[clarke68]

Quote:

Originally Posted by fewtch I see in your equipment list that you're using an external DAC. Some sort of jitter between the card and DAC at higher sampling rate modes could be producing artifacting.

I was wondering about that myself...if outboard processors (DACs, jitter filters, re/upsamplers) would cause artifacts with this file, or if it will just reveal problems coming from the PC.

Somebody wanna unplug their fancy setup and let us know?

 

[clarke68]

Quote:

Originally Posted by roadtonowhere08 Question: what's the point of all this testing? If you like what your headphones are putting out, enjoy your music. It seems like some of you are spending more time on testing your equipment than actually using them for what they are for, to listen to music with.

I completely agree with your sentiment here, although I think there is some value in testing. Presumably, setting up a system that performs well in a test will produce a system that is more enjoyable/musical/whatever when you're using it for what its meant for.

On the other hand, testing could cause stress (and money loss) if it reveals "problems" you didn't know you had, and you were enjoying your system just fine before the test.


Quote:

Originally Posted by roadtonowhere08 These results are confirming what we already know, good sound cards don't distort and poorly designed ones do...

Not exactly...and my own set up is proof. My results were fine (no artifacts) using Foobar with no upsampling or additional processing of any kind, and the stock sound card that came with my work PC. All this testing is making me question whether or not I want to upgrade at all!

 

[halcyon]

Quote:

Originally Posted by fewtch I see in your equipment list that you're using an external DAC. Some sort of jitter between the card and DAC at higher sampling rate modes could be producing artifacting.

Not using an external DAC.

 

[fewtch]

Quote:

Originally Posted by halcyon Not using an external DAC.

Just out of curiosity, what kind of artifacts are you hearing with upsampling... what does it sound like? (I don't have a link to the HA thread).

 

[halcyon]

Low freq hum or really high freq chirping (modulated distortion) at really loud playback levels:

http://www.hydrogenaudio.org/index.p...ic=9772&st=75&

cheers,
halcyon

 

[Jasper994]

Quote:

Originally Posted by Wodgy By the way, if your hearing is very good and you turn up the volume enough, you may also be able to hear a faint high-pitched sound (around 20kHz). This is normal. The problem sounds, if they occur, are much more obvious.

Well when I had SSRC and several other things turned on there was a distinct high pitched zing-zing-zing-zing noise. When I turned all those things off it went away... until I turned the volume up... then I could hear it again, but at a much much higher pitch.

apparently the soft clipping limiter was causing the major distortion...

 

[fewtch]

Quote:

Originally Posted by halcyon Low freq hum or really high freq chirping (modulated distortion) at really loud playback levels: http://www.hydrogenaudio.org/index.p...ic=9772&st=75& cheers, halcyon

I'm not hearing it at all... must be related to your gear, or perhaps other output plugins (if applicable). The signal is clean with my card (M-Audio AP 2496) and the WinAMP WaveOut v2.0.2a SSRC plugin upsampling to 24/96 with triangular shaped dither, triangular distribution -- no other output plugins.

Edit -- or maybe it's something to do with my headphones (Senn HD-600) but I wouldn't have a clue as to why. Will try it with some low impedance cans.

Edit2 -- OK, I got something that sounds like *very* slight distortion at a high volume with low impedance cans (Denon AH-D550). I'm not convinced though, as it seems to follow along with the high pitched tones (almost like a sort of hissing), and I didn't want to compare it to unresampled too often (my ears are already ringing

). No low frequency hum/distortion at all, if anything it's a very slight deviation in the higher frequencies. It could have been the noise shaping itself I was hearing, with cans that sensitive at a high volume. I'm not trying it again, my poor ears...

Reading what you said on HA, I honestly wouldn't jump the gun and assume this is intermod distortion due to upsampling (tho it could be). Sound pressure at that loudness and frequency could be causing the drivers themselves to distort, or it could be noise shaping (if you were using it), or even just an illusion if one has to listen that hard for a "masked" chirping sound. Or it was the Audigy (Creative Labs... uggh).

 

[Wodgy]

Quote:

Originally Posted by Jasper994 Well when I had SSRC and several other things turned on there was a distinct high pitched zing-zing-zing-zing noise. When I turned all those things off it went away... until I turned the volume up... then I could hear it again, but at a much much higher pitch. apparently the soft clipping limiter was causing the major distortion...

I think that's one of the best things about this test, is you can see for yourself how transparent SSRC, the soft limiter, and the advanced limiter, etc. really are. Some people are probably going to stop using SSRC because of this (I too can detect a bit more distortion when upsampling using SSRC, though this observation won't change the way I listen because I haven't been using it in the past). For others, especially the Creative Labs users, they'll find SSRC more transparent than not using it.

 

[Wodgy]

Quote:

Originally Posted by fewtch I can't think of any logical reason why upsampling would produce artifacts

In an ideal world, upsampling will not produce artifacts. But we're not talking about an ideal implementation here. SSRC takes a number of shortcuts because doing upsampling directly is too computationally intensive for general purpose processors. Going from 44.1kHz to 96kHz directly requires 320 times oversampling, FIR filtering the result, and then decimating by 147 times. This is too expensive to implement. Even with custom hardware, it is hard to implement this directly. Asynchronous upsamplers like the AD1896 use a method where they keep a table of "predicted" FIR filter coefficients and use this to generate a result. This shortcut approach can be implemented in custom hardware, but the implementation details determine the DNR. AD1895 and AD1896 use the same basic algorithm, but AD1896 is more accurate. Because SSRC is undocumented, it's unclear exactly what shortcuts it takes, but we can be sure it takes some, otherwise it would not run as fast as it does (slow, admittedly, but tractable). Also, while SSRC does let you choose the quality of its FIR filter (this is the "slow" versus "fast" setting), filter design is complicated and always involves tradeoffs. (I have taken a grad-level course in digital signal processing, so I'm not just blowing hot air here.) I wouldn't be surprised to see subtle ringing even with the slow setting in SSRC.

