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

[flkin]

2 hours ago, Dev said:

 

Ok, got it. @flkin, are you using Pink Faun to up-sample ? if so, with HQP or Roon ?

Hi @Dev, I was using the Pink Faun 2.16x to upsample and tried with both Roon and HQP. It quickly becomes clear that HQP is the way to go as it offers a large selection of filters/dithers/modulators to choose from and although harder to use, I found it generally sounded better. These options matters and definitely affects the final outcome of the sound.

 

In a nutshell, upsampling tends to focus the sound. If you have soundscape images today that are large or fuzzy, then using upsampling will tend to make them tighter. This gives the impression of clarifying the soundstage and help with the 3D being there effect. Smeared background sounds will tighten up and may resolve to becoming a coherent and recognisable sound eg like a distant bell or soft timpani drum. 

 

If on the other hand your soundscape today is already very clear and focused, using upsampling may tighten the stage too much and make the music sound too thin. And the air sound the soundstage might tighten up and disappear giving the impression that the entire field has shrunken. If you like a larger, warmer enveloping sound, then upsampling might not work for you and reduce this to a clearer, perhaps more accurate field in front of you rather than around you.

 

So that said, if you still head towards HQP, there are a large set of combinations (not all of which work for all file types - which sometimes makes it maddening to use) to choose from. HQP requires a lot of computing resources to work fully and the largest filters today are still beyond my reach even though I have an AMD Rizen 8-core player. But the filters I have found to work well in my system are the following:

 

If you stick with upsampling in PCM (less loading on CPU):

 

Filters

poly-sync-ext2

poly-sync-xtr-lp (for desktop and not embedded HQP the same is ploy-sync-xtr)

closed-form

closed-form-M

Sinc-M

Dither

NS5

Gauss1 (an improved version of the TPDF which is the industry standard)

 

If you want to upsample and convert to DSD files (heavy loading on CPU):

 

Filters

poly-sync-ext2

poly-sync-xtr-lp

closed-form

closed-form-16M

Sinc-M

Modulator

ASDM7

AMSDM7 512+fs

 

 

The main difference between these filters are that some are apodizing and some aren't. The non-apodizing ones I have marked in italic.  These to me are incredible sounding but theoretically don't sound best with bad recordings (my experience is the are still pretty good anyway). But the apodizing filters like the 'sync' ones generally sound great with all types because they can apparently improve recordings with bad filters. So the above is a good place to start. It's likely that you may like a set of filters for the week and change you mind the following, I've done that countless times. It's pretty sensitive to system changes.

 

Right at the beginning I said "I was using..." and it's because over time I am starting to move towards not upsampling and staying with NOS straight out from Roon. I am finding that over time, my Pink Faun 2.16x is becoming more focused by itself for whatever the reason. I suspect it's to do with the clocks inside which has taken quite some time to settle in after burning. So take upsampling as a variable to be checked and changed from time to time with other system changes and not a total solution by itself.

[flkin]

I have a retraction and apology to make regarding what I mentioned during a review of the Pink Faun 2.16x streamer I made a month ago.

 

At that time, I felt that convolution was a good tool to complete the sound by managing room frequency and phase issues and recommended it's use. It certainly has its advantages of removing the room from the equation (especially bass nodes) and making the sweet spot larger as I had described. At that time I was still adjusting the filters and had not come to a final set but felt it was pretty close to what I wanted.

 

But over time and longer listening, I started to question my earlier findings on whether there was a downside to using convolution. I started to feel that the dynamics and immediacy of the music was affected. Camouflaged by the more solid bass and less muddied lower mids that convolution offered, the downside effects were beginning to be clear to me.   

 

 @romaz commented to be wary of using convolution as he found that soundstage and emotional impact was affected. Was it this that I was hearing?

 

So I started a thread in Roon communities https://community.roonlabs.com/t/convolution-with-null-filters-are-audible/51732/39 to see if I could isolate the problem to the process of convolution rather than the adjustment of frequency/phase. I wondered if the action of making adjustments to the music file caused the defects I heard heard. By using a NULL convolution file I found I could still hear the difference with and without convolution and the defects are similar to what I heard with full frequency/phase adjustments. 5 others there found the same. So it might be the process that may be causing the sound defects not the adjustments. 

