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

[numlog]

Has it been established through listening tests how important the temperature stability of a clock is?

Some will argue high temperature stability isnt important at all for audio, 50ppm being the ''redbook standard''.

The highend clocks I seen were all OCXOs, but with insanely low jitter anyway.

 

The best affordable choice for low jitter clocks are some of the Crystek line, but there some similarly affordable TCXOs with much higher temperature stability and worse, but still good, jitter.

[sandyk]

24 minutes ago, numlog said:

Has it been established through listening tests how important the temperature stability of a clock is?

Some will argue high temperature stability isnt important at all for audio, 50ppm being the ''redbook standard''.

The highend clocks I seen were all OCXOs, but with insanely low jitter anyway.

 

The best affordable choice for low jitter clocks are some of the Crystek line, but there some similarly affordable TCXOs with much higher temperature stability and worse, but still good, jitter.

 

I got great results with my old highly modified Musical Fidelity X-DAC V3 by replacing the 50PPM oscillator with first a .3PPM TCXO , then a further small improvement with a .1 PPM 5V 24.576MHZ TCXO. The improvement was most marked with 24/96 and 24/192 , but not so much with 16/44.1 You could hear the SQ jump up a notch after a couple of minutes from switch on.

 A friend of mine reported the same improvement with his X-DAC V3 when he fitted my previous .3PPM TCXO

 I also powered them using a much lower noise power supply. A LT3045 PCB may work well there too, without overheating from a +12V supply as the current draw should be much less than 50mA ?  

[Iving]

1 hour ago, Dev said:

I am using NUC i7DNHE in a Akasa Plato xd case and I am hearing squeaking noise as if its running a fan inside if I put my ears close to case. This may sound a bit weird since there isn't any moving parts when I remove the fan to install it inside the fanless case. My other NUC CJY is dead silent like it should be. Does anybody has any experience like this ?

In a no moving parts pc a mobo whine near the cpu can be the result of antagonism between power settings in bios and the o/s respectively. I have changed mobos thinking them faulty or nffp before discovering this. Hope you have less trouble.

[mourip]

7 hours ago, Iving said:

In a no moving parts pc a mobo whine near the cpu can be the result of antagonism between power settings in bios and the o/s respectively. I have changed mobos thinking them faulty or nffp before discovering this. Hope you have less trouble.

 

When I first started building fanless audio PCs I had one that had an audible physical whine. Those boards have inductors which I always suspected as being the culprit by physical vibration/oscillation. I was able to fix it buy replacing the DC-ATX converter. Oddly the most noise came when I had a monitor hooked up to them and was using a mouse to work in the OS. Perhaps this caused more current to pass through them. The NUC does not have the same separate ATX converter but it probably has inductors on the board.

 

What particular power settings did you find as the culprit and were you able to successfully eliminate the noise by making adjustments?

[Iving]

24 minutes ago, mourip said:

 

When I first started building fanless audio PCs I had one that had an audible physical whine. Those boards have inductors which I always suspected as being the culprit by physical vibration/oscillation. I was able to fix it buy replacing the DC-ATX converter. Oddly the most noise came when I had a monitor hooked up to them and was using a mouse to work in the OS. Perhaps this caused more current to pass through them. The NUC does not have the same separate ATX converter but it probably has inductors on the board.

 

What particular power settings did you find as the culprit and were you able to successfully eliminate the noise by making adjustments?

Any "High" (or equiv. bespoke) Power Settings in Windows (CP/Power/Adv/Proc. Pwr. Management/Min proc. state) keep Min CPU state at 100%. If you have any kind of SpeedStep (Asus) or Power management Settings (e.g. C States etc) ON in BIOS they try to alter CPU state to <100%. The royal battle that ensues (coils working harder) is audible and fantastically annoying on an audiophile PC otherwise silent. I guess yes coils working harder/vibrating. Some folks say you can fix them with glue. But that of course would be hiding >1 problem for us. (I imagine NUC benefit is mainly small mobo limiting electronic turbulence btw.) Turning all Power settings in BIOS OFF is the solution for me. I imagine for someone else it could involve making sure Windows settings will be friendly with regulating CPU in BIOS i.e. vice versa to what I do. (But as I have said before in this thread I think low Power in the PC is a false religion and SQ is best when the CPU isn't throttled either way.)

 

My prev. post didn't assume Windows as you can see. If BIOS "struggles" against AL in parallel way - same point.

[Dev]

13 hours ago, Dev said:

I am using NUC i7DNHE in a Akasa Plato xd case and I am hearing squeaking noise as if its running a fan inside if I put my ears close to case. This may sound a bit weird since there isn't any moving parts when I remove the fan to install it inside the fanless case. My other NUC CJY is dead silent like it should be. Does anybody has any experience like this ?

 

I have narrowed down the issue with the Optane memory. Once Linux loads and starts accessing them, the squeaking noise starts. I tried playing with the power settings in the Bios but no luck. Not sure if its a board issue or something with the memory.

