chiron
Offline
• • • • •
Upload & Sell: On
|
 p.4 #14 · The main (theoretical) differences between APS-C and Full-Frame? | |
tschopp wrote:
No, the effect is pretty simple, say the dual gain comes on at 400 iso. The goal of the circuit is to amplify everything by a factor of 4. The Sony uses a 14 bit analog to digital converter. Properly exposed iso 100 will have some pixels that use the most significant bit ( the brightest highlights) and might have some shadows that have pixels in the least significant bit. This would give you 14 bits dynamic range. If you are at iso 200, you shift everything up by 1 bit, but now you only have 13 bit dynamic range. At iso 400 the analog amplifier turns on and shifts the signal by a factor of 4. It will boost the signals below the least significant bit and make them readable. The reason to do this is to recover some dynamic range. If you just shifted the data by 2 bits you would have 12 bits, but by turning on the analog amp you get closer to 14 bits again, not quite because it did introduce some noise. Then as you increase to 800 iso you loose another bit of dynamic range and so on as iso increases.
The canon iso used to increase the analog gain at each step of iso. But it is easier to make a low noise fixed gain amplifier than a low noise variable amplifier, so the Sony gets better performance by a single low noise fixed gain amplifier instead of a noisier variable gain amplifier like the old canon cameras....Show more →
---------------------------------------------
tschopp wrote:
Ruthenium is correct about the amplifier or gain circuit reveling the noise in a small signal. But the amplifier also adds some noise as well. Think of a cheap radio amplifier as you turn up the volume you hear the noise in the signal, but you may also hear a hum from the amplifier.
In the case of the camera the photon shot noise depends on the number of photons. If you amplify a small number it will have more noise, but the amplifier will also increase noise a bit. Typically just picking up electrical noise from the environment or the rest of the circuits in the camera...Show more →
---------------------------------------------
dclark wrote:
Signal and noise that is produced prior to the gain stage is amplified and noise that is produced downstream of the gain stage remains fixed. Shot noise is produced prior to the gain stage. Read noise has contributions both upstream and downstream of the gain stage. In older generations of sensors and sensor electronics there was a significant contribution of electronic noise downstream of the gain stage. As the gain was increased the noise contribution from downstream was overwhelmed by the amplified signal and noise from upstream. Consequently, as ISO was increased the input-referred read noise decreased. Today's sensors and electronics are much improved, and the noise contributions are dominated by shot noise and, in very low light, read noise from upstream of the gain stage. Consequently, input-referred read noise is constant as ISO increases.
The slope of the PDR vs ISO curve reflects the noise performance. If the sensor is dominated by shot noise (read noise is negligible) the PDR will drop 1 stop as ISO is doubled. This is because as the signal is reduced by a factor or two (i.e. the ISO is doubled), the noise is reduced by a factor of the square root of two, so the SNR is reduced by SQRT(2), which is one stop. If you check the PDR data for the Sony A1 (the camera I use), above ISO 500 the curve is essentially perfectly 1 stop per ISO doubling, even at very high ISO. You can more easily see this by looking at the Input Referred Read Noise curve. For the A1 above ISO 500 the curve is at a very low level (~1 electron!) and essentially flat. This is an indication that the read noise is dominated by contributions before the gain stage. Read noise in the best sensors today (e.g. the A1) is dominated by residual kTC noise that the correlated double sampler, CDS, does not completely eliminate. The very low level of input-referred read noise is an indication the CDS is nearly perfect and the noise downstream of the gain stage is negligible. If you check this for some older sensors you will see that read noise was much higher and not constant.
I have used the term "gain stage" which may be a bit misleading and have not said anything about the big step at ISO 500. The step change at ISO 500 is not due to amplifier gain, it is due to the reduction of the capacitance at the sense node where the photoelectrons are deposited to produce a voltage that is digitized. In physics there is something called the equipartition theorem that says there is noise due to the temperature of the capacitor. This is the infamous reset noise which the CDS is meant to reduce. The larger the capacitance, the larger the reset noise the CDS has to overcome. As the number of photoelectrons is reduced, a smaller capacitance can be used at the sense node, which means the kTC noise can be reduced. For the A1 at ISO 500 the capacitance of the sense node is reduced so there is a big drop in input-referred read noise. Reducing the capacitance is in effect "gain" (more voltage for the same number of photoelectrons) but it's not gain due to an amplifier.
Dave
...Show more →
Thank you dclark and tschopp for your explanations and comments. It makes it very clear to me that I really don't understand these issues and that I probably won't until I have a bachelor's or master's in electrical engineering with a concentration in sensor design--and that isn't going to happen. Dave, I appreciate the effort it took to craft your long and very detailed response,
tschopp gave his take earlier on the theoretical differences for image characteristics between aps-c and full frame. Dave, if you wanted to add your thoughts on the main differences, that would be very welcome.
|