 

 

Given this, at the present time, I now feel that convolution cannot be used as a way forward to push the boundaries of sound quality. Although it does proved massive improvements in certain parts of the sound spectrum, the defects it adds to my system are audible and not subtle enough to ignore. Back to physical room changes.... ?

 

I've started another thread here to further discuss this and see if there is a way around the sound defects caused by convolution. https://www.computeraudiophile.com/forums/topic/54773-convolution-with-null-filters-are-audible/

 

[flkin]

41 minutes ago, diecaster said:

 

This is absolutely the case. Even with all filters and noise shapers turned off, Roon/HQPe/NAA sounds better than Roon/RAAT. I would love a technical explanation as to why.

 

I've always found HQP to sound better than Roon but recently that's changed. 

 

I believe believe once you reach a certain level, touching the digital signal as little as possible is better than manipulating it for whatever the reason.

 

Roon/Raat is now exceeding HQP for me.

It's system dependent. 

[flkin]

9 minutes ago, Dev said:

 

So how do you reach a certain level and what do you exactly mean by touching the digital signal as little as possible ? We are no talking about up-sampling neither in Roon nor in HQP here.

 

We all trying to push our systems higher, everyone have their own way. With my Pink Faun single box with a customised Audiolinux installation and internal clocks, I'm at the stage where my tX-USBultra and clock makes no difference anymore, actually makes the sound worse.

 

I'm finding that Roon with no upsampling now sounds better than HQP both with or without upsampling. And since both sounds different without upsampling, the signal is being processed by HQP in some way. I'm suggesting to handle the signal as little as possible.

[flkin]

 

14 minutes ago, diecaster said:

 

My PS Audio DirectStream DAC has test you can do with a specific test track to see if the track is being sent to the DAC completely unaltered. I turned off all the upsampling, filters, and noise shapers in HQPe and played the test track. The DirectStream reported on its display that the track was "Bit Perfect" which means the track is *NOT* being processed by HQPe in any way.

 

 

Can you explain how that works? What are you comparing a track played by hqp  to? 

 

Is the same test possible for Roon? 

 

That would be interesting to see what your DAC says

[flkin]

30 minutes ago, Dev said:

 

Ok, I see. I think, however, which sounds better also has to do with the DAC in the system. I have noticed my use case with two DACs and they both sound better with HQP/NAA but the comparison was made before I had the JCAT net card. I will have to repeat the experiment this weekend. I doubt the result will change but who knows ?

 Very true, for sure the DAC matters and the rest of the system too. I've been using HQP on and off for many years. It does solve an irritating bump in sound with Roon when it switches formats from DSD back to redbook.

 

I'm in my off cycle ? and suspect this might last for a while 

[flkin]

2 minutes ago, diecaster said:

It's a test file that the DirectStream DAC is coded to see and report on the DAC display if it is being seen as bit perfect. Yes, it would work with Roon too and I have tested it in the past. It also reports as bit perfect in Roon if I turn off all DSP functions including volume leveling.

 

Does that mean that Roon and HQPlayer should sound the same if we keep DSP off?

 

I'm finding it doesn't. I'm going to have to revisit this again or see a shrink ?

[flkin]

1 hour ago, greenleo said:

What are the clocks that you're using?

The 3 OCXO clocks are inside the streamer for the motherboard, processor and USB cards. The 4th clock on the LAN card is on the way

[flkin]

4 minutes ago, LTG2010 said:

This is true, but the digital signal is manipulated quite a bit in mastering processes, engineers tweaking the recordings to their taste and equipment, some of those studio monitors very unbalanced. My Audeze LCD XC headphones just sound too bright in my setup so a little dsp in Roon helps balance this out.

 

Its true the signal before arriving at the streamer has been massaged a lot for sure. My thinking is to arrange my system play whatever was the intent of the  mastering engineers. Might be different to what I prefer but it's true to the intent and that makes it interesting to hear and sometimes reflects the era of the recording too. I'm thinking this is achieved with a truly neutral system and then everything sounds good. But for sure a bit of DSP adjustment to correct issues in the rest of the system is probably the only solution before finding more permanent solutions.

 

One thing that is interesting to read is how we talk a lot about the front end sound but the streamer and DAC is only part of the entire system and there are so many more parts to optimise apart from the source. If we don't work on this at the same time we might be working on producing defects to compensate for other defects in the rest of the system. - and - to make a + sort of thing.

