The strangest sound ever, or ? Q5= ...

[PeterSt]

For Stanley, but also to others :

If it would be true that with this the balance between left and right can change, I dare (softly) say that it is another testimonial of samples being able to get skipped or lost.

Of course now I am again the conspiracy thinker but I have been thinking this forever. It is only that I can't find the technical proof of it, with for example digitally looping back and observing bit perfectness. But then digitally looping back is not exactly the same as putting all through to the (beginning of) the D/A converter. But I don't know other/better means.

And *if* this were true it would
- clarify how I myself have been puzzled how such huge changes ever can be still bit perfect (this is one of them);
- far batter explain how software can influence SQ which is supposed to happen in-DAC only;
- which then is not true anymore of course. Instead it's just "manipulation" in the PC;
- explain how in the world that Left/Right behavior in W8 can be so way different than in e.g. W7.

So, what's up really ?

Peter

[PeterSt]

What Q5=1 does is causing "intervention" by other OS processes to be completely impossible. Well, if I have done this job right. So what I am supposing is that samples may get lost when other processes interfere. And notice that in itself a sample getting lost is inaudible, but general SQ will change. For that see my first post about the "resolution" to transients and what it could imply.

May I also refer to the large W8 topic ("My first Windows 8 Experience") where at some stage I suggested that the samples for left and right were not pushed out "together" (a chunk of them for left, then a chunk of them for right) which is how the Left/Right experience could emerge.

Supposed we all start to feel that this is the way to go ... I think I can improve on it. So, no other OS processes interfere ... but one : the track part loading (which is a continuous process at SFS=2 levels). Thus, the Audio Engine actually interferes with itself. I think I can move that out of the way ...

Peter

[acg]

Anthony, your sig says "Asus" X79 etc ...
That would be ASRock, right ?

You got me there...I will ammend my sig.

[acg]

Anthony, thank you for your clarifications.

How easy will it be to observe these clicks ? I mean, do I have a chance that I miss them ? Does it need more silent music (piano perhaps) ?

In the mean time I wonder whether you can observe the interval of the clicks changing with the SFS changing. I am afraid you won't have much headroom though (the larger the SFS the less chance it will work at all). And of course it requires a fast interval to observe in the first place (I think SFS=2 implies track part loads of 3x per second. SFS=6 simply triples that to once per second - just watch your CPU graph).

Peter

Peter, the clicks are quite easy to hear and vary in size from barely audible to a little pop like vinyl to a temporary almost complete silence where a lot of data is lost.  Looking at the Phasure Driver Control Panel the errors show up about a second or so after the irregularity is heard.  The individual errors vary in size from 1 to about 1200.

I will look at the SFS stuff later on.

Anthony

[PeterSt]

Anthony, what happens when you upsample to 384 instead of 768 ? I think this will do it !
Watch those FiFo errors ...

Peter

[PeterSt]

Btw, I forgot to mention one other required setting : Logging must be Off.
I added it to the original post.

[PeterSt]

Anthony,

I tried a few things a/w changing something to the Appointment Schemes so I can "see" better ...

First of all, at least with 4ms in the NOS1 Driver Control Panel - no dice. Errors all over (all sorts except ASIO of course).

Secondly, with 16ms in there, no problems with 44.1 based.

Thirdly, with 48 based and 768 output, errors. But in my case merely WASAPI/WDM errors than FiFo errors.

Lastly, with 48 based and 384 output, no errors anywhere ever.

And because of my changes to the Schemes I can nicely see how it is caused. Maybe it is solvable for 48 based and 768 output, but what should be solvable for sure is the stopping when the PC was out of use for too long. I think I have that already working largely.

Something else is that with Attended playback is just interrupted when Q5=1 doesn't work well. So, does not stop.


Summarized :
24/768 has always been very critical for the NOS1 and I think I just learnt that something in the OS is actually causing it to easily fail. It will be (critical) driver related and in the end is caused by "me" hogging too much; just that tad too much for 24/768.
This is my conclusion because the buffer size in the Driver Control Panel being low(er) also doesn't work; how this is a "similar thing" is a bit complicated, but it is. Think like the response at 4ms must be 4 times faster than 16ms, but things are hogged to long to service that.

Q5=1 *is* a super trick in the first place, me thinking that it never could work, but now I know it can. It already does, but for example I have now seen how it can work with the higher SFSes just the same.

