Measured evidence that bit perfect playback software alters the analog output of DAC's?

[esldude]

There are many playback softwares available. They are said to sound different.

 

Some load everything into RAM for playback or make that an option. Some reduce activities of the source computer or have multiple options in that regard. It is said playback software sounds different with different settings. All of these I am referring to can be bit perfect or bit correct. Meaning the bits sent to the DAC are exactly the same as the source file regardless of the various settings, activities of the source computer, or playback software in use. Which leads me to wonder why they sound different.

 

Many ideas have been conjectured like software jitter, noise from the activities of the computer, etc. etc. What I haven't seen are measurements showing the analog output of the DAC has changed in an audible manner. Hopefully we can agree if the sound you hear physically changes then the signal output of the DAC must have changed or been interfered with. So if you know of such measurements that corroborate such changes I would like to learn about it.

 

I am not looking for anecdotes of sighted listening impressions. I am discounting them to keep the thread focused, clean, civil and useful. If you disagree with me discounting such evidence then fine, but don't post here to argue about it. My goal is to document and understand what kind of changes in sound quality are caused by playback software. Then one could avoid or optimize the various factors for best playback quality.

 

If you still wish to argue the relative merits of measures vs ears please start your thread on that topic.

 

Thank you in advance.

[Miska]

I didn't try measuring strictly between playback softwares because I focus mostly on my own. But I've been measuring between different playback hardwares and there are measurable differences. (same software and same DAC, different computer hardware)

[acg]

Nice thread Elsdude but I doubt you will get many measured differences at the output of the dac unless guys like Miska and PeterSt want to do some work.

 

Miska, I think that your hardware change affecting the dac output is very relevant and I will try to explain why. PeterSt has written this playback software for windows called XXHighEnd. There are many different settings that can be adjusted or toggled and many of them influence the behaviour of the computer. I have recently embarked on a power supply project for my w8 computer and during the first steps I measured the amperage draw of ALL of the ATX rails into the P24, P8, SSD, USB Card etc. whilst running XXHE in some of its various 'states'. Sound quality wise, one of the biggest toggles in the software is the "Minimise OS" button that shuts down just about all of windows services to leave a core that is the bare minimum to run the Playback Engine and the Library software. When you press play XXHE shuts down even more services to only leave those required for the Playback Engine. Here are my measurements for amperage draw of the cpu before and after the os has been minimised:

 

[TABLE=width: 500]

[TR]

[TD]O/S Mode[/TD]

[TD]Computer Startup[/TD]

[TD]System Idle[/TD]

[TD]Playing Music[/TD]

[TD]Computer Shutdown[/TD]

[/TR]

[TR]

[TD]NOT-Minimised[/TD]

[TD]3.78A peak[/TD]

[TD]1.15A peak[/TD]

[TD]3.3A peak[/TD]

[TD]4.85A peak[/TD]

[/TR]

[TR]

[TD]Minimised[/TD]

[TD]3.74A peak[/TD]

[TD]0.9A peak[/TD]

[TD]1.2A peak[/TD]

[TD]2.62A peak[/TD]

[/TR]

[/TABLE]

 

 

You will notice that minimising the os the way XXHE does affects the peak cpu draw everywhere but at startup. When playing music the draw is almost one-third lower with the minimised os. What these numbers do not show you is how the power cycled when playing music. Not minimised the draw when playing music was rythmic and cycled as low as 1.2A and then generally up to 2.5A and occasionally up to 3.3A: I could see my lab psu readout cycling in a reasonably reliable pattern. With the os minimised, power draw during playback cycled from 1.05A to 1.2A (0.15A compared to a swing of 2.1A). Since these measurements were taken I have underclocked the cpu and changed some other settings in XXHE and the draw is even more stable and does not break 1.0A at any stage during playback.

 

Do I need to tell you whether the minimised os sounds better or not? You can guess what I think. How XXHe manages the hardware (the cpu in this case) subjectively changed the sound (I cannot measure the output of the dac) and in this situation I am convinced that it is largely due to more stable processor activity (no spikes - low power draw) and a theoretical reduction of noise that can make its way into the usb cable or to ground or both. The bits don't change: I can watch the dac driver control panel and there are zero KS/WASAPI errors, zero FIFO errors so the dac is getting a bit-perfect feed.

 

I hope this helps the conversation, and Miska I am interested in your measurements at the dac output due to changes in computer hardware.

