The No.1 Website for Pro Audio
Experience with Gain-Ranging Converters (Stagetec)
Old 29th June 2020
  #31
Lives for gear
 
tourtelot's Avatar
I am no electronics engineer but it does seem safe to say that complicated conversions would have, probably, more chance of artifacts.

D.
Old 29th June 2020
  #32
Lives for gear
Quote:
Originally Posted by voltronic View Post
You are only limited by the max level of the mic input stage and the analog noise floor of the mic preamps. Anywhere within that, you have the full dynamic range available, and can shift everything up and down in post with no added noise.
To be frank, this is the case with any 24-bit medium, as the medium SNR and dynamic range greatly exceeds the dynamic range of all real world playback scenarios.
Old 29th June 2020
  #33
Gear Maniac
 

SNR and dynamic range are issues of their own - I believe we are thinking 'analogue' if we concentrate on terms such as SNR and dynamic range. We need to look at avoiding unpleasant artifacts caused by truncation and calculation errors in lsb's during processing. With high-quality digital we aren't mostly usually talking 'noise' we are talking 'errors'.

This SNR thinking dates back a long way to when some designers believed dither was of no importance in a mixdown or after processing because the s/n of the analogue signal was significantly within a 24bit range. In practice we found that dither made an audible difference not just a theoretical one, even if the analogue broadband noise-floor of a recording was even 40dB above the noise-floor of a pure digital 24bit system.

I see the marketing appeal of 32bit recording, but in practice if the a/d is driven into clip it will still be clipped - and the noisefloor will be set by the weakest link in the system in the presence of signal. If we end up with audible calculation errors in mixing and processing, we might be better sticking to 24bit and sorting out dither.
Old 29th June 2020
  #34
@ voltronic , thank you for the links. One of the posts from Tapersection had a video from Merging in regards to their new I/O cards:

https://youtu.be/IMkUcWr2-cg?t=629

This video from Merging certainly has some exemplary measurements in it. Their previous cards were also exceptional, and are regularly used for the highest end work.

In my own studio, when I switch from a high end analog monitor controller to going digital all the way to my Genelec 8351a's, I received a very noticeable increase in imaging, while the noise dropped at the same time. All the non-musicians in my house commented and asked what I did. It was not subtle to pull out a huge stack of analog cables.

I am now at a place where microphones go into preamps and direct into converters and then are not analog again until inside the Genelec's on their way to my ears. Part of my wondering is if I can simplify yet again - microphones into a preamp/digitization box that actually improves the sound quality by eliminating analog electronics. I am thrilled with the quality of the Sonosax SX-R4+, which essentially works this way. It is an excellent sound, and it seems that if you only amplify the signals enough for the converters (which don't need +24 or +28 dbU to operate), and that is optimized exactly for the conversion, that a lot of circuitry is eliminated, with a corresponding benefit to the recording transparency.

It seems that in the Merging implementation, they have a higher level path at +4 dbU and a lower level path at -17dbU. After conversion, the higher level path is reduced by 21 db so that it is the same level as the lower path. This should drop any higher noise floor at the +4 level by the same amount, and eliminate any modulation of the noise floor. I'm guessing that all of the gain ranging converters do this?

It also seems that Merging offers a +12db analog boost to raise very low levels before the converters, and then has digital gain after the signals are re-combined. It is a clever system.
Old 29th June 2020
  #35
Lives for gear
 
voltronic's Avatar
 

Quote:
Originally Posted by TonyF View Post
I see the marketing appeal of 32bit recording, but in practice if the a/d is driven into clip it will still be clipped - and the noisefloor will be set by the weakest link in the system in the presence of signal. If we end up with audible calculation errors in mixing and processing, we might be better sticking to 24bit and sorting out dither.
Tony, I am not sure if your criticism is directed at 32-bit float recording, or at gain-ranging multiple ADCs?

By my reasoning, such a system which uses both of these makes clipping the ADC much less likely. One of those ADC chips is only being fed the loudest signal range, the top end of which would result in digital levels far about 0 dBFS. If that output is packaged in a 32-bit float container, you won't have any clipping. I have tried this myself with test recordings that peak upwards of +50 dB, and once I turn down the level in post there is no clipping.

