The No.1 Website for Pro Audio
 Search This Thread  Search This Forum  Search Reviews  Search Gear Database  Search Gear for sale  Search Gearslutz Go Advanced
Float > 24 outboard & back - what's necessary? Z-Systems EQ
Old 4 weeks ago
  #1
Gear Nut
Float > 24 outboard & back - what's necessary? Z-Systems EQ

I've spent the past couple weeks looking for a definitive answer to a question about bit-depth variation / need for dither when going from my DAW to an external AES/EBU device - I've talked to a lot of people, I've read endless articles, but, in the end, I'm now more confused than ever - can someone set the record straight!?

Here's the scoop: I recently won a Z-Systems EQ on eBay, I use REAPER for mastering and prior to this new-to-me EQ all outboard I've used has been analog, the EQ is strictly digital (AES/EBU). As I understand it AES/EBU is a 24-bit I/O, within REAPER I work at 32-bit or even 64-bit float, using their ReaInsert plugin I can set up a processing loop at any point in a project - this is my typical method of going OTB, then back. My feeling was that I would have to dither to 24 just prior to the send - some info I've come across / folks I've talked to said the same, but I've also found info suggesting it isn't necessary, suggesting 32-bit float only differs from 24 fixed in +8 exponent bits and that these do not factor into the auditory portion of the stream and would be dropped inherently and these dropped bits would be of no detriment so dither would not be necessary. I was told, as well, that the same would go for 64-bit but on further reading I've found that 64-bit float is not simply a 24-bit with exponent but is 53 significand bits + 11 exponent and, it would seem to me, in that case, dither would be necessary. On top of all of this, though the I/O is 24-bit, the internal processing is 40 float on the EQ. All of this has confused me in a terrible way and I'm tired of thinking about it. Can anyone set the record straight here so I can get onto other things in life

Thanks
Old 4 weeks ago
  #2
Lives for gear
 
SmoothTone's Avatar
 

It's funny, but dithering at 24bit wordlength remains quite a contentious issue. I don't know why. Here is a summary of the facts as I understand them (copied from a previous post):

Quote:
Opinions vary but the facts don't:

Truncation occurs whenever wordlength is reduced.

Truncation creates distortion - it sounds nasty.

Dither prevents the distortion (it doesn't just mask it).

At 24bit, truncation distortion is inaudible. So is dither noise. This is where opinions differ: if you can't hear it, whether you dither or not makes no audible difference...

AS LONG AS NO FURTHER PROCESSING IS APPLIED!

Distortion accumulates with multiple truncations.

Cumulative truncation distortion has a higher amplitude than cumulative dither.

Dither completely removes truncation distortion and results in a nice even noise floor. It also preserves a little bit of dynamic range that gets lost in truncation.

In short: dither leaves the audio in better shape for further processing and has a less detrimental cumulative effect than truncation distortion. (There's always further processing in the digital delivery age).
I would dither.
Old 4 weeks ago
  #3
Gear Nut
OK, so . . . detrimental truncation DOES occur on a 32-bit float signal upon passing a 24-bit hardware input? By 'detrimental' I mean "impacting auditory aspects of the signal". I had several discussions where it was suggested this would not be the case - that the 24-bits that constitute the auditory portion of a 32-bit float signal are identical and would pass through unaltered as the 8 bits that remain are exponent bits which are related only to dynamic range, and only these would be dropped, not by purposeful truncation but by a blindness to their existence at the 24-bit input - the only complication of this transition being decreased dynamic range BUT if the internal processing of the EQ is 32+ bits float (as it is, in this case) it wouldn't matter in the least as the exponential bits would be recovered internally within the EQ then dropped again at the hardware output only to return the instant the signal was returned to the DAW, culminating in 0 alteration to the signal from start to finish and therefore no need for dither. All these discussions tended to come with an 'as I understand it' proclamation -that always makes me a bit uneasy.
Old 4 weeks ago
  #4
Quote:
it wouldn't matter in the least as the exponential bits would be recovered internally within the EQ then dropped again at the hardware output
Fixed point and floating point representation differ wildly. You can't turn a 32bit float into a 24bit fixed by shaving off the most significant 8bits! You don't have to trust me either, just enter any number here: https://www.h-schmidt.net/FloatConverter/IEEE754.html and turn off the exponent bits. What happens to your number?


About your main question, I'd say it doesn't matter. Dither or truncation distortion, the level is incredibly low, likely far lower than the signal's natural noise floor - which would simply make it self dither. And no, truncation distortion does not accumulate that much. It does more like noise, because each truncation is uncorrelated to the previous.

