65dB. Is it strange that I can
have both high dynamic range and high read noise? I know the two are
somewhat decoupled, because the full well is quite high, but something
seems a little off here.

Far off, if you ask me. High dynamic range doesnt increase the readout
noise in this amount. Sensors with such a high fullwell have sometimes
two different readout modes, one with a higher v/charge output and
decreased fullwell - but a slightly lower read-noise, and one with
the maximum fullwell and somewhat higher readout-noise.
But the values for good CCDs are somewhere about 5 and 8e_rms,
dependend on the speed and method of readout, higher than 20e should be
avoided.

I think you should use a acquisition board, designed or handmade for
the sensor.

Best regards,
Jens

Hi Patrick,

Thank you again for the excellent insight.

We are not doing strict CDS, because of the nature of the CMOS device.
The reset and video levels are sampled and stored at different times
on the device (as in most CMOS imagers). Also, there is only one
output on the device.

Ok, i am not very familiar with cmos devices, so my insight belongs
more to CCDs.
Nevertheless (is this a english word?), readout noise should be lower.

As far as acquisition board goes, I think you are exactly right, the
acquisition board may be what is limiting me. I will be moving to a
more sophistiated setup soon.

So the readout noise is more a quantization noise.

But does what I see so far at least advance the theory that I'm never
in a shot-noise limited region because at maximum signal, my noise is
only 2.7 times the read noise?

My not very professional diagram, depending on 280e readout noise and
500ke fullwell looks like this:
http://freenet-homepage.de/JDierks/curve.gif

I dont know how much saturation can flatten the slope, but 0.2?

But im quite sure that a lower noise level should give you a much
better approximization.

Best regards,
Jens

Hi Jens,

So the readout noise is more a quantization noise.

I believe some of it may be.

But does what I see so far at least advance the theory that I'm never
in a shot-noise limited region because at maximum signal, my noise is
only 2.7 times the read noise?

My not very professional diagram, depending on 280e readout noise and
500ke fullwell looks like this:
http://freenet-homepage.de/JDierks/curve.gif


That drawing IS good. It is quite similar to what I see. So, it looks
like I just never reach the shot-noise region.

I dont know how much saturation can flatten the slope, but 0.2?


I would expect the characteristic steep fall off of the curve at
saturation.

But im quite sure that a lower noise level should give you a much
better approximization.

Agreed. Thanks again.

Best regards,
Patrick

Hi Patrick,

So the readout noise is more a quantization noise.

I believe some of it may be.

If it doesnt decrease with another board, you have to rethink the
way you gain the data.
I dont know how the darklevel of each pixel is sampled in a cmos
sensor. But in my understanding, it has to be sampled just close
bevor the lightlevel, or has to be integrated for a longer time
to get rid of the low frequencies of the amp-noise.


That drawing IS good. It is quite similar to what I see. So, it looks
like I just never reach the shot-noise region.

Yes, but in the case that the read-noise increases with a longer
exposure time (and the brighter images were captured with longer
exposures), it would maybe pointing to a wrong darklevel acquisition.


I dont know how much saturation can flatten the slope, but 0.2?


I would expect the characteristic steep fall off of the curve at
saturation.

Yes, if the converter is saturating bevor the sensor gets close
to the fullwell.
Otherwise it shouldnt fall off so quick, dependend on the sensors
characteristic. Ok, 760e noise_rms is nothing against 500ke fullwell,
so it wouldnt flatten the curve a lot, too.

Best regards,
Jens