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

[dmance]

I stumbled upon PGGB during my weekly audio forum visits and downloaded it. After a few days, here are my comments:
First ... a huge congratulations to ZB and the group that made this happen. Thanks to Ray G for (again) being involved at the forefront of digital audio and his effort to help communicate this amazing development to a wider audience. And thanks for all the fish.

Introduction:
I jumped on the Chord DAC bandwagon with hearing Rob Watts' large tap count sinc filters and his excellent analog section design. I sensed that Rob's approach was the right way to build a DAC and his millon tap MScaler confirmed (at least for me) that his mathematically perfect conversion could also be emotionally satisfying. Of course, digital processing is not the entire story - a DACs final D/A stage and output clocking is also subject to many and myriad influences of EMI and such. I started Audiowise Inc (www.audiowise.ca) with OPTO-DX optical isolation of the Chord DX interface and I’ve evolved a range of solutions to address the detrimental effects of RF noise. 

In 2019 I prodded Jussi L. to push the envelope with his HQPlayer PCM filters and noise shapers to meet or exceed MScaler sound quality. His 2M tap Sinc-L filter and 15th order LNS15 noise shaper then became my reference for up-sampling. It was clear that more detail retrieval was possible with more taps and worthwhile to adopt for playback - despite the one second delay issues. So, since early 2020 my current chain is attached, IMO, an extremely well RF isolated DAVE configured with my full kit of products that incorporate everything I’ve learned about the insidious nature of RF noise.

If I had not heard PGGB, I would have been quite satisfied with the sound from HQP's 2 million taps. Now, after only a few days with PGGB, it's apparent that 2 million is not enough. Nowhere close to it. Go big or go home, I say.

PGGB Usage:
For PGGB installation I used a 2019 vintage Lenovo ThinkPad P71 laptop with 16Gb ram and a fast SSD. Processed WAV files were copied to my NUC and I used a remote desktop session to HQPlayer or JRiver MC for my tests. Previously, with HQP upsampling, my NUC was at 40-50% CPU usage during playback.  Now, with the complete separation of the heavy lifting of upsampling away from actual playback, my NUC trickles by on 5-10% CPU ...and likely far less RF emissions.

I'm quite satisfied with the PGGB user interface - mostly because ZB's 'Auto' modes are not compromised but actually provide the optimal setting. So, I set Auto taps, Auto Noise Shaping and Auto Gain. I’m also quite satisfied with the performance; however, I would like more descriptive status ...especially during periods of inactivity and, on launch, the long delay between the disappearance of the startup screen and appearance of GUI interface regularly had me concerned that PGGB had crashed.

I set PGGB upsampling to 705.6/768 16fs and bit depth to 24-bits ... because that's what I used with HQPlayer and it's what MScaler's DBNC interface is. I use SRC-DX USB-to-DBNC bridge to avoid USB noise so I have not tried a 32bit file direct via USB. However, I believe USB’s 32bits is truncated to 24 at the DAVE’s signal interface anyhow.

With HQP upsampling, the need to set a global -3dB (or -4dB) attenuation to avoid inter-sample overs bothered me for the reduction in volume and dynamic range. So, I am very happy that PGGB Auto Gain minimally attenuates yet perfectly avoids inter-sample overs. As the status messages whiz by I am seeing far less than -1dB of applied gain and typically less than -0.5dB so everything is louder and closer to master 0dB volume. The compromise of minor variation in gain on a per-track basis is not noticeable and, anyhow, each album is produced with a different average volume. This is no issue for me as I have the remote volume control in my hand full time and am always tweaking volume anyhow.

I set TRANSPARENCY to Natural and PRESENTATION to Transparent. Regarding noise shaping...the PGGB Auto noise shaping works very well to minimize noise - I trust ZB that this is optimal ...sure sounds like it. I set the HF NOISE FILTER to 'Minimal' since, IMO, it sounds best. DAVE does its final noise shaping after its own application of an HF Filter and its excellent volume control.