[flkin]

2 hours ago, greenleo said:

I c.  Thought you used Ref 10 or cybershaft.

 

Yes I have used both with the external tX-USBultra to distribute the clock goodness in a spaghetti setup. This was before moving to a single box streamer with clocks built in. Initially during the burning in process the external clock still helped. But after the Pink Faun settled down, I found that the internal clocks had exceeded the sound of the external reference clocks didn’t need them anymore.

 

The discussion of using a NUC as a Roon end point has got me interested in using external clocks again. I wonder if Larry can assembe a AL unit that accepts an external reference clock signal? Even Jord had said ultimately a two box solution was best. He uses two 2.16 machines as the streamer ?

[flkin]

17 hours ago, greenleo said:

Thanks for the info flkin.  May I know which clock did you use, the Ref 10 or the Cyber and in which grade?

 

For the mod of the mobo to change the clocks, is it the regime of the SOtM?  I look forward to hearing your mod results.

 

Regards.

 

I used both. The Mutec ref 10 was on loan and the Cybershaft is the OP14, one of the last to be made apparently.

 

SOtM clock mods generally use their distribution board (the sCLK-EX) to split an external clock ref signal into what ever frequencies you need. Up to 4 different output frequencies can be programmed. So say a streamer had the sCLK-EX board inside you would still need to feed it from an external clock. In my case the clocks are actual OCXO clocks at the final use frequency and attached to the location they are required. No reference clock signal nor frequency programming is needed. It's part of Pink Faun's streamer design to offer these as optional clocks and wasn't a mod as such.

 

There are others offering similar devices, like the JCAT LAN or USB femto cards with good clocks on board already. The motherboard and processor clocks are more tricky and was matched specifically for the AsRock board and AMD Rizen processor Pink Faun used. I suspect the Audiolinux operating system and even bios was tweaked for this. I say this because apparently others like Innuos and The Linear Solution  have tried to reclock the processor and had faced issues with stability.

[flkin]

One can always add an external fan to suck out the hot air. Powered separately, it would add no electrical noise. And the fan at the low setting is very quiet, basically inaudible at normal listening distances.

 

https://www.amazon.com/gp/product/B012CL2V3I

[flkin]

7 hours ago, lmitche said:

How do you know this? Have you heard this solution?

 

Yes I have ? and it's good! ?

 

7 hours ago, greenleo said:

..

flkin posted his impression here.  Incidentally, I read his PF thread (each of the posts) few days ago that started at around June 2018, much earlier than our very heated discussion in this thread on the AL+NUC solution recently.

...

 

I don't mean to take credit for the recent thread discussions about Piero's AudioLinux as AL has been around for a long time, but it's one of the design features in the Pink Faun 2.16x streamer that makes it sound sooo good. And this was described in detail in both my reviews of the PF where I explained that Jord worked with Piero to create an optimised version of the AL for the streamer that emphasised on low latency. Actually Piero talks about this in his website also shown below (didn't ask for permission to clip this, hope it's ok).

 

 

I'm hopeful I contributed to this thread by kickstarting the AL cause or at least nudging it along in the early discussions but even if not, glad to see the benefits people are experiencing now.

 

Larry, you've got me interested in seeing how your NUC/AL might server as an Roon endpoint with my PF as a streamer. I already tried it with the SonicTransporter i5 (and the usual ultraRendu, sMS-200 etc) but it didn't work out. Perhaps the difference is in the OS. I wonder how much it would cost to build one with say the SOtM sCLK-ex card inside with a ref clock input? I'm loath to switch back to an end point without clock options.

 

But it's true that a double box 2.16 would sound better - I don't have personal experience of this - but Jord informs me that this is how he plays his own system today and recommends it. I wonder how many people could go this route given the costs! ? 

 

What's interesting is what Pink Faun is planning but still sometime to go :

 

Quote

• "We're planning a high-end build of the Scion within a couple years, this high end version can be used right after the streamer 2.16, sort of a dedicated dual setup!"

A high end Roon endpoint based on an extreme version of their Scion. An ultra-capacitor version matched to their 2.16 streamer. It would still have to be reasonably priced to make it work though.