 

Can someone with Optane installed on i7 NUC confirm or deny if they hear any noise please ?

[Dutch]

29 minutes ago, Dev said:

 

I have narrowed down the issue with the Optane memory. Once Linux loads and starts accessing them, the squeaking noise starts. I tried playing with the power settings in the Bios but no luck. Not sure if its a board issue or something with the memory.

 

Can someone with Optane installed on i7 NUC confirm or deny if they hear any noise please ?

 

If it’s not a property of this type of memory modules (I can’t test since I have no Optane memory) the cause could also be the lack of sufficient power (causing a voltage drop under load). Does this also happen with the standard NUC power supply?

[Dev]

1 minute ago, Dutch said:

 

If it’s not a property of this type of memory modules the cause could also be the lack of sufficient power. Does this also happen with the standard NUC power supply?

 

Good point. It may not be the memory module itself since they are solid state devices but could be the electronics around it. Its powered by JS2 which I believe is more than sufficient. I haven't tried the NUC smps but will shortly. I have also ordered another one from Amazon to compare.

[Dutch]

15 minutes ago, Dev said:

 

Good point. It may not be the memory module itself since they are solid state devices but could be the electronics around it. Its powered by JS2 which I believe is more than sufficient. I haven't tried the NUC smps but will shortly. I have also ordered another one from Amazon to compare.

 

The JS2 should indeed more than powerful enough but indeed you could check anyway with the default PSU. I do know the thing called ‘coil whine’ if that’s what you hear (my music server (non-NUC) has it too though at a very low SPL) and mourip probably also referred to it. It could be hard to fix but if you Google the term you could perhaps find some strategies on how to deal with it.

[austinpop]

Mission Possible: The Quest for Optimal Glitch-free SQ with Roon/Squeezelite

 

Hey Everyone,

 

This is a call for Volunteers for an experiment I call Mission Possible: The Quest for Optimal Glitch-free SQ with Roon/Squeezelite.

 

Your mission - should you choose to accept it - is to explore the parameter space of squeezelite (SL) for buffer settings that both sound better than the default, AND do not cause the skipping, freezing, or gap-ful playback. You can do this on AL, Euphony, or whatever OS you happen to prefer.

 

To set some context: I revisited the settings I'd been using on AL: -b 2097152:2097152 -a 52428800:4:: because I wanted to evaluate the SL buffers setting on Euphony. Here are the key takeaways:

• Even with default settings, SL exceed Roon Bridge in SQ by a small degree
• That said, my custom buffer settings noted above do improve SQ much further
• There are 3 settings: -b <input buffer>:<output buffer> -a <ALSA buffer>:4::
• The settings I've been using, -b 2097152:2097152 -a 52428800:4::, correspond to:
  • input buffer IB = 2GB
  • output buffer OB = 2GB
  • ALSA buffer AB = 50MB
• I tried reducing these values, but sadly found that it had an immediate impact on SQ. The only value I could reduce safely with minimal reduction of SQ was output buffer to 1GB.

Now the next step is to understand at what values the SL glitches set in. That is going to take me some time to get to, so I am crowd sourcing it here!

 

Our goal is to find a set of values for (IB, OB, AB) that sounds better than default, but still functions flawlessly. Go forth and listen!

[mozes]

24 minutes ago, austinpop said:

buffer settings that both sound better than the default

Do you know what the default settings are?

[austinpop]

11 minutes ago, mozes said:

Do you know what the default settings are?

 

No, I don't.

[mozes]

I will try playing with the settings, but I am using LMS to SL so may not be applicable to Roon to SL.

[austinpop]

2 minutes ago, mozes said:

I will try playing with the settings, but I am using LMS to SL so may not be applicable to Roon to SL.

 

Interestingly, I observe no glitches with my SL settings with LMS - only with Roon. 

 

But do experiment... you may find even better SQ with different parameters.

[mozes]

Just now, austinpop said:

 

Interestingly, I observe no glitches with my SL settings with LMS - only with Roon. 

 

But do experiment... you may find even better SQ with different parameters.

Agree I have no glitches whatsoever 

[mozes]

Hi Rajiv

 

I am playing with the SL buffer settings but what I am finding so far is the opposite of you found. The smaller the buffer the more spacious the sound. I tried so far 3GB, 2GB, 1GB, 500MB and 100MB for both input and output settings. I left the ALSA buffer constant at 50MB. This is with LMS running with Win10 on my server and SL is running with Euphony on  the end point. I will be validating over the coming days.

[austinpop]

How much RAM on your endpoint? You need to have at least 8gb.

[vortecjr]

29 minutes ago, austinpop said:

 

Interestingly, I observe no glitches with my SL settings with LMS - only with Roon. 

 

But do experiment... you may find even better SQ with different parameters.

