A toast to PGGB, a heady brew of math and magic

[Miska]

29 minutes ago, Zaphod Beeblebrox said:

It is easy to design a noise shaper that behaves similar to what you have shown in your measurements, but I am not interested in doing so

 

Likewise it is easy to design a noise shaper that behaves similar to yours. Or any arbitrary amount of dB, as long as the numbers can fit in available RAM and you can spend unbounded amount of time doing the processing. No reason to restrict into 3-figure dB numbers when you can go for four or five figures, not a problem.

 

I'm interested in improving what comes out of DAC in real physical analog world. When chasing numbers becomes counter productive to that goal it is not my thing.

 

29 minutes ago, Zaphod Beeblebrox said:

But I do not feel the urge to jump into every software thread to argue the benefits or downsides of one approach over the other. I understand people have preferences.

 

I only comment when people talk/ask in relation to HQPlayer. Other than that, I don't really care.

 

As plain figures, as asked earlier, noise level (SNR) of 32-bit integer is -192.67 dB, and for 64-bit float it is -319.09 dB.

 

Quote

the intermediate samples can have a higher resolution

 

Of course, but physical world puts limits on that higher resolution as well.

 

[Zaphod Beeblebrox]

37 minutes ago, Miska said:

I'm interested in improving what comes out of DAC in real physical analog world. When chasing numbers becomes counter productive to that goal it is not my thing.

I respect your stance. I am interested in what comes out of the DAC too. It is not about chasing numbers for me either. For example, yes, it is possible to reduce the noise floor beyond where it is now or add more precision to processing. I have tried that, but as you say there is no point in throwing more resources if there are no audible benefits.  Not everything can be measured, measurements can take us only so far and serve as a guide.  So, I trust my ears more when there are audible benefits and stop pushing it when I do not hear a difference or if it sounds worse. Bias and preference are at play here too, so listening is crowd sourced to a group of listeners whose ears I trust.

[mpaulson540]

I have been away from internet access today - lots to digest. 

[Miska]

7 hours ago, Zaphod Beeblebrox said:

For example, yes, it is possible to reduce the noise floor beyond where it is now or add more precision to processing. I have tried that, but as you say there is no point in throwing more resources if there are no audible benefits.

 

Especially because pushing the noise floor down in one place keeps pushing it up elsewhere... As I have shown above. Which in turn will degrade the sound quality. As usual with anything in life, it's about balance.

 

[Zaphod Beeblebrox]

3 hours ago, Miska said:

 

Especially because pushing the noise floor down in one place keeps pushing it up elsewhere...

 

Yes, nothing we did not know already, and it can easily be seen/shown when analyzing the signals going into the DAC (no measurements necessary), as I show below.

 

I am using a 20 bit input signal at 705600Hz. Going from the noise shapers used with 64-bit precision (magenta) to 256-bit precision (orange trace) reduces the quantization noise in the audible range by 250dB. Yes, it also increases the noise way beyond the audible range (60000 Hz) because the noise pushed out from the audible range needs to go elsewhere. I have also drawn a red dashed line to indicate the measured analog noise floor for the Holo Audio Spring 3 DAC, and it is easy to anticipate what to expect in the measurement, which I am aware of.

 

 

 

 

3 hours ago, Miska said:

As I have shown above. Which in turn will degrade the sound quality. 

This is where I disagree. You equate increased noise way beyond the audible range to automatically translate to a decreased sound quality (or the possibility of increased thermal noise affecting sound quality).  But many here, including myself prefer the improved sound quality from higher precision despite the increased noise beyond the audible range. In my experience, decreasing the quantization noise in the audible range has only ever resulted in improved sound quality.

 

Processing with higher precision is moot if the noise shapers cannot retain the additional information, which is why I design noise shapers with much lower quantization noise in the audible range with increasing precision. 

 

P.S. The FFTs used to generate the plots also use 128bit or higher precision, else the noise floor will be limited by 64bit precision.

 

3 hours ago, Miska said:

As usual with anything in life, it's about balance.

I agree, and where that balance is, is in the eye of the beholder (or the ears of the beholder in this hobby). 

 

[Miska]

3 hours ago, Zaphod Beeblebrox said:

I am using a 20 bit input signal at 705600Hz. Going from the noise shapers used with 64-bit precision (magenta) to 256-bit precision (orange trace) reduces the quantization noise in the audible range by 250dB. Yes, it also increases the noise way beyond the audible range (60000 Hz) because the noise pushed out from the audible range needs to go elsewhere. I have also drawn a red dashed line to indicate the measured analog noise floor for the Holo Audio Spring 3 DAC, and it is easy to anticipate what to expect in the measurement, which I am aware of.

 

On purpose I have designed mine such way that the noise-shaper does the intended purpose of removing non-linearity distortions, while keeping the overall noise floor below the analog noise floor.

