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

[Miska]

8 hours ago, Zaphod Beeblebrox said:

If you have set dither = None on HQP Embedded, then all this is doing is zero padding and what you are hearing is the original 16bit output from PGGB and you likely prefer 16bits to 24bits.

 

You have to be pretty careful with that, if the output doesn't happen to be identical with input you have a nice distortion generator. There are lot of settings that could cause that not be the case.

 

[ASRMichael]

On 3/25/2022 at 11:24 PM, Zaphod Beeblebrox said:

When you go from 44.1 => 352.8, what setting do you use in PGGB for dither/noise shaping? Have you tried both Noise shaping and Dither only options? 

 

When going from 44.1k -> 352.8k, the 16bit input is converted to 64bit first and all the processing is done, then converted back to 24 bit or the output bit depth you desire (16 bit in your case). To reduce and/or decorrelate noise during this conversion from 64bits back to 24 bits, you apply dither or noise shaping. The lesser the output bit depth, the more the noise. By choosing to keep the initial bit dept out of PGGB at 16 bit, you are increasing the noise (vs say 24 or 32 bit).

 

By doing this in two steps (by adding one more step of dither/noise shaping to a higher bit depth) you are at risk of adding even more noise and /or distortion depending on the settings you chose at the second step.

 

While there is nothing wrong in liking a specific sound signature, you may want to try:

• Try setting the output to 64bits in PGGB and then use HQP for dithering/noise shaping
• Try using 16bit output bit depth for PGGB and set HQP to pass through or (i.e no dither) and set output to 24bits

Hi just so I understand you correctly

 

I should upsample my 384kHz to 64bit & adaptive noise shaping off? Then let HQP DSD do the rest? HQP will then change dit depth back to 32bit?

[Zaphod Beeblebrox]

30 minutes ago, ASRMichael said:

Hi just so I understand you correctly

 

I should upsample my 384kHz to 64bit & adaptive noise shaping off? Then let HQP DSD do the rest? HQP will then change dit depth back to 32bit?

It depends on what you are trying to do. Are you trying to od the PCM upsampling in PGGB then use HQP to do DSD upsampling? If yes, then sure set PGGB to output 64bit (it will automatically turn off dither and noise shaping).

[ASRMichael]

49 minutes ago, Zaphod Beeblebrox said:

It depends on what you are trying to do. Are you trying to od the PCM upsampling in PGGB then use HQP to do DSD upsampling? If yes, then sure set PGGB to output 64bit (it will automatically turn off dither and noise shaping).

Yes that’s what I’m doing. Another question? Is there any reason why One would not upsample at maximum frequency 1411kHz using PGGB then convert to DSD? 

[Zaphod Beeblebrox]

6 minutes ago, ASRMichael said:

Yes that’s what I’m doing. Another question? Is there any reason why One would not upsample at maximum frequency 1411kHz using PGGB then convert to DSD? 

You can even go twice that, but the limitation will be file size, processing time, RAM etc.

[ASRMichael]

56 minutes ago, Zaphod Beeblebrox said:

You can even go twice that, but the limitation will be file size, processing time, RAM etc.

Ok for experiment sake…I upsampled to 2822kHz @ 64bit. Then convert to DSD 64! The sound is insane. I also notice when using HQP the CPU usage is very low. Presume this has something to do with PGGB upsampled to 2822. 
 

If google would lend me a server farm to store the files I would convert everything.

 

Fun experiment to do! 

[ASRMichael]

Has any other users here tried this? 
 

Upsample to 2822/kHz @ 64bit then play through HQP? 
 

Thoughts please? If you can please try it & let me know your thoughts? 

[kennyb123]

9 hours ago, ASRMichael said:

Has any other users here tried this? 
 

Upsample to 2822/kHz @ 64bit then play through HQP? 

I want to start by mentioning that the diagram of your system in your profile is really well done.  I checked there to figure out which DAC you were using to shed light on why you might be doing this.  It appears that the chips in your DAC can’t do DSD natively (at least according to the HiFiPlus review).  I’m thus curious why you went down this path.  
 

In my case HQPlayer isn’t the least transparent of my playback options, so I wouldn’t have been inclined to try this.  But you piqued my interest since you said the sound is insane.

[ASRMichael]

4 hours ago, kennyb123 said:

I want to start by mentioning that the diagram of your system in your profile is really well done.  I checked there to figure out which DAC you were using to shed light on why you might be doing this.  It appears that the chips in your DAC can’t do DSD natively (at least according to the HiFiPlus review).  I’m thus curious why you went down this path.  
 

In my case HQPlayer isn’t the least transparent of my playback options, so I wouldn’t have been inclined to try this.  But you piqued my interest since you said the sound is insane.

