Canon 5D Mark II Audio Exposed; with BeachTek DXA-5D and juicedLink CX231

[Jon Fairhurst]

I've started testing and production on a new P3Pictures video: Canon 5D Mark II Audio Exposed.

For recorders, I'll test the 1.1.0 firmware, as well as an alpha version of Magic Lantern. I'll also compare it to the MicroTrackII, and I've put out a call for a Zoom H4n recorder in the Portland area. http://www.dvinfo.net/conf/all-thing...-portland.html

Regarding XLR inputs, I'll test the BeachTek DXA-5D and the juicedLink CX231. I'll also try the MicroTrackII (and maybe the Zoom) with a direct mic connection.

For the microphone, I'll use the Rode NT1-A for the noise tests. It's not really a video mic, but it has a self-noise of only 5 dBA, which is about as good as it gets. We will know that the noise is from the electronics, rather than the mic. I will also do a real-world test with an AT815b shotgun. I'll record dialog at a moderate distance and subtle sounds with that setup.

I'll do my best to keep everybody up to date with my progress...

[Chris Barcellos]

Wish I could be there to see the process, Jon. Definitely keep us posted..

[Jon Fairhurst]

I've been working on calibration. I'm using a -18dB 500 Hz signal and close mic'ing the speaker. (I don't use 1kHz, because that's my crossover frequency.)

I started with the MicroTrack at 50% volume. I matched the juicedLink with MagicLantern (Hi setting, pot at 1:30.) I ran the same signal into the BeachTek and the 5D2 with standard firmware, and, surprisingly, the signal was 12 dB hotter. Being passive, I thought that it would be on the low side, but the camera gain is pretty high by default.

BTW, the pilot tone from the BeachTek seems to be just under 20kHz. And that 16 kHz rolloff I had seen from stuff on the web? That might have been due to compression pre-filtering.

One thing that will make calibration a bit tough is that the BeackTek doesn't really disable the ALC. It limits it. For most dialog, ALC should be inactive, but if you take the system to a metal concert, the ALC will come back to life. In effect, there is still an active compressor/limiter. The ALC turns down the volume quickly, but the volume increases very slowly during quiet moments.

Oh well. It's 1:13. Time to sleep on it...

[Tramm Hudson]

Originally Posted by Jon Fairhurst

I started with the MicroTrack at 50% volume. I matched the juicedLink with MagicLantern (Hi setting, pot at 1:30.) I ran the same signal into the BeachTek and the 5D2 with standard firmware, and, surprisingly, the signal was 12 dB hotter. Being passive, I thought that it would be on the low side, but the camera gain is pretty high by default.

The Canon firmware has the analog MGAIN register in the AK4646 set to +20 or +29 dB by default and the ALC was typically selecting digital gains in the +20 dB range.

One thing that will make calibration a bit tough is that the BeackTek doesn't really disable the ALC. It limits it. For most dialog, ALC should be inactive, but if you take the system to a metal concert, the ALC will come back to life. In effect, there is still an active compressor/limiter. The ALC turns down the volume quickly, but the volume increases very slowly during quiet moments.

The ALC recovery timers and rates are configurable in the IC. Since I'm just turning it off, I haven't tested any of the different values. The default appears to be increasing the gain 0.375 dB every 2.9 ms when sampling at 44.1 kHz.

Good luck with your tests today! I'm eager to hear the results.

[Glen Elliott]

IMHO, going through great lengths to try and capture audio on-cam may not be the best approach in terms of overall production workflow for this camera. It bulks the camera up and relies on sub-par internal audio circuitry.

Regardless I appreciate what you are doing and look forward to your results Jon.

[Matthew Roddy]

"sub-par internal audio circuitry"

I thought the camera was CAPABLE of capturing HQ 48/16 audio. I thought that once AGC was defeated and audio level could be set manually, we'd have audio comparable to any HD camera (Uncompressed 48/16).
Is this not the case?

[Tramm Hudson]

Originally Posted by Matthew Roddy

I thought the camera was CAPABLE of capturing HQ 48/16 audio.

The AK4646 will do 48 KHz, but it isn't clear if the DIGIC will process it at that rate. That is a good future project for the Magic Lantern firmware following the initial release.

I thought that once AGC was defeated and audio level could be set manually, we'd have audio comparable to any HD camera (Uncompressed 48/16).
Is this not the case?

