Topic: Zoom F6 (32-bit float equipped)

[aaronji]

'trim' was set to +12 dB in 24 bit case, and of course is disabled in 32 bit case

Per the specs, minimum gain is 12 dB; is this minimum gain or minimum gain plus 12 dB?

I have to admit the stats are a little perplexing to me. I guess the differences are due to the way gain is applied in 32-bit mode? It would be interesting to see what levels in 24-bit yield the same output levels for both 24-bit and 32-bit for various SPLs.

any benefits of 32-bit FP disappear.

What benefits are those, other than not needing to set levels?

[voltronic]

any benefits of 32-bit FP disappear.

What benefits are those, other than not needing to set levels?

You say that like it's insignificant.  It might not be a big thing for you in your workflow, but for someone like me it is huge.  I admit that my recording situation may be unusual compared to most on this board, but this format has been very helpful for me.

Just to be clear: I'm not speaking about 32-bit FP generically, but specifically the way it is implemented in these new recorders with multiple auto-ranging ADCs.  The combination of the two things is what allows you to have the larger dynamic range, and be able to recover recordings that have levels way over or under without adding significant noise.  That room for error certainly has limits between analog noise floor and input stage overload, but is still way bigger than you would have otherwise. 

Secondary benefits to this are no need for safety tracks, saving storage and battery, and limiters (which are almost always audible when they engage.  Analog limiters sound much less offensive than digital, but are more expensive to implement.  If you consider the input stage overload as your upper limit with a multi-ADC 32-bit FP implementation, a limiter is not going to help you there anyway; you'd need a pad.

I'm not parroting some marketing copy; I'm telling you what my personal experience has been.

These threads are probably a better place to debate the usefulness of the format itself.
https://taperssection.com/index.php?topic=192112.0
https://taperssection.com/index.php?topic=192100.0

[jerryfreak]

'trim' was set to +12 dB in 24 bit case, and of course is disabled in 32 bit case

Per the specs, minimum gain is 12 dB; is this minimum gain or minimum gain plus 12 dB?

again it doesn’t have an analog gain stage, that’s all post-sdc so it incorporates any noise previously accumulated. at some point i’ll do a high-gain test of v3 vs stock input

[Gutbucket]

Well, it most certainly has an analog input stage, just not one that features user adjustable trim of gain.  Could be unity gain, or whatever is needed to provide the appropriate signal level to the ADC, at minimum it acts as a buffer.  I suspect the padding which happens when line-input and/or the advance limiter is in use is implemented via a gain adjustment of that input stage.

[aaronji]

You say that like it's insignificant.  It might not be a big thing for you in your workflow, but for someone like me it is huge.  I admit that my recording situation may be unusual compared to most on this board, but this format has been very helpful for me.

No, not at all. If you need it you need it. I was just checking to see if there was some other benefit of which I was unaware. Apparently not.

As for limiters, if you set the levels conservatively (based on mic specifications and input sensitivity) on one of these high dynamic range recorders, if you do hit the limiter, it will generally only be on brief, sporadic spikes and will be totally inaudible. At least that's the case with the MixPres.

'trim' was set to +12 dB in 24 bit case, and of course is disabled in 32 bit case

Per the specs, minimum gain is 12 dB; is this minimum gain or minimum gain plus 12 dB?

again it doesn’t have an analog gain stage, that’s all post-sdc so it incorporates any noise previously accumulated. at some point i’ll do a high-gain test of v3 vs stock input

So in 24-bit mode (as per the quote of your post), there is still no trim? That doesn't really square with what I have read about this recorder. I thought the "no trim" only applied to 32-bit.

[jerryfreak]

well im just basing it on block diagram, it seems that there isnt true analog gain. just one fixed level set to be the best match to input into the ADC(s). and the pad when needed. (I'll do some more detailed measurements with the pad on and off this evening)

it seems that at the end of the day that there is a fixed range that an input stage can have

at high levels, signal voltage goes up pretty dramatically
0dBU= 0.77V
+4 = 1.23V
+6 = 1.54V
+12 = 3.08V
+24 = 15.5V
+30 = 24V

which is probably why we dont see gear that can take signals much hotter than that. I may be getting out over my skis on circuit knowledge but i would imagine hot signals like that would demand completely different design in regard to components that are rated for higher voltage, as well as probably increased spacing of traces, and other issues.

