AGC Disable - Magic Lantern vs. Juicedlink?

[Jon Fairhurst]

No. The Canon firmware doesn't adjust analog gain. It leaves it fixed at +31dB. The AK4242 chip allows analog gain to be adjusted in large steps. Magic Lantern enables that adjustment.

Regarding digital gain, the AK4242 is interesting.

Case 1: You have a file that clips at 0 dBFS in your NLE. You adjust the gain down in your NLE (digitally) by 6 dB. Now the signal clips at -6dB. The signal in the original file didn't gain any headroom, because it was already clipped.

Case 2: You have a signal in the AK4242 chip that clips. You reduce the digital gain in the AGC circuit by 6dB. The signal gain is now reduced, but it doesn't clip at -6dB. The signal excursions still go all the way to 0 dBFS. Adjust it down far enough, and the signal no longer clips.

Case 3: You own an M-Audio Multitrack II. The signal is too hot. You reduce the gain by 6dB. Now the signal clips at -6dB(!) The Multitrack II gain adjustment is digital, and it is implemented like the NLE case (1) above. I don't recommend the Multitrack II. ;)

So, even though the AK4242 chip adjusts the gain digitally, it has additional bits of headroom built in. In essence, it works like an analog gain control. However, the physical analog gain is adjusted higher than it need be with the Canon firmware. So the results are not as quiet as with Magic Lantern, where you can set the digital gain to 0 and the analog gain to +10dB or +17dB as needed.

[Tom Morrow]

I can't find anything googling for an AK4242 chip... anyone have a link to the spec sheet? I'm assuming it's the ADC chip that converts the analog signal into digital in the Canon DSLRs.

Case 1: You have a file that clips at 0 dBFS in your NLE. You adjust the gain down in your NLE (digitally) by 6 dB. Now the signal clips at -6dB. The signal in the original file didn't gain any headroom, because it was already clipped.

It doesn't really make sense to me to say you have a file that clips... the signal the file was recorded from may have been clipped in order to fit into the gain range of the file, but once the signal is in the file it is what it is... clipping is a verb not a noun. A file is a static thing not a process. You could say the file is clipped, but not that it clips.

The reason I mention this is I want to clarify that the situation you are talking about where a clipped file opened in an NLE cannot be restored to unclipped is a completely different situation from what I was talking about when I mentioned NLE clipping behavior. I was referring to the situation where a full range signal with peak at 0dBFS will get clipped if NLE gain is increased, and will gain headroom if NLE gain is decreased.

Case 2: You have a signal in the AK4242 chip that clips. You reduce the digital gain in the AGC circuit by 6dB. The signal gain is now reduced, but it doesn't clip at -6dB. The signal excursions still go all the way to 0 dBFS. Adjust it down far enough, and the signal no longer clips.

When you say you have a signal in the AK4242 that clips, I assume you mean that you feed an unclipped signal into the AK4242 (say peaking at -10dBFS) and the digital gain causes clipping at the digital output of the chip. The fact that you can stop this clipping by reducing the digital gain is just good chip design... if it didn't work this way then I would consider that a chip design fail.

So, even though the AK4242 chip adjusts the gain digitally, it has additional bits of headroom built in. In essence, it works like an analog gain control.

Like an analog gain control, you get clipping when the input signal times gain exceeds the output signal max level. All properly designed non-buggy digital gains should work this way... the only difference between digital and analog gain is that digital gain increases the internal preamp noise by the same number of dB as the signal noise, while analog gain increases the signal more than it increases the level of the preamp noise.

Sorry for being so wordy about this... to summarize this post, I think ML digital clipping behaves exactly like an NLE in that a full range analog signal with peak at 0dBFS will get clipped if NLE/ML gain is increased, and will gain headroom if NLE/ML gain is decreased.

[Tom Morrow]

I found some good info in the

Magic Lantern 0.1.6 User Manual - Magic Lantern Firmware Wiki

Analog gain
This selection sets analog gain levels. Typically you would want this to be as low as possible if you are using a high quality pre-amp that can drive a hot signal into the camera. The 5D's preamps are noisy when turned about +20 dB.
L.Gain
This selection sets left digital gain level
R.Gain
This selection sets right digital gain level
...
Best audio is obtained by use of a preamp system fed to the camera. As a general rule, the use of a quiet preamp to send the signal to the camera will result in better the sound recorded in camera. Use of a preamped XLR adapter like the Juiced Link CX 231 or a field mixer will give superior results.
...
Configuration file snippet:
# Audio data
# mgain is according to this table:
# 0 == +0 dB
# 1 == +20 dB
# 2 == +26 dB
# 3 == +32 dB
# 4 == +10 dB
# 5 == +17 dB
# 6 == +23 dB
# 7 == +29 dB
# mgain affects both channels equally.
audio.mgain = 4

# dgain is in dB. If you want separate gains per channel,
# use audio.dgain.l and audio.dgain.r to set their gains.
audio.dgain = 18 # dB
...

