Jump to content

Zoom F6 (a 32bit recorder!)


IronFilm

Recommended Posts

I'm also sceptical of Zooms marketing speech, and I think 24bit is enough to adequately capture sounds in extremely high dynamic situations, but one thing which is interesting about 32bit floating point is that a file format could be used that gets rid of the idea of an absolute maximum value, thus no clipping ever (this is done on the image side with the EXR file format).

 

with 32bit floating point one could use a file format that allows to record normal dialog at -12db digital scale and if somebody starts screaming that would be recorded at +15db which could be brought down below 0db later in post (or a gunshot at +30db).

with traditional 24bit capture we'll have to record at very low levels if we want to have a lot of headroom without hitting the limiters or worry about clipping (say dialog at -30db in the above example to leave room for the screaming, or -45db for the gunshot) and the editor has to rise the volume of all the clips.

so while we kind of can already do this, a lot of clients would be rather worried if I handed them audio with levels at -30db or -45db.

 

with good implementation of dual preamps and ADC this could become very convenient since you never would have to worry about clipping anymore (I guess zaxcom users will say that's old news, I have no experience with their products but I thought you'll still have to deal with the low levels somehow).
chris

 

Link to comment
Share on other sites

It's interesting. I can see potential issues in the editing stage tho. When editors print a 24 bit omf/aaf without adjusting the volume. 

 

Looking forward to a new F8 that incorporate this. Then I could happily switch to that as a recorder when many channels are needed if this dual adc trickery is as magic as they claim. 

Link to comment
Share on other sites

I expect 32 bit to be useful at low levels, not really for high level gain setting.  An F8n line input set to -10 gain will still clip at about -5.5dBFS with a line source, so the input stage is still the overload point rather than the converter.  Running F8n on rock band instrument recordings, anything like an MKH40 or KM140 gain stages correctly set on line input at 0dB gain setting.  

Link to comment
Share on other sites

6 hours ago, Constantin said:

 

.There‘s more to A-D conversion than just bit-rate. Most (or none)  24-bit converters don’t achieve the theoretical maximum dynamic range of 144dB and with 32-bit converters I also only know of one that achieves 148dB, although it’s not a dual converter...

32-bit converters also have lower self-noise, better anti-aliasing filters, etc., which all justify the use of 32-bit converters even if the actual dynamic range cannot be significantly increased. 

 

I'm not familiar with the claim that 32 bit converters have better filters (why?) or lower self noise so I can't really comment on that.

Instead I'll be paraphrasing (and borrowing , quoting from)  Dan Lavry here and would recommend looking him up whenever questions about digital audio conversion arise:

One should always differentiate between processing bit depth and conversion bit depth. The lower bits out of a 24 bit converter carry noise, not audio data, due to mic and micpre generated noise, so in reality 20 bits is about as good as one can expect. on the DA conversion side it's the same since "real hardware can't even do 24 bits, because the lower bits are buried in noise......fundamentally there is nothing to be gained by inputting more then 24 bits. More bits would be a waste of space, and no sonic improvement".

 

24 bits is more than enough for conversion yet not enough for processing to maintain a full 24 bit resolution..

6 hours ago, Constantin said:

How do you know it was the pres and not the mic?

because the same mics (dpa) into SD 664s line input during an identical setup didn't clip or sound "pinched". The dpa's can easily handle the SPL.

 

4 hours ago, Vincent R. said:

Ok so to be fair, it is actually not the "preamp", but the fact the padding saved you, and it has the phantom option whilst padding to line acceptance? 

As far as actual clipping goes the answer is yes but like I mentioned, even on passages that didn't outright clip the Zoom pres sounded non-linear, raspy, edgy, etc. which makes me wonder how figures for max input signals are measured over at Zoom. I've experienced this with pretty much every budget interface/ recorder, the sound quality suffers as you get closer to actual outright clipping.