again im just speculating but i doubt most (any?) ADC chips are really able to take double-digit voltage inputs, so at some point we are 'funneling' our signal down to fit the ADCs parameters

likewise, taking +4dB (1.23V) as our hypothetical upper limit, if we are assuming a max of 130 dB dynamic range, -126dBU would correspond to 3.88 x 10-7!
So at some point just via physics we run into inevitable circuit noise

Recall that conventional recording wisdom is to capture the range of sound in a room with the appropriate microphone and use a gain stage to match that range to an ADC ore recording media

a studio may be 20dB background noise level. a quiet concert hall with respectful crowd probably 40dB. rock concert? maybe 60ish dB with all the people

in each case we may hit 70dB of dynamic range (assuming the studio and concert halls will be in a listenable mid 90s, and an ear-splitting rock concert in the 120s

in the zooms case, when the EIN of the analog input approaches 130dB, and a 24-bit word can handle 144 dB of range (and ideally matched on the top end), youre gonna run into the noise floor of the input first in almost every case.

Now that is analog noise and just because its higher than the quantization noise of the LSB in the 24bit word....doesnt mean that it will completely mask the quantization noise. It seems from the few people (myself included) who have listened, that the 24 bit and 32-bit samples sound"different", though nobody can really peg a "better"

remember that at the end of the day, a 32-bit float still only has a 24-bit mantissa in which to express the data, and the exponent is a sliding scale of sorts.

I'm not convinced the 768dB dynamic range offered by the 32bit float container really offers substantial advantage over the 144 dB of a 24-bit binary signal for capturing real world dynamic range of sounds, that appears to be well under 100dB range in all cases. Thats what we are attempting to determine. The 4006 is an extremely quiet mic, 15 dBA self noise with a maximum SPL of 143.

The zoom has a stated noise of "−127 dBu or less (A-weighted, +75 dB input gain". which is pretty low, at that high gain level. But what does it mean? does that mean that we can amplify a signal contianing 120dB of data 75dB and the full dynamic range is preserved? or that when amplified 75 dB we have a 127 dB container to represent that, and what noise did it bring along with it? the former scenario would assume the analog input circuit has a full dynamic range of 190+ dB which i dont think is achievable

Well, it most certainly has an analog input stage, just not one that features user adjustable trim of gain.  Could be unity gain, or whatever is needed to provide the appropriate signal level to the ADC, at minimum it acts as a buffer.  I suspect the padding which happens when line-input and/or the advance limiter is in use is implemented via a gain adjustment of that input stage.

[jerryfreak]

So in 24-bit mode (as per the quote of your post), there is still no trim? That doesn't really square with what I have read about this recorder. I thought the "no trim" only applied to 32-bit.

there is a trim function (to alter signal level written to file), but its done in DSP. According to the block diagram the analog inputs feed right into the ADC.see block diagram on page 197

https://www.zoom-na.com/sites/default/files/products/downloads/pdfs/E_F6_3.pdf

i assume the 'ADC' in that block diagram is a black box containing their multiple ADCs and assumedly some sort of analysis/autoranging

it is unclear where the 20dB pad for line-in operates

heres some info from zoom support, somewhat cryptical to me in parts
---
zoom:

The Trim on the F6 acts as a digital stage. The dual ADC is used when recording in 32 bit float and when recording in 24 bit.

The Trim allows you to set the level that will be recorded to the 24 bit file. This makes it so that if the signal clips in 24 bit mode, it is due to the file clipping from the limitations of the 24 bit integer format, not from clipping at the converter.

Since linear formats cannot accommodate the wide dynamic range of the dual ADC on the F6, a software Trim is necessary to set the level of the recorded file.

-----

me:
so in other words, there is no analog gain or attenuation stage whatsoever before the ADC

and thus the best noise performance in all cases would be to feed it a signal as close as possible to its maximum +4dBU input clipping level?

----
zoom:
Essentially, yes, any louder signals will always have better noise performance, as long as at least one of the converters is not clipped.