The main concern in our conversation on TS was about overloading the analog input stage with mics that put out very hot levels, as that seems to be the weakest link. One of the members there has done classical piano recording with Josephson C617s into his F6, and everything has come out fine for him, even though the F6 input stage maxes out at a rather low +4 dBu. The Sound Devices implementation lets you go another 10 dB above that.

I have no idea about the dither situation; maybe it is worth asking reps from these companies because that does not seem to be published. All I know is that my F6 has a decently low noisefloor, so if there are digital errors being introduced I am not hearing / seeing them.
Old 29th June 2020
  #36
Lives for gear
 
voltronic's Avatar
 

Quote:
Originally Posted by niversen View Post


voltronic, thank you for the links. One of the posts from Tapersection had a video from
It seems that in the Merging implementation, they have a higher level path at +4 dbU and a lower level path at -17dbU. After conversion, the higher level path is reduced by 21 db so that it is the same level as the lower path. This should drop any higher noise floor at the +4 level by the same amount, and eliminate any modulation of the noise floor. I'm guessing that all of the gain ranging converters do this?

It also seems that Merging offers a +12db analog boost to raise very low levels before the converters, and then has digital gain after the signals are re-combined. It is a clever system.
I don't know if that's what the others do, but that seems like a smart way to do this. It actually sounds similar to how digital limiters work.
Old 29th June 2020
  #37
Lives for gear
 

no doubt they are of high quality and (possibly) very clever in technical terms*; it's just that i don't like the 'sonic footprint' much, especially the a/d side...


* i don't know enough to comment on this but i'm not getting any satisfaction from the assumption alone that i might be using the latest and greatest technology: it's gotta sound right to my ears - i that respect, evaluating gear remains highly personal [and reminds me of discussion years ago: dsd just had to sound better as it was so ridiculously expensive - well, i didn't (and still doesn't), besides the technical boundaries in terms of processing, lack of acceptance in the market, issues with distribution etc.]
Old 29th June 2020
  #38
Gear Maniac
 

I guess that in a way the 'dither issue' relates to both propositions.

Within SADiE we have had the option to archive a mix undithered at 32bit resolution. The logic being that it is a 32bit float DAW and that represents the cleanest audio you can archive with extra headroom and footroom - but you would have to dither it to deliver it or to play it through a 24bit dac. So I cannot see any real profound point, unless you want to raise the peak level above notional peak bits, and crush a few momentary peaks using a clipper of some kind - thereby increasing the perceived loudness (peak to mean ratio) of your recording by squashing one or too unusually loud peaks before dithering to 24bit.

With gain-ranging multiple adc's the gear changes between adc's must be managed with dither, and even then that still does not deal with the offsetting of THD+N background as you switch converters. For example: consider a system combining two 16bit a/d converters, one set to clip at a 18dB lower level than the other. In the presence of peak bits audio the overall s/n ratio is still that of a 16bit a/d converter. Remove the audio input driving the a/d converter to peak bits, and the DSP will switch over to the other a/d converter and hey presto the background improves instantly by 18dB. Wow. So which measurement do we take to define a published signal-to-noise? - the one which is 18dB greater of course - even if it isn't really in context. Are we convinced that switching converters within music around -18dB will be totally transparent and inaudible? I believe not. It sounds like a convoluted almost Class B concept. The audio's characteristics are being modulated as a function of level. How closely matched are the a/d converters going to be? Finally - why? - when some sensible management of headroom would do a better job. You could also do better using digital gain cautiously.

Most of us doing live recording for a living leave headroom in the recording in any case, not least because modern adc's have such low inherent noise you can throw away 6dB or 10dB if you have to with no overall degradation in the end result. If you record your mix of ranging adc system not worrying about headroom at 32bit safe in the knowledge that nothing will actually clip if you go over nominal peak-bits, then you still have to dither the audio to 32bit, only including audio which exceeds the normal peak bits of a 24bit system. Then you have to dither it again later as a 24bit deliverable after setting the overall level to where it needs to be.

The original Sony PCM1 (1977) used a 12bit plus ranging bit a/d. The a/d chip in the PCM-F1 had effectively two converters within the chip, thermally coupled. There is little new in the concept.

Last edited by TonyF; 29th June 2020 at 07:40 PM.. Reason: error
Old 29th June 2020
  #39
Lives for gear
 
voltronic's Avatar
 

I think those are all good points, and we won't have real answers to them until we hear directly from manufacturers on how they are dealing with this, and we have more testing on this tech in the field with actual music recording. One would think the people who make this stuff have tested it thoroughly, but the question is: with what material?

