A novel way to massively improve the SQ of computer audio streaming

[Qhwoeprktiyns]

On 7/4/2020 at 7:58 AM, austinpop said:

I know you were being mostly facetious, but it’s a great question. Trust me, I’ve asked myself that many times, and of late I’m convinced it’s “whenever you decide.” There IS no end, but we all can and should stop or pause and enjoy our systems as they are.

 

Sorry to chime in here,  for a brief comment. The answer is obvious: it will stop when noise can be prevented from distorting digital to analog conversion.

 

Until then, have fun! 

[lmitche]

18 hours ago, kennyb123 said:

Great project status report!

 

Would you remind me again of the devices into which you are plugging those SFPs?  
 

You indicated that the transceivers you tried are all 10GB/s SFP+.  I missed why you went with SFP+ instead of gigabit SFP.  

Hi Kenny,

 

Thanks for your comment.

 

Give me a few weeks and I will share more details. There are a few device configuration permutations that need to be checked before providing more details.

 

Larry

[FredM]

11 hours ago, flkin said:

For the question I asked a few days ago above, I believe the drink in the track is “Evian”. I hear it being ordered, bartender confirming it and then once more perhaps by the customer. 😄

 

29” Customer: politely waves at the Bartender and orders an Evian for his girlfriend 

30” Bartender: looks at customer, raises his eyebrow, asks: Evian?

31” Customer: nods and says Evian to confirm

 

And then, a little later:

40” Bartender: slides the water over de bar to the beautiful girl

41” Customer: puts two aspirins next to the glass 

42” Girlfriend: looks at boyfriend and says: what are these for? I don't have a headache

43” Customer: Gotcha! 😄

[altsouza]

Hello people,

I recently made some improvements to my Roon Server. I installed Audiolinux, which gave an excellent improvement in Sound, and I also installed a 400w Dc-Atx HDplex in conjunction with Dell's 330w SMPS. I also installed a Ladaptor LPS on the Allo USB Signature Streamer.

The Roon Server is made up of >>> Gigabyte B360n MOBO / 8gbRam / i5 processor / 4tb Hdd.

The Network Chain is: Fiber ISP Modem >> 20m Cat6 >> Tplink Archer 3200 wifi Router >> 20m Cat6 >> Regular Switch (with Lps) >> Roon Server   >>  Allo USBSig streamer. I use Tidal with Roon

Following guidelines of this topic, I will make the following improvements in the Chain, in order of priority.
Any help or opinion in changing the order of priority or changing the components is very welcome:

1) Buy a "Hdplex 300w" (4 rail Lps) for the Roon Server (19v atx and 12v direct to Eps) and eventually use a rail for the Allo USB Sig and another rail for upcoming Uptone etherRegen.
2) Buy Uptone ether Regen for Roon Server (A side)  and Allo USB Sig (B side)

3) Buy a HDPlex H3 case to go Fanless
4) Include a Jcat net board in Roon Server

5) Replace the Itx board with a more expensive mAtx board (Asus Rog or Gigabyte Aorus Line)
4) Replace Allo USB with the new Jcat usb xe, and make the Roon Server an All in one Server/end point unit.

During the upgrade process, I can also bring the ISP Modem near to the Roon Server, to avoid long cables, and I can also use a Hdplex rail to power it. But this modification, despite being free, will give me a lot of work to do.

Any help is welcome.

Thanks

Antonio

[cat6man]

regarding audio-linux v. euphony

 

i see that many folks have moved from audio-linux to euphony over the past months with reports that euphony sounds better...............so i got a trial of euphony to check it out...........and the results are very interesting.

 

common baseline:

running HQPplayer embedded in server to NAA running on opticalRendu, much network optimization.

HQP running pcm only, no upscaling, no dither, just bit exact (theory being that having renderer at server and minimal processing at NAA is a good thing to do).  music on HDD attached via usb to NUC running music server

 

audio-linux running in ramroot, realtime set to extreme

euphony running on usb stick (in ramroot, it doesn't see HDD, so couldn't test ramroot)

 

tests:

audio-linux 2.5 -- meh

euphony 20200114v3 -- meh

audio-linux 1.4 -- wow!