Hi thanks. I actually need to update my diagram. Why? I can across a post saying PGGB can complement HQP. So I thought I would try. 
 

Regarding Dop / Natively? Does this actually matter? Too many variables to know if makes a difference or not. What I do know is it sounds really good. I prefer PGGB v HQP. But using PGGB with HQP sound really, really good. Probably DAC dependant for sure. If you have PGGB convert a track to maximum rate & bit depth & play through HQP. See what you think. There’s trial versions for both PGGB & HQP. Really keen to see if others think it sounds better? 
 

Thanks

[kennyb123]

12 hours ago, ASRMichael said:

Regarding Dop / Natively? Does this actually matter?

I think what the DAC chip does can matter.  I used the word “native” because the review I read used that term when mentioning the DAC chip.  I don’t believe they were referring to how the DSD is passed in as that’s more of a function of the USB interface.

 

My DAC has ESS chips.  I think it converts both DSD and PCM to SDM.  Regardless, to my ears PCM sounds a lot better.  I haven’t tried your method though and may do so later this week.

 

[ASRMichael]

24 minutes ago, kennyb123 said:

I think what the DAC chip does can matter.  I used the word “native” because the review I read used that term when mentioning the DAC chip.  I don’t believe they were referring to how the DSD is passed in as that’s more of a function of the USB interface.

 

My DAC has ESS chips.  I think it converts both DSD and PCM to SDM.  Regardless, to my ears PCM sounds a lot better.  I haven’t tried your method though and may do so later this week.

 

So many variables involved. Like, personal preference, speakers, etc etc. love to hear you thought? Use 2822 output rate & 64bit then use HQP. Some combination! 
 

Only issue is 1 album with 15 songs was 157gb! Worth creating few playlists mind you. 

[seeteeyou]

Assuming that all tests will be converted to DSD 64 by HQP, are you expecting each candidate to sound any different from the others?

1. 705.6kHz @ 32bit
2. 705.6kHz @ 64bit
3. 1411.2kHz @ 32bit
4. 1411.2kHz @ 64bit
5. 2822.4kHz @ 32bit
6. 2822.4kHz @ 64bit

Just trying to find out if the law of diminishing returns were still applicable in this case, storage devices do get more affordable down the road but we might also be interested in knowing where the sweet spot is IMHO.

[ASRMichael]

44 minutes ago, seeteeyou said:

Assuming that all tests will be converted to DSD 64 by HQP, are you expecting each candidate to sound any different from the others?

1. 705.6kHz @ 32bit
2. 705.6kHz @ 64bit
3. 1411.2kHz @ 32bit
4. 1411.2kHz @ 64bit
5. 2822.4kHz @ 32bit
6. 2822.4kHz @ 64bit

Just trying to find out if the law of diminishing returns were still applicable in this case, storage devices do get more affordable down the road but we might also be interested in knowing where the sweet spot is IMHO.

Yes & No, My audio memory would really struggle with so many comparisons. So I’m comparing the following.
 

32bit - 384kHz v 64bit - 705.6kHz 

64bit - 705.6kHz v 64bit - 1411kHz

64bit - 1411kHz v 64bit - 2822kHz

 

The above then converts to HQP 64bit. Haven’t even messed with HQP filters yet. EXT2 is what I used previously. 

 

The higher I go the better it sounds. Probably because at 2822kHz @ 64bit HQP has very little to do. Minimal processing and temperature only go up 2-3c. The lower output rate the more HQP increases CPU processing, in turn increased temperature. 

Then there’s tracks being split into parts. This messes with meta data with Euphony. Then as you say, the cost of storage. 
 

I’ve spent a lot of time today trying to find the sweet spot. 
 

I’ve got 8tb NAS drive spare, so enough for a few playlists! 

 

[seeteeyou]

10 minutes ago, ASRMichael said:

The above then converts to HQP 64bit.

 

Please correct me if I were mistaken, did you actually mean that HQP was always configured to output DSD 64 @ 1bit for every single test?

 

In other words, HQP should be essentially downconverting each PCM track from a relatively high sampling rate to a (much) lower one @ DSD 64

 

https://aurender.zendesk.com/hc/en-us/articles/360026675554-Why-does-the-sample-rate-of-a-DSD-file-show-176-4kHz-or-352-8kHz-on-my-Aurender-

Quote

Aurender uses DoP (DSD over PCM) to send DSD files to DoP-compatible digital-to-analog convertors. Since DoP uses a PCM data stream, a DSD64 file will display at 176.4kHz, and a DSD128 file will display 352.8kHz on the Aurender Conductor app readout. Note that the receiving DoP-compatible DAC still reads the DSD file natively, as the PCM stream is simply acting as a transport for the original data.