 

[flkin]

On 11/14/2018 at 12:24 PM, flkin said:

I have a retraction and apology to make regarding what I mentioned during a review of the Pink Faun 2.16x streamer I made a month ago.

 

At that time, I felt that convolution was a good tool to complete the sound by managing room frequency and phase issues and recommended it's use. It certainly has its advantages of removing the room from the equation (especially bass nodes) and making the sweet spot larger as I had described. At that time I was still adjusting the filters and had not come to a final set but felt it was pretty close to what I wanted.

 

But over time and longer listening, I started to question my earlier findings on whether there was a downside to using convolution. I started to feel that the dynamics and immediacy of the music was affected. Camouflaged by the more solid bass and less muddied lower mids that convolution offered, the downside effects were beginning to be clear to me.   

 

 @romaz commented to be wary of using convolution as he found that soundstage and emotional impact was affected. Was it this that I was hearing?

 

So I started a thread in Roon communities https://community.roonlabs.com/t/convolution-with-null-filters-are-audible/51732/39 to see if I could isolate the problem to the process of convolution rather than the adjustment of frequency/phase. I wondered if the action of making adjustments to the music file caused the defects I heard heard. By using a NULL convolution file I found I could still hear the difference with and without convolution and the defects are similar to what I heard with full frequency/phase adjustments. 5 others there found the same. So it might be the process that may be causing the sound defects not the adjustments. 

 

 

Given this, at the present time, I now feel that convolution cannot be used as a way forward to push the boundaries of sound quality. Although it does proved massive improvements in certain parts of the sound spectrum, the defects it adds to my system are audible and not subtle enough to ignore. Back to physical room changes.... ?

 

I've started another thread here to further discuss this and see if there is a way around the sound defects caused by convolution. https://www.computeraudiophile.com/forums/topic/54773-convolution-with-null-filters-are-audible/

 

 

A quick note to inform that a bug in Roon’s convolution module has been found in the handling of resampling impulse response and acknowledged by Brian at Roon Labs. Details in the link above. Once fixed, I’ll revisit Roon’s convolution again, perhaps it’s still an option after all. ?

[flkin]

6 hours ago, afrancois said:

The system goes automatically in trial because the license is not recognized.

 

 

Did you change the Bluetooth or Wan/wifi settings? This helped a license issue I had with HQP embedded. 

[flkin]

6 hours ago, seeteeyou said:

http://www.audio-linux.com

 

https://www.computeraudiophile.com/forums/topic/43629-pink-faun-216/?page=5&tab=comments#comment-886404

 

CX+EX combo wasn't doing too well despite their rather "special" motherboards from Asus, though the operating system itself might be nothing special

 

https://antipodes.audio/server-upgrades.html

 

Yeah, I did a comparison of my Pink Faun 2.16x (uses a PF version of AudioLinux) with a Antipodes Cx+Ex in a detailed review a while ago - link is above, not in the index. 

 

I believe that Antipodes discovered the benefit of 2 boxes a long while ago and has been experimenting to optimise different custom boards for server and renderer functions along with adding good power to it. It feels to me they have taken it to the next step compared with the simpler NUC. I understand they even have a placement strategy for the boxes which optimises for EMI during playback. The CxEx plays to a high level but didn't manage to top my PF during the tests. 

 

Not sure if we can extrapolate this to how a NUC with AL would sound against the CxEx combo. 

 

I am converting my SonicTransporter i5 into an AudioLinux Roon endpoint to see how that compares with my all in one streamer under AL. I didn't have much success with the STi5 with the original SonicOrbitor OS which was clearly worse. 

 

@Kritpoon is experimenting to try and convert his Ex box to a AL endpoint to see if it's the OS in the CxEx combo that is the bottleneck now. The CxEx claims to use their own OS but Khun Aek suspects it might be only a tweaked version of the SonicOrbitor.

 

[flkin]

Will do, thanks for the suggestion.

 

Did you ever try it as an Roon endpoint rather than a file server? As a file server, where did you keep the music?

[flkin]

There's a difference between bass-shy and thinner sound which can sometimes come across as bass-shy. Might be worth considering this?

 

A few things I have found in the past that causes thinner sound are insufficient power, silver wires and HQPlayer.