 

in my system, the old build of audio-linux is vastly superior to the newer builds of both audio-linux and euphony and it isn't close.  i have not chosen to compare the first two as who cares if mehA is superior to mehB when you have wow as an option.

 

since YMMV, i'd love to hear from other audio-linux and euphony users.  try comparing your current servers with the old audio-linux 1.4 and let us know if you agree/disagree

 

something is going on here but damned if i know what............scheduling/queuing algorithms in the realtime scheduler, arch-linux build variations/issues, ?????.....................but whatever it is, it isn't minor.

i can't listen to the first two for more than a few minutes before the difference is obvious and i'm back to the audio-linux 1.4 build.

 

stay safe out there!

 

p.s. will be upgrading music server from hdplex200 to hdplex300 this week

 

[drjimwillie]

18 minutes ago, cat6man said:

regarding audio-linux v. euphony

 

i see that many folks have moved from audio-linux to euphony over the past months with reports that euphony sounds better...............so i got a trial of euphony to check it out...........and the results are very interesting.

 

common baseline:

running HQPplayer embedded in server to NAA running on opticalRendu, much network optimization.

HQP running pcm only, no upscaling, no dither, just bit exact (theory being that having renderer at server and minimal processing at NAA is a good thing to do).  music on HDD attached via usb to NUC running music server

 

audio-linux running in ramroot, realtime set to extreme

euphony running on usb stick (in ramroot, it doesn't see HDD, so couldn't test ramroot)

 

tests:

audio-linux 2.5 -- meh

euphony 20200114v3 -- meh

audio-linux 1.4 -- wow!

 

in my system, the old build of audio-linux is vastly superior to the newer builds of both audio-linux and euphony and it isn't close.  i have not chosen to compare the first two as who cares if mehA is superior to mehB when you have wow as an option.

 

since YMMV, i'd love to hear from other audio-linux and euphony users.  try comparing your current servers with the old audio-linux 1.4 and let us know if you agree/disagree

 

something is going on here but damned if i know what............scheduling/queuing algorithms in the realtime scheduler, arch-linux build variations/issues, ?????.....................but whatever it is, it isn't minor.

i can't listen to the first two for more than a few minutes before the difference is obvious and i'm back to the audio-linux 1.4 build.

 

stay safe out there!

 

p.s. will be upgrading music server from hdplex200 to hdplex300 this week

 

Did you see that AudioLinux had posted the wrong 2.5 they had posted the 2.5 for pink faun.    now they have posted the correct 2.5

[austinpop]

Roon Update Build 571 on Euphony OS

 

As many of you know, my primary music player is Stylus on Euphony OS. I play music both from local files stored on a NAS, and from Qobuz. I do from time to time use Roon, especially when I am in discovery mode, as one of my favorite activities is grazing in the Qobuz new classical releases section, and Roon is so much better as a user experience.

 

Of late, I've been feeling that the SQ with Roon streaming Qobuz has sounded very satisfying, so it got me wondering if this was just a feeling or whether there was something going on. I did a quick check with this album, which is identical on Qobuz and my local copy on NAS:

 

In all cases, I was booted into Euphony OS, version 20200430, in RAM root mode. I compared the SQ of the following:

• Roon streaming Qobuz
• Roon from NAS
• Stylus (with cache, buffer queue to RAM) from Qobuz
• Stylus (with cache, buffer queue to RAM) from NAS.

To my surprise, Roon from Qobuz sounded significantly better than Roon from NAS! Overall, the SQ progression was:

• Roon from NAS < Roon from Qobuz < Stylus from Qobuz < Stylus from NAS.

For the math-challenged: "<" refers to "less than" or really, "worse sounding than."

 

Also interesting was that the difference in SQ between Roon from Qobuz and Stylus from Qobuz was not that big. So I was not imagining what I heard with Roon from Qobuz.

 

Analyzing Roon network behavior

 

To investigate further, I looked at the network behavior in the Roon cases, using the nifty network monitor in Euphony (click on the word REGISTERED on the bottom right of the web UI). Here's what I observed.

 

Roon build 571's network behavior is different based on source:

• When does preload start: when you hit play
• Qobuz/streaming sources: 
  • download current track at network speed
  • appears to preload enough of next track to maintain gapless, but full track download appears to only happen when the track becomes current. This is different from Stylus (below).
• Local network files from a NAS or SMB share:
  • no preload!!
  • Appears to be doing normal file reads, which results in some read-ahead, but not a full download.
• Local files on SSD
  • did not test, but not relevant to network traffic.

