Topic: 32Bit Float recording - The Technical view

[morst]

Solution: a compact field recorder that accepts AES42 (digital microphone) signals. No more 32FP! ))

Aaton Cantar mini has 4... too big?
https://shop.cinneshop.com/product/aaton-cantar-mini/


Oops company went into receivership a year and three days ago...
https://jwsoundgroup.net/index.php?/topic/62483-aaton-bankruptcy/

[Billy Mumphrey]

Aaton Cantar mini has 4... too big?
https://shop.cinneshop.com/product/aaton-cantar-mini/


Oops company went into receivership a year and three days ago...
https://jwsoundgroup.net/index.php?/topic/62483-aaton-bankruptcy/

Ah naturally. I still want one though.

I think the Sound Devices 788t accepts AES3 and AES42 signals....

[dogmusic]

Does anyone know if the Tascam FR-AV2 uses two ADCs for the external in (as well as for the XLR in), or is it crippled like the X8 external input?

I asked Tascam and got this reply:

“The EXT input has the same circuit path (post mic-pre) as the main inputs.”

[adrianb]

Does anyone know if the Tascam FR-AV2 uses two ADCs for the external in (as well as for the XLR in), or is it crippled like the X8 external input?

I asked Tascam and got this reply:

“The EXT input has the same circuit path (post mic-pre) as the main inputs.”

Post mic-pre?

I was disappointed to discover that the 3.5mm input on my Sound Devices MixPre 3II doesn’t use the same Kashmir pre-amps as the XLR inputs. That, together with the low 2.5v PIP means that I never use it for lavalier type mics.

There’s a lot of love for the Tascam FR-AV2 as a stealth recorder because it removes the need for setting levels and a battery box. I’ve not used mine for that yet, but do hope it’s using the same pre-amp as the XLR inputs.

[WiFiJeff]

Does anyone know if the Tascam FR-AV2 uses two ADCs for the external in (as well as for the XLR in), or is it crippled like the X8 external input?

I asked Tascam and got this reply:

“The EXT input has the same circuit path (post mic-pre) as the main inputs.”

Many thanks!!

[Ozpeter]

I have just uploaded some tests on YouTube which attempt to show how well three recorders perform when recording something very loud (rock band) and something very quiet (fly) in the same clip.  I did the test with low cost recorders as presumably that's where the problems would be most evident.  And currently, as I wanted to compare dual ADC with single ADC, only lower cost devices have single ADC circuits.

I chose the Zoom H2n to represent the 24 bit option, the H2e to represent the single ADC 32 bit float option, and the M2 to represent the dual ADC 32 bit float option.  Obviously there were chosen largely because I have them in the cupboard!

To record the loudest possible sound which doesn't quite clip the H2n, its level control has to be set to zero.  The other two recorders have no gain control.  Can all three handle the same loud music without distortion?  What does the silence sound like when the music stops and only the fly can be heard? (Actually there's no fly, I just recorded a few words in a quiet room, and normalised the results).

And then what happens if I only want to record the imaginary fly in its own recording, band-free?  For that the gain on the H2n has to be set to maximum (10).  but the other two have no gain control so can they capture that tiny sound as well as the 24 bit device? Or in this practical experiment, the sound of quiet speaking?

Finally what difference is there between the results from the dual ADC M2 and the single ADC H2essential?

The tests were carefully done but there are a lot of variables involved, so maybe the playing field isn't completely level.  For instance in the quiet room tests, the M2 waveform looks noisy but actually its mics record more LF level from the very faint suburban traffic noise outside.  But still, I think the outcome is interesting.

https://youtu.be/5bqYQ87e64Y

I don't make money from my channel so I'm not profiting from this post.

[Rairun]

I have just uploaded some tests on YouTube which attempt to show how well three recorders perform when recording something very loud (rock band) and something very quiet (fly) in the same clip.  I did the test with low cost recorders as presumably that's where the problems would be most evident.  And currently, as I wanted to compare dual ADC with single ADC, only lower cost devices have single ADC circuits.