 

DSD tracks are supposed to be a single-bit format all the way from DSD 64 to DSD 2048 while 64 simply meant the multiples of 44,100

 

BTW, I wasn't expecting anything close to a critical analysis of comparing stuff like 1411.2kHz versus 2822.4kHz or 32bit versus 64bit etc.

 

IMHO it's more like whether we're merely talking about subtle differences (i.e. maybe we'll just go as far as converting our favorite tracks) if we're going all the way up to the highest sampling rate available, or a reasonably significant delta that's actually worth all that effort (i.e. let's start redoing the entire music library from now on) to dive right in.

 

Many thanks for sharing the results of your experiments with everyone, this project should be meant to benefit those DACs with “NOS” options the most but now maybe we're finding new ways to achieve even greater gains for other options out there.

[ASRMichael]

5 minutes ago, seeteeyou said:

 

Please correct me if I were mistaken, did you actually mean that HQP was always configured to output DSD 64 @ 1bit for every single test?

 

In other words, HQP should be essentially downconverting each PCM track from a relatively high sampling rate to a (much) lower one @ DSD 64

Yes that’s correct. 
 

See CPU load when using HQP. I may test with DSD128 in time, but keeping to the above test for now. DSD128 is maximum my dac goes to. 
 

I look forward to you trying this? 

[ASRMichael]

22 minutes ago, seeteeyou said:

 

Please correct me if I were mistaken, did you actually mean that HQP was always configured to output DSD 64 @ 1bit for every single test?

 

In other words, HQP should be essentially downconverting each PCM track from a relatively high sampling rate to a (much) lower one @ DSD 64

 

https://aurender.zendesk.com/hc/en-us/articles/360026675554-Why-does-the-sample-rate-of-a-DSD-file-show-176-4kHz-or-352-8kHz-on-my-Aurender-

 

DSD tracks are supposed to be a single-bit format all the way from DSD 64 to DSD 2048 while 64 simply meant the multiples of 44,100

 

BTW, I wasn't expecting anything close to a critical analysis of comparing stuff like 1411.2kHz versus 2822.4kHz or 32bit versus 64bit etc.

 

IMHO it's more like whether we're merely talking about subtle differences (i.e. maybe we'll just go as far as converting our favorite tracks) if we're going all the way up to the highest sampling rate available, or a reasonably significant delta that's actually worth all that effort (i.e. let's start redoing the entire music library from now on) to dive right in.

 

Many thanks for sharing the results of your experiments with everyone, this project should be meant to benefit those DACs with “NOS” options the most but now maybe we're finding new ways to achieve even greater gains for other options out there.

Sorry one last thing. My first test was using only HQP DSD64 from 44.1kHz versus PGGB 32bit - 384kHz - output to DSD64. This was the discovery moment where I said mmmmm. Then I started working my way up.

 

The could all be ear, DAC dependant but certainly worth a go. 

[ASRMichael]

26 minutes ago, seeteeyou said:

 

Please correct me if I were mistaken, did you actually mean that HQP was always configured to output DSD 64 @ 1bit for every single test?

 

In other words, HQP should be essentially downconverting each PCM track from a relatively high sampling rate to a (much) lower one @ DSD 64

 

https://aurender.zendesk.com/hc/en-us/articles/360026675554-Why-does-the-sample-rate-of-a-DSD-file-show-176-4kHz-or-352-8kHz-on-my-Aurender-

 

DSD tracks are supposed to be a single-bit format all the way from DSD 64 to DSD 2048 while 64 simply meant the multiples of 44,100

 

BTW, I wasn't expecting anything close to a critical analysis of comparing stuff like 1411.2kHz versus 2822.4kHz or 32bit versus 64bit etc.

 

IMHO it's more like whether we're merely talking about subtle differences (i.e. maybe we'll just go as far as converting our favorite tracks) if we're going all the way up to the highest sampling rate available, or a reasonably significant delta that's actually worth all that effort (i.e. let's start redoing the entire music library from now on) to dive right in.

 

Many thanks for sharing the results of your experiments with everyone, this project should be meant to benefit those DACs with “NOS” options the most but now maybe we're finding new ways to achieve even greater gains for other options out there.

Reading your post above. If DSD64 is x64 44.1 & I’m upsampling to 2822kHz then is there no down sampling involved? Only down sampling with anything below 2822. 

[Zaphod Beeblebrox]

1 minute ago, seeteeyou said:

Please correct me if I were mistaken, did you actually mean that HQP was always configured to output DSD 64 @ 1bit for every single test?

