[UltMusicSnob]

I did open up both the Redbook-ripped 44.1/16 wav file and the iZotope 64-bit SRC upsampled files in SoundForge 10, and zoomed all the way in, to where the display starts showing points connected by straight lines, and looked at various corresponding points.
 
Yep, there were visible differences: zero crossings happening at slightly different points, slightly changed peak points, slightly different curves drawn.--TINY TINY differences.
 
The upsampling algorithm has to put an additional point **somewhere**. As it does so, the algorithm and parameter settings used determine where those extra points go--it doesn't just interpolate on straight lines. And then there are additional corresponding differences again, when the D/A to analog Line Out converts to points to a continuous electrical signal. Under a High_Resolution setting, the Voxengo SPAN frequency analysis tool shows visible differences in the spectra--very small, but SPAN at least knows they are there. Don't have the graphic here, but I'll post it later.

 

[xnor]

Quote:

Yep, there were visible differences: zero crossings happening at slightly different points, slightly changed peak points, slightly different curves drawn.--TINY TINY differences.

That seems more like a non-integer group delay. The resampler shouldn't do that.
 
 
Quote:

The upsampling algorithm has to put an additional point **somewhere**. As it does so, the algorithm and parameter settings used determine where those extra points go--it doesn't just interpolate on straight lines. And then there are additional corresponding differences again, when the D/A to analog Line Out converts to points to a continuous electrical signal. Under a High_Resolution setting, the Voxengo SPAN frequency analysis tool shows visible differences in the spectra--very small, but SPAN at least knows they are there. Don't have the graphic here, but I'll post it later.

Yes, it interpolates using an adjustable low pass filter. But all information that is "added" is redundant, in fact, some information gets actually removed in the process (the low pass needs some room to work).

 

[dizzyorange]

What kind of resampling are you guys using?  Is there any preference between the PPHS and SoX resamplers for foobar?
 
Also, what is your opinion of non-oversampling DACs using those old TDA chips?  I had a DAC-AH a few years back and it was the only DAC I felt really had a distinct sound signature. 

 

[stv014]

Quote:

That seems more like a non-integer group delay. The resampler shouldn't do that.

 
Yes, the iZotope resampler does have a small amount of delay even with a linear phase filter, this has been analyzed in earlier posts. For consistency, I have replicated the same delay in the other resampled and recorded files.

 

[stv014]

Quote:

What kind of resampling are you guys using?  Is there any preference between the PPHS and SoX resamplers for foobar?

 
UltMusicSnob's "original" 192/24 file was created with the iZotope sample rate converter in SoundForge.
 
The other files were processed with my converter (there are some tests of it here), mostly with settings that try to "emulate" the iZotope resampler or sound card DAC filters. It was also used to accurately synchronize and pitch correct sound card recordings.
 
I think both the PPHS and SoX converters are good with the right settings, but SoX is faster at the same quality and is more flexible/configurable.

 

[stv014]

Quote:

Also, what is your opinion of non-oversampling DACs using those old TDA chips?  I had a DAC-AH a few years back and it was the only DAC I felt really had a distinct sound signature. 

 
Well, they have some high frequency roll-off (at least ~-3 dB at the Nyquist frequency from the simple "sample and hold" conversion alone, possibly more from any analog lowpass filters). Also, there is a substantial amount of ultrasonic and even RF imaging, which can be difficult for the analog circuits to handle, and often results in relatively high IMD. Additionally, the low level linearity also tends to be worse than that of the newer oversampling converters, but not necessarily to an extent to be a real problem in practice.
 
With a DAC that supports high sample rates, it is possible to "emulate" a non-oversampling DAC, even if not perfectly due to the limited bandwidth.

 

[nick_charles]

Quote:

 
Well, they have some high frequency roll-off (at least ~-3 dB at the Nyquist frequency from the simple "sample and hold" conversion alone, possibly more from any analog lowpass filters). Also, there is a substantial amount of ultrasonic and even RF imaging, which can be difficult for the analog circuits to handle, and often results in relatively high IMD. Additionally, the low level linearity also tends to be worse than that of the newer oversampling converters, but not necessarily to an extent to be a real problem in practice.
 

 
Are there any technical advantages in NOS DACs , i.e is there any area in which they performs better than oversampling cousins ?