 

Yours is just swallowing highest harmonic frequencies of hires content into the increased noise...

 

[kennyb123]

I’m old enough to remember when objective performance was said to be this we should assess the quality of a filter.  Funny how things change once graphs start being produced showing spectacular objective performance. 
 

All I can say is that the proof is in the pudding.  It’s easy to hear improvements from PGGB’s 256 bit precision its 64 bit precision even on a modest setup like the one I use at work (Poly/Mojo 2/Aeon Noire with tracks played off an SD card).  

[Zaphod Beeblebrox]

3 hours ago, Miska said:

On purpose I have designed mine such way that the noise-shaper does the intended purpose of removing non-linearity distortions, while keeping the overall noise floor below the analog noise floor.

 

Yes that is obvious and that you do not believe in reducing the noise floor any further in the audible range. 

 

3 hours ago, Miska said:

Yours is just swallowing highest harmonic frequencies of hires content into the increased noise...

 

Again priorities, I am more concerned about what is in the audible range. There is not much content, let alone audible harmonic content beyond 40kHz, even with Hires, most of it is ADC noise. Also, since the reconstruction has already been done at this point, any information that was present in these harmonics, that are required to reconstruct the upsampled signal in the audible range has already been used. For those who want to retain more of the harmonic content (and who think they are being swallowed), there are the alternate and the extended noise shapers PGGB where the noise floor is extended well beyond the audible range. Though I do not prefer using them myself.

 

 

[taipan254]

Hi @LowOrbitand @mpaulson540.  What settings are you using in HQP to upsample to Dsd? Curious about the filter settings in particular. I know the half band filters preserve the original samples and I would assume preserving the pggb samples is of paramount importance but I thought I’d ask you before experimenting with my Cayin RU7 (discrete 1 bit dsd dongle dac). 

[mpaulson540]

2 hours ago, taipan254 said:

Hi @LowOrbitand @mpaulson540.  What settings are you using in HQP to upsample to Dsd? Curious about the filter settings in particular. I know the half band filters preserve the original samples and I would assume preserving the pggb samples is of paramount importance but I thought I’d ask you before experimenting with my Cayin RU7 (discrete 1 bit dsd dongle dac). 

 

I am currently using 1x: Poly-sinc-gauss long Nx1: Poly-sinc-gauss-hires-lp with Asdm7ecv3, DSD output is at 48k x 256.

HQPlayer processed on Mac Mini M1 to UP Gateway NAA/Denafrips Ares II

These are my normal HQPlayer filter settings and have not found anything better.

I try dabbling with other modulators but generally return to one of the Asdm7ec's.

 

I am still using PGGB in trial mode and have currently settled on 16fs 64-bit output with 64 bit precision. 

I have also switched from Roon to HQPlayer/Client for playback to facilitate use of 64-bit files. 

 

Primary tracks for evaluation have been "Sera Una Noche" and "Bruce Dunlop - About Home" (both at 44.1).

They are very neutral and simple mic'ed with natural room sound with good dynamics. At this point and to my ears, these recordings sound much better with pre-processing through PGGB and DSD upsampling/playback through HQPlayer.

 

I am trying to give myself time to make sure I am not being seduced by a new sound that may not prove to be better over repeated listenings. I have fallen prey to this impulsiveness before. 

 

Based on comments from the software developers, this methodology (PGGB->HQPlayer DSD) may be producing unwanted noise and distortion. Based on comments from some users on AS, this unwanted noise and distortion may actually sound pleasing. 

  

I'm not sure if measurements have actually been made with a PGGB pre-processed file upsampled in HQPlayer to DSD. If so, I would be curious to see the results. 

 

Hope this is helpful and good luck on your journey.    

[LowOrbit]

I use either Poly-sinc-gauss-xl or Closed Form Fast with PGGB files at 16 or 32fs/64bit. Volume control in HQPlayer is deactivated.

 

Typically I use DSD512 to feed my DAC200, but with 16fs input files I can get reliable replay at DSD1024 also. Not sure it really makes any in-band difference.

 

I no longer fuss about these settings too much. I get little enough time for music listening, as long as I don't get interrupted playback I am satisfied. 

[Zaphod Beeblebrox]

11 hours ago, mpaulson540 said:

I'm not sure if measurements have actually been made with a PGGB pre-processed file upsampled in HQPlayer to DSD. If so, I would be curious to see the results. 

These plots are done easily by anyone if you use RASA. Of course, these are not measurements done from the output of the DAC but instead plot the spectrum of upsampled DSD 256 signals and are good for comparison purposes.