In contrast, let's look at Euphony/Stylus. The behavior here is also different based on source:

• When does preload start: 
  • Either when you hit play, OR
  • Click on "Buffer Queue to RAM." This allows you to wait for network activity to quiesce before you hit play.
• Tracks from Qobuz/streaming sources: 
  • download current and next track at network speed
  • when next track becomes current, download the following track at network speed, and so on
  • In other words, at most 2 tracks (current and next) are buffered
• Local network files from a NAS or SMB share:
  • download all tracks at network speed, up to limit of RAM buffer (tunable option).
• Local files on SSD
  • did not test, but not relevant to network traffic.

More experiments with Roon file playback

 

To see how much of the Roon file playback degradation was due to my files being on NAS, and the Core's not preloading the entire file like it was doing on Qobuz, I compared the same track, and varied the file location:

• original NAS location
• local SSD
• in RAM.

To achieve the above, I first fired up Stylus and cached and buffered the album. Then switching back to Roon, I added the following data locations in the Storage settings:

 

Cached files are stored on my boot Optane SSD in /data/Music/E_CACHE, and the files buffered in RAM actually live in /tmp/EUPHONY_BUFFERED. The reason that the /tmp files are considered in RAM is because /tmp is a filesystem of type tmpfs in Linux, which is a flavor of RAM disk.

 

Listening results

 

As expected, the SQ was worst from the NAS location, and best in RAM. However, what is interesting is I still felt the playback from Qobuz sounded better. So in SQ terms:

• Roon from NAS < Roon from local SSD < Roon from in RAM < Roon from Qobuz.

My best guess is that even in the "in RAM" case, the Core is still making filesystem calls to access the file, whereas in the Qobuz case, the entire file is cached within the Core process (note: this is a guess, not confirmed).

 

Relating this back to the Taiko Extreme Roon "degradation" experience

 

The above finding of disappointing results with Roon file playback at least loosely correlates now to the complaints of Roon SQ degradation reported on the Extreme forum.

 

Emile is claiming, and Extreme owners are reporting, that his latest tunings mitigate the problem, and that Roon now sounds better than ever. This to me suggests that there are OS tunings possible that can close this gap. I have some ideas, but this is one of the huge challenges with Roon.

[kennyb123]

26 minutes ago, austinpop said:

To my surprise, Roon from Qobuz sounded significantly better than Roon from NAS

 

Interesting finding.  I had in the past encountered edge cases like this before, but always assumed it had something to do with my source file.  What was the source for the file on your NAS?  Did you rip it from CD yourself?  If so you could eliminate source file differences by purchasing this album directly from Qobuz.  But - I doubt that's the issue and it has more to do with what you reported.  

 

I do recall claims in the past that the same CD but from different manufacturing plants were found to sound different.  Some also speculated that files delivered direct from the studios to streaming sources might have an advantage over those that went though the process of being manufactured onto a CD and then ripped from it.  Your findings suggest it's more about how the player handles these files, which sounds more plausible. 

[Dev]

31 minutes ago, austinpop said:

My best guess is that even in the "in RAM" case, the Core is still making filesystem calls to access the file, whereas in the Qobuz case, the entire file is cached within the Core process (note: this is a guess, not confirmed).

 

Roon makes incredible filesystem I/O continuously, god knows for what reason. Earlier when I was running AL in Ramroot, I had a USB3.0 external storage drive powered by LPS directly connected to the server. The USB drive has a blue HDD access led and those were constantly blinking during playback indicating Roon constantly kept accessing the storage. I believe NAS access is no different.

 

 

35 minutes ago, austinpop said:

Emile is claiming, and Extreme owners are reporting, that his latest tunings mitigate the problem, and that Roon now sounds better than ever. This to me suggests that there are OS tunings possible that can close this gap. I have some ideas, but this is one of the huge challenges with Roon.

 

Any hint from anybody what it could be ? I believe Extreme uses Windows. Maybe (or maybe not) we could map the tweak in Linux.