 

Cheers,

 

Anthony

[Superdad]

Good post Anthony, but I think Dennis is looking for someone to measure differences in the ANALOG output of the DAC--based on changes in software. A tall order, but based on the sometimes radical differences I have heard, I bet that some things CAN be measured. I just don't expect the differences to be in something simple like frequency response. Transient response and other temporal differences are easy to hear but a bit harder to measure.

 

Regards,

Alex C.

[One and a half]

Good post Anthony, but I think Dennis is looking for someone to measure differences in the ANALOG output of the DAC--based on changes in software. A tall order, but based on the sometimes radical differences I have heard, I bet that some things CAN be measured. I just don't expect the differences to be in something simple like frequency response. Transient response and other temporal differences are easy to hear but a bit harder to measure.

 

Regards,

Alex C.

 

To measure the analog output for a number of terms, THD, IMD could be accomplished with simple sine waves. This won't be enough to compare to music signals, you would have to measure missing bits out of the various caches or number of retries. Even if you could quantify that, how does this influence the character of the sound output, would it result in grain, or detail loss, and if so by how much and how could you define the yardstick to begin with. Should THD be 0.000001 or 0.0000001 ? Miska might produce many graphs with noise figures and various waveforms that validate measurement difference between software players, but what if any result would that benefit other than to give hard work away.

 

Specifications and numbers are difficult to translate to sound quality, always have been and in this case it's no different.

 

Software is easy to install if you like it, keep it and pay for it. Remove software if you don't like the results you hear, just apply the same criteria to the other parts in the audio chain that works for you.

[bibo01]

I believe measuring different media players that are all bitperfect at DAC output is a fools errand using current tests commonly done. I think that there is not much point to measure static signal in the analog domain. Furthermore, it is complicated to setup a testing system. When testing media players you may understand more about the differences they make if you use a high grade, high speed multichannel 'scope to monitor the power supply lines (and ideally an RF analyser attached at various ports) to observe if there are differences in RFI and PSU Noise, than looking at the DAC output.

 

Please take a look at this null test difference system which was setup by a friend of mine (it's a Google translation). It has some pros and cons, and some limitations (Delta Sigma ADC), but it is still a tool. If you find it interesting, I can post his test on 4 common players.

[esldude]

I believe measuring different media players that are all bitperfect at DAC output is a fools errand using current tests commonly done. I think that there is not much point to measure static signal in the analog domain. Furthermore, it is complicated to setup a testing system. When testing media players you may understand more about the differences they make if you use a high grade, high speed multichannel 'scope to monitor the power supply lines (and ideally an RF analyser attached at various ports) to observe if there are differences in RFI and PSU Noise, than looking at the DAC output.

 

Please take a look at this null test difference system which was setup by a friend of mine (it's a Google translation). It has some pros and cons, and some limitations (Delta Sigma ADC), but it is still a tool. If you find it interesting, I can post his test on 4 common players.

 

By four common players, do you mean common software players? If so yes, I am interested.

 

Again, in other replies I am seeing lots of we probably could, or it changes PS noise etc. If it doesn't change the DAC analog output there is nothing for us to hear different. In some instances one can change things that change sound output not one little bit. I am looking for a change in analog output that we can hear between players. Whether that is change in conventional measures or nulling tests or any such change at the DAC output.

 

Transient and temporal response are quite easy to measure with difference testing.

 

Still looking for measurements that show those differences in playback software.

[bibo01]

By four common players, do you mean common software players? If so yes, I am interested.

 

Again, in other replies I am seeing lots of we probably could, or it changes PS noise etc. If it doesn't change the DAC analog output there is nothing for us to hear different. In some instances one can change things that change sound output not one little bit. I am looking for a change in analog output that we can hear between players. Whether that is change in conventional measures or nulling tests or any such change at the DAC output.

 

Transient and temporal response are quite easy to measure with difference testing.

 

Still looking for measurements that show those differences in playback software.

Then perhaps you wouldn't like the result of the following test...

 

It is all in the first message. To properly understand the graph you may need to look at the explanation on the testing system in my previous message.

However, a few things have to be kept in mind: the server was a highly optimized PC, both as OS and, most importantly, power supply, so results on a normal PC may differ; the spectrum is 20-20kHz.

 

 

BTY, I don't think this is the final word on media players differences.

[esldude]

Then perhaps you wouldn't like the result of the following test...

 

It is all in the first message. To properly understand the graph you may need to look at the explanation on the testing system in my previous message.