I processed dozens of tracks in trial mode (the first 5 tracks from each album I selected)- My playback was done manually with HQPlayer 4.10.3 and JRiver MC. I used remote desktop from my laptop to my NUC for control. I made 100% sure the audio files were untouched during playback.

SQ Impressions:
Stunningly good. But you might have to experience it with 16fs or better to have the full transparency reveal itself. Some tracks present with just more nuance and realism; some just sound more natural. All sound better and increase my experience of Qualia …which should be everyone’s goal. From an Audiowise perspective, PGGB is a much better tool than HQP to help me plumb further depths of sound quality and make better products.  I am so glad this has been realized. Again, kudos to everyone involved. I hope Jussi L. has downloaded PGGB and hears how off-line processing with ‘practically' infinite taps sounds better than his best real-time approach.

I'd like to ask ZB to report on any round-trip analysis: like redbook upconversion to a float output (WAV 32 Float?) followed by downconversion to the original. Ideally this process is transparent and helps confirm that PGGB is not adding euphonics but actually demonstrating that the reconstruction is optimal to the theoretical. If confirmed, ZB should publish a paper and secure implementation patents for having realized Nyquist-Shannon. Rob Watt’s original estimate of only 1M taps being needed for redbook roundtrip was never was publicly validated with his Davina project and now seems a tad on the low side.

Workflow:
A PGGB file is not stream-able today but the SQ is so compelling that I want to move ahead with a workable environment to enjoy both my own files and streamed content. WAV files are big but they have format ubiquity on playback software. I was happy to get back to using JRiver MC for playback since I have mixed feelings about Roon. Monkeys Audio is a great lossless compression program that maintains all WAV metadata. (here) and MC is a champion with full support.  I converted all my PGGB'd .WAV files to .APE for seamless playback via MC.  I was seeing 25-30% file sizes. That is, my typical 1000Mbyte .WAV files became ~270Mbyte .APE files. Some of this is due to my 24-bit sample leaving an empty 8-bits in WAVs 32-bit PCM word. For me, the good news about .APE compression is that the 2TB SSD in my NUC can now hold a ~7500 track PGGB music collection !! …which is plenty. FYI - the decoding from APE to PCM in MC takes about 0.5sec with prebuffering of the track. So, for the lossless purists, all PCM samples are just read from RAM during playback.
ZB…PGGB has to be integrated into playback software (please!) for automatic conversion and file management. In the interim, I will work on my own workflow components for my continued testing.  

Video Files:
Audiophiles, in general, seem to shun consuming video media with their music and both HQPlayer and ROON have stated they have no intention of adding video support. In my case, I find that watching short videos of live or studio recordings is a very compelling way to consume music. I have tested custom integrations of MC and HQP to achieve this and experimented with HQP-Pro to integrate WAVs into video files. The much-improved SQ of PGGB will now have me revisit this. 

I’m sure not every users audio setup is geared for video playback -> to a nice monitor and audio playback -> to a DAC ...but mine is. I’ve remixed PGGB’d WAV audio from a few 4K YouTube clips into MKV multiplexed video files. The results dramatically increase the feeling of being present : 4K video with 16fs remastered audio in perfect sync is amazing. For those interested, I have three clips being downloaded here (usage under fair use). You will need JRiver MC for playback of these files. Note that local audio playback will down-convert the audio to typically 48k so make sure MC is setup properly to preserve the full resolution PCM bitstream to your DAC.

i'll keep reporting back ...definitely will purchase! Great job ZB and team!

 

Dan
 

[dmance]

I ran out of space in Dropbox for my posted PGGB'd music videos link. Use this Google Drive link here.

Thanks,

Dan

 

 

[dmance]

13 hours ago, Progisus said:

I hooked my SRC-DX to my H2 and sent my 32bit files to it. Everything reported 32 bits and sounded excellent. Is there a 32b to 24b coax conversion in the DX? Should I be hearing a sound degradation by not using 24b pggb files?