 

In other words, HQP should be essentially downconverting each PCM track from a relatively high sampling rate to a (much) lower one @ DSD 64

While I am unfamiliar with the internal operations of HQP, merely packaging DSD64 as DoP does not constitute to downs sampling. DSD64 has the same bit rate as 176.4kKz at 16 bits and that may be the reason for Aurender to display 176kHz and it is not really down sampling. Also 2822kHz can be directly converted to DSD without need for any upsampling.

[ASRMichael]

9 minutes ago, Zaphod Beeblebrox said:

While I am unfamiliar with the internal operations of HQP, merely packaging DSD64 as DoP does not constitute to downs sampling. DSD64 has the same bit rate as 176.4kKz at 16 bits and that may be the reason for Aurender to display 176kHz and it is not really down sampling. Also 2822kHz can be directly converted to DSD without need for any upsampling.

That clearly explains the low CPU processing going on at 2822kHz. Thanks

[Miska]

3 hours ago, seeteeyou said:

In other words, HQP should be essentially downconverting each PCM track from a relatively high sampling rate to a (much) lower one @ DSD 64

 

No, if you input 2882.4 kHz for DSD64 conversion, then there is no need for sampling rate conversion at all. Just delta-sigma modulation.

 

DSD64 sampling rate is 2882.4 kHz. Which is very low though. By today's standards you should be doing at least DSD256 which is 11289.6 kHz. With many DACs you can do DSD1024 which is 45158.4 kHz. When you use HQPlayer to convert from PCM to these rates, the digital filter you have selected is used to convert to that final output rate before the modulator.

 

So you can have millions of taps being computed at tens of MHz sampling rates.

 

[seeteeyou]

7 hours ago, ASRMichael said:

I look forward to you trying this?

 

Maybe, maybe not.

 

I'm quite loyal to the PGGB family while the 64-bit PCM milestone could very well be reached within this year or 2023

 

https://www.akm.com/content/dam/documents/products/audio/audio-dac/ak4499exeq/ak4499exeq-en-productbrief.pdf

https://audioxpress.com/news/akm-developed-new-flagship-velvet-sound-ak4499exeq-stereo-d-a-converter

Quote

When used in conjunction with the AK4191 the AK4499EX is the ideal analog conversion solution for playback of ultra-high-resolution sound sources, up to 64-bit PCM with 1536kHz sample rate, or 44.8MHz sample rate DSD (DSD1024). The AK4191EQ is a new concept multi-bit sigma delta modulator data converter employing a built-in digital filter - effectively, six types of 64-bit digital filters - and 256 times oversampling processing that minimizes the effects of digital noise within the analog output, resulting in a perceived improvement of the ratio of signal to noise.

 

Though we'll also need USB interfaces that could actually support 64-bit PCM, not to mention stuff like operating systems / device drivers / software players etc.

 

Obviously not everything listed above would appear right away, let's just wait and see then

 

https://resoundsound.com/sample-rate-bit-depth/#comment-171884

Quote

Bouncing the Virus TI into audio (online render – realtime that is) and listening through Audeze iSine10s. At 48Khz and 16 bit, the stem sounded flat and sterile – lifeless. The same for the 24 bit stem – no life, as in, the sound didn’t move or appear to have the same dynamics as the source material. The 32 bit stem captured it. The 64 bit file did even better and at this point I can’t believe the fact I can hear the difference. Then, Reaper offers some 2 or 3 other formats which sound like the 64 bit file.

 

[CNoblet]

Just an update on my experience with PGGB and Chord Mojo dacs.  The USB micro input on my Mojo finally crapped out, so I purchased a Mojo 2.  Listening with no EQ applied, at same volume setting, I find it significantly better than the original Mojo.  For me, the upgrade in sound quality is well worth the cost increment of the purchase.  For anyone buying the Mojo 2 as their first Chord DAC, it is an amazing bargain.  Just for example, Hotel California had become really old stuff to my ears over the years, but listening to the 32/765 PGGB file (from 24/96) on my main system was a riveting experience.  Large, meaningful improvement over the original Mojo.  BTW, I have 13 TB of PGGB files at this point, with another 6-7 TB needed to complete the conversion of my digital collection.  With PGGB (32-bit, 16 fs) and the Chord Mojo 2, I am in Seventh Heaven listening-wise.

[Zaphod Beeblebrox]

Just a heads up PGGB V3.0 is around the corner; I suggest holding off on any large batch conversions...

[Peti]

V3, hmmm sounds enticing. What will be the improvements over the current version?

[Zaphod Beeblebrox]

7 hours ago, Peti said:

V3, hmmm sounds enticing. What will be the improvements over the current version?

If you have current PGGB offline license and have access to a PC with 64GB of RAM (or a Mac with 128GB of RAM) and interested in beta testing V3, PM me.