I chose the Zoom H2n to represent the 24 bit option, the H2e to represent the single ADC 32 bit float option, and the M2 to represent the dual ADC 32 bit float option.  Obviously there were chosen largely because I have them in the cupboard!

To record the loudest possible sound which doesn't quite clip the H2n, its level control has to be set to zero.  The other two recorders have no gain control.  Can all three handle the same loud music without distortion?  What does the silence sound like when the music stops and only the fly can be heard? (Actually there's no fly, I just recorded a few words in a quiet room, and normalised the results).

And then what happens if I only want to record the imaginary fly in its own recording, band-free?  For that the gain on the H2n has to be set to maximum (10).  but the other two have no gain control so can they capture that tiny sound as well as the 24 bit device? Or in this practical experiment, the sound of quiet speaking?

Finally what difference is there between the results from the dual ADC M2 and the single ADC H2essential?

The tests were carefully done but there are a lot of variables involved, so maybe the playing field isn't completely level.  For instance in the quiet room tests, the M2 waveform looks noisy but actually its mics record more LF level from the very faint suburban traffic noise outside.  But still, I think the outcome is interesting.

https://youtu.be/5bqYQ87e64Y

I don't make money from my channel so I'm not profiting from this post.

Thanks for this! It's an interesting comparison of the devices as a whole, but it's hard to tell whether the noise during the quiet parts is due to the preamps/ADCs or the microphones themselves (I presume they don't share the same internal mics?).

One interesting thing about the Zoom H2n's 'clicks' is that many years ago, while running a very similar experiment between a Zoom H1 and a Roland R-05 (both fed by the same external mic), the Zoom H1 also displayed the same pulsating noise! It was interesting because the Roland R-05's noise floor was actually louder, but it was a constant, uniform hiss. The Zoom H1 had quieter hiss between each click (by something like -8 dB compared to the R-05), but the clicks were louder than the R-05's hiss by at least 5 dB.

[Ozpeter]

I'm glad it was of interest.  The clicks would only be heard when massively amplifying audio recorded with the gain on zero (well, I guess I haven't checked at 1 and 2...) which normally would not be something you would do.  Perhaps I should have done less normalizing and let people check the quiet tests at the level they were recorded at, but then I'm not sure that would have been a level playing field either.  I shall think of what I could try next... 

Also it's hard to find a location which is really quiet enough for the quiet tests - which perhaps illustrates that maybe when it comes to real-world recording, we are getting too fussed about system noise (within limits).  As for the difference between dual and single ADC devices, well, overall the M2 came out best imho overall, but then it's probably a better quality device end to end (mic, preamp, and the twin converters, and fully integrated because you can't connect any external source).   So again, the playing field isn't level.  Hmm.

[Ozpeter]

"sync sound" has reported on the reddit thread that I started that he queried Zoom about the H2e and its single converter, and got this reply -

"Me: "Since it's a single 24 bit converter, does that mean that the total dynamic range of the recorded file can't exceed the dynamic range of that converter?"

Zoom: "Yes this is correct, and the built-in microphone itself will clip before the dynamic range would be reached."

VERY interesting.  On the face of it, and in the light of my tests, what they could be doing is taking the 24 bit output from the converter, translating it to 32 bit float, and adding about 15dB to every sample.  So it looks clever when you see it in your DAW like a brick wall and you have to bring it down 15dB to remove the digital distortion, but it's just playing around with the samples to make it appear to be 'real' 32 bit float, when actually it remains real 24 bits.  Fortunately that messing with the samples will not affect the final quality - it's just maths - but it still leaves a funny taste in the mouth.  Or the ears.

Once again, I'd say don't bother reading the 32 bit float claims.  Focus on the number of converters claimed - until someone from one of the companies convinces us that there is an advantage in 32 bit float, single ADC.  And of course focus on the end-to-end sound - a crap preamp or noisy mic will be very accurately captured by either real or pretend 32 bit float.

