32 bit input


#1

Hi @tedsmith

Is the DS’s Bridge II a 32 bit input and all other DS inputs are 24 bit?

Cheers


#2

Obviously not Ted, but in my experience, no. I discovered this while trying to convert my DSD 128 files via Roon to an acceptable PCM equivalent because I use BII, which doesn’t work nicely with DSD 128. So Roon tried to convert to 32/192 resulting in digital hash. Once I restricted conversion to 24 bit in the settings, it went away.


#3

As far as the FPGA is concerned there are two kinds of inputs - I2S or S/PDIF like.

S/PDIF, AES/EBU and TOSLink only support 16, 20 or 24 bits in the raw data but almost any driver will properly accept 32 bit data and properly align it in the 24 bit field so there’s no real problem in practice.

I2S is designed to accept a wide variety of bit widths and hence the FPGA will accept any bit width at all and things are “left” aligned and zero padded on the right, i.e. the first bit in every sample is the sign, he 2nd bit represents 1/2 of full scale, the 3rd 1/4 of full scale, etc…

So in essence the FPGA only sees 24 bit data. It then reports how many bits out of those 24 bits are changing sample to sample, e.g. 16, 18, 20 … or 24.

The Bridge accepts files of data that may be formatted many ways and various renderers also repackage many different sources: I honestly have no idea if there are any common files that can’t be accepted based on a weird sample width, but I doubt that that’s a problem.

The USB input accepts the standard USB Audio packets, but also declares to the source which sample rates and formats it accepts. The MAC OS’s USB drivers seem to be a little more sensitive exact combinations than the PC USB driver that PS Audio supplies. (Tho, on the PC, ASIO accepts all bit widths, WASAPI gives an error for 32 bits, etc.)

I guess what I’m saying is that the further upstream the 32/24 bits is handled the more variability.


#4

Thanks Ted.

Sometimes, when I’m not doing critical listening and I don’t want to adjust my amps volume, I turn on my software’s (Roon’s) auto volume leveling - I know know, SNR gets reduced with digital volume control, but I use it for non-critical listening.

So in Roon (though not specific to Roon but I could be wrong in generalizing so please correct me) if I apply -18db volume leveling, then I’ve read that I could lose 3 bits.

So for my 24 bit music, if the Bridge II accepts 24 bits (like the other inputs) and I am playing 24 bit music and I apply -18db volume levelling, then I will lose 3 bits of music info - is that correct?

The reason I ask is because the web page for the Bridge II says “192kHz 32 bit compatible”

http://www.psaudio.com/products/perfectwave-bridge/#tab-features

So back to the DS, I thought that if the Bridge II accepts 32 bits (as per the website) then with Roon able to spit out 32 bits for all it’s PCM DSP, if I used -18db volume leveling then the 3 bits of loss would only be the zero padded stuff - no music bits lost.

I may not have fully grasped your reply above, in case you already answered this, so apologies if I didn’t.

So even though this is mainly for non critical listening, I just wanted to ask if all the music bits would still be there, when playing 24 bit music with -18db volume leveling (for example), assuming Roon can spit out 32 bits with the padding zeros

Many thanks in advance


#5

Since 24 bits (144dB) accommodates a dynamic range from a single molecule hitting the ear drum to standing in a jet engine exhaust there’s no reasonable need for more than 24 bits as the final delivery method. If you are doing filtering, or other DSP then using more than 24 bits makes sense, but there’s really nothing to be gained by using more than 24 bits after the last DSP before the DAC.


#6
Ted Smith said Since 24 bits (144dB) accommodates a dynamic range from a single molecule hitting the ear drum to standing in a jet engine exhaust . . .
This should cover the dynamic range of any Shelby Lynne recording.

#7
Ted Smith said Since 24 bits (144dB) accommodates a dynamic range from a single molecule hitting the ear drum to standing in a jet engine exhaust
Ha but everything matters Ted!! I joke. Thanks for the info. Always nice to have some big picture context.

#8

I do not know how Roon implements volume control. At least previously (a couple of years ago), Roon had no internal volume control but relied on the driver/DAC to which Roon simply sent signals (depending on compatibility). Thus, the volume control was as good as that contained in the driver you were using.

Of course, a good volume control is transparent; it does not throw away bits. But even if the digital control you are using throws away 3 bits from 24, you are not going to hear it. The thermal noise of your system alone obscures the last three or four bits in any event.

Keep in mind a really good classical vinyl record possesses around 60dB - 65dB dynamic range, a really superb 45 RPM specially pressing perhaps 75dB at the absolute best. Few are unhappy with the dynamic range of a good record.


