[nick_charles]

Quote:

Originally Posted by leeperry /img/forum/go_quote.gif you're only as strong as your weakest link, and I'm afraid any comparison would just boil down to measuring the poor clocking of that Edirol USB soundcard you'd need more serious equipment to conduct this kind of tests IMHO..

The Edirol is certainly limited

I was toying with getting an E-Mu 0404 but I am relcutant to shell out $185 for something I, in all likelihood won't use after this exercise and unless it acts as a pure passthrough the same problem may occur

but in this case the bigger problem is with the PC digital output. The PC digital stream is showing the higher level of deviation compared to the CD player digital stream.

 

[b0dhi]

Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif I downloaded a demo version of Autosignal. It is a serious package. I could not find a way to get the FFT size down, it kept giving me 81K samples so the text files were enormous and the results were a bit dubious.

There's nothing whatsoever dubious about those results. Audacity uses an apodizing transform. All apodization windows have limitations as far as their amplitude and/or frequency resolution. Using a Hanning window with 2048 or 1024 point FFT to detect differences below 0.01dB is akin to an astronomer smearing oil over the lens of their telescope, then trying to detect the presence of an asteroid in the Kuiper belt.

I'd be happy to test using FlexPro 8, which is another analysis package. Please upload the "sample" file and I will post the results of analysis with FlexPro 8 and we can compare. This should be the same sample file which was used in your last post re. AutoSignal.

In any case, the previous test of Audacity's FFT resolution should be enough evidence that it is not suitable for the purposes of your experiment. That test is easily replicable. The 1khz and 13khz tones have a known equal amplitude, yet Audacity, at its best, reports them to be 0.9dB different. Using it to detect differences in the order of 0.01dB is not reasonable.

Without resolving this I don't believe the experiment will have any validity.

 

[nick_charles]

Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif There's nothing whatsoever dubious about those results. Audacity uses an apodizing transform. All apodization windows have limitations as far as their amplitude and/or frequency resolution. Using a Hanning window with 2048 or 1024 point FFT to detect differences below 0.01dB is akin to an astronomer smearing oil over the lens of their telescope, then trying to detect the presence of an asteroid in the Kuiper belt. I'd be happy to test using FlexPro 8, which is another analysis package. Please upload the "sample" file and I will post the results of analysis with FlexPro 8 and we can compare. This should be the same sample file which was used in your last post re. AutoSignal. In any case, the previous test of Audacity's FFT resolution should be enough evidence that it is not suitable for the purposes of your experiment. That test is easily replicable. The 1khz and 13khz tones have a known equal amplitude, yet Audacity, at its best, reports them to be 0.9dB different. Using it to detect differences in the order of 0.01dB is not reasonable. Without resolving this I don't believe the experiment will have any validity.

I'll upload the files at the weekend. If Audacity really is no good then I am out.

 

[TwoTrack]

Interesting test Nick. I look forward to seeing the results.

 

[leeperry]

Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif The Edirol is certainly limited I was toying with getting an E-Mu 0404 but I am relcutant to shell out $185 for something I, in all likelihood won't use after this exercise and unless it acts as a pure passthrough the same problem may occur but in this case the bigger problem is with the PC digital output. The PC digital stream is showing the higher level of deviation compared to the CD player digital stream.

a not so high end Edirol USB soundcard and a freeware audio editor(Audacity is really not world class) might really not be the best tools to conduct such precise tests

what about WaveSpectra? generate some 15/20kHz sinewaves w/ SineGen, get them through your j..... roller coaster, and measure the THD/THD+N/SNR afterwards? it'll be more accurate than Audacity IMHO(you can see the FFT size up to 131072 samples): WaveSpectra 1.40

but you'll just be measuring the poor clock of the Edirol anyway..it'd need something like AP to give proper results IMHO.

 

[leeperry]

one thing's for sure, when you were talking about roll off...you might very well be completely right.

I use this cable(the top one): SHIN KIN - GLASS TOSLINK CABLES

and just switched between a CMI8768 board(that has one single 14.318MHz PLL) using those 100% bit-perfect drivers: cmediadrivers

to an AudioTrak Prodigy HD2(running two discrete clocks, one 22.58MHz PLL/one 49.152MHz oscillator): ) using those 100% bit-perfect drivers:

I like to EQ down middle ear resonances to get a flatter FR: http://www.head-fi.org/forums/6508490-post409.html

and the same preset very much does NOT work on both cards...bit-perfect, yeah right.

the difference I'm hearing is much bigger than what I'd expect from opamp rolling

 

[b0dhi]

This might be helpful. In particular, see the table near the bottom comparing the type of signal to be determined and the window that would be appropriate for it.