So far I have heard no artifacts or noisefloor shifts, though I do not know how many of my recordings have caused much shifting to happen from one ADC to another. I don't do this professionally, nor do I have occasion to record a large symphony orchestra very often.
Old 30th June 2020
  #40
Gear Maniac
 

Agreed.

All just feels a bit of a specmanship gimmick and potentially a can of worms, but I must be missing a point somewhere. Zoom did it, so Sound Devices had to, now we have this discussion. A long while back Deutsche Grammophon devised a concept called 4D which worked along similar lines based around Yamaha tech., combining two a/d converters so when one clipped, the other one automatically switched in suitably padded down to save your life.

6dB of digital gain in the appropriate place would give the adc 6dB of headroom without any complications of 32bit management, though it would degrade the noise specification with a shorted input - still the noise would be well below any acoustic audio signal.

I don't think AES3 (AES/EBU) handles audio wordlengths greater than 24bit. Do all popular plugins work with 32bit i/o? I don't know. MADI and AES67 audio networking have maximum bitdepth of 24bits, and audio over IP is a big deal. I have been reading about 64bit float in DAW's to complicate the issue further.

More anon I am sure.
Old 30th June 2020
  #41
Lives for gear
 

Quote:
Originally Posted by TonyF View Post
(...)I don't think AES3 (AES/EBU) handles audio wordlengths greater than 24bit. Do all popular plugins work with 32bit i/o? I don't know. MADI and AES67 audio networking have maximum bitdepth of 24bits, and audio over IP is a big deal (...)
i think you just answered your own question about potential dithering of 32-bit converters: digico for maybe two years has a multichannel preamp/32-bit converter which fits into their remotely controlled stageboxes:

https://digico.biz/rackmodules/sd-32-bit-mic-pre-amp/

since the format to transport audio to/from their desks (which houses the dsp which runs on fpga chips) is madi, signals must be 24 bit then!
Old 30th June 2020
  #42
Gear Maniac
 

There's a thought.

Truncating to 24bit without dither is not such a good idea sonically.

The 32bit 'thing' in a DAW like SADiE is set up to give you headroom above the usual peak-bits as per 24bit, plus footroom below the 24th bit. You sort the levels and dithering out later when you come to make up the final 24bit or 16bit deliverable.

Do desks like Stagetech and Digico use 32bit signal-paths to give the extra headroom above normal peak-bits? If they do, then transmission down a 24bit MADI connection would hard clip any levels greater than the usual peak-bits. Yuk.

I shall ask one of my live mixing friends about this and report back.
Old 30th June 2020
  #43
Lives for gear
 

dunno about stagetec, lawo, calrec, digico, ssl etc. - my studer desk uses this:

Quote:
The Infinity Core DSP engine provides 12 A-Link fiber digital audio interfaces, which deliver more than 5,000 inputs and outputs. A fiber-based audio protocol, A-Link uses a 3GBit/s data rate to provide 1,536 audio channels per connection with 32-bit standard AES timeslots.
Old 1st July 2020
  #44
Lives for gear
Quote:
Originally Posted by TonyF View Post
...We need to look at avoiding unpleasant artifacts caused by truncation and calculation errors in lsb's during processing. With high-quality digital we aren't mostly usually talking 'noise' we are talking 'errors'.
But in sampling, quantization errors manifest as broadband noise. That is the only 'artifact' of the sampling process. There are no other 'errors' that manifest sonically. If there are, what would they be?

Quote:
Originally Posted by TonyF View Post
Do desks like Stagetech and Digico use 32bit signal-paths to give the extra headroom above normal peak-bits? If they do, then transmission down a 24bit MADI connection would hard clip any levels greater than the usual peak-bits. Yuk.
It's well established that basically all mix engines use higher internal bit depths. Whether they truncate or dither down 24 bit transmissions (MADI) is anyone's guess, but it could possibly be tested with simple test signals. IMO dithering 32>24 is good practice, but not something to worry about if it's not done. We're talking -144 dBFS noise, that is probably already masked by the analog electronic noise of the individual channels. A concert probably has 40-60 dB dynamic range, less than an LP. That's what I mean by real world.
Old 1st July 2020
  #45
DAH
Lives for gear
 
DAH's Avatar
Quote:
Originally Posted by brew View Post
But in sampling, quantization errors manifest as broadband noise. That is the only 'artifact' of the sampling process. There are no other 'errors' that manifest sonically. If there are, what would they be?