Keep in mind that the F6 still has a very low noise floor at essentially all signal levels. Since the device is constantly switching converters depending on the input level, low level signals are always recording with the best noise level possible on the device, which is comparable to a typical 24 bit recorder/preamp operating at the most optimal setting possible (which is often difficult to predict).
-----

so from a design perspective, there technically should be a substantial advantage, its surprising we cant measure a dynamic range that is significantly broader in 32-bit float mode vs 24-bit.

does zoom have a patent we can unearth, similar to what was found for sound devices' implementation?

[aaronji]

^ That's very interesting. Thanks for posting your communication with Zoom! I am really surprised that there isn't any analog gain stage even in 24-bit. I guess, in essence, 24-bit mode is just handicapped 32-bit...

[jerryfreak]

^ That's very interesting. Thanks for posting your communication with Zoom! I am really surprised that there isn't any analog gain stage even in 24-bit. I guess, in essence, 24-bit mode is just handicapped 32-bit...

well i havent opened mine up, but TBH, between the battery sled, display/controls, and volume required for the XLR inputs, theres not much room left. Its probably a single board design on the size of say an average handheld

heres the teardown for the mixpre3. i cant find the FCC id for my zoom but i imagine there is something similar
https://fccid.io/2AKLX-739M3/Internal-Photos/Internal-Photos-3297643

i would guess the board real estate is half the size of the mixpre and not a lot of room for analog circuitry

[voltronic]

Huh.  I never really compared the block diagrams of the fixed and float point modes until now.  It sure looks like there is no analog gain control; it's all post-ADC digital level control no matter what mode you are in.  The difference is where it happens:

In fixed point and dual modes, the level control is before the HPF, limiter and phase controls.

In float point mode with the trim pot setting as REC LEVEL, the level control is after all of those things (except the limiter which is disabled).

In float point mode with the trim pot setting as REFERENCE LEVEL, you have no control at all over the levels being written to the ISO tracks.  You only have fader controls for the L/R downmix.


I wonder how many other recorders out there don't really have an analog gain stage.  If Paul is about, I'm sure he will tell us how that works on the MixPre-II.

[jerryfreak]

just because its not adjustable, doesnt necessarily mean it doesnt have at least a basic op-amp stage. most handhelds have at least a basic gain circuit. i imagine at a certain level (size, cost, amount of gain, the particular sdc you are feeding), that digital gain can actually be superior to an analog circuit

when i tore open my d:vice and sent pictures to doug he said:
They use a real preamp chip with THD+N down @ -136dB. I use that one in some of my MODs. That should sound close to the AD2K.


which seems like a bold call, but im compelled to share this other info we discussed, emphasis mine

I agree, anything over 100dB SN and dynamic range is more than adequate for an A/D chip. What's far more critical is the analog circuity supporting it. A lot of real estate was dedicated to presenting the ADC with as perfect a signal as possible in the AD2K. It's only in the last 2 years that single op amp stage analog circuity could come close. Even so, when I hear what can be done with multiple stages in state of the art converters, I realize that while today's single stage op amp circuits can do what yesterday's state of the art multistage circuits could do, they don't do state of the art sound. If you  get a chance, check out a Cary Audio or EMMLabs DAC. You may realize just how good your recordings are.


so point being, weve come a long way in both analog and digital chips.... the d:vice manages 114 dB of dynamic range and that board is the size of a half dollar.

Huh.  I never really compared the block diagrams of the fixed and float point modes until now.  It sure looks like there is no analog gain control; it's all post-ADC digital level control no matter what mode you are in.  The difference is where it happens:

In fixed point and dual modes, the level control is before the HPF, limiter and phase controls.

In float point mode with the trim pot setting as REC LEVEL, the level control is after all of those things (except the limiter which is disabled).

In float point mode with the trim pot setting as REFERENCE LEVEL, you have no control at all over the levels being written to the ISO tracks.  You only have fader controls for the L/R downmix.


I wonder how many other recorders out there don't really have an analog gain stage.  If Paul is about, I'm sure he will tell us how that works on the MixPre-II.

[jerryfreak]

more recordings coming soon... birds were going off at sunrise, i wish i woulda got em as its raining again. I wish i had a basic LDC like a rode NT1, would buy me 10 dB lower noise over any of the SDCs i have. anybody want to loan one?