[flkin]

Removed

[flkin]

Please be very careful when you set your pre-amp to 100%!  😮

 

Very easy to blow up your speakers and ears if you are not careful...

[flkin]

On 2/10/2019 at 2:16 AM, auricgoldfinger said:

UPDATED COMPARISON:  sMS-200ultra vs. NUC7PJYH

 

Today I upgraded my sMS-200ultra to the latest Eunhasu image v.0.4.56 for the purpose of comparing this new and improved version to my NUC7PJYH.  By way of reminder, my sMS-200ultra has some unique attributes that are described in my initial comparison to the NUC7PJYH.

 

 

Since that time, I have made the following changes to my chain:

 

1. I removed the FMCs. 

 

2. I upgraded the tX-USBultra to the “SE” version with the Evox capacitor.  In addition, I further improved the internal DC cable from the SOtM silver wire upgrade to a cable made from a pair of 20 AWG solid core Neotech silver wires with JSSG 360 shielding. (The standard SOtM silver DC upgrade uses unshielded 22 AWG stranded silver of lesser quality.)  Although it is outside the scope of this review, I would like to note that the tX-USBultra now sounds like a completely new device which is vastly superior to the original model.  The shielded 20 AWG solid core silver Neotech internal DC cable has lowered the noise floor, which has the effect of magnifying the improved dynamics and musicality arising from the Evox capacitor.

 

3. I purchased a Sony SR-64HXA “low noise” microSD card for the sMS-200ultra. 

 

I began by listening to the sMS-200ultra.  The latest image has made a nice improvement to the sound quality.  Detail and dynamics are better than I remembered them.  My initial reaction was that the gap with the NUC7PJYH had been closed.

 

When I switched to the NUC7PJYH, however, I heard an obvious difference in sound quality.  I only needed to listen for a few seconds to know that the NUC7PJYH still outperforms the sMS-200ultra in terms of detail and dynamics.  I also feel like the NUC7PJYH has more body, better tonality, and sounds more natural overall.

 

 

In my experience silver wires generally produces thinner, harder less natural sounds and might be the reason the tx sounds this way compared to your NUC?

[flkin]

17 hours ago, auricgoldfinger said:

 

I'm not really sure I understand your comment since the sMS, not the tX, is being compared to the NUC.  In any event, my experience with high quality silver wire is exactly the opposite of what you describe.  While some people may prefer the tonality of copper over silver, I find your characterization of silver producing thinner, harder less, natural sounds to be completely inaccurate.  Perhaps lower quality silver wire has these attributes, but I have no such wire anywhere in my system.

 

Ah, ok so the tx was not paired with the sMS200 and then compared with the NUC without tx. Not sure why I mis-read that.. In that case you are right, the silvered tx wouldn't have any effect on the sound at all. My apologies @auricgoldfinger ☺️

 

[flkin]

Hi guys, I’ve had a chance to install the new extreme USB clock board in my Pink Faun 2.16x streamer replacing the older one. The new clock improves the phase noise by -22dbC/Hz.

 

My impressions of the new sound is here :

 

 Have a read if you’re interested. Thanks.

[flkin]

My single Pink Faun 2.16x with 4 extreme OCXO clocks sounds pretty good to me. Can’t imagine what another external box would bring apart from more CPUs, power supplies, connectors, jitter and noise.

 

The new extreme clocks really made a difference - subjectively I’d say something like 30% more pleasure. Not necessary to upsample with the clocks anymore. Everything is so clear and natural already, I’m finding upsampling now only brings a sense of unnatural massaging to the music.

 

One thing about a better clocks, i’ve never found them to have any downsides, only adds to the sound.

 

If you wanted to separate the audio functions, streamer from renderer, the split CPU cores functions in AL has that option. Shocking what details can be extracted with this function but unfortunately the overall presentation becomes too hard and digital sounding for long time listening. 

[flkin]

4 hours ago, cool_chris said:

So tell me why the manufacturer used TWO Pinkfauns 2.16x at Munich show ?

I am sorry to tell you but your imagination has nothing to do with what is real.

Also this is the pace were we post What we know .

 

 

I have no idea why Jord used two 2.16x machines although he did tell me that's the next upgrade. He might be right but my pocket has difficulty believing it 😅

 

I repeat, I really can't imagine how much better things can get with another set of components massaging the signal. 😄😂