I know I have a book here that goes into the math and most importantly shows the amplitude error margins of each of these windows but I can't for the life of me remember where it is. Will post those details when/if I find it.

Edit: found something here: http://www.bksv.com/doc/bv0031.pdf. See table 2 on page 12. The "ripple" column shows the maximum amplitude error.

 

[nick_charles]

Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif This might be helpful. In particular, see the table near the bottom comparing the type of signal to be determined and the window that would be appropriate for it. I know I have a book here that goes into the math and most importantly shows the amplitude error margins of each of these windows but I can't for the life of me remember where it is. Will post those details when/if I find it. Edit: found something here: http://www.bksv.com/doc/bv0031.pdf. See table 2 on page 12. The "ripple" column shows the maximum amplitude error.

Thanks for the information. Okay Audacity is out of the question. However I do have a better package Cool Edit Pro which supports a variety of FFT windowing (Hanning, Blackmann-Harris, Blackmann, Hamming, Welch, Triangular) up to 65K this seems a lot more precise, would that work ?

 

[b0dhi]

I think that will work. Been doing some tests and I got roughly +/- 0.3dB amplitude accuracy with a ~65k FFT and 4-term Blackman-Harris window having ~-90dB side lobe. Blackman (3-term) also worked within that accuracy.

I tested with various signals of known amplitude and noise 20, 40 or 60dB down to imitate broadband music.

One thing I should point out is that none of my software lets me choose a FFT length smaller than the number of samples in the data, so I don't know what effect that will have. If your sound samples are longer than 1.486 seconds @44.1khz (i.e., 65535 samples) what I said above might not apply.

 

[jcx]

no mention of Audio DiffMaker comparison software? - I'd be interested in hearing peoples experience with it

Audio DiffMaker

the theory presented in the refs looks like they know what they're doing but complicated sw can have implementation or even just UI probelms


I have seen dropout problems with Audacity at higher sample rates
also a few years ago I was very put off by by the arrogance of the developers when I mentioned it was simply wrong to Silently limit the tone generator to 20KHz instead of fs/2 - you type in a higher number and at the time it just created a 20KHz tone in my 96K file without a warning message - trying to make the point that this was very bad UI behavior I was told - "audio stops at 20KHz"


if you are using limited time samples free SciLab, a MatLab "workalike", can be OK for signal processing up to 10s of seconds of audio wav files - but inefficient 24 bit wav file reading means you actually speed up by setting Audacity to create 32 bit wav files - you need to look for a "toolbox" for extended res wavread function

 

[upstateguy]

Nick

I was wondering if the measured difference between your computer and your CD player was audible?

USG

 

[nick_charles]

Quote:

Originally Posted by b0dhi /img/forum/go_quote.gif I think that will work. Been doing some tests and I got roughly +/- 0.3dB amplitude accuracy with a ~65k FFT and 4-term Blackman-Harris window having ~-90dB side lobe. Blackman (3-term) also worked within that accuracy. I tested with various signals of known amplitude and noise 20, 40 or 60dB down to imitate broadband music. One thing I should point out is that none of my software lets me choose a FFT length smaller than the number of samples in the data, so I don't know what effect that will have. If your sound samples are longer than 1.486 seconds @44.1khz (i.e., 65535 samples) what I said above might not apply.

I'll trim my killer sample to 1.4 seconds, it is three cymbal crashes, I daresay 1 will do just was well...

 

[nick_charles]

Quote:

Originally Posted by upstateguy /img/forum/go_quote.gif Nick I was wondering if the measured difference between your computer and your CD player was audible? USG

I'll try a DBT on a pair of samples this weekend.

 

[leeperry]

Quote:

Originally Posted by nick_charles /img/forum/go_quote.gif I'll try a DBT on a pair of samples this weekend.

they'll be "blurred" by your USB soundcard output IMHO..if the input clock sucks, the output does too.

 

[nick_charles]

Quote:

Originally Posted by leeperry /img/forum/go_quote.gif they'll be "blurred" by your USB soundcard output IMHO..if the input clock sucks, the output does too.

I have a separate PC sound device for playback,a Musiland USD01 that converts USB to SPDIF and thence to an external DAC, I do not use the Edirol for playback. I also have a direct SPDIF output from the PC.