However, a few things have to be kept in mind: the server was a highly optimized PC, both as OS and, most importantly, power supply, so results on a normal PC may differ; the spectrum is 20-20kHz.

 

[ATTACH=CONFIG]10015[/ATTACH]

 

BTY, I don't think this is the final word on media players differences.

 

 

Okay, but first an explanation of what is being graphed please?

 

And reading the link from the earlier post, was this not a difference test with two different DAC's?

 

I am looking for differences where the DAC and upstream hardware is unchanged, but the software is the only difference.

 

Differences where the DAC is unchanged and the feeding computer is changed would be interesting too.

 

It also appears in the link given above the test is comparing different rate MP3 files. Those would not be bit perfect comparisons. Or am I missing something?

[Miska]

Sound quality wise, one of the biggest toggles in the software is the "Minimise OS" button that shuts down just about all of windows services to leave a core that is the bare minimum to run the Playback Engine and the Library software.

 

Rather than doing that, I decided to separate the actual audio reproduction away from the player software to a dedicated device running minimal Linux and no display output or other interfaces than network and USB. So the player software is kind of split out to separate "player" and "audio output engine". Another reason was that USB has a problem that it is not easily isolated (especially ground connection), while copper ethernet is transformer isolated and it can be further isolated by using optical ethernet. USB can be isolated at the I2S side, but it is not very frequently done.

 

If DAC doesn't have galvanic isolation, I would say it is more likely to have impact with different computer hardware and software configurations. Just compare "with galvanic isolation" and "without galvanic isolation" results here

exaSound Audio Design > e20 DAC > Measurements

...and you could assume that without the isolation there could be significant differences... And with the isolation I would find it unlikely that there would be notable differences between player computers or software (when bit-perfect).

 

I hope this helps the conversation, and Miska I am interested in your measurements at the dac output due to changes in computer hardware.

 

Generally I don't like posting these kind of results, because they are always so much dependent on the combination of playback computer hardware, DAC and everything else like peripherals connected to the computer.

 

But here are two examples. I don't know if it's good idea to post these, but oh well... I just want to emphasize that you should not and cannot draw any generic conclusions from any set of measurements. The results apply only to that particular configuration.

 

With direct connection (without NAA):

 

With split-out configuration (with battery powered NAA):

 

(this is DAC's noise floor when playing back dithered silence)

[esldude]

Rather than doing that, I decided to separate the actual audio reproduction away from the player software to a dedicated device running minimal Linux and no display output or other interfaces than network and USB. So the player software is kind of split out to separate "player" and "audio output engine". Another reason was that USB has a problem that it is not easily isolated (especially ground connection), while copper ethernet is transformer isolated and it can be further isolated by using optical ethernet. USB can be isolated at the I2S side, but it is not very frequently done.

 

If DAC doesn't have galvanic isolation, I would say it is more likely to have impact with different computer hardware and software configurations. Just compare "with galvanic isolation" and "without galvanic isolation" results here

exaSound Audio Design > e20 DAC > Measurements

...and you could assume that without the isolation there could be significant differences... And with the isolation I would find it unlikely that there would be notable differences between player computers or software (when bit-perfect).

 

 

 

Generally I don't like posting these kind of results, because they are always so much dependent on the combination of playback computer hardware, DAC and everything else like peripherals connected to the computer.

 

But here are two examples. I don't know if it's good idea to post these, but oh well... I just want to emphasize that you should not and cannot draw any generic conclusions from any set of measurements. The results apply only to that particular configuration.

 

With direct connection (without NAA):

[ATTACH=CONFIG]10016[/ATTACH]

 

 

 

With split-out configuration (with battery powered NAA):

[ATTACH=CONFIG]10017[/ATTACH]

 

(this is DAC's noise floor when playing back dithered silence)

 

So are these examples you posted with or without galvanic isolation? And what is NAA?

[esldude]

Then perhaps you wouldn't like the result of the following test...

 

It is all in the first message. To properly understand the graph you may need to look at the explanation on the testing system in my previous message.

However, a few things have to be kept in mind: the server was a highly optimized PC, both as OS and, most importantly, power supply, so results on a normal PC may differ; the spectrum is 20-20kHz.

 

[ATTACH=CONFIG]10015[/ATTACH]

 

BTY, I don't think this is the final word on media players differences.

 

Sorry in my first reply to you, I looked at your first link, and not the second one.

 

Am I understanding right that it shows no difference in the 4 software players to speak of?