I built an experiment. It's a simple very low level sine wave (-58dB fs), to make sure it "provokes" trouble when being truncated.



Let's truncate it to 16bit (more obvious than 24bit), without any dither:



And now a dithered 16bit truncation:




So far so good. No surprises, the dither removes any spectral patterns and even (minutely) reduce the maximum level of these partials.


Now comes a surprise. this time, we do 3 truncations in a row, the first to 24bit, the next to 20bit, and the final one to 16bit:



Please compare to the above! There's almost no difference.

And now the same three truncations, but with the last one (to 16bit) dithered:



No relevant difference to the one time truncation to 16bit.

I guess you got it, truncation DOES NOT accumulate dangerously. In fact, at extremes, the opposite happens! The more numerous the truncations, the more likely the chance that they will self dither (in sum, as a system, with many many many little truncations. Only reason we don't see this beneficial effect here is because we're using a pure sine).



Attached Thumbnails
Float > 24 outboard & back - what's necessary? Z-Systems EQ-clean.jpg   Float > 24 outboard & back - what's necessary? Z-Systems EQ-16bit_truncation.jpg   Float > 24 outboard & back - what's necessary? Z-Systems EQ-16bit_truncation_dithered.jpg   Float > 24 outboard & back - what's necessary? Z-Systems EQ-24bit_20bit_16bit_serial_truncation.jpg   Float > 24 outboard & back - what's necessary? Z-Systems EQ-24bit_20bit_16bit_serial_truncation_dithered.jpg  


Last edited by FabienTDR; 4 weeks ago at 01:35 PM..
Old 4 weeks ago
  #5
Gear Nut
Quote:
Originally Posted by FabienTDR View Post
You can't turn a 32bit float into a 24bit fixed by shaving off the most significant 8bits! You don't have to trust me either, just enter any number here: https://www.h-schmidt.net/FloatConverter/IEEE754.html and turn off the exponent bits. What happens to your number?
Thanks for the detailed explanation Fabien! I AM glad for this morning's edit as last night I'm almost positive 'can't', in the paragraph above, read 'can' and I was ready to tell you that there was a wild difference between the number after nixing the significant bits - much like the difference when you add a 't' to 'can'.

I recreated your experiment myself - I like hands-on demonstrations so that really helps - and yes, the difference between truncation and dithering is indeed negligible.

This brings up the obvious other question then . . . why bother dithering to 24-bit at all? Why not simply render a 24-bit PCM and have done with it?
Old 4 weeks ago
  #6
Quote:
Originally Posted by brijem View Post
This brings up the obvious other question then . . . why bother dithering to 24-bit at all? Why not simply render a 24-bit PCM and have done with it?
Comes down to personal preference. Not dithering 24bit is certainly no blasphemy, it's just somewhat irrelevant compared to the other "problems" an audio engineer has to mess with. It's also somewhat irrelevant to following truncations (as these happen at far higher levels than the previous).
Old 4 weeks ago
  #7
DAH
Lives for gear
 
DAH's Avatar
Reverb tails and space differ with dither to 24 bit on vs off in my world.
Old 4 weeks ago
  #8
Lives for gear
Your reverb tails are down at -144dbfs?
Old 4 weeks ago
  #9
DAH
Lives for gear
 
DAH's Avatar
Quote:
Originally Posted by scraggs View Post
Your reverb tails are down at -144dbfs?
Obviously, not and dither affects signal higher than LSB
Old 4 weeks ago
  #10
Lives for gear
 
loji's Avatar
IMO, this 100% falls under the umbrella of 'cumulative effect' and will be based on taste per-song.

I have a ULN8 with 8x channels of AES (24-bit) splitting out to digital hardware Junger Accent, Zsys EQ, and L2 ... all 24-bit I/O


I have a dither block after each AES return, as well as on the path to the DA. (32-bit float from reaper, up to 80-bit (long double fixed) within the MIO hardware, down to the physical 24-bit chip in the DAC)

the dither on the DAC is almost always on. the ones after each AES insert ... it depends.

sometimes the very slight 'edge' of truncation at that stage gives a beneficial bite, sometimes the extra 'smoothness' of adding dither is better.

part of the chain is bypassing each dither block, or combination. It makes a very very subtle difference. If I forgot to bypass it a track would never be ruined. But by utilizing dither in a very deliberate (and technically correct) manner, it's one more piece of the chain to reach that last 1.5%

so yes, adding or not adding dither will change the sound. technically it should be added. . experience has proven to me that it's not always best to add it, although sometimes can is.
Old 4 weeks ago
  #11
DAH
Lives for gear
 