Link to comment
Share on other sites

I'm gonna paraphrase a Ray Charles quote (I think) that Jeff W has in his signature (I think); "i don't care if it has 99 bits, what does it sound like baby?"

 

I mean 32 bits sure sounds impressive. But it doesn't actually tell us anything about the preamps... It's just a number at this stage. And  a marketing scheme 

Link to comment
Share on other sites

51 minutes ago, Werner Althaus said:

I've experienced this with pretty much every budget interface/ recorder, the sound quality suffers as you get closer to actual outright clipping

I yet have to dive deeper in this, but truth to be told, the main intention of these devices (both let us say a 664 and a Zoom f8/n) are to record location audio for cinema/tv/docs etc. So yeah, in the realms of the high fidelity the F-series (or mix pre) might not be the cream of the crop, but to put it simple; who cares, in that/our line of work. I yet have to clip a preamp (beyond user error) on one of these devices due to one of the mics we typically use on set, or that the scene/location is causing me that much trouble SPL wise.

So yeah I might not be using an F series recorder for classical music recording / "taping" a grand piano...

And to come back full circle to the 32 bit feature; I don't think it is something I specifically "need". I think it might actually be a option for the "novice" in audio recording, the one man band dude with the camera and this new F6 thingy strapped to the bottom of it, just plugging in a mic and not caring at all if it is "gain staged" correctly. 

Link to comment
Share on other sites

I don't think this is actually a 32 bit converter. I think this is taking the two 24 bit streams and assembling a unified 32 bit float file. So the classic debates of analog performance bottleneck aren't really applicable to the converter side of this discussion. It seems they must be running one mic pre>converter at unity gain and the other at max gain, then stiching them together at some "mid-point". I don't know that for a fact, but given that they are claiming your gain setting is irrelevant, then the only logical way to achieve the noise performance of a 70db mic pre would be to use a 70db mic pre. And the opposite end is true too, the only way to achieve the headroom of a unity gain mic pre is to use unity gain. 

 

Also when they talk about the "deficiencies" of 24 bit, they don't say it's inadequate, they just say that this is better. Dan Lavry's comments on bit depth are in relation to fixed point conversion. Sure there is no point in 32bit fixed converters. But, that doesn't mean better resolution is not possible. What they are attempting to address in their marketing here, is the problem of fixed point bits being applied in a linear fashion to a logarithmic signal. That's why the "float" file is essential. If they can do that it will be a huge leap forward in conversion technology. But doing so with a dual conversion system seems to imply what they are actually doing is similar to what a "log taper" potentiometer does. Meaning it's two linear scales cut together to approximate log. I'm interested to hear what it sounds like.

 

I found this thread looking to see if anyone was talking about the Zaxcom issue, but after reading and writing my reply here I am willing to bet that there won't be a patent issue as long as they don't focus too hard on the "impossible to clip" aspect. Zaxcom's tech is all about overload protection and dynamic range. If zoom talks about their tech as being about applying logarithmic bits to audio conversion, and making gain staging redundant, then they might be able to argue a parallel tech.

 

Anyway, blah blah blah, let's wait to hear it.

Link to comment
Share on other sites

21 hours ago, Werner Althaus said:

I'm not familiar with the claim that 32 bit converters have better filters (why?) or lower self noise so I can't really comment on that.

 

I am not the greatest expert on this, but from what I‘ve read, the higher bitrate allows a higher factor of oversampling and this eases the burden on the anti-aliasing filter, which is analog and doesn’t need to have such steep cut-off curves to bandpass filter the desired audio (because Nyquist would be much higher). Many a/d converters augment this with an additional digital filter. Oversampling also improves the SNR. 

Also a higher order of modulation as well in combination with a higher bitrate reduces the quantization noise of the converter. 