 

I am curious, I have done difference testing with two of the 4 and gotten lower nulls than shown. Still if I understand the explanation of that post, he is showing the results of playback from 4 different software players and they all overlap except at the very lowest levels when level and time is compensated for. I am taking from that the residual is primarily the noise in the test system.

[Miska]

So are these examples you posted with or without galvanic isolation? And what is NAA?

 

This is using MuFi V-Link192 and to DAC using AES cable. So AES is transformer isolated, but the AES cable had pin 1 connected and thus ground is not isolated. (I have since cut out pin 1 wire from the DAC side cable connector, but didn't redo this particular measurement setup)

 

NAA (Network Audio Adapter) is a split-out minimal Linux that I explained in the upper paragraph. In this case it was running on a BeagleBone (White).

[bibo01]

Sorry in my first reply to you, I looked at your first link, and not the second one.

 

Am I understanding right that it shows no difference in the 4 software players to speak of?

 

I am curious, I have done difference testing with two of the 4 and gotten lower nulls than shown. Still if I understand the explanation of that post, he is showing the results of playback from 4 different software players and they all overlap except at the very lowest levels when level and time is compensated for. I am taking from that the residual is primarily the noise in the test system.

You are correct.

 

Like Miska, I emphasize that the results apply only to that particular configuration.

 

NAA stands for network audio adapter and it's usually a tiny computer box that acts as a buffer connected by ethernet to main server and by USB to DAC. In my tests also the quality of NAA's power supply affects the sound.

[esldude]

This is using MuFi V-Link192 and to DAC using AES cable. So AES is transformer isolated, but the AES cable had pin 1 connected and thus ground is not isolated. (I have since cut out pin 1 wire from the DAC side cable connector, but didn't redo this particular measurement setup)

 

NAA (Network Audio Adapter) is a split-out minimal Linux that I explained in the upper paragraph. In this case it was running on a BeagleBone (White).

 

So your graphs are the noise floor of the DAC output?

 

And what is the difference between the two, just the software player?

[mayhem13]

Something to consider.....if the measurements Dennis is looking for DO NOT exist, how did the software developers engineer, proof and finalize the software players? Did they simply 'listen' to different configurations and decide one sounds better than another?

[Miska]

So your graphs are the noise floor of the DAC output?

 

And what is the difference between the two, just the software player?

 

Yes. Player itself is the same, but in the second case the output part it is split out to a separate device. So while the player runs on Windows, in second case the output is coming from ARM board running Linux and only a minimal output software.

[Miska]

Since both are well below the threshold of audibility, I guess what you're saying is it doesn't make any difference.

 

I wouldn't claim anything about threshold of audibility. This is anyway only for one case and the differences tend to have side-effects like changes in jitter characteristics and then threshold of audibility of that is yet again subject to another debate. So I rather go objectively and see if I can make measurable difference and how. And when I've exhausted this measurement method I already know how to dig out even more from seemingly pure noisefloor.

[sbgk]

"My goal is to document and understand what kind of changes in sound quality are caused by playback software. Then one could avoid or optimize the various factors for best playback quality."

 

The changes in SQ I have heard from different players are sibilance, detail, depth, height, width, bass, timing, muddy sound etc, how you would measure these I don't know.

 

Usually you would observe something and then try and understand it via measurement and theory, all you have said is that others report that players sound different, do you find any difference between foobar wasapi vs asio vs ds vs memory play vs different buffer sizes ? I do.

 

 

[Miska]

Usually you would observe something and then try and understand it via measurement and theory, all you have said is that others report that players sound different, do you find any difference between foobar wasapi vs asio vs ds vs memory play vs different buffer sizes ? I do.

 

Now the big question is if the effect is the same with all possible computer + DAC combinations...

 

I think the measurements are important to find and eliminate the root causes...

[esldude]

"My goal is to document and understand what kind of changes in sound quality are caused by playback software. Then one could avoid or optimize the various factors for best playback quality."

 

The changes in SQ I have heard from different players are sibilance, detail, depth, height, width, bass, timing, muddy sound etc, how you would measure these I don't know.

 

Usually you would observe something and then try and understand it via measurement and theory, all you have said is that others report that players sound different, do you find any difference between foobar wasapi vs asio vs ds vs memory play vs different buffer sizes ? I do.

