[rawr]

Its harder to compress noise, because noise is so random?

Depends I guess on the noise source and type?

I haven't studied this much, so I apologise if this sounds stupid. One thing that has me puzzled is that we are often told high ISO performance above say ISO1600 is achieved via digital boost. So in theory a 6400 ISO RAW file should just have the image data from a 1600 ISO exposure trapped in there, with a instruction in the RAW header that simply says to a RAW processor 'boost signal X 3 on processing'. But instead it seems like the RAW file really does capture all of the noise and other information that you would expect to be captured from an actual exposure above ISO 1600, not just a digital boost result.

[Na Horuk]

Well if you take a photo at ISO 1600 and ISO 6400 and same aperture/shutter, you can import the ISO 1600 into Lightroom or something and use the Exposure slider to make it as bright as the ISO 6400 photo. Now you will notice that they look virtually the same. To me, the end result is what matters most, I dont really know if higher ISO is achieved by increasing the sensitivity on the sensor or boosting the signal or whatever. But I still use the ISO that I want, I try to make properly exposed photos in-camera so I have to play with exposure as little as possible in post.

[rawr]

What puzzles me is that while a 1600 ISO DNG from the K-5 may be a 20MB file, a 6400 ISO DNG from the K-5 of the same scene may be 30MB (or more).

If the 6400 ISO scene was being rendered merely via digital boost in-camera, not any additional information coming in via the lens, it is hard to understand why file size should vary so much from a 1600 ISO capture.

[lister6520]

I very much doubt that ISO over 1600 is done by digital boost. I don't know about the K-5 because I don't have one but for the K-30 as far as I can tell from some simple tests it is only the step that are done by digital boost., that is from 12,800 to 25,600.

If the boost were applied digitally for anything beyond 1600 then it would mean that you could always shoot raw at 1600 and boost in PP but in practice I found that a picture shot at 12,800 is far better than one shot at 1600 and then boosted in PP to 12,800. With further PP it is possible to extract much more detail (and low noise) from the picture shot at 12,800 than the one shot at 1600. The advantage seems to vanish when comparing 12,800 and 25,600.

The 'boost' over most of the range is usually achieved by analog amplification before digitisation (and/oror by reduction of the ADC reference voltage). Normally when one refers to the native maximum ISO and the 'boosted' ISO it is when one applies analog amplification to the point when the actual noise becomes significantly higher than the 'quantisation noise' though I don't think there is any specific number for 'significantly higher'.

There might be a flaw in my analysis if the K-30 sensor digitises in 14 bits, since the camera only records 12 bits. That would mean that it is losing two stops of ISO when recording the image so a dark photo would indeed be a good candidate for digital 'boost' since that would recover those two bits which would otherwise have been discarded. If that is the case then it means 'digital boost' is being applied already at 3200 but it would still be resulting in a better picture because more information is being recovered from the original image that the sensor captured.