Link to comment
Share on other sites

1 hour ago, Constantin said:

 

I am not the greatest expert on this, but from what I‘ve read, the higher bitrate allows a higher factor of oversampling and this eases the burden on the anti-aliasing filter, which is analog and doesn’t need to have such steep cut-off curves to bandpass filter the desired audio (because Nyquist would be much higher). Many a/d converters augment this with an additional digital filter. Oversampling also improves the SNR. 

Also a higher order of modulation as well in combination with a higher bitrate reduces the quantization noise of the converter. 

 

Hmm. I think you are confusing bit rate with resolution.

 

Bit rate is the product of resolution (bits/sample) by sampling frequency (sample/second). But both factors are not interchangeable. Increasing resolution won't affect the Nyquist frequency which is determined *only* by the sampling rate.

 

It's true that perfect analog filters do not exist, so the behavior of the filter is not so good at frequencies close to the Nyquist frequency. So yes, oversampling is a solution to that. You can use a simpler analog filter with the cutoff frequency much higher than the intended maximum frequency of your application, and add a digital filter (which is cheaper and more effective to implement) to "clean" the bandwidth portion between your intended maximum frequency and the actual Nyquist frequency of the oversampled converter.

 

Trying to make it clearer. Let's imagine an A/D converter with an intended band pass of 20 KHz (good old audio) and a "visible" sampling frequency of 48 KHz.

 

We can make it the straightforward way, adding an analog filter in front of the converter. Let's say the cutoff frequency is 24 KHz. But it will affect the phase of a signal being sampled depending on its frequency. The closer to the limit, the worse. And of course it won't cut frequencies right above the cutoff frequency very well.

 

Now, let's do 2x oversampling. We sample at 96 KHz, so we can put the filter on, say, 48 KHz. So in the digital domain we have a sampled signal with a bandwidth of 48 KHz. We still need to clean up it before down converting to 48 KHz sampling or aliasing will happen. 

 

The analog filter is always necessary. Aliasing is an irreversible phenomenon. Once a signal outside of the Nyquist limit has been sampled it becomes indistinguishable of a signal below it. That's the reason why ultrasound can create aliasing. But with oversampling you can relax the analog filter specifications a lot and rely on a cheaper to implement and more flexible digital filter.

 

Reality is a bit complicated and the actual problem with aliasing is to pretend to sample frequencies *both* above and below the Nyquist limit, but this phenomenon mostly has applications in other fields. You can search for information about "Nyquist Zones" but unless you are into software defined radio or other similar applications it won't have much interest for you :) The first Nyquist zone would be from 0 to Fs (Fs = sampling frequency), the second zone would be Fs to 2Fs, etc. So, you can sample signals between Fs and 2Fs as long as your anti aliasing filter is a (Fs, 2Fs) band pass filter. 

 

 

 

 

 

 

Link to comment
Share on other sites

2 hours ago, Constantin said:

 

I am not the greatest expert on this, but from what I‘ve read, the higher bitrate allows a higher factor of oversampling and this eases the burden on the anti-aliasing filter, which is analog and doesn’t need to have such steep cut-off curves to bandpass filter the desired audio (because Nyquist would be much higher). Many a/d converters augment this with an additional digital filter. Oversampling also improves the SNR. 

Also a higher order of modulation as well in combination with a higher bitrate reduces the quantization noise of the converter. 

Myself also being far from an expert I recall that the bitrate has nothing to do with the kind of anti-aliasing filter being employed.

 In this day and age I'm under the impression that this particular nut (pre and post ringing, phase shift) has been cracked, a big reason why many of the past's most ardent devotees of analog gear have adopted a ITB (In The Box) workflow as of late. As you delve into the more budget friendly gear this may not be the case yet.

Dropping the quantization noise to even lower theoretical levels seems irrelevant when the analog circuitry (the weakest link in the chain as far as noise is concerned) can't possibly exceed 21 bit dynamic range, and that is state of the art analog circuitry at low gain. I've heard it expressed like this: it makes no difference if your calculator computes the amount owed to you to the 20th decimal point, in the end you'll still only receive dollars and whole cents, the currency's maximum "analog" physical resolution.