 

 

 

Yes, your reports of what sounds different are similar to many others. If the sound coming from the DAC has changed, then somehow the signal differs. It should in principle at least be measureable. I don't necessarily have to know how to measure for sibilance, detail, depth etc. etc. I just have to know how to measure something as different about the resulting signal.

 

I have mostly compared wasapi, and asio, but not very much kernal streaming. I sometimes thought I heard differences, but more careful listening never really seemed to pin anything down as consistently different. Buffer sizes and memory play didn't sound different either.

 

So then I measure them, and also get no difference. Null testing showed pretty much nothing either. So is there some measures that correlate? There have to be if the software is changing the signal that creates the sound. Maybe I am just not hearing it or maybe a most focused cleaning up of it would make it apparent to me.

[esldude]

Now the big question is if the effect is the same with all possible computer + DAC combinations...

 

I think the measurements are important to find and eliminate the root causes...

 

Yes, I agree with this.

 

So far your measures are showing me different noise levels with the same player on different hardware. Not exactly what I was looking for, but one I am interested in as well. And also one I wondered about. I use an Audiophilleo SPDIF converter. Maybe it is better than average isolating the DAC from upstream issues. It never seems to let junk off the USB show any effect on the DAC output. I know the noise on USB differs from one machine to the next, what it is doing, whether there is a powered hub in between and how good the PS is on it. But none of those things alter the output when using the Audiophilleo. I have nearby friends with two versions of the MuFi Vlink and one with Berkely Audio DAC. I haven't done any measures of those. Maybe the Vlink lets more stuff through to effect the output. Not sure. If some devices do and others don't seems an important thing to test when people review such products.

 

But does anyone have measures of two different software players working through the same hardware and DAC altering the analog signal from the DAC?

 

I haven't used Miska's software, but my limited understanding is it does much more than just play bits. It allows configuration of various filters and other things. I assume it isn't bit perfect doing that or am I confused there somehow?

 

I also have the question Mayhem13 asked which is if guys making bit perfect software can't measure the results how do they go about design? My guess is mainly two ways. The assumption lowering demands on the computer, shutting down other services, and such reduce noise and variation in results. More or less on faith. The choices then guided by sighted listening to evaluate which areas make the most difference.

[mayhem13]

I can accept a lower noise floor as a direct result, but that's it as that's all I could detect through listening, polars , FR and HD sweeps. But typically the noise floor is well below the threshold of hearing, no less the typical ambient room noise level. As I've mentioned before my dedicated HT has all the gizmos for low noise, hung walls and ceilings, fiberglass ducts and damping, etc. I've really tried to appreciate or detect noise......and I can't. Another interesting point for the low noise crowd to digest......I'm also running extremely high efficiency at 95db/2.83 so even 2-3 watts with a recorded empty signals should reveal something as far as noise goes.......yet it doesn't......but I'm still wiling to accept the fact that some may claim to appreciate the reduced noise........but as to imaging components, there's nothing even remotely capable of backing those claims.

[Miska]

But none of those things alter the output when using the Audiophilleo. I have nearby friends with two versions of the MuFi Vlink and one with Berkely Audio DAC. I haven't done any measures of those. Maybe the Vlink lets more stuff through to effect the output.

 

How do you know? Did you try measuring it with different computer and DAC combinations? But in any case it is at least five times more expensive than V-Link192...

 

V-Link192 is USB bus-powered, so it is somewhat likely to have different jitter patterns with different computers (since it's the clock-master). DACs that have built-in USB and don't have galvanic isolation tend to be sensitive too. But good thing is that V-Link192 has isolating output transformers, while the non-192 V-Link doesn't.

 

But does anyone have measures of two different software players working through the same hardware and DAC altering the analog signal from the DAC?

 

Obviously it would be bad idea for me to start posting any measurements "my player vs competitors"... So I stick to my own things.

 

I haven't used Miska's software, but my limited understanding is it does much more than just play bits. It allows configuration of various filters and other things. I assume it isn't bit perfect doing that or am I confused there somehow?

 

It can be configured for bit-perfect output too, but it is not intended to be used like that. It would waste 75% of the code in there...

[Jud]

Still if I understand the explanation of that post, he is showing the results of playback from 4 different software players and they all overlap except at the very lowest levels when level and time is compensated for.

 

Transient and temporal response are quite easy to measure with difference testing.

 

 

Hi Dennis, a question about difference testing as you implement it:

 

How do you distinguish between error in measuring temporal response that should be compensated for (as you mention in the first quote above) and actual transient/temporal response of the equipment that should not be compensated for?