One day, someone will implement a 44.1kHz -> 88.2kHz upsampling plugin with no FIR filtering. This is a much easier task, and it would be more transparent, but it would also end up with a lot of ultrasonic aliasing. Pick your poison, I guess.

 

[fewtch]

Quote:

Originally Posted by Wodgy In an ideal world, upsampling will not produce artifacts. But we're not talking about an ideal implementation here. SSRC takes a number of shortcuts because doing upsampling directly is too computationally intensive for general purpose processors. Going from 44.1kHz to 96kHz directly requires 320 times oversampling, FIR filtering the result, and then decimating by 147 times. This is too expensive to implement. Even with custom hardware, it is hard to implement this directly. Asynchronous upsamplers like the AD1896 use a method where they keep a table of "predicted" FIR filter coefficients and use this to generate a result. This shortcut approach can be implemented in custom hardware, but the implementation details determine the DNR. AD1895 and AD1896 use the same basic algorithm, but AD1896 is more accurate. Because SSRC is undocumented, it's unclear exactly what shortcuts it takes, but we can be sure it takes some, otherwise it would not run as fast as it does (slow, admittedly, but tractable). Also, while SSRC does let you choose the quality of its FIR filter (this is the "slow" versus "fast" setting), filter design is complicated and always involves tradeoffs. (I have taken a grad-level course in digital signal processing, so I'm not just blowing hot air here.) I wouldn't be surprised to see subtle ringing even with the slow setting in SSRC.

I'm really just saying that I can't hear any... or anything worth noting anyway, that doesn't take intense concentration and about a dozen repeats to "maybe" hear some subtle intermodulation in the provided file.

Fwiw, upsampling to 24/96 in SSRC's "slow mode" requires ~40-60% of the CPU time of my 1.2 GHz processor. General-purpose or not, that's some pretty hefty processing power being put to work.

 

[Stephonovich]

So take off Slow Mode. It does diddly crap anyway. Even the author says it's not noticeable unless you're doing 8000-48000 or something similar.

(-:Stephonovich

 

[Wodgy]

Don't get me wrong, you definitely want slow mode. All I was saying was that even slow mode involves tradeoffs.

Fewtch, even though 50% CPU usage seems like a lot, you'd need much more processing power to implement upsampling in the obvious way. We're talking on the order of 10's of gigaflops.

My personal recommendation for people with good, non-resampling hardware like the M-Audio cards is simply not to use upsampling. The slight "smoothing" effect you get is just as likely to be SSRC munging the high frequencies as it is likely to be the result of forcing the DAC to use a different digital filter. With an external DAC, upsampling is a good strategy for attenuating jitter, but with computer sources this isn't an issue.

 

[halcyon]

Quote:

Originally Posted by fewtch It could have been the noise shaping itself I was hearing, with cans that sensitive at a high volume. I'm not trying it again, my poor ears...

Try it without dither then (although with re-quantisation I wouldn't recommend it, but you'll learn more).

BTW, it's not practical to talk about upsampling and re-quantisation without dither, as one get's even more artifacts (on audible level) without dither.

Quote:

Originally Posted by fewtch Reading what you said on HA, I honestly wouldn't jump the gun and assume this is intermod distortion due to upsampling (tho it could be).

Well, that is easy to verify. Try it without upsampling and re-quantisation with your non-resampling hardware (I have already done this). Hear any problems?

It's also quite of an ear opener to try various different upsampling algorithms.

Quote:

Originally Posted by fewtch Sound pressure at that loudness and frequency could be causing the drivers themselves to distort, or it could be noise shaping (if you were using it), or even just an illusion if one has to listen that hard for a "masked" chirping sound.

That is of course very possible. IMD in elements is a known and studied phenomenon.

However, it happens in the exactly same manner in drivers, regardless of sampling levels. If the analog input to drivers is the same, so is the IMD (hence you'd hear it with and without upsampling).

If you are saying that the analog input is significantly different, well then you are giving quite of a support argument to my case, I'm afraid.

Quote:

Originally Posted by fewtch Or it was the Audigy (Creative Labs... uggh).

Nope. RME and various other non-resampling hardware tested (since that HA thread).

regards,
Halcyon

PS To put things into perspective. Aliasing artifacts are a known problem with asynchronous upsampling. However with this sample and using SSRC the effects are so small as to be practically non-existant and I don't think anybody should be worried about them in terms of ruining musical enjoyment.

 

[Wodgy]

Quote:

Originally Posted by halcyon PS To put things into perspective. Aliasing artifacts are a known problem with asynchronous upsampling

All asynchronous upsamplers that I know of filter out the ultrasonic images, so there are no aliases, and hence no aliasing artifacts. No one would use AD1896, etc. if this were not the case.

 

[Jasper994]

ahhh very nice... on my home rig w/out soft clipping limiter I can only barely hear that obnoxious 19khz noise and the phone tones... nothing else...