CCD Readout Noise vs Temperature

[Drakkith]

TL;DR Summary

CCD Read Noise vs Temperature

Hey all. I tested my astro camera a while back and found what looks like a temperature dependent readout noise value.
Which is odd, as everything I've ever read has told me that readout noise is independent of sensor temperature.

I took 50 bias frames at 5°c intervals, from -15°c to +20°c and recorded the value of a single pixel in each frame (the same pixel in all frames).
I found the standard deviation of these pixel values in ADU's and then converted that to e- for each temperature.
Results are as follows:

Temp in c	-15	-10	-5	0	5	10	15	20
Stdev in e-	8.368	9.056	9.492	9.525	10.502	10.278	12.106	14.979

Camera: SBIG ST-2000xm, with KAI-2020M sensor.
Specs can be found here: http://www.company7.com/library/sbig/sbwhtmls/st2000xm_new.htm

I'm wondering if I'm doing something wrong in my analysis or if my understanding of how CCD's function is incorrect. Any ideas?

 

[glappkaeft]

``````````````````````````````````````````````````````````````````````````````````````````````````````````````````That sounds a little odd indeed. The noise in a CCD image comes from the signals it records. Since the signals are all Poisson distributed statistical processes the noise from each signal source is the square root of the signal.

The readout noise, often called bias noise since it comes from the bias signal that is added to the the rest of the signal (thus biasing it) to create a minimum level (a pedestal) that makes sure that the values after the A/D converter are positive, should as you say be independent of temperature.

How long are the exposures of your bias frames? What are the average pixel vales i.e. do they also increase in the expected way (square of noise) with the increasing noise?

ETA: pintle ->pedestal

 

[Drakkith]

How long are the exposures of your bias frames?

Essentially zero.

What are the average pixel vales i.e. do they also increase in the expected way (square of noise) with the increasing noise?

Here are the average pixel values:

Avg Pixel Value (ADU's):	1023	1021.42	1017	1028.02	1028.5	1043.1	1073.88	1125.44
Temp in c:	-15	-10	-5	0	5	10	15	20

As you can see, there is negligible or no increase until around +10 C is reached, at which point it rapidly increases. Note that the amplifier circuitry is uncooled and remained at ambient temperature the entire time.

 

[phyzguy]

While you kept the off-chip circuitry at a constant temperature, the on-chip amplifier is at the same temperature as the CCD. This amplifier has some non-zero noise associated with it, and to me it is not surprising that the noise introduced by this amplifier decreases as the temperature decreases. In principle, you could use your data to extract this component of the noise and separate it from the Poisson noise, as discussed by @glappkaeft above.

 

[glappkaeft]

Looking at your additional data I agree with @physguy that something in the CCD camera system is adding noise in a non-Poisson fashion. I'm not a CCD sensor expert but the amplifier sounds like the most likely suspect.

Considering the relatively low levels of the noise and the way low noise sources sort of disappear when added to larger sources (typically the photon and dark current signals) I doubt it really has a big effect on your images but it could in principle be modeled. The simplest way would probably be to use the temperature dependent bias calibration frames you already have made.

It would be interesting to see if it has an non-linear effect on dark frame calibration.

 

[trurle]

Camera: SBIG ST-2000xm, with KAI-2020M sensor.
Specs can be found here: http://www.company7.com/library/sbig/sbwhtmls/st2000xm_new.htm

I'm wondering if I'm doing something wrong in my analysis or if my understanding of how CCD's function is incorrect. Any ideas?

KAI-2020 sensor has a built-in amplifier. It is natural for such amplifiers to become noisier at high temperature. And noise you reported is under sensor specs. The rapid noise increase from 10C is likely self-heating effect when high-current output stage of video amplifier is heating the input stage. It is commonly happens when chip maker overdo "cost-performance optimization" by cramming too much circuits in too small area.

 

[Drakkith]

And noise you reported is under sensor specs.

What do you mean? Sensor specs are about 7.6 to 7.9e- rms. All of my measurements are above that. Almost double at higher temps.

The rapid noise increase from 10C is likely self-heating effect when high-current output stage of video amplifier is heating the input stage.

Hmmm. Why would this be a larger problem at slightly higher temps? I should also note that all of my frames were taken with a small pause in between in order to allow for the amplifier to cool back to ambient temp after readout.

KAI-2020 sensor has abuilt-in amplifier. It is natural for such amplifiers to become noisier at high temperature.

While you kept the off-chip circuitry at a constant temperature, the on-chip amplifier is at the same temperature as the CCD. This amplifier has some non-zero noise associated with it, and to me it is not surprising that the noise introduced by this amplifier decreases as the temperature decreases.

Ah, okay. I wasn't aware that there were on-chip amplifiers in addition to the off-chip amplifier.

 

[phyzguy]

Ah, okay. I wasn't aware that there were on-chip amplifiers in addition to the off-chip amplifier.

The signal being read out of the CCD is extremely small, often just a few electrons. There is no way to pipe this tiny signal off chip without losing it or introducing a lot of noise. So there is an amplfier on chip (often just a singe transistor) that converts this charge signal into a voltage signal. This voltage signal is then routed off chip where it is amplified further by the off-chip circuitry.

 

[Drakkith]

Well, I guess this is just more reason to keep my chip cooled and hope I can afford a new camera in the near future. My ST-2000XM was probably about 5 years old when I bought it used in 2010. Been itching for a long time to get a new camera. Newer ones commonly have less than half the read noise as mine and their sensitivity is double or more in some areas of the spectrum. It would be nice to have more than ~33% QE in the Hydrogen-Alpha band.

 

[trurle]

What do you mean? Sensor specs are about 7.6 to 7.9e- rms. All of my measurements are above that. Almost double at higher temps.

Which specs? I refer to
onsemi.com/pub/Collateral/KAI-2020-D.PDF
which lists 16-20 e- "readout noise".

Hmmm. Why would this be a larger problem at slightly higher temps? I should also note that all of my frames were taken with a small pause in between in order to allow for the amplifier to cool back to ambient temp after readout.

Because video amplifier cores are commonly driven with constant current, the thermal dissipation grows typically as T^2.5, where T-absolute temperature. Also, the thermal conductivity of chip fall at higher temperature, providing a positive temperature feedback. In worst case of self-heating effect, step-like performance dependence on temperature or even temperature-performance hysteresis may happen.

 

[Drakkith]

Which specs? I refer to
onsemi.com/pub/Collateral/KAI-2020-D.PDF
which lists 16-20 e- "readout noise".

I don't think that's accurate for my chip. If I understand it correctly, the readout noise depends on how fast the pixels are read out, with decreasing noise the slower the readout. The 16-20 is listed for 20 and 40 MHz readout rates I believe, but I think mine operates at around ~400 KHz, as it takes about 4.5 seconds to read out the entire 1.9 megapixels.

Because video amplifier cores are commonly driven with constant current, the thermal dissipation grows typically as T^2.5, where T-absolute temperature. Also, the thermal conductivity of chip fall at higher temperature, providing a positive temperature feedback. In worst case of self-heating effect, step-like performance dependence on temperature or even temperature-performance hysteresis may happen.

Got it. Thanks!

 

[Tom.G]

Yes, circuit noise is definitely temperature dependent. In imaging sensors it is called 'dark noise.'

Serious astronomical imaging sensors are always cooled. For a few decades there was a company, Santa Barbara Instrument Group (SBIG) (since purchased by diffractionlimited.com), that was THE supplier to the advanced amateur and professional astronomer market of thermo-electrically cooled cameras.

If there is an Astronomy group near you, ask them about the SBIG cameras, and what replaces them.

Cheers,
Tom