Regards,
Peter

[listening]

Peter,

I have to increase processor clock to 1.1 GHz - so hopefully I do not compare apple and oranges. First impressions with the system here:

- Music is quieter with the same XXHE volume position
- The first 5 minutes of the listening I was missing dynamics
- After 20 minutes of listening  - the stage is backwards greater, instruments are moving backwards too and sounding a bit quieter, what should be the right thing.
- There is additional "space" between "sound fragments" - if listening to a tabla I can better distinguish between the fingers drumming or the swing out of sitar tones is clearly heard even when new strings are plucked.
- All with naturalness and without superficiality

Georg

[AlainGr]

Hi Peter,

While I cannot pretend I hear something odd happening, I suppose that if your processor frequency is very low and that errors are happening when some interruption is induced, maybe you should adjust the frequency to be a little higher ? Provided that an i7 is tuned to work normally at 3.2ghz, it provides a lot of headroom for reduced speed, but the "sweet spot" could be somewhere between what you have achieved and the nominal frequency ? Just a suggestion of course and without any pretention, as a few things are completely over my head...

Alain

[PeterSt]

For those who can't get this going (for a longer time) without stops or interrupts, this is solved for the next version (beyond 0.9z-9b).

The issue with FiFo (etc.) errors for 48KHz based tracks is also solved (quite unexpectedly actually).

Peter

[acg]

Anthony, what happens when you upsample to 384 instead of 768 ? I think this will do it !
Watch those FiFo errors ...

Peter

I did this test Peter and all errors are gone.  Note that I never had FIFO errors, only KS/WDM errors.

Anthony

[acg]

For those who can't get this going (for a longer time) without stops or interrupts, this is solved for the next version (beyond 0.9z-9b).

The issue with FiFo (etc.) errors for 48KHz based tracks is also solved (quite unexpectedly actually).

Peter

Good news Peter.  I love these new settings and I love my 48 based music as well so it would be nice if they could be used together.  When is the new version due?

Regards,

Anthony

[Nick]

Hi,

Iv been trying the Q5 setting this week. In terms of sound quality I agree with what has been said above, there is a beautifully analogue presentation without harshness and edge. Here I do also find that there is a slight drop in pace and drive in the music (I feel the need to use a more forward set of interconnects than the ones usually used to compensate). It might be that I am getting used to the new Q5 setting but I think that transients do not have quite the same weight which results in a slight loss of presence and engagement with the sound. Overall though its a big step forwards.

What has been fascinating is that I have the feeling I have heard almost the same change in sound characteristics before. I think the change is very similar in character to that experienced from changing the 24mhz USB clocks and applying bypassing caps on the NOS1 USB board.

Regards,

Nick.

[PeterSt]

Yes Nick. And you will find it very hard to believe I think, but somehow I expected it to be very similar. This is only (small) partly from your descriptions but I'm merely thinking about what this actually causes and how it could be dealt with by electrical means. And to be honest I was really waiting for your judgement on this, so thanks for trying it and the feedback of course.

In the mean time I have been working myself on exactly those possibly just too much lack of transients but maybe this is too hard to explain once again. But let's say that the "blending" ("from all angles") I talked about earlier, also exhibits a "towards more grey" for the worse. And of course we don't like that.

Not to miss (for the below interpretation) : I most probably have changed more (code) than you out there can copy for the same behavior which I think contributes not only to the better "integration" again, but also made the downsides worse. To counter attack that (and this may work for you out there just in the same (good !) direction), I did this after 90 minutes of trying to find settings :

Set Q1 to 6 so SFS can be 0.8.
WARNING : As far as I can recall Q1 can not be 6 (at xQ1=1 and DevBufSize=4096) but now this works. So, not sure at all whether you can do this without ticks (different ticks, just caused by a too low buffer size and btw talking about the NOS1).

What happens with this is crazy;
All the grey turns into bells and tinkles and the IMO right rendition of all which is "highs". Metal is beautifully metal and the same transients we talk about work out to just against the threshold of "too much". So, all just good.


Meanwhile I am almost sure that the protection of SFS being very low at a too high Q1 (you will see the messages when Unattended is attempted) is unjustified because in Attended they just work and the protection should be about Phase Alignment which I am not using. IOW : More is to be achieved if I only remove that protection.
If that is out an SFS=0.2 will work just the same (in Attended it already does) with my now new goal of ... well I won't tell yet. But the approach is quite different from all the years before and it could "justify" how the lower SFS can work out for the better while I said a 100 times at least it should not but agree that it does.
So, to be continued.

Peter

[Arjan]

Hi Peter,
Yes, this is much better, the SF=0.8 does the trick! All details are back!
I was already on Q1=6 and SFS=1.2 and that was an improvement. In the past going below 1.2 was not possible for me so did not try that.
My clock resolution is on 5ms. I can not have 0.5ms, it sound compressed.

I can say that the SQ has never been so good!