[austinpop]

9 minutes ago, kennyb123 said:

Did you rip it from CD yourself?  If so you could eliminate source file differences by purchasing this album directly from Qobuz.  But - I doubt that's the issue and it has more to do with what you reported.  

No CD involved. This is a very recent 24/48 release. I bought the download files from the label. Can’t imagine they would be different on Qobuz, as this is too new to possibly have other masterings in the wild.

[cat6man]

3 hours ago, drjimwillie said:

Did you see that AudioLinux had posted the wrong 2.5 they had posted the 2.5 for pink faun.    now they have posted the correct 2.5

 

yup..........a non-issue really since the 'wrong' one wouldn't have loaded anyway

[bobfa]

1 hour ago, austinpop said:

To my surprise, Roon from Qobuz sounded significantly better than Roon from NAS! Overall, the SQ progression was:

• Roon from NAS < Roon from Qobuz < Stylus from Qobuz < Stylus from NAS.

For the math-challenged: "<" refers to "less than" or really, "worse sounding than."

 

Also interesting was that the difference in SQ between Roon from Qobuz and Stylus from Qobuz was not that big. So I was not imagining what I heard with Roon from Qobuz.

You seem to have found part of the mystery here. I saw all these posts saying the Roon started sounding bad. I do a lot of my listening when I am testing using Qobuz as I do not have downloads for my playlists ( some from your other playlists you have shared).  

 

I started messing with GentooPlayer to see what I could find there.  (No conclusions yet)

 

I need to swap back over to Euphony again.

[ted_b]

5 hours ago, cat6man said:

regarding audio-linux v. euphony

 

i see that many folks have moved from audio-linux to euphony over the past months with reports that euphony sounds better...............so i got a trial of euphony to check it out...........and the results are very interesting.

 

common baseline:

running HQPplayer embedded in server to NAA running on opticalRendu, much network optimization.

HQP running pcm only, no upscaling, no dither, just bit exact (theory being that having renderer at server and minimal processing at NAA is a good thing to do).  music on HDD attached via usb to NUC running music server

 

audio-linux running in ramroot, realtime set to extreme

euphony running on usb stick (in ramroot, it doesn't see HDD, so couldn't test ramroot)

 

tests:

audio-linux 2.5 -- meh

euphony 20200114v3 -- meh

audio-linux 1.4 -- wow!

 

in my system, the old build of audio-linux is vastly superior to the newer builds of both audio-linux and euphony and it isn't close.  i have not chosen to compare the first two as who cares if mehA is superior to mehB when you have wow as an option.

 

since YMMV, i'd love to hear from other audio-linux and euphony users.  try comparing your current servers with the old audio-linux 1.4 and let us know if you agree/disagree

 

something is going on here but damned if i know what............scheduling/queuing algorithms in the realtime scheduler, arch-linux build variations/issues, ?????.....................but whatever it is, it isn't minor.

i can't listen to the first two for more than a few minutes before the difference is obvious and i'm back to the audio-linux 1.4 build.

 

stay safe out there!

 

p.s. will be upgrading music server from hdplex200 to hdplex300 this week

 

Great post, but this would be even more relevant in the "Shootout at the Linux Corral: Audiolinux vs Euphony" thread, no?

 

[tims]

On 7/3/2020 at 9:32 PM, tgb said:

Thanks for the confirmation John.

After 24h "burning", the SQ of the GS2024 seems a bit better (long process for this big piece). It's going the right way. I admit I was quite desappointed with the top-harsh sound of the cold machine yesteday. Still using its internal SMPS, cause I don't have proper LPS yet for this unit.

 

Regarding the external power with 3 PS @ 1 - 1.5 - 3.3VDC, I'll have a try within 24hours & post the results (good or bad). 

Rgds

Hi @tgb

Did you get a chance to try external power at the three dc inputs (1 - 1.5 - 3.3) on your GS2024? 

If so, any changes/improvements?

Thanks

[tgb]

hi Tims,

sorry i did not give answer since the thraed went another way... & though it was no interest.

 

I tried this WE : i couldn't manage to boot this bloody switch...

Control panal with all LEDs lit... that's all I could get....

 

A few weeks ago, I did it on my TP-Link SG5412F : straight forward trick : you plug the external power at the pins of of cap after the regulation steps (3.3V / 1.8V 1V in this case) => power on with the 3 external PS => the switch boot & run... easy. The SQ improves a lot on 3D / placement etc. typical in this mods.