#9
Elk said Of course, a good volume control is transparent; it does not throw away bits.
I'm not sure this is correct. Applying -18db digital volume leveling in a Software player, to a 24 bit music file to a 24 bit input DAC, there will be bits lost - there's no such thing as a free lunch as they say. Of course as Ted hinted earlier, we probably won't hear the difference, especially with 24 bit music.

#10

It depends on how the software player performs volume changes.

And there are DACs, such as the DirectStream with a lossless digital volume control.

Mi2016 said Of course as Ted hinted earlier, we probably won't hear the difference, especially with 24 bit music.
And as I stated explicitly. :)

No system possesses 126dB dynamic range. Thermal noise alone guarantees this, let alone the other sources of noise in a system.

Not to mention you will never find a recording with a 126dB dynamic range. In the real world of recording, even in a studio environment, it is a challenge to capture over 60dB dynamic range. My recordings of symphony orchestras, chamber orchestras, etc. in concert rarely exceed a dynamic range of 60 dB, easily captured by 16 bits (I record either DSD or 24 bit). And as I mentioned above, a really good classical vinyl record possesses around 60dB – 65dB dynamic range.

As a practical matter, the “loss” of four to six bits of a 24-bit file is much more a theoretical concern than practical.


#11
Elk said

It depends on how the software player performs volume changes. And there are DACs, such as the DirectStream with a lossless digital volume control.

Sincere apologies, there seems to be a misunderstanding. I was asking Ted about software playback digital volume control, not the DS's internal digital volume control. I'm sure Ted will correct me but it doesn't matter what kind of magic the software says it can perform. My understanding is that a software player playing a 24 bit music file to a 24 bit input DAC with -18 dB volume levelling , there will be 3 bits lost. The rule of no such thing as a free lunch applies regardless of the software player, for this specific example.

#12

Digital volume controls got a bad reputation in the past (deservedly so at the time) for at least three reasons:

  1. There were digital volume controls that did 16 bit math, i.e. if you lowered he volume by 18 dB you’d be down to 13 bits S/N.

  2. There were digital volume controls that did math with appropriate precision to keep a full 16 bits of accuracy and used bad dither to go back to 16 bits. The signature of some bad dither is audible, especially if one later turned the volume back up, either digitally or with an analog preamp, etc. downstream. Note that good dither can (statistically) keep the full 16 bits of resolution even with some loss of bits at the expense of adding some random noise that often isn’t noticed or if noticed doesn’t detract from the system.

The last reason is a little harder to explain.

Digital audio isn’t done with integers, instead it’s done with pure fractions: if there are 8 bits they represent the values from -128/128 to 127/128 if there are 12 bits they represent the values from -2048/2048 to 2047/2048. 16 bits -> -32768/32768 to 32767/32768. Notice the slight asymmetry, there’s one more negative value than positive value representable. The funny thing is that value is always -1. A perfect -1.

Now if you naïvely use, say 16 bit signed math to represent the volume for a 16 bit volume control or 24 bit signed math to do a 24 bit volume control you can’t represent +1 - i.e. you can’t represent no volume change! you’ll always change bits even at full volume. These changes can be noticeable with if implemented with bad dither in 16 bits even when you are playing at full volume. The trick is to use negative signed numbers to represent the volume so full volume is represented by -1 which is representable and making sure that there are an even number of digital gain stages in your system (often with DSP there are more than one multiplication in a digital algorithm.) Or if the system is balanced it can do a final inversion by swapping two wires to match an extra digital gain using the -1 trick.

Good DSP could also get around the problem by having at least one more bit on the top end which allows the more straight forward representation of 1 for unity volume.

These days there’s no excuse for a digital volume control that messes with the bits at unity volume or for limiting the output to 16 bits and needing dither to keep the extra resolution (with some extra noise.)

Personally for my own private listening I have no compunctions about using a modern digital volume control that outputs 24 bits or more even with 16 bit inputs.

Still for the purists among us (myself included) the DS keeps all of the bits of the result of the multiply of the input and the volume to the very end of processing: e.g. with 30 bit inputs and a 20 bit volume the DS does all math down stream at at least 50 bits…


#13
Ted Smith said Personally for my own private listening I have no compunctions about using a modern digital volume control that outputs 24 bits or more even with 16 bit inputs.
Wow Ted, do you enjoy displaying how little the rest of us (myself first and foremost) know!? laugh

I joke of course - all said in good humor. I asked for the facts and certainly got it. This was awesome reading and learning for me.