DAH's Avatar
I am talking about Daws last insert before the dac. I have tried several combinations of settings as well as dither off ooking for the best space and timbral accuracy and body/definition on a classical record confirmed later by checking PF's Hey You and my zoom recorder samples and a Vst reverb on a snare like synth sound.
Old 4 weeks ago
  #12
Lives for gear
 
Jerry Tubb's Avatar
 

Verified Member
I do love my little Z-Sys EQ, the 96k version.

although the knobs are a bit ‘fiddly’ (to quote Jack the Bear)

but OT, i often convert 32-bit to 24-bit without dither...

sounds fine, too many layers of dither can have a cumulative effect,

as can truncation distortion i suppose.

so it depends...

~ in all things audio, let the ear be the judge ~ Lord Railey

best, JT

Last edited by Jerry Tubb; 3 weeks ago at 02:29 PM..
Old 3 weeks ago
  #13
Lives for gear
 
loji's Avatar
Quote:
Originally Posted by Jerry Tubb View Post
I do love my little Z-Sys EQ, the 96k version.

although the knobs are a bit ‘fiddly’ (to quote Jack the Bear)
me too, great sounding EQ... with a HUGE bandwidth range. Mine is the silver faceplate version .... haven't had an issue with the encoders yet . . . but I do know that eventually happens.

There's also a common problem with the PSU transformer. it gets bad solder joints where it joins the PCB, that happened to mine. Easy enough to just open it up and reflow the solder on the power transformer. You can sometimes gets these EQ's for peanuts if it has a power problem .. often that will fix it.


Do you know what type of dither in in the Zsys EQ? ... I usually just have it set to regular 24-bit, not 24dith. I prefer trianglular dither, and I seem to remember they used a noise-shaping one ?
Old 3 weeks ago
  #14
Lives for gear
 
loji's Avatar
Quote:
Originally Posted by brijem View Post
within REAPER I work at 32-bit or even 64-bit float, using their ReaInsert plugin I can set up a processing loop at any point in a project - this is my typical method of going OTB, then back.....On top of all of this, though the I/O is 24-bit, the internal processing is 40 float on the EQ. All of this has confused me in a terrible way
First off, congrats on a great EQ.

Let's look at your chain :

We'll keep it super simple, In this example we're just important a 24-bit Source Mix a client sent, doing 1x volume change in Reaper, sending to the ZSYS for EQ with a ReaInsert, then hitting 1x final limiter = exported master file. Pretty simple master: volume, EQ, limit. done.

So we've got :

Source file (24-bit) > Reaper processing (let's assume you made a volume adjustment or something) (64-float) > ReaInsertI/O > ??

this is where you need to look at your specs. Are you using a firewire sound card? a USB one? etc ... whatever the connection to your AES interface, this will be either 24-bit or 32-float from the driver audio layer, into whatever interface (if any) your sound card has, down to the physical chip of the AES device. (this is 100% 24-bit fixed)

So in my case, my FW connection from reaper to the soundcard is 32-bit float. . . look at yours

so at the output of the ReaInsert, you're going from 64 float to 32 float. (double precision to single precision).

reaper is likely rounding this, not truncating.

from the 32-float going in your FW/USB cable it's hitting the AES chipset, which is almost 100% truncating to 24-bit. (not necessarily a bad thing) In my case, I have the Metric Halo hardware, which allows me to place a dither block there if I choose .... you may or may not have this ability in your hardware. Arguably doesn't really matter.

So now we're at the physical chip of the AES output ... you hit a cable. (110ohm digital cable on XLR) I prefer Apogee wyde-eye ... if you're using adapters/transformers/etc.. here, that's another layer we'll just ignore right now since it probably doesn't apply, but you should be using XLR not adapters.

from the cable you're hitting the input chip of the ZSYS EQ. This hits a PPL, and recovers the clock signal (minimizing jitter) assuming your clocked from this AES signal, not wordclock or something ... you make an EQ change here , and now your at 40-bit FIXED point within the SHARC processor in the ZSYS EQ.

that SHARC processor hands off the 40-bit fixed word to the AES 24-bit output chip. your ZSYS has settings here to choose what you want to do with that hand-off. (24-dith/24-trunc/20-dith/20-trunc/16-dith/16-trunc) ... leave it on 24-bit dither or not, whichever you think sounds best. I generally prefer not.