 

With this Buffalo, there is a trick womewhere... is there is a feedback control to check if the 12VDC is still there ? I don't know.

 

Theere are numerous points to measure 3.3 / 1.5 / 1V on the PCB.

Power the control panel thru the 3.3V is easy.

Regarding the 1.5V for instance : tricky. If you give 1.5V on the PCB without 3.3V on => no power delivered. The 1.5V powers on IF you supply 3.3V besides.

Quite disappointed because you have on the PCB very nice & clear marks - / + with soldier on 1V & 1.5VDC parts, but when you plug on ithem : no boot... 

 

To be clear, I received the GS2024 5days ago, it burns.... burns... & still sounds awful compred to my current optical network. Thus I tried these tweaks without being "invasive" on the PCB (no soldier etc) because as this switch is a big step back vs SQ for me, shortly I will have to send it back to the vendor to get my money back.

Maybe a deeper & more invasive tweak could make it boot... don't know...

 

Have a try. Hope you find the trick !

BRgds

[cat6man]

43 minutes ago, ted_b said:

Great post, but this would be even more relevant in the "Shootout at the Linux Corral: Audiolinux vs Euphony" thread, no?

 

 

excellent point...........off there i shall go

[altsouza]

On 7/6/2020 at 2:54 PM, austinpop said:

Roon Update Build 571 on Euphony OS

 

As many of you know, my primary music player is Stylus on Euphony OS. I play music both from local files stored on a NAS, and from Qobuz. I do from time to time use Roon, especially when I am in discovery mode, as one of my favorite activities is grazing in the Qobuz new classical releases section, and Roon is so much better as a user experience.

 

Of late, I've been feeling that the SQ with Roon streaming Qobuz has sounded very satisfying, so it got me wondering if this was just a feeling or whether there was something going on. I did a quick check with this album, which is identical on Qobuz and my local copy on NAS:

 

In all cases, I was booted into Euphony OS, version 20200430, in RAM root mode. I compared the SQ of the following:

• Roon streaming Qobuz
• Roon from NAS
• Stylus (with cache, buffer queue to RAM) from Qobuz
• Stylus (with cache, buffer queue to RAM) from NAS.

To my surprise, Roon from Qobuz sounded significantly better than Roon from NAS! Overall, the SQ progression was:

• Roon from NAS < Roon from Qobuz < Stylus from Qobuz < Stylus from NAS.

For the math-challenged: "<" refers to "less than" or really, "worse sounding than."

 

Also interesting was that the difference in SQ between Roon from Qobuz and Stylus from Qobuz was not that big. So I was not imagining what I heard with Roon from Qobuz.

 

Analyzing Roon network behavior

 

To investigate further, I looked at the network behavior in the Roon cases, using the nifty network monitor in Euphony (click on the word REGISTERED on the bottom right of the web UI). Here's what I observed.

 

Roon build 571's network behavior is different based on source:

• When does preload start: when you hit play
• Qobuz/streaming sources: 
  • download current track at network speed
  • appears to preload enough of next track to maintain gapless, but full track download appears to only happen when the track becomes current. This is different from Stylus (below).
• Local network files from a NAS or SMB share:
  • no preload!!
  • Appears to be doing normal file reads, which results in some read-ahead, but not a full download.
• Local files on SSD
  • did not test, but not relevant to network traffic.

In contrast, let's look at Euphony/Stylus. The behavior here is also different based on source:

• When does preload start: 
  • Either when you hit play, OR
  • Click on "Buffer Queue to RAM." This allows you to wait for network activity to quiesce before you hit play.
• Tracks from Qobuz/streaming sources: 
  • download current and next track at network speed
  • when next track becomes current, download the following track at network speed, and so on
  • In other words, at most 2 tracks (current and next) are buffered
• Local network files from a NAS or SMB share:
  • download all tracks at network speed, up to limit of RAM buffer (tunable option).
• Local files on SSD
  • did not test, but not relevant to network traffic.

More experiments with Roon file playback

 

To see how much of the Roon file playback degradation was due to my files being on NAS, and the Core's not preloading the entire file like it was doing on Qobuz, I compared the same track, and varied the file location:

• original NAS location
• local SSD
• in RAM.