Two comments about this:

1. The fact that the ML manual says "Best audio is obtained by use of a preamp system fed to the camera" seems to disagree with people who think that ML alone is best.

2. The gain bands that Jon is talking about are probably those listed in the mgain table above. This is good evidence that the Canon firmware gain menu probably adjusts both analog gain and digital gain in some combination.

[Jon Fairhurst]

1. The fact that the ML manual says "Best audio is obtained by use of a preamp system fed to the camera" seems to disagree with people who think that ML alone is best.

ML alone? That would be a poor choice, unless you're running a mic like the VM Pro, which has a built-in preamp.

The idea is to use ML to reduce the gain to a low level. To make up for it, you need to send in a strong signal. Besides, the best mics have XLR connections and often require phantom power. ML alone doesn't support such mics.

ML plus a clean preamp with additional gain - and a good XLR mic - are the keys to good in-camera, DSLR audio.

I have the datasheet here someplace, but have upgraded PCs and moved hard drives, so I need to dig around to find it. Unfortunately, this is a very busy week for me...

[Jon Fairhurst]

The part is the AK4646. Sorry about quoting the wrong number from memory above. Also, the analog gain wasn't fixed at +31dB. It's 32 or 29dB, but I can't recall which.

Here is the datasheet: http://www.asahi-kasei.co.jp/akm/en/...k4646_f05e.pdf

Note that the ACG works in steps of .375dB. Test the Canon manual gain and you will find that the steps are in multiples of 0.375dB.

[Tom Morrow]

ML alone? That would be a poor choice, unless you're running a mic like the VM Pro, which has a built-in preamp

Aha... I suppose I was thrown off by the thread title which made me think of ML without a preamp, but upon rereading I see that both Jon and the juicedlink folks were suggesting use of ML along with an external preamp like a juicedlink.

Here is the datasheet: http://www.asahi-kasei.co.jp/akm/en/...k4646_f05e.pdf

Note that the ACG works in steps of .375dB. Test the Canon manual gain and you will find that the steps are in multiples of 0.375dB.

Good info in there. It confirms the preset mic gain levels available (+32db, +29db, +26db, +23db, +20db, +17db, +10db, 0db) as the same values listed in the ML user guide snippet above. The quoted S/N ratios in the 80's seem decent although not anywhere near great.

The sense I'm getting is that using the lower half or third of the Canon firmware gain range is not something to worry too much, but that the higher gain amounts may be problematic.

With S/N of 80+db, the noise is not going to be an issue so long as the signal is decently strong.

[Jon Fairhurst]

Ideally, the Canon firmware would have used the 0.375 dB digital steps for a while, then take a large analog step, compensate with a large set of digital steps, and continue fine tuning the digital gain. I think they didn't do that because of the variability in analog. With a step-wise approach, the gain might have big jumps or even change in the wrong direction as you go between certain ticks. The advantage would be lower noise. I think Canon valued predictable UI behavior over highest quality.

That said, the noise isn't terribly worse with the Canon firmware, though it is certainly not as clean as it can be with ML. From memory, I was getting about a 9dB difference.

[Tom Morrow]

Thanks for all the info Jon. I feel like I have a real sense now of how the gain works in the Canons now and the more info you provide the more "sense" it all makes. You are probably right that a cheap chip might not have accurate analog gain steps, meaning that someone might call Canon support and say why does my gain go up when I click it down or vice versa. Canon designed this audio system more for simplicity than for quality. So although I will probably attempt some measurements of my 60d myself, I'm no longer doubtful that Canon really did fix the analog gain in their firmware.

It seems to me that most signals only have 20-40db of real dynamic range before mic or room noise gets in the way, so the 80+db (minus 31ish db for the hardwired Canon firmware gain)=50+ is still enough to record a decent signal, so long as you get your levels adjusted well. But if you start having your levels 10-30 or more db less than optimal, which would be common without a preamp, then you could very well run into noise. So the takeaway for me is that a preamp should be used for any critical audio, but that it's not totally insane to just plug a mic directly into the Canon mic jack and set the Canon firmware gain to lowest, and record without worry.

In fact I did something similar when I was walking around with my camera looking for b-roll footage a few months ago. Not having a microphone with me, nor having time to set levels, I just recorded some street music with the internal mic and gain all the way down. I was expecting horrible audio, but it was actually surprisingly usable, probably largely because it was recorded in a tunnel that increased level and whose reverberations hide the noise floor. At any rate it's good to know I was actually right to not mess with the gain during filming.

With this new knowledge that Canon's gain controls are all digital, I don't see any reason to ever adjust them above all the way down. It gives me some security knowing that even if not optimal I can get usable audio without a preamp, level controls, external metering, or ML in a pinch.

[Tom Morrow]

[removed message]

[Tom Morrow]

I started a new thread with my measurements of the noise on my 60d:

http://www.dvinfo.net/forum/all-thin...ml#post1674809

In short it appears that something other than simple digital gain is happening with the Canon firmware.