I’m the same in this regard - I also have no issues with using modern digital volume control.

Personally I would say ~70% of my music listening is with Roon’s automatic volume leveling enabled (even with 24 bit music) and the advantages far outweigh any disadvantages that you mentioned above.

But for the ~30% of the time where I want to go full ‘Everything Matters Mode’, then I certainly am not questioning the DS’s internal volume - I just make sure Roon is set to bit perfect mode and let Huron wash over me.

I know the Bridge II is not your product, but since the FPGA sees 24 bits, then I wonder if the Bridge II’s website should be revised where it currently says “192kHz 32 bit compatible” because the 32 bits doesn’t really help? @dennis

Thanks again for the knowledge sharing.


#14

The documentation on the Bridge should (IMO) talk about the formats that it accepts not about the format it puts out. The problem is that that would require a big list of bits/sample rates for each file it understands directly (e.g. flac, wav…) and yet another for UPnP (perhaps more than one depending on options) and (perhaps I don’t understand these correctly) for what ever protocols it supports for over the net playing… I’m all for accurate specs but this might be a case of overload and further it may also depend on the Bridge software release :slight_smile:


#15
Ted Smith said playing... I'm all for accurate specs but this might be a case of overload and further it may also depend on the Bridge software release :)
Not a problem. Makes perfect sense.

#16
Mi2016 said
Ted Smith said Personally for my own private listening I have no compunctions about using a modern digital volume control that outputs 24 bits or more even with 16 bit inputs.

Wow Ted, do you enjoy displaying how little the rest of us (myself first and foremost) know!? laugh

I joke of course - all said in good humor. I asked for the facts and certainly got it. This was awesome reading and learning for me.

I’m the same in this regard - I also have no issues with using modern digital volume control.

Personally I would say ~70% of my music listening is with Roon’s automatic volume leveling enabled (even with 24 bit music) and the advantages far outweigh any disadvantages that you mentioned above.

But for the ~30% of the time where I want to go full ‘Everything Matters Mode’, then I certainly am not questioning the DS’s internal volume - I just make sure Roon is set to bit perfect mode and let Huron wash over me.

I know the Bridge II is not your product, but since the FPGA sees 24 bits, then I wonder if the Bridge II’s website should be revised where it currently says “192kHz 32 bit compatible” because the 32 bits doesn’t really help? @dennis

Thanks again for the knowledge sharing.


The bridge is only 24 bit. I’ll pass that along to the people who can change the web site.

Dennis


#17
Mi2016 said
Elk said

It depends on how the software player performs volume changes.

And there are DACs, such as the DirectStream with a lossless digital volume control.

Sincere apologies, there seems to be a misunderstanding. I was asking Ted about software playback digital volume control, not the DS's internal digital volume control.
I was addressing exactly what you asked. You just do not like the responses. laugh

I also happened to mention the DirectStream as an example of a lossless “playback digital volume control.” Ted kindly explained the software aspect.

As I already noted, Roon, at least in the past, had no internal volume control but relied on the driver/DAC to which Roon simply sent volume change requests. Thus, the volume control was as good as that contained in the driver/DAC you were using.

Accordingly, you may want to set up Roon where it tells the DirectStream to change volume, thereby taking advantage of the DS’s lossless digital volume control. This would render your concern even more moot.


#18
Elk said you may want to set up Roon where it tells the DirectStream to change volume, thereby taking advantage of the DS's lossless digital volume control.
Sincere apologies again but again this is not correct. I was asking Ted about the convenience automatic volume leveling where you don't need to touch any volume control for the DS, software or amp. Not asking about the DS's internal volume control.

I hope Ted’s post helps all those using software playback, to learn more, especially if using the convenience of auto volume leveling with any player that supports it.

This was fascinating reading and learning for all of us.


#19

Yes, I understand what you are asking. You want to rely on Roon to automatically volume level, akin to replay gain. Roon can do so on its own by changing its own volume input. We know this.

BUT If Roon can do so by telling the DS to change the DS’ volume output you both get what you want and alleviate your concern that you are somehow losing sound quality through digital volume attenuation. In the past, Roon was able to control other devices.

Roon may not be able to do this with the DS, but it is worth checking.


#20
Elk said Roon may not be able to do this with the DS, but it is worth checking.
Roon can access the DS's internal volume control, that's not the issue at all. Even my iPhone running mConnect and Tidal can do that. But Roon can't do this in tandem with it's automatic volume leveling feature.

Does anyone know, for certain, of any player that can? I have a feeling in my waters that foobar may but can’t recall for certain.