To achieve the above, I first fired up Stylus and cached and buffered the album. Then switching back to Roon, I added the following data locations in the Storage settings:

 

Cached files are stored on my boot Optane SSD in /data/Music/E_CACHE, and the files buffered in RAM actually live in /tmp/EUPHONY_BUFFERED. The reason that the /tmp files are considered in RAM is because /tmp is a filesystem of type tmpfs in Linux, which is a flavor of RAM disk.

 

Listening results

 

As expected, the SQ was worst from the NAS location, and best in RAM. However, what is interesting is I still felt the playback from Qobuz sounded better. So in SQ terms:

• Roon from NAS < Roon from local SSD < Roon from in RAM < Roon from Qobuz.

My best guess is that even in the "in RAM" case, the Core is still making filesystem calls to access the file, whereas in the Qobuz case, the entire file is cached within the Core process (note: this is a guess, not confirmed).

 

Relating this back to the Taiko Extreme Roon "degradation" experience

 

The above finding of disappointing results with Roon file playback at least loosely correlates now to the complaints of Roon SQ degradation reported on the Extreme forum.

 

Emile is claiming, and Extreme owners are reporting, that his latest tunings mitigate the problem, and that Roon now sounds better than ever. This to me suggests that there are OS tunings possible that can close this gap. I have some ideas, but this is one of the huge challenges with Roon.

I have noticed the same buffering from Tidal in the recent days. Do tou think this applies also to Tidal?

[Nenon]

Buffalo switches - does the 8-port model sound like the 16-port model? Time to find out.

 

 

But first, some serious burn-in time needed.

[edwardsean]

I'm looking to add optical isolation to my USB chain. 

 

The Monoprice slimrun is hard to find. 

 

Is the one made by FIBBR essentially the same cable achieving the same sound quality? Or, is it compromised like the Corning model?

 

https://www.amazon.com/FIBBR-Female-Active-Extension-Optical/dp/B07BW91RHZ/ref=sr_1_4?dchild=1&keywords=FIBBR&qid=1594161696&sr=8-4

[edwardsean]

Also, I would be using this in-between a tx-USBUltra and a Chord Dave. Since both are USB 2.0 devices and the Monoprice and FIBBR are USB 3.0, will these not work properly? 

[cat6man]

1 hour ago, edwardsean said:

 

 

The Monoprice slimrun is hard to find. 

 

 

 

just recently back in stock.

got mine yesterday.

[edwardsean]

Did you get it from Monoprice?

 

Also, do you know if this works with USB 2.0 devices? 

[tgb]

3 hours ago, Nenon said:

Buffalo switches - does the 8-port model sound like the 16-port model? Time to find out.

But first, some serious burn-in time needed.

Hi Nenon,

How long was it for your 2016s ? How long is "serious time" :-) ?

My 2024 has been burning for 6 days now : still far from being acceptable, & far behind my previous network setup based on optical LAN with modded switch & FMCs.

 

The life of my GS2024, streaming music, is in your hands. In 2 days time, if it is still harsh, I get my money back and the Buffalo will go back to the vendor b4 ending its life in a datacenter or anywhere, but no more music to stream for it ! :-))

Rgds

[Nenon]

3 minutes ago, tgb said:

How long was it for your 2016s ? How long is "serious time" 🙂 ?

 

3 weeks. 

[Exocer]

1 hour ago, tgb said:

Hi Nenon,

How long was it for your 2016s ? How long is "serious time" 🙂 ?

My 2024 has been burning for 6 days now : still far from being acceptable, & far behind my previous network setup based on optical LAN with modded switch & FMCs.

 

The life of my GS2024, streaming music, is in your hands. In 2 days time, if it is still harsh, I get my money back and the Buffalo will go back to the vendor b4 ending its life in a datacenter or anywhere, but no more music to stream for it ! :-))

Rgds

 

@tgb, With the BS-GS2016, the bottom ports are preferred. Not sure which ports are audibly superior on the 24 port model. Which ports are you using?

 

I have a bone stock BS-GS2016P and I do not consider it to add harshness to the sound I get with my LPS 1.2 powered Etherregen downstream. This is only with the "proper" ports in use. The top ports sharpen the sounds too much.