 

[stv014]

Well, they allegedly sound better because of having a shorter impulse response (lack of ringing; note however that there can already be ringing in the digital signal from the A/D converter anyway). They are also said to be less sensitive to clock jitter, and easier to implement. There is no ultrasonic/RF noise from noise shaping either (then again, there is high frequency content from imaging due to the lack of a digital reconstruction filter), nor other artifacts associated with delta-sigma modulators. At least these are the advantages claimed by those who prefer NOS DACs, not that I necessarily agree with their practical usefulness.
 
NOS DACs are sometimes used in combination with digital upsampling. So, the signal is in fact "oversampled" then, but without noise shaping.

 

[xnor]

shorter impulse response = violating the sampling theorem resulting in loads of aliasing
 
RF noise from noise shaping: isn't that largely filtered out using an analog low pass filter?
 
Digital oversampling first, then non-oversampling D/A-conversion: What's the point? Afaik the simple zero order hold or maybe even linear interpolation requires a crazy oversampling ratio in order to get anywhere near 16 bit performance.

 

[stv014]

Disclaimer: the above is not my opinion, but a summary of arguments I have seen in favor of NOS by people who think it is better.

 

[UltMusicSnob]

Stirring the pot again. This isn't Redbook versus 192, btw, its Redbook versus MP3 320. Happy to split thread if that's appropriate, moderators.
 
foo_abx 1.3.4 report
foobar2000 v1.2.8
2013/09/16 19:01:08
File A: C:\Users\KiarkAudio\Documents\Ravel_Test_File_2.wav
File B: C:\Users\KiarkAudio\Documents\Ravel_Test_File_1.mp3
19:01:08 : Test started.
19:01:27 : 01/01  50.0%
19:01:34 : 02/02  25.0%
19:01:42 : 03/03  12.5%
19:02:15 : 04/04  6.3%
19:02:49 : 05/05  3.1%
19:03:01 : 06/06  1.6%
19:03:52 : 07/07  0.8%
19:04:06 : 08/08  0.4%
19:04:28 : 09/09  0.2%
19:04:38 : 10/10  0.1%
19:04:51 : 11/11  0.0%
19:05:28 : 12/12  0.0%
19:05:44 : Test finished.
 ----------
Total: 12/12 (0.0%)

This test is from an excerpt of Ravel's Daphnis et Chloe (full version, not the
Suite). This was a planned set of 12 rounds.
The .wav file above was ripped directly from the music CD using SoundForge 10.
SoundForge 10 exported the file to MP3 at 320 kbs setting.

 

[stv014]

Well, successful ABX of lossless vs. 320 kbps MP3 is not that surprising with the right samples. However, if you uploaded the files, maybe others could try to compare them, too.

 

[UltMusicSnob]

  Well, successful ABX of lossless vs. 320 kbps MP3 is not that surprising with the right samples. However, if you uploaded the files, maybe others could try to compare them, too.

Forgot to post the links, sorry. Here are the files:
WAV version
MP3 Version (320 kbs)
 
This is a very soft passage from Ravel's "Daphnis et Chloe", so it should in fact be a stress test for the functionality of the algorithm at these low signal levels. This kind of signal turns up all the time in classical, I would say never in pop or rock music.

 

[nick_charles]

  Forgot to post the links, sorry. Here are the files:
WAV version
MP3 Version (320 kbs)
 
This is a very soft passage from Ravel's "Daphnis et Chloe", so it should in fact be a stress test for the functionality of the algorithm at these low signal levels. This kind of signal turns up all the time in classical, I would say never in pop or rock music.

 
I tried and failed miserably, going to try again with different headphones. Not helped by the fact that for me FooBar blips going from sample to sample !

 

[UltMusicSnob]

   
I tried and failed miserably, going to try again with different headphones. Not helped by the fact that for me FooBar blips going from sample to sample !

You do a classic recording studio A/B, switching from one to the other during continuous playback? I tried that when I first got the tool installed, and as you say, the blip FooBar makes when making the switch ruins that approach entirely for me--I might as well just stop the playback manually, it's actually less noisy that way.
 
For all the results I've posted, I uncheck the "Keep position" box, I set a Start and End segment, and then listen repeatedly to that segment by clicking as necessary. For these hard ones, I try to 'acclimate' my ears to the sound of 'A' and 'B' first, then move to the unknowns.