[adrianb]

Zoom: "Yes this is correct, and the built-in microphone itself will clip before the dynamic range would be reached."

Maybe this is why their cheaper 32-bit recorders with built in mics have single ADCs whilst the, also cheap, H1 XLR had dual ADCs?

They figure that the mics will clip anyway so nobody will notice that we’re cheating?

I’m still quite impressed with the overall performance of the H1 XLR for the price sold.

[Rairun]

24-bit has always exceeded the dynamic range of most (all?) microphones and analogue stages. That's never been the issue. The issue is that the AD converter has its own inherent ANALOGUE noise floor (not talking about quantisation errors here) - it needs to be fed a signal that is strong enough to surpass that noise floor by a wide margin, or else this shows up in the recording as hiss.

Since Zoom decided not to include gain control on the device, it makes perfect sense for them to match the strongest signal the ADC can take to the strongest signal the microphone can output without distorting. The 24-bit digital container has more than enough dynamic range to contain every last bit of useful signal it converts, i.e. from the loudest signal the ADC can take to the ADC's own analogue noise floor. Quantisation errors would be quieter than the ADC's noise floor.

This does make the device very fool-proof. The user is never going to be able to set gain high enough to clip digitally when the mic is producing a non-distorted signal. Likewise, the user is never going to set gain so low that the ADC's analogue noise floor becomes lower than the 24-bit digital noise floor (i.e. quantisation/dithering noise), which WAS possible to do in devices that allowed you to set digital attenuation instead of giving the option of using an actual pad before the analogue stage.

But again, while 24 bits is more than enough to store all that data, it needs to be stressed that this data includes the ADC's own analogue noise floor! That analogue limit has ALWAYS been the issue. So what Zoom are saying here is a bit disingenuous - by focusing on clipping levels (which no one doubted they could deal with in the first place), it is easier to ignore what happens on the other end of the range. And while their statement is also correct about the digital limitations of the quieter end of the 24-bit container (i.e. they are not an issue), it conveniently ignores why users used to accidentally clip in the first place: they added analogue gain to quiet microphone signals so that they would be as far away from the ADC's analogue noise floor as possible, and then a sudden louder noise would clip.

If you can't clip, then that also means you can't boost quiet signals far above the ADC's analogue noise floor - that is, unless you make clever use of a parallel analogue stage/ADC OR a variable analogue gain/compensation technique with a single ADC, which they are almost certainly not doing because they keep focusing on the limits of the digital dynamic range instead.

[Rairun]

It must be said though that the internal microphone's self-noise, under the setup I described above, might well still be louder than the ADC's analogue self-noise. In that case, Zoom are absolutely right: there is NO need for a second ADC. At all. I don't know if this is the case, but it's a possibility (the mics would need to be fairly noisy). If so, fair play to them - they have designed the perfect recorder for that specific microphone. All bets are off if you're using better external microphones, though.

[Gutbucket]

Yes that's essentially it. 

Forgive me for making a few nit-picky corrections in differentiating the analog noise-floor from the digital noise-floor.

[snip..] The issue is that the AD converter has its own inherent ANALOGUE noise floor (not talking about quantisation errors here) - it needs to be fed a signal that is strong enough to surpass that noise floor by a wide margin, or else this shows up in the recording as hiss.

The analog noise floor of the recorder comes from from the analog circuitry ahead of the ADC (analog 'hiss'). The digital noise-floor comes from the ADC itself (quantization noise).  We want both to be low enough in level, but we want the quantization noise to be lower than the analog noise, because "smooth low level hiss" is far more perceptually benign than "nasty" quantization noise.  So we want the analog noise to be low, but not overly low.  Dither is the intentional introduction of noise at the analog stage just ahead of the ADC which is just slightly higher in level than the onset of quantization noise.

