I just spent all morning measuring the noise on my Canon 60d DSLR, and I'd like to share what I learned for other Canon DSLR users.

This thread

http://www.dvinfo.net/forum/all-things-audio/498727-agc-disable-magic-lantern-vs-juicedlink.html
seems to indicate that the Canon DSLR firmware only adjusts gain with digital gain (ie multiplying the numbers after sampling), and that there is no way of adjusting the analog gain (before sampling) in the Canon firmware like Magic Lantern enables.

However my measurements appear to contradict this; if there was only digital gain then the noise would stay at the same level relative to the signal for all gain settings on the Canon. But I'm seeing the noise level (relative to signal) drop by about 10-12 db for high Canon gain versus for low Canon gain. I'd like to hear comments or suggestions for further experiments that could help determine once and for all whether Canon adjusts analog gain in the 60d firmware.

The excel graph of my results are attached.
To obtain this graph, I fed a -75dbu 1khZ sine wave from my NTI minirator MR2 into the left channel of the 60d, and left the right channel empty so I could observe noise level. I made one .mov file for each of the 63 gain settings, and manually played each one back in Audition while observing the onscreen level meter.

So I had tone level and noise level for all 63 of the level marks (except no noise data for marks below 11 because the noise didn't register on the -60db Audition level meter). I determined by trial and error that treating each level mark as 0.875db yielded a straight line for my tones. That number is different from the 0.375 that Jon predicted based on the chip data sheet... Hmm.

This graph below should theoretically just be two straight horizontal lines if gain was pure digital, but the slope of the noise line indicates to me that something other than simple digital gain is going on. If anyone wants the original spreadsheet data let me know and I'll send it. Please don't read too much into this data, as the manual measuring process means any value is likely to be off by 1-2db. And due to some mistakes keeping track of which click I was on, I ended up redoing the measurements between level marks 24 and 35; I'm guessing battery levels or room temperature changes are the reason for the shallow plateau in the tone levels in that region, rather than any real effects. However the overall trend seems clear... increasing gain lowers the noise level.

Very interesting. Thank you for sharing your results.

Just as a check on my experimental method, I tried applying "digital gain" in Adobe Audition to the file recorded at level mark 12, gaining it up by 38db in Audition to match the tone level in the file recorded at level mark 57 (higher Canon firmware gain). In that case the noise level was -20dbFS in the file resulting from my adding digital gain in Audition, but -31dbFS in the file with the Canon gain. 11db difference is enough to make me prefer canon gain.

I suppose it's possible that Canon realized the issues with digital-only gain in the 5d, and added some analog gain when designing the later 60d.

At any rate, a takeaway from my testing is that at the lower levels of gain and signal I have about 20db between signal and camera preamp noise, pretty marginal. Higher Canon gain on the same signal give about 30db between signal and camera preamp noise, which is enough for pretty good results. Of course best of all is to leave camera gain low and apply a strong clean signal with a preamp, which can give much more than 30db between signal and camera preamp noise.

Very interesting details that you provided. However, there are number of things that I don't quite understand about what you presented.


"However my measurements appear to contradict this; if there was only digital gain then the noise would stay at the same level relative to the signal for all gain settings on the Canon."

I would be surprised if adjustment of the canon firmware changes anything but the digital gain. I think there's a concept that you're not quite understanding, based on your statement above.

Digital gain does not change the signal-to-noise ratio, itself. However, digital gain in combination with the backend headroom limitations, does effectively change the signal-to-noise performance. The more back end digital gain that you use, means that can then have to start limiting yourself in the signal level that you can inject into the ADC input (to avoid clipping in the back end of the codec). So, as you are forced to decrease the signal level going into the front end of the ADC, and your signal to noise deteriorates as your signal level gets closer to the one bit quantization noise of the ADC.


I don't know if that's exactly what you're looking for. But, I hope this helps ...

Thanks for the reply Robert.

We're in complete agreement that it's better to put a clean hot signal into the front end of the ADC (with digital gain minimized) than to try and make up for a low signal with digital gain in the ADC or "back end" (NLE). And of course a juicedlink is well suited for this purpose.

The only time I would put the Canon firmware gain above the lowest mark (+1) would be if I needed to go without an external preamp, just using say a dynamic mic straight into the 60d for some reason.

So I think we're on the same page there about not wanting to use digital gain in the camera... there's just no need for it when you can always make up the gain in post.

What I can't understand though is why my measured S/N ratio could be anything but constant at all levels of Canon firmware gain, if that gain is just digital as you theorize. My current thinking is that perhaps Canon enabled several levels of analog gain in the ADC, and transitions between them with digital gain. That would explain the higher-slope parts of the noise graph above.

Perhaps if you could try explaining how digital gain could yield anything other than a straight line for the noise graph above, that would put us on the same page. Say you apply 2x digital gain in the ADC. That effectively just shifts the bits right by one position; the noise is increased by exactly 2x just like the signal. So flat line, not the noise sloping down to the right like my measurement graph actually shows.

Thanks for taking the time to help us learn!

Nice test, Tom.

You might take a close look at the signal levels. Normalizing to 0.875 dB steps may not be the way to go. My data (can't find it...) showed the gain taking steps of 2, 2, 3, 2, 2, 3... (0.750, 0.750, 1.125dB...), which averages out to 0.875dB per step.

>Normalizing to 0.875 dB steps may not be the way to go.
>My data (can't find it...) showed the gain taking steps
>of 2, 2, 3, 2, 2, 3... (0.750, 0.750, 1.125dB...),
>which averages out to 0.875dB per step.

Jon-
Why wouldn't it be the way to go if it corresponds to your data?

I'm still very curious to hear response to:

> What I can't understand is why my measured S/N ratio
> could be anything but constant at all levels of Canon firmware gain

The bottom line is that when feeding the camera with a clean preamp with gain, we don't adjust the gain using the camera. We set the camera at one tick over the minimum (with the Canon FW) and adjust the gain using the preamp. The registers (as read by Tramm Hudson on the 5D2 during Magic Lantern development) showed that the analog gain was set to a high level in that condition.

It would be interesting to have some of the ML developers read the registers of the 60D in that condition. Why use black box testing, when you can peek inside? :)