LMS has a configuration page set up in a matrix. One one side is locally supported formats and on the other there are possible formats supported by the various endpoint. This was done because the various SqueezeBox device supported different decoding and different sample/bit rates. The Roon guys bought all the SqueezeBox devices so they could know what is best to send to each. SqueezeLite is not a SqueezeBox device so it needs to send a generic stream. Anyway, my point is that a LSM stream might be every different than a Roon stream based on the rules in the configuration matrix. Using SqueezeLite with Roon is really a loop hole more than feature. 

 

I generally don't get support issues with the default settings in SqueezeLite which are nothing close to your settings. Real time Kernels are also notorious at causing dropouts and add to that your proposed settings and I'm not surprised you are getting drop outs. To me drop outs = bad sound quality. 

[ray-dude]

I remember when I was doing my SL experiments, I inexplicably was hearing differences between LMS and Roon as a server, but only for certain tracks.  I tracked it down to some transcoding that was happening (hidden in the LMS configuration matrix).  Once I turned off transcoding, order returned to the universe.

 

To echo the experience of others, I don't have any skipping with LMS and SqueezeLite, only Roon Core.  Alas Rajiv, my experience on how low I can take buffers very closely mimics yours: Lots of sensitivity to input buffers and ALSA buffers, but less so to output buffer.

 

As an aside, I'm currently running Roon Bridge on my NUC endpoint, but feeding it with two ISO Regen's in series (LPS 1.2 powering both).  That has hugely narrowed the SQ gap between Roon Bridge and SL with large buffers for me (almost equivalent).  Lots more experimentation to do, including rerunning a lot of my previous listening tests to see what may have changed as the rest of my chain has evolved, but an interesting hint (for me) of what may be going on with large buffers in SL.

[vortecjr]

3 minutes ago, ray-dude said:

I remember when I was doing my SL experiments, I inexplicably was hearing differences between LMS and Roon as a server, but only for certain tracks.  I tracked it down to some transcoding that was happening (hidden in the LMS configuration matrix).  Once I turned off transcoding, order returned to the universe.

What exactly did you do to turn off transcoding?

[mozes]

2 hours ago, austinpop said:

How much RAM on your endpoint? You need to have at least 8gb.

I have 8GB

[ray-dude]

7 minutes ago, vortecjr said:

What exactly did you do to turn off transcoding?

 

It was on the screen that had the different file types, with the transcoding target to different file types.  I had to go through one by one to make sure my main audio formats (flac, mp4, wav, dsf, etc) were getting passed through as native files.  

 

If helpful, I can fire up LMS and look for the specific admin screen (let me know).  Alas, I didn't find a switch to make a global change

 

[bobfa]

4 hours ago, austinpop said:

Now the next step is to understand at what values the SL glitches set in. That is going to take me some time to get to, so I am crowd sourcing it here!

 

Our goal is to find a set of values for (IB, OB, AB) that sounds better than default, but still functions flawlessly. Go forth and listen!

So my first test will be to compare Roon Bridge to SL with your buffer size set.  (Euphony OS).  I have 32GB of RAM on my NUC.  Do you think there is a specific direction that warrants research.  My DAC is only PCM so (YggY). Is this of any help to you?

[vortecjr]

20 minutes ago, ray-dude said:

 

It was on the screen that had the different file types, with the transcoding target to different file types.  I had to go through one by one to make sure my main audio formats (flac, mp4, wav, dsf, etc) were getting passed through as native files.  

 

If helpful, I can fire up LMS and look for the specific admin screen (let me know).  Alas, I didn't find a switch to make a global change

 

I think it would be helpful to others here. 

[austinpop]

11 minutes ago, bobfa said:

So my first test will be to compare Roon Bridge to SL with your buffer size set.  (Euphony OS).  I have 32GB of RAM on my NUC.  Do you think there is a specific direction that warrants research.  My DAC is only PCM so (YggY). Is this of any help to you?

 

Good questions. Let's start with the defaults. I did some poking around. Going by this file:  https://github.com/ralph-irving/squeezelite/blob/master/squeezelite.h

• #define STREAMBUF_SIZE (2 * 1024 * 1024)
• #define OUTPUTBUF_SIZE (44100 * 8 * 10)
• #define ALSA_BUFFER_TIME  40
• #define ALSA_PERIOD_COUNT 4

This suggests the defaults are:

• IB (inputstream buffer) is 2MB
• OB (output buffer) is ~3MB
• -a 40:4::

Note that the first value in the -a option has a change in units at 500. For 0-499 it has units of ms, and 500 or greater, it's in bytes.

 

So back to your question, Bob. We know that the high settings sound great, but are glitchy:

• input buffer IB = 2GB
• output buffer OB = 2GB
• ALSA buffer AB = 50MB

We know the defaults have lower SQ but no glitches:

• input buffer IB = 2MB
• output buffer OB = ~3MB
• ALSA buffer AB = 40ms

So the search space for best SQ while avoiding glitches lies between these values.