Now you're back at the 24-fixed AES input to your soundcard, which sends the 24-fixed word to your ReaInsert I/O , and it continues on it's journey .. (which is likely another 'something' ITB processing, like a limiter... which will pop it back up to the internal depth of 64-float within Reaper's engine, or to the max depth of 32-float of the VST you're using .. depends on what tools you've chosen)

SOoo ... the whole chain when you pop in the ReaInsert is :

24-fixed (Source) >
64-float (Reaper) >
32-float (USB) *>
24-fixed (AES) >
40-fixed (Zsys) *>
24-fixed (AES) >
24-fixed (USB) >
24-fixed (reaInsert return) >
64-float/32-float (Reaper ITB process/3rd party VST) >
*Final Dither to Master Depth (ex: 16-bit for CD)

the spots I marked with an asterix are the technically correct places where dither could be applied. Remember the drop in bit-depth from double to single precision is being rounded, so dither is not needed there. The only one that is required is the final dither to your destination depth. In this example, the RedBook CD requirement of 16-bit.

it looks much more complicated than it really is. . . but it's just a basic signal flow. Look at it in your system until you understand exactly what's going on. a tool like Bitter will help you visualize and test.



... Man, that was a lot of words to say, flick dither on and if you don't like it, flick it off.
Old 3 weeks ago
  #15
Gear Nut
Wow! Thanks loji - I'll be back to this in detail tomorrow - workin' today / tonight but tomorrow is free. THIS is just what I needed to know. This Z-Sys is my 1st all-digital outboard device so a whole new batch of questions and considerations arose. I have heard all sorts of great things about this EQ (and can add you and Jerry Tubb to that list now) - I got it at quite a steal - $230, shipped. I read about the fidgety knobs - I've turned it on to check to see if it has that known issue & yes, it does. Unfortunately, I haven't had a chance to HEAR it in action yet - my interface is out for repair at Lynx - I have a Hilo USB, and my stand-in is a DAC only, no inputs, so have to wait before I can actually use it! Will look over the signal path soon - such a big help. Thanks again
Old 3 weeks ago
  #16
Lives for gear
 
loji's Avatar
Quote:
Originally Posted by brijem View Post
- such a big help. Thanks again
Great! I was worried that was too convoluted to be any assistance. I'm glad it was helpful to you.

... and I'm sure you know this, but never hurts to be extra clear. Only use ReaInsert in serial not parallel.

All digital hardware has at least 2-sampels of delay (1 sample buffer to the input/output AES chips) most have more , plus any delay your interface has
Old 3 weeks ago
  #17
Lives for gear
 
Jerry Tubb's Avatar
 

Verified Member
Quote:
Originally Posted by loji View Post
me too, great sounding EQ... with a HUGE bandwidth range. Mine is the silver faceplate version .... haven't had an issue with the encoders yet . . . but I do know that eventually happens.

There's also a common problem with the PSU transformer. it gets bad solder joints where it joins the PCB, that happened to mine. Easy enough to just open it up and reflow the solder on the power transformer. You can sometimes gets these EQ's for peanuts if it has a power problem .. often that will fix it.


Do you know what type of dither in in the Zsys EQ? ... I usually just have it set to regular 24-bit, not 24dith. I prefer trianglular dither, and I seem to remember they used a noise-shaping one ?
yes, 24-bit, no dither is my favorite as well.

always like the “ sound “ of it better than any ITB plug-ins.

it does a nice job of Mid-Side EQ’ing as well.

but it takes more time & effort to operate.

so Z-sys may have become another casualty of the ITB trend, dunno.

best, jt
Old 3 weeks ago
  #18
Gear Nut
Quote:
Originally Posted by loji View Post
. . . sometimes the very slight 'edge' of truncation at that stage gives a beneficial bite, sometimes the extra 'smoothness' of adding dither is better.
This sums things up for me - I DO like to know the technical reasons, which have been covered here quite well and I feel much more aware of what's happening along the route (thanks!!), but the final judgement is going to be auditory and artistic, especially at this minute level of 'harm' - or possibly even, enhancement!

Thanks to everyone for so graciously chiming in and taking the time!
Old 2 weeks ago
  #19
Gear Nut
I finally got my Hilo back so was able to connect the Z-Sys today and hell, yes - this is one beauty of an EQ! The knobs are jumpy as others have described and this does seriously hinder the smoothness of working with it (wonder if there's any sort of fix?) but it's not a deal-breaker, sounding as it does - I've only had it up & running for an hour and I'm positively thrilled.
Topic:
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…
Thread Tools
Search this Thread
Search this Thread:

Advanced Search
Similar Threads
Thread
Thread Starter / Forum
Replies
Red Dog / So Much Gear, So Little Time
1
Stevedresser83 / Low End Theory
6

Forum Jump
Forum Jump