It really depends on how much attention Canon paid to good analog circuitry leading from the mic input jack to the AK4646. If they have traces too close to digital circuits, or if the quality of the analog components aren't top-notch, the unbalanced signal can pick up lots of noise. Jon's tests will give us an idea of the noise floor that the camera can actually achieve.

[Jon Fairhurst]

Glenn, this is exactly what I'm trying to confirm: can the 5D2 be used to capture good, if not excellent, audio?

Anyway, here's the signal I'm using to calibrate levels (right-click and download):
http://p3pictures.com/audio_5d2/test...els_-18-60.wav

I'm positioning the microphone close to one woofer and muting the other speaker. I then record this into each preamp/recorder combination - without touching the mic. I adjust levels as needed to get them to record the same levels, as allowed by adjustment ranges. I'll note any differences (small differences - or large ones because of running out of adjustment range) and account for these digitally in post. That should level the playing field.

[Chris Barcellos]

Having worked with a high frequency tone on one track, to disable AGC, I am feeling pretty good about what this set up will do. You will have to add a preamp system like with the Juiced Link CX231, but I have already seen improved capability with that set up.
With Jon testing Tramms set up, we should get a pretty good idea where its going.

All of us have different levels of need of course. Some of us will be satisfied with sound that we have been able to get into our prior DV and HDV cameras. For those looking for something more spectacular, you may be disappointed. But the initial reports I have heard from Jon, and my own testing lead me to believe we should be able to get decent in camera audio. But lets wait and see..

[Jon Fairhurst]

Here's what I have calibrated so far...

CALIBRATED

* The juicedLink CX231 into the 5D2 with Magic Lantern firmware. The juicedLink is cranked to 100% on high. Tramm's software sets the analog gain in the camera to 0dB, and pushes the digital gain by 12 dB. There will be adjustments in the future, but I have no more gain available at this time, so this set the reference levels. From experience, this is a practical real-world level, so this is reasonable.

* The BeachTek DXA-5D into the 5D2 with Magic Lantern firmware. Since the firmware is gain limited and the BeachTek is passive, the gain is ridiculously low. I boosted the signal 18dB in post. In the production firmware, the in-camera gain would be increased. This is not a real-world scenario.

* The BeachTek DXA-5D into the standard 5D2. The camera's analog gain is high, and the digital gain of the ALC (automatic level control) isn't fully squashed by the BeachTek's pilot tone, so I actually had to turn the BeachTek down to calibrate things. I set it at 9 clicks below full scale. (I like that the knob clicks. It's easy to repeat gain levels.)

* The MicroTrack II. It's set at 2 major ticks past mid-way, and two quick taps down. (I wish the firmware provided numbers for easier repeatability. The dots are tiny.) I'm running 44.1 kHz, 24-bits, stereo with the limiter off.

* The juicedLink into the MicroTrack II. From experience, the best setting on the MicroTrack II is 50%. Any lower and I can't get get to full digital levels. I'm running 44.1 kHz, 24-bits, stereo with the limiter off. The juicedLink is set at one dot past noon.


NOT IN THE TEST PLAN - YET

* The BeachTek into the MicroTrack II doesn't make sense. The BeachTek is passive, so it won't let me turn down the MicroTrack II preamps to reduce noise levels.

* I'm not going to bother with the juicedLink into the 5D2 without a pilot tone to disable the ALC.

* I'll try the juicedLink into the 5D2 with a pilot tone at a future date. I want to optimize my tone setup, before I do critical measurements.

* Nobody has offered an H4n yet in the Portland area. This would be another good item to test in the future.

Okay. Step 1 completed. I've got my first round test cases calibrated.

[Tom Daigon]

Jon:
In laymens terms...based on what you have seen thus far....can I use my Sennheiser ew100 mikes (assuming I use the Magic Lantern when released) straight into the 5D for a relatively
decent signal. Or do I need to spend money I dont have on the juiced link box you are testing? Im just trying to anticipate an audio configuration that will work.
Tom

[Jon Fairhurst]

Originally Posted by Tom Daigon

...can I use my Sennheiser ew100 mikes (assuming I use the Magic Lantern when released) straight into the 5D for a relatively decent signal.

Yes. The mic preamp is part of the wireless system. You shouldn't need an additional preamp after the receiver. As I recall, the ew100 has a line-level output. If anything, you'll need a pad on the output to bring the signal down to (a hot) mic level.