Quantisation errors would be quieter than the ADC's noise floor.

^That is the purpose of dither.  Just enough added noise to effectively randomize the value of the least significant bit through the ADC, which effectively eliminates quantization noise.  The question then becomes if the noise-floor of the analog circuit path ahead of the ADC can effectively act as dither itself, or if additional dither noise need be added.

This does make the device very fool-proof. The user is never going to be able to set gain high enough to clip digitally when the mic is producing a non-distorted signal. Likewise, the user is never going to set gain so low that the ADC's analogue noise floor becomes lower than the 24-bit digital noise floor (i.e. quantisation/dithering noise), which WAS possible to do in devices that allowed you to set digital attenuation instead of giving the option of using an actual pad before the analogue stage.

My claim is that any digital recorder that allows you to record in such a way that quantization noise noise becomes apparent in low level signals is improperly designed.  Either dither noise or the input stage noise-floor acting as dither (both of which are forms of analog noise) should define the low level dynamic limit of any analog input digital recorder.  That noise should never be lower than the least significant bit of the digitized output.

[Rairun]

The analog noise floor of the recorder comes from from the analog circuitry ahead of the ADC (analog 'hiss'). The digital noise-floor comes from the ADC itself (quantization noise).  We want both to be low enough in level, but we want the quantization noise to be lower than the analog noise, because "smooth low level hiss" is far more perceptually benign than "nasty" quantization noise.  So we want the analog noise to be low, but not overly low.  Dither is the intentional introduction of noise at the analog stage just ahead of the ADC which is just slightly higher in level than the onset of quantization noise.

I think we are just slightly disagreeing on what counts as the analogue circuitry before the ADC and the ADC itself. As well as the quantisation noise, the ADC's circuitry produces its own analogue self-noise due to thermal noise, having its own input stage, etc. This should always be higher and more relevant than the quantisation noise (if it isn't, there is something wrong with the design). I like thinking of it as the ADC's noise (rather than, say, part of the preamp stage) because you can't skip it - it doesn't matter if you use Line-In or Mic-In, nor if you use internal or external gain. For our purposes, we have to make sure the resulting signal from the microphone + preamp stages reaches the ADC at a level that makes this noise insignificant.

When I mentioned that you can run into quantisation/dithering noise, I was thinking of situations where the recorder allows you to set digital attenuation after the entire analogue chain and digital conversion. The more digital attenuation you add, the more of the signal you "submerge" into dithering/quantisation noise waters. But a recorder should not be designed in a way where digital noise is higher than the ADC's own analogue self-noise.

Does that make sense?

[Gutbucket]

It does and I think we are essentially saying the same thing.  I'm just specifying some points of definition I think will be conceptually helpful to others.

Any analog circuit will have a noise-floor that cannot be lower than that of the thermal noise, including an ADC (the "A" in ADC).  But since dither is analog noise and quantization noise is a digitization artifact I think it best to not lump the two together. Conceptually, when I speak of ADC quantization error I'm talking about the decimation and digital part of the conversion and I think it helps to consider the analog circuit noise and dither noise as being separate from that.  Dither is compensation for quantization noise, necessarily made before quantization artifacts can occur in the digitization process, applied in an entirely different "analog realm" prior to the "digital realm" conversion steps, even if both happen within the same physical ADC chip. 

I do disagree with the wording here-

[snip..] The more digital attenuation you add, the more of the signal you "submerge" into dithering/quantisation noise waters.

Once digitized, the noisefloor becomes fixed relative to the rest of the signal.  We can alter level digitally and the noise-floor shifts along with the signal. If we amplify a lot so that low level signals become perceivable, the noise floor is amplified along with the signal and can also become perceivable.  The boat rises along with the tide, both are being raised or lowered by the same amount.  The signal "submerging" into noise (into analog thermal or dither noise in a properly designed ADC rather than into quantization noise) all happens upon or prior to the digitization process.