[carpa]

I'm following this interesting thread, trying to understand....I can't get all of what's discussed here due to my lack of knowledge, so there are some point/questions I'd like to make clear, stated that - if I'm right - there is no analog setting of the level possible before the ADC. What I'm trying to figure out is how owners' behavior will change while using this Zoom vs. "traditional" stuff. Sorry for being so "basic" but I really can't get to the point; for most of you this is probably dumb questions but it may help someone like me to understand.

1) When recording  we have some dynamic expectations. A classical guitar is very different from a drum set so we get prepared to it also trying to match the settings ( If I place the mic at the mouth of a trombone I'll not crank all the gain up!).  We still may end up with a soft and conservative level to be raised in post (risk of noise floor increase), with a clipped or badly limited recording, or - still possible - with a good setting not needing further treatment level-wise.   F6 "doesn't know" wether I'll be whispering at distance or I'll be exploding a bomb close to it; so, no clipping risk but a recording which will in every case need to be adjusted to be usable?   
Some YouTube videos I've seen ( i.e. Curtis Judd) show a treatment in post of a very low signal- then raised 40 or 50 dbs- vs. another ultra-clipped signal - then lowered without consequences. How's it possible? The idea of fixed gain suggests me to think about an average level decided by the recorder itself, probably all a bit too hot to a bit to low...I don't understand ho he came to have such low and such clipped signal from the same mic and recording situation.

2) According to the previous point, in an average scenario, i.e. a piano and clarinet or violin with a stereo mic set placed - let's say - from 1 to 2 mt. from the players - what the recording would be right out of the box? Just listening on headphones what's been recorded, how far will it sound from a "traditional" recording?

3) A slightly different aspect which has been treated in this thread concerns the risk of clipping the input stage with very hot signals, not being possible to trim he gain before ( even 24 bit mode, if I understand, acts the same way). In this scenario only the mic sensitivity will rule, so it would just be necessary to exclude the use of microphones according to their specs or it is more complicated than that?

Thank you for your help

[Gutbucket]

The differences in working procedure are basically this:

1) Choose the appropriate microphone(s) and place/arrange them like you otherwise would (this part does not change).

2) Record. If using 16 or 24 bit modes you adjust trim as usual.  If using 32bitFP mode you do not need to adjust input trim (this is different).

3) Afterward.. If you recorded using 16 or 24bit modes you may wish to re-adjust levels somewhat, but they will likely be close to correct and usable "as is" due to you having manually set input trim prior to recording.  Using 32bitFP mode you will need to re-adjust levels to get appropriate output because you did not manually set input trim prior to recording (this is different).

4) In addition to 3, some post-recording procedures involving handling the 32bitFP file format will be different.

That's essentially it.

How's it possible?

That's what our extended discussion has partly been about, both here in this thread, and more specifically in the other thread intended for more in depth exploration-
32Bit Float recording - The Technical view. Your post is a good example of why we should try to keep the technical discussion in that thread, so that F6 operation information in this one remains clear.

[jerryfreak]

ok more tests
ch1 (L): 4011A
ch3 (R): CMC64

Source material: Time Difference by Hiromi's Sonicbloom, SACD
playback: Benchmark DAC2 > balanced out > single Dynaudio BM5A monitor at low volume

source material volume was the same for all tests. Volume level (SPL) estimated at 64-70 dB peak, 44-50dB RMS*

room noise estimated at less than 41-47 dB peak, 28-34 dB RMS*. it was difficult to tell if i was running into the noise floor of the recorder though

I also measured room noise at 31 dBA with iphone app. Room was quieter than my tinnitus. Room is in quietest corner of house, nothing electrical in the room except for the dynaudio monitor,  the system feeding it via xlr was 2 rooms away.

on one sample (car_noise.flac) you can hear a vehicle accelerating through its gears all by its lonesome at 2AM. Thats through a closed window on a highway about a 1/3 mile away. Long story short, you would be hard pressed to be recording any source this quiet outside of nature settings

recorder was mounted right on top of stand, mics sticking right out the side, about 16" from the monitor