Chord DACs are 24 bits internal. So even a 32bit USB setting will waste 8 bits. Of course DX is just dual SPDIF which is 24 bit in the signal. As mentioned by others, set PGGB to output 24 bits for best sq to Chord DACs

[dmance]

I fell in love with PGGB after upsampling the first few tracks with the demo version. After purchasing the licensed version I knew I wanted to use the command line version for better integration into my AudioWise product testing environment. I then evolved a very useful workflow utility ...something that would make selecting folders and managing batch PGGB processing easier. I decided that it was too good not to share with other PGGB users. So today I have posted, as a $free product, PGGB-IT! Workflow App on my AudioWise site. It's a Windows 10 console application that does PGGB batch processing on a queue of folders stored in a Windows Explorer library. The included companion QUEUE-IT! application allows you to add folders to the queue using per-folder drag-drop from Explorer or per-album 'Send to' directly from JRiver Media Center. PGGB-IT! then grabs the list of input folders from the library and runs PGGB (programmatically) on them to the output folder - using FQN pathing.  And while you are waiting for the current batch

to finish, you can load up more folders in the queue and PGGB-IT! will process those as well. The diagram below says it all

If there are any Window10 users out there playing with PGGB, please download and give it a try!

 

Dan

 

[dmance]

31 minutes ago, hanshopf said:

Hello everyone,

 

I would like to add something to my recent suggestion for a proof of concept experiment for PGGB.

 

Thanks to the new cloud based option of upsampling files with PGGB, I today had the opportunity to try it out for myself. I played the upsampled files through Audirvana and sent them directly to DAVE. The original reference files for comparison were as well played through Audirvana and sent to MScaler. 

 

I am a trained listener, recording concerts with classical music on a regular base myself (not as a sound engineer but as a video director), but still did not expect to make out a clear difference because I am not able to reliably discern things as subtle as some in this forum seem able to.

That said, I was stunned how obvious the difference between Mscaler and PGGB was. First I perceived a greater depth in soundstage and air between the instruments. But later I very clearly heared the PGGB sound as being brighter and sharper than with MScaler. Trumpets, Violins and soprano voices had an artificial bright touch. After my initial enthusiasm I therefore clearly find Mscaler sounding more natural or, if you like, neutral.

 

Is PGGB then maybe acting like some kind of presence filter, giving the illusion of more space and air?

 

The only way to find out what is going on, I guess, would be the kind of proof on concept I recently suggested: comparing an analogue mastertape with a digitized file through MScaler as well as PGGB. One of these two should sound more similar to the analogue tape. My bet would be MScaler, even though I can see that there's potential for improvement. 

 

 

 

 

 

 

When doing an apples/apples comparison its important to ensure the galvanic signal path and inputs for both is the same.  You could try repeating this test by sending the PGGB file thru MScaler (it will go into bypass if the input sample rate is 16fs) and DBNC into Dave.  Then do the normal MScaler upsample to 16fs and DBNC into Dave.  In both cases make sure any USB cable is removed from Dave (as well as any other coax inputs you may have). 

[dmance]

@seeteeyou @Zaphod BeeblebroxWAV is uncompressed. If you are referring to WAVPACK, PGGB-IT! uses the FFMPEG internal code base, not the wavpack libs. So files are exact lossless but the bitstream is slightly different.  FYI, this is also the case with FLAC. If your sequence is PGGB-IT! -to- WAV -to- FLAC (via XIPH exe) and compare to PGGB-IT! -to- FLAC (direct), the files will be different but the contents are lossless.

 

[dmance]

Regarding DSD over DOP via SRC-DX. SRC-DX was only specified to support DOP DSD128 ...as this is a single SPDIF stream and the DOP mechanism is defined.  The CT7601 chip used does have the internal DOP support up to DSD256  but the AES dual-date-date mode to support DBNC does not define DoP.  If it does work ...its ad-hoc and fortunate. I need to ask a SRC-DX/Chord DAVE owner to try this out. *thanks*