It's well established that basically all mix engines use higher internal bit depths. Whether they truncate or dither down 24 bit transmissions (MADI) is anyone's guess, but it could possibly be tested with simple test signals. IMO dithering 32>24 is good practice, but not something to worry about if it's not done. We're talking -144 dBFS noise, that is probably already masked by the analog electronic noise of the individual channels. A concert probably has 40-60 dB dynamic range, less than an LP. That's what I mean by real world.
32fp dithered vs truncated to 24integer can still be heard on real-world material, and yes, I have tested it myself in ABX.
Old 2nd July 2020
  #46
Gear Maniac
 
Mr P's Avatar
 

I use the stagetec nexus's a bit and can answer a few questions.

Nexus backplane is 24bit, not 32bit. Digital gain adjustments on the card can allow "access" to the last 2 bits of the input signal (the xmic+ works out to be around 26bits of dynamic range from memory). Not sure about signal paths inside their desks however. No idea if the output from mic card to backplane includes dither TBH.

I don't see the principal value of this design being for controlled recording applications but for safety and operational flexibility in dynamic live / broadcast environments. In a straight up recording application like being discussed here, I am not sure they would be compellingly better than other good options - though to be clear I haven't made any such comparisons and don't plan to. I think the mic inputs sound good, but don't use them for chamber / orchestral applications.

The value for us in the system is the operational flexibility of the router and control system, in conjunction with the wide array of available IO (we use mic, line, aes, madi, and dante).

As far as the proprietary fiber network that was mentioned, that is principally of value in larger installations for linking multiple base devices. For a smaller setups there is no need for it, just use your recording format of choice (madi, dante, aes67, etc).

Last edited by Mr P; 2nd July 2020 at 08:05 AM..
Old 2nd July 2020
  #47
Gear Maniac
 

Thanks everyone for the enlightening info.. I contacted one of my friends who does heavy-duty live mixing and he passed on similar observations. In most live setups the audio at the back end going to the speaker systems is either 24bit digital, or analogue from a dac connected to the 24bit audio. To dither or not is not a question asked often at all - there are so many other urgencies in live sound. Like crackles, bangs, hum, dropouts to be avoided at all costs.

I had thought the arguments over whether dither was necessary to optimise sound quality were long gone. But I guess the length of time explains that the arguments have been forgotten. Truncation makes nasty sound quality, and it is fixed with dither. Applying logic from the analogue world is unhelpful, unfortunately. We might have 144dB to play with in theory, but abuse of digital audio produces all sorts of side-effects less benign than a bit of added broadband noise. Aliases and spuria appear out of nowhere as a result of truncation and play tricks with the sound quality. We have a clock in the mix, which analogue audio does not.

The first generation of Sony PCM converters were undithered because quite reasonably at the time none of us knew much better. It was assumed the background noise in the input analogue signal would overwhelm or mask any problems. I plugged an analogue audio oscillator into an early PCM1610 and listened to the output via an external dac. At high levels, what came out sounded pretty similar to what I had put in. At low levels, sadly this was not the case. Sweeping the frequency dial on the oscillator delivered all sorts of extra birdies, especially when you went up to higher frequencies. Some swept down as I swept up. Dither fixed the problem at no serious penalty of extra audible noise - and that was in the world of 16bit audio, and we have been well beyond that since decades.

We make assumptions, and we can be wrong. When Decca was building its own digital audio gear there was discussion about whether their homebuilt digital mixer should include dither or not in the output. After all most analogue classical recordings have a noise-floor (venue, microphones, etc.) well above what a digital audio system can deliver. Tony Griffith told me he had done a blind listening test with a colleague from their accounts department (non-technical, non-musician, non-audiophile) switching dither in and out, and the accountant in his sixties could hear a difference and preferred the dithered audio every single time.