 

I chose the track 13 Volver in the album Sera Una Noche. I used your settings (Poly-sinc-gauss long/Poly-sinc-gauss-hires-lp) with Asdm7ecv2 instead of Asdm7ecv3 (as I do not have access to Asdm7ecv3). But these comparisons are still valid since your interest is to see if there is any obvious increased out of band noise. I used 16fS 64bit output from PGGB done at 128 bit precision to match your settings.

 

I think the graphs are self-explanatory, the out of band noise for PGGB 16fs -> DSD 256 is right on top of the noise of 44.1k->DSD 256, i.e., I do not see any elevated noise and I am not surprised given the very low out of band noise from 16fS 64bit track (which I have also plotted at the end).

 

 

Since   PGGB 16fs -> DSD 256 is right on top of the noise of 44.1k->DSD 256, I removed the orange trace below to illustrate it better.

 

The main difference we see in the plots between the two conversions is near the end of audio band. This is expected because of the different filter PGGB uses (non apodizing), I have zoomed it in further.

 

 

Finally, here is the plot comparing the original track to 16fS 64bit PCM converted at 128bit precision.

 

 

These graphs only tell you half the story, let your ears decide what you like best.

[Miska]

25 minutes ago, Zaphod Beeblebrox said:

The main difference we see in the plots between the two conversions is near the end of audio band. This is expected because of the different filter PGGB uses (non apodizing), I have zoomed it in further.

 

Those frequencies near 22.05 kHz are just junk, mostly aliasing from the half-band ADC decimation filter... Makes it sound shrill. One could quickly mistake it for "detail" because it sounds overly bright, until at closer listen of the highs it makes the highs shimmery and distorted. Like sunlight coming through a scratchy car windshield.

 

This will also trigger ringing of extremely long filters.

[Zaphod Beeblebrox]

28 minutes ago, Miska said:

Like sunlight coming through a scratchy car windshield.

 

Yes, we know your preference of not using apodizing filters and you have beaten that horse to death, like you did a few times in this thread (quoted below for reference). Perhaps it is time for you to move such discussion to a different thread.

 

On 4/14/2023 at 9:33 AM, Miska said:

It is like looking through worn out car windshield with a lot of micro scratches that diffract light. With light coming from front it also shines brightly due to the diffraction. When it gets cleaned, the transients snap clearly and properly in focus, without "digitus". Although there are other sources for "digitus" as well.

 

I was responding to a simple question about possible increased out of band noise and also the differences in the plot without trying to impose my own subjective opinion.

[Miska]

25 minutes ago, Zaphod Beeblebrox said:

I was responding to a simple question about possible increased out of band noise and also the differences in the plot without trying to impose my own subjective opinion.

 

Your plot was with 64-bit output files, while earlier you recommended 32-bit output files. And I said it is better to use 64-bit files since they have lower noise. And I earlier posted results with 20-bit noise-shaped output to R2R section of the DAC.

 

OTOH, the modulator noise-shaper rising slope is designed to match the D/A conversion filter's falling slope. For example with Holo DACs the result is flat analog noise floor throughout at DSD512:

 

This is with peak-hold, no averaging.

 

Similar with T+A:

[Miska]

41 minutes ago, Zaphod Beeblebrox said:

Yes, we know your preference of not using apodizing filters and you have beaten that horse to death

, like you did a few times in this thread (quoted below for reference). Perhaps it is time for you to move such discussion to a different thread.

 

 

I was responding to a simple question about possible increased out of band noise and also the differences in the plot without trying to impose my own subjective opinion.

 

And I was referring to hard objective technical data and trying to describe what it possibly sounds like. Something that does exist in the source data. Yes, I work based on measurements from the ADC input devices, DAC outputs, and analysis of source content and recording equipment.

 

Of course I comment when HQPlayer is being referred to. Perhaps you stop talking about HQPlayer in this thread?

 

You are referring to subjective number games all the time combining subjective claims with plots. And next claim subjective differences in something that is not possible to exist in physical analog world due laws of physics, by several orders of magnitude.

 

[ray-dude]

Folks, as OP, I think we're into non-productive territory here.  There is a PGGB vs HQP thread if you want to compare and contrast the different design approaches and intent, but these have been covered multiple times here.  

 

There are differences in approach and design philosophies, and that's OK.  Both PGGB and HQP very generously offer free trials so folks can come to their own conclusions about what works best for their ears and preferences.  

[Zaphod Beeblebrox]

@mpaulson540 If space is a constraint, you could still opt for 16fS, 32bit output from PGGB. Below I have compared the results of converting 16fS 32 bits (orange) and 16fS 64 bits (yellow). Blue trace is the direct conversion from 44.1kHz to DSD 256. As you can see, there is no increase in the out of band noise between 32 bit and 64 bit PGGB files, they are still right on top of each other. Again, I am not surprised given the out of band noise is quite low to begin with.