[Jon Fairhurst]

I set up my NT1-A in a basement bathroom that has no outside walls or windows. I suspended the mic in the middle of a blanket-walled PVC frame. I shut the door. I turned off all of the electronics in the neighboring room. The house is out in the hills. I didn't record when some planes passed. It was pretty quiet in there. ;)

After capture, I ran every clip through a 20 Hz, -Inf, 24 dB/oct, shelf EQ to remove any DC offsets. This doesn't affect the sound of the noise, but it calms the meters.

For listening tests in Vegas, on the BeachTek DXA-5D only (to calm the pilot tone), I used EQ to apply a notch filter at 19,740 Hz, -Inf, 0.1 octave bandwidth. I didn't need this in SoundForge, but Vegas' meters didn't like the pilot tone. I can't hear anything up there, so again, it doesn't affect the audible sound.

The clear winner was the juicedLink CX231 into the Canon 5D Mark II with Tramm's Magic Lantern Firmware. The noise is even and continuous. In SoundForge, the meter showed -67 dB. Even after the EQ, there's a little low frequency junk driving the meters. From 1kHz up, the noise is down near -96 dB, which is the limit for 16-bit audio. There's a 2dB bump at 18kHz, and a spike up of -79 dB up at 20.3kHz.

In second place was the juicedLink into the M-Audio MicroTrack II. The MicroTrack has a repetitive noise, but with the juicedLink, I can turn down the recorder gain enough to help hide it. The meter reads -58dB. This combination hits the -96 dB limit around 1.5kHz, but increases with a ripple as the frequency rises. By 16kHz, the noise peaks at -82 dB. We see a bump at 20.3 kHz, which probably comes from the juicedLink, but would be inaudible.

It's possible that the juicedLink/Microtrack would beat the juicedLink/5D when capturing quiet sounds or an insensitive mic. I still have nearly a half-turn available on the JL gain control into the Microtrack. On the 5D, the JL is maxed out. Any additional gain must come from the camera or in post.

It's harder to place the non-juicedLink competitors. The BeachTek/5D combination has the most noise - either with the pilot tone, or with the Magic Lantern and the required +18dB of additional gain in post to match levels. The MicroTrack is quieter, but has an annoying repeating pattern - is sounds like a mechanical sprinkler is running in the background! Though quieter, it might be harder to remove the MicroTrack sound with noise reduction software.

The thing that surprised me about the BeachTek/5D recordings is that they sound virtually identical regardless of the firmware! I thought I might have overwritten a file, but I can clearly see that the pilot tone is only present in one of the clips. The hiss is a bit high, but its even and consistent.

The BeachTek DXA-5D into the 5D with Magic Lantern firmware shows -54dB on the meter. On the frequency plot, there's almost no additional low-frequency noise. It peaks around -74 dB. The noise plot is very smooth, but never dips below -82 dB. It's -79 dB at 16 kHz. Though the noise at any one frequency isn't high, they are combining to create the -54 dB level on the meter.

The BeachTek DXA-5D with the pilot tone into the 5D2 1.1.0 firmware shows -56dB on the meter. The noise is very smooth at around -81 dB. One thing to note is that the ALC isn't fully defeated. It still pumps when you get a very loud sound. That could be good for hands off operation, or bad if you want to control the levels yourself. There is another 9dB or so of gain available above my calibration, so this would likely beat the BeachTek/MagicLantern combination when you need more gain.

The MicroTrack II alone is a bit odd. It is quieter than the BeachTek at low frequencies with -90dB by 900Hz. The noise rises and ripples with frequency. By 16kHz the ripples peak at -71dB and trough at -83dB. Having used this on a short film submitted to an international film festival (we didn't get in - there were 3,000 submissions for 150 slots), the MicroTrack II alone isn't adequate, unless you use a preamp or have a very sensitive mic.

To summarize, the low-noise of the active juicedLink is excellent. I can turn it full bore and get better than 16-bit performance at most frequencies. It's a good match for the 5D2 with Magic Lantern firmware, though we might need to add some gain in the camera to capture low-level sounds or a dull mic. The BeachTek5D combos are noisy, but smooth. The MicroTrack II really needs an active preamp to mask its patterned noise.

Now, on to listening tests...

[Tom Daigon]

Thanks for the info John! Its nice not to need to buy something else for a change.

[Chris Barcellos]

I think I am very happy.....!!