Forgetting about dither makes noise specs look better, but in the worst cases it can make the sound quality like breaking glass.
Old 3rd July 2020
  #48
I guess I consider dither solved science. My DAW does this automatically. I can adjust it, but its fine on defaults. Wouldn't any digital engineers worth employing take care of this automatically? Any live digital mixer runs a mix bus that has much more than 24 bits (often 40-90 bits, floating-point). They don't sound fragile, harsh, or any of that, so presumably the dither is all handled in the DSP/FPGAs. My Allen & Heath SQ-5 sounds indistinguishable from the DAW, that much I do know. Are we suggesting that high-end fully professional audio companies would just truncate data on their finest audio systems? It's possible, but seems very unlikely. Merging, Sonosax and the other vendors on this thread are held in the highest regard by international classical recordists. If there were obvious artifacts, their customer base would rebel in an instant.
Old 3rd July 2020
  #49
Lives for gear
 
Plush's Avatar
Yes, but the consoles you speak about don't all dither by default. they have to be set to do certain kinds of dither. Also one finds that different outputs on the console are set for different types / levels of dither. Often there is no global dither on the consoles.

Yes, setting the dither is called "engineering."
Old 4th July 2020
  #50
Dither is not configurable on my SQ-5 - admittedly many steps below a Stagetec, but representative for Allen & Heath. A search of the Digico SD-7's software reference manual turns up no instances of the word dither. The Yamaha CL reference guide has no mention. I opened up the SSL SOLSA application, and there's no dither settings. The Sonosax we both enjoy using has a 40-bit internal engine. But there is no dither setting in the UI.

I fully agree about it's importance as an engineering practice. It is certain that every digital mixer performs mathematical operations that extend the word length beyond 24 bits. I can understand why it might be a configurable option, but I don't know of live consoles where it is. I am not familiar with broadcast consoles (Lawo, Stagetec, Calrec, Studer, etc) so I'll defer to other's knowledge - it very well may work that way. But the reason I said that I considered it handled science is because it is not configurable in a big swath of professional live sound. It isn't configurable in plugins (which would also have long word lengths) - maybe they get a pass because they are passing results back to the mix engine running at a higher bit depth, and then the mix engine has to deal with it. But I kind of have to assume the engineers implemented this in an acceptable way. I have no knobs to change it or even to inspect it until the final export stage.

Of course, dither is configurable in the DAW for final export, and I set it appropriately, but I've never owned converters or stage boxes that had an option for this. This later makes sense - if the converters output 24bits, there is no need to dither. There is no change in wordlength till someone does math. But the 32bit recorders all write a 32-bit wave, don't they? As long as there's no word length expansion/reduction, there would be no need to dither.

I think no one thinks about dither because it is fully absorbed into live-mixing FPGA/DSP math so no one can forget to do it in the heat of the moment. I trust companies like Merging and Sonosax are doing whatever is necessary in their digital code to do the math right in lieu of giving me configuration options.

I do think we are drifting away from the practical concerns surrounding gain-ranging converters, unless someone has evidence that high-end gain-ranging products are being mis-engineered, and we users are being denied proper bitstreams. It seems more and more of the industry is experimenting with these designs (or maybe a patent expired somewhere), so I think there are a lot more ears checking out these designs. If there are problems, wouldn't there be complaints? Is this a concern end-users should have, or should digital device manufacturers be expected to do their sums properly before outputting a bitstream?
Old 5th July 2020
  #51
Gear Maniac
 
Mr P's Avatar
 

My sincere apologies, I've been passing off bunk info. I double checked the Nexus technical reference. The back plane bus is 30 bit it seems.
📝 Reply

Similar Threads

Thread / Thread Starter Replies / Views Last Post
replies: 200 views: 18191
Avatar for Patzerat
Patzerat 5th August 2005
replies: 106 views: 9133
Avatar for ramallo
ramallo 28th August 2019
replies: 2 views: 1098
Avatar for whippoorwill
whippoorwill 4th December 2018
replies: 1115 views: 115739
Avatar for MeBeMe
MeBeMe 4 hours ago
Post Reply

Welcome to the Gearslutz Pro Audio Community!

Registration benefits include:
  • The ability to reply to and create new discussions
  • Access to members-only giveaways & competitions
  • Interact with VIP industry experts in our guest Q&As
  • Access to members-only sub forum discussions
  • Access to members-only Chat Room
  • Get INSTANT ACCESS to the world's best private pro audio Classifieds for only USD $20/year
  • Promote your eBay auctions and Reverb.com listings for free
  • Remove this message!
You need an account to post a reply. Create a username and password below and an account will be created and your post entered.


 
 
Slide to join now Processing…
🖨️ Show Printable Version
✉️ Email this Page
🔍 Search thread
🎙️ View mentioned gear
Forum Jump
Forum Jump