I am still curious about the mic pres and (lack of) gain staging in the F6, the pres are supposedly the same as the F8. Are they fixed gain at 10dB, the lowest setting on the F8 or continuous like stagetec's true-match ADCs? How do I monitor a mic signal without sufficient analog gain applied during normal operation, cranking the fader which controls the record level? Remember there are no prefade ISOs here, yes, the recorded file may be clean but the isos will have all potentially faulty fader moves baked in.  If then confronted with excessive spl will the headphone amp clip if my fader is too high? Can I trust that the recording was unclipped despite clipping the headphone amp? In a highend digital mixing console with tons of internal headroom these things are computed in real time to guarantee the output and monitoring remains free of clipping at all time regardless how the internal levels are, is that the case here? How will the super low level files integrate into a realistic post workflow? How susceptible to RFI/EMI are the un-amplified mic level signals within this tightly packed "portable computer" really? I mean the reason we apply gain via mic pres is to optimize audio levels in terms of noise before they are distributed and manipulated. And lastly , what's keeping me from using the same theoretical approach of feeding "raw" mic signal into a conventional 24 bit converter and enjoy what has previously believed to be unobtainable 144 dB of dynamic range?

Link to comment
Share on other sites

2 hours ago, borjam said:

 

Hmm. I think you are confusing bit rate with resolution.

 

 

It‘s always possible that I‘m confusing things, but I don’t think this. 

It‘s what I had read that some new 32-bit A/D converters offer higher oversampling rates. Because of/despite their higher bitrate

1 hour ago, Werner Althaus said:

Myself also being far from an expert I recall that the bitrate has nothing to do with the kind of anti-aliasing filter being employed.

 

 

Not directly, no. Anyway, maybe it’s just that 32-bit converters come with more powerful chips that are capable of higher oversampling rates, I don’t know

Link to comment
Share on other sites

15 hours ago, Constantin said:

 

It‘s always possible that I‘m confusing things, but I don’t think this. 

It‘s what I had read that some new 32-bit A/D converters offer higher oversampling rates. Because of/despite their higher bitrate

 

Not directly, no. Anyway, maybe it’s just that 32-bit converters come with more powerful chips that are capable of higher oversampling rates, I don’t know

 

Hope I wasn't too blunt! I was just thinking about beginners getting confused with the terms.

 

That would make sense, modern A/D converters implementing better digital filtering. 

 

This one is used in MixPre-3:

 

https://www.akm.com/akm/en/file/datasheet/AK5558VN.pdf

 

And it has some nice features such as four different types of digital filters. I was puzzled when I read that they used 32 bit converters, maybe that's the reason. 

 

Also I noticed that it can use a neat trick (summation) to increase S/N ratio. Cirrus Logic explain it very well here. I had never heard of this, you can learn new tricks everyday :)

 

https://statics.cirrus.com/pubs/appNote/AN331REV1.pdf

 

Link to comment
Share on other sites

9 hours ago, borjam said:

 

Hope I wasn't too blunt! I was just thinking about beginners getting confused with the terms.

 

That would make sense, modern A/D converters implementing better digital filtering. 

 

This one is used in MixPre-3:

 

https://www.akm.com/akm/en/file/datasheet/AK5558VN.pdf

 

And it has some nice features such as four different types of digital filters. I was puzzled when I read that they used 32 bit converters, maybe that's the reason. 

 

 

The mixpre3 uses the smaller 6 channel AK5576EN in a 6-to-3 mode giving it better dynamic range.  

 

Link to comment
Share on other sites

23 minutes ago, Display Name said:

 

The mixpre3 uses the smaller 6 channel AK5576EN in a 6-to-3 mode giving it better dynamic range.  