 

Subjectively, I do not know if the 64bit file will sound better than the 32bit file on your setup, so I leave that to you. The other advantage you may have with going to 32bit files is you can now play them in Roon.

 

[seeteeyou]

Pardon me for not understanding why the same dude could keep getting the free pass, over and over again

 

https://audiophilestyle.com/forums/topic/14016-lumin-audiophile-network-music-player/page/3/#comment-194307  

On 1/5/2013 at 11:36 PM, The Computer Audiophile said:

Also, be very careful when talking about your product and competitors' products. A couple rules have worked well here on CA. 1. You can talk about your product when directly asked a question about it and 2. When in doubt don't comment.

 

Feel free to contact me if you aren't sure if you're breaking the rules.

 

Many manufacturers have learned the hard way that discussing their own products in a sketchy fashion backfires in a big way.

 

https://audiophilestyle.com/forums/topic/57498-lt3045-paralleling-ilim-circuit-pros-and-cons/#comment-1000325  

On 10/31/2019 at 3:15 AM, The Computer Audiophile said:

No, only because written rules, while they sound great, are always used by people to find loopholes and see how close they can come to the edge of what's allowed. Our rule is to not talk about your products unless specifically asked about them, don't talk about competitors or competitor's products, and if you aren't sure, just don't do it. Always ask if needed.

 

Plus, I guarantee you that 100% of the time when a member of the industry tries to break these rules with a sneaky sentence or two, it always backfires.

 

Actually it ain't even the first competitor's product IIRC, the same kinda world-class behavior already appeared on yet another thread

 

https://audiophilestyle.com/forums/topic/46501-chords-new-m-scaler/page/14/#comment-856355  

On 8/6/2018 at 4:18 AM, Miska said:

OK, so we cannot discuss any upsampling technicalities except complete isolation from each other?

 

And of course I spend my time and energy on my own product, and have been doing so for past 20 years.

 

https://audiophilestyle.com/forums/topic/46501-chords-new-m-scaler/page/14/#comment-856388  

On 8/6/2018 at 5:26 AM, austinpop said:

When you do it here, you call into question your motives, since you are in a thread devoted to a competitor's product. Whether you like it or not, you are synonymous with HQPlayer, the product. Do you really want to alienate future customers who may be turned off by what they perceive as you trashing a competitor's product on that product's thread?

 

Why is this so difficult to understand?

 

Straight from the horse's mouth, IMHO it would be really nice to spend someone's time and energy on someone's own product for the next 20 years as well.

[The Computer Audiophile]

Wow, I missed some stuff until just now :~)

 

Let's take a step back. Both @Zaphod Beeblebrox and @Miska have tons of info to offer and different products. As I read the recent discussion, I thought to myself, it's actually cool these two disagree and have different approaches. This gives us as music lovers two options. What more could we want. If both of them believed the same thing and created the same product, that would be less than fun. 

 

Anyway, let's keep this thread about PGGB. The other thread to discuss the HQP and PGGB exists. I know the world is full of shades of gray, so it's hard to completely separate discussions, but all of you guys are above average and I'm sure you can find your way without further moderation. 

 

When in doubt, ask.

[MemoryPlayer]

On 12/17/2023 at 9:27 PM, davide256 said:

Audirvana

I'm curious to know why upsample to 64Bits if the DAC accept only 32Bits?

I also own a Audirvana Origin license...

Did you compare the same file upsampled to 32 and 64 bits?

BTW is there any available DAC that supports 64Bits input?

[Zaphod Beeblebrox]

2 hours ago, MemoryPlayer said:

I'm curious to know why upsample to 64Bits if the DAC accept only 32Bits?

I also own a Audirvana Origin license...

Did you compare the same file upsampled to 32 and 64 bits?

BTW is there any available DAC that supports 64Bits input?

I did not see any mention of 64bits, only 24bits, where did you get that from?

[MemoryPlayer]

37 minutes ago, Zaphod Beeblebrox said:

I did not see any mention of 64bits, only 24bits, where did you get that from?

Your newest PGGB version doesn't  upsample to 64 bits?

Audirvana can play 64/768  wav, but output to DAC is 32/768!

[Zaphod Beeblebrox]

15 minutes ago, MemoryPlayer said:

Your newest PGGB version doesn't  upsample to 64 bits?

Audirvana can play 64/768  wav, but output to DAC is 32/768!

Yes, that is not meant for direct play back to DAC but to enable further processing by other players (like DSP), this way the upsampled files can have a higher resolution and allow the player to process at 64 bits.

[MemoryPlayer]

2 minutes ago, Zaphod Beeblebrox said:

Yes, that is not meant for direct play back to DAC but to enable further processing by other players (like DSP), this way the upsampled files can have a higher resolution and allow the player to process at 64 bits.

So, playing 64/768 sounds the same as 32/768?