*the reason you see a 6dB range in estimated SPL is surprisingly the SPL calibrator i used read 6dB different on the schoeps and DPA caps (likely because of capsule back venting design, it is not super accurate for testing absolute measurements on different styles of directional mics, as i believe gutbucket pointed out to me along time ago. What was odd though, despite the mics reading so differently with the calibrator, they were actually equally sensitive within about 1 dB on all the tests. so the '94 dB levels' you see referenced in each test were the levels on the recorded tracks when the calibrator was used on each mic. likewise the calculated 'max SPL at FSD' is a range guessed from this. in other words, at those particular recording settings, you would need to hit that SPL before you clip the ADC. Of course some of the numbers are ridiculously high and the analog input would be beyond overloaded at 180-200+ SPL(!). id have to do some math to figure out what the actual clipping SPL numbers would be with each mic (i.e. what SPL causes +4dB/+24dB signal into the input). maybe i can try the calibrator with an omni and report back on that

anyway, on with the show. I was trying to test the following variables with the zoom
low vs high recording level where applicable
24bit vs 32 bit files
mic vs line in

i found it educational, i hope you do as well

test 1:
24 bit
mic-in
trim=+12dB (the minimum)
fader =-48dB  (the minimum)
94dB levels = -44dB, -36 dB
corresponding SPL at FSD: 138dB, 130dB
peak program level = -74dB
RMS program level = -94dB
file was normalized approximately +73dB

test 2:
24 bit
line-in
trim=-8dB (the minimum)
fader =-48dB  (the minimum)
94dB levels = -64dB, -56 dB
corresponding SPL at FSD: 158dB, 150dB
peak program level = -94 dB
RMS program level = -114dB
normalized approximately +93dB

test 3:
32 bit float
line-in
trim= -48dB (the minimum)
94dB levels = -124dB, -117 dB
corresponding SPL at FSD: 218dB, 211dB
peak program level = -154 dB
RMS program level = -173dB
normalized approximately +153dB


test 4:
32 bit float
mic-in
trim= -48dB (the minimum)
94dB levels = -104dB, -97 dB
corresponding SPL at FSD: 198dB, 191dB
peak program level =-134dB
RMS program level = -154dB
normalized approximately +132dB

test 5:
24 bit
mic-in
trim=+12dB (the minimum)
fader =+24dB  (the maximum)
94dB levels = -44dB, -36 dB
corresponding SPL at FSD: 138dB, 130 dB
peak program level = -74dB
RMS program level = -94dB
normalized approximately +73dB

test 6:
24 bit
mic-in
trim=+75dB (the maximum)
fader =-48dB  (the minimum)
94dB levels = (over FSD)
corresponding SPL at FSD: 80-85dB est.
peak program level = -11dB
RMS program level = -31dB
normalized approximately 10-11 dB

test 7:
tascam DR100mkiii
mic in
pad off
gain set to minimum (+20dB)
94dB levels = -22dB, -16dB
corresponding SPL at FSD: 116dB, 110dB
peak program level = -51dB
RMS program level = -70dB
normalized approximately 50dB


test 8:
tascam DR100mkiii
mic in
pad off
gain maximum (+56dB)
94dB levels = (over FSD)
corresponding SPL at FSD: est. 75-80 dB
peak program level = -15dB
RMS program level = -34 dB
normalized approximately 14dB

test 9:
tascam DR100mkiii
mic in
pad on
gain set to maximum (+31.5dB)
94dB levels = -10dB, -4dB
corresponding SPL at FSD: 104dB, 98dB
peak program level = -40dB
RMS program level = -58dB
normalized approximately 38dB

test 10:
tascam DR100mkiii
mic in
pad on
gain set to minimum (0)
94dB levels = -41dB, -34dB
corresponding SPL at FSD: 135 dB, 129dB
peak program level = -71dB
RMS program level = -90dB
normalized approximately 70dB

test 11
V3 with 30 dB of gain
peak program level = -66dB
RMS program level = -85dB
est. SPL at FSD: 125-130dB
normalized approximately 65dB in post


test 12
V3 with 70 dB of gain
peak program level = -27dB
RMS program level = -46dB
est. SPL at FSD: 85-90dB
normalized approximately 26dB in post

here is a folder with all six of the zoom tests and what i thought were the best test each from the Tascam and the V3. they are all cut to within a second of each other on start point so you can AB if you wish. imo its kinda all about the noise vs program material in the first few seconds, you'll be able to discard a few right off the bat. Then you can listen more critically to the ones which remain

https://drive.google.com/open?id=13nG-Z4kluspAzYrdrkO36glBGDrOAiS7