 

Are you sure? Has there been a change during production? I ask, since the teardown photos show the AK5558VN as per Borjam's post: https://fccid.io/2AKLX-739M3/Internal-Photos/Internal-Photos-3297643.iframe

 

Cheers,

 

Roland

Link to comment
Share on other sites

Yes the mixpre3 photographed in the FCC docs differs as they used the poorer speced 5558VN during the certification. 

 

I believe the 5576EN was not yet released during development. 

 

The 5558VN can not reach the specified 120dB S/N unless you do 8-to-2 summing and then you are one channel short. 

Therefor it seems logical that the MP3 follow the same design solution as the 6 and 10 series which use the 557x and 8-to-4 summing to reach the 120 dB spec. Also the AK5576EN was released later then the 8 channel ADC and there for not available early in the dev process. 

Link to comment
Share on other sites

Here's a bit of an overview of the current Sonosax preamp design which incorporates post (dual) ADC gain and has an option to record 32bit files:
http://rtsound.net/no-gain-no-pain-sonosax-r4-and-ad8-gain-structure/

I've asked Sonosax a couple of questions about this since this discussion about the F6 has started, whether recording in 32bit produces float or integer files, and whether they can 'save' any overmodulated tracks

Link to comment
Share on other sites

Regarding the 32-bit vs 24-bit (as well as recording in higher sampling rates), I always like to quote the following article:
https://people.xiph.org/~xiphmont/demo/neil-young.html

Specifically, the paragraph below:

Quote

When does 24 bit matter?

Professionals use 24 bit samples in recording and production [14] for headroom, noise floor, and convenience reasons.

16 bits is enough to span the real hearing range with room to spare. It does not span the entire possible signal range of audio equipment. The primary reason to use 24 bits when recording is to prevent mistakes; rather than being careful to center 16 bit recording-- risking clipping if you guess too high and adding noise if you guess too low-- 24 bits allows an operator to set an approximate level and not worry too much about it. Missing the optimal gain setting by a few bits has no consequences, and effects that dynamically compress the recorded range have a deep floor to work with.

An engineer also requires more than 16 bits during mixing and mastering. Modern work flows may involve literally thousands of effects and operations. The quantization noise and noise floor of a 16 bit sample may be undetectable during playback, but multiplying that noise by a few thousand times eventually becomes noticeable. 24 bits keeps the accumulated noise at a very low level. Once the music is ready to distribute, there's no reason to keep more than 16 bits.

 

While I think the idea of 32-bit float recording is nice in theory, I also don't think it's a necessary feature for production sound recording. In fact, it may create more confusion and issues in post, particularly when the AE ingests our files and syncs them, and they see that dialog ins't gain staged as normal (or at least it wouldn't look good if I'm having to explain why gain staging properly didn't matter since they're 32-bit float files to the producer and/or editor who are calling me inquiring about them).

 

24-bit provides enough dynamic range to capture the entire usable dynamic range of any microphone (from above its self-noise to its Max SPL). IMO the only place where higher dynamic range (and higher sampling rates) is really useful, is in post, for DSP.

 

I'm sure for consumers and prosumers alike, the idea that they can record in 32-bit float files and not have to worry about properly gain staging during recording (the whole "fix it in post" mentality) may sound like a great thing. For the rest of us (or at least for me), I think while cool, not really needed.

Link to comment
Share on other sites

I think one of the biggest reasons why Zoom is using 32bit floating-point is because of the digital gain knobs on the F6.

If you turn up the digital gain it is possible that the signal would go above 0dBFS.

With 24bit fixed-point 0dBFS is the maximum value and everything above that simply clips.

32bit float on the other hand allows the signal to go above 0dBFS and can still capture it just fine.

You could then turn down the audio in post and save all the audio above 0dBFS with the 32bit float file, which wouldn't be possible with 24bit file.

So, the 32bit float ensures that even when the digital gain on the device wasn't set properly, the whole signal is captured without clipping.

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

×
×
  • Create New...