confused about -3.5db @20k on nos dac build

[hbc]

Hi
I am building a NOS power dac, I have dutifully taken on board the perceived wisdom that it should be down 3.5 dB by 20kHz. I can measure with the ARTA, using m-audio transit at 44.1K and I get -3.5dB at 20K, measuring same signal with Sound Technology 1700B distortion analyser I get -0.6dB.... on the scope the pk-pk is constant for all frequencies.


f _______ST______pkpk__arta____arta normalised
100Hz___0db______8.6_ -16.8 fs__0dB
1kHz____0dB______8.6_ -16.8 fs__0dB
10kHz___-0.2dB___8.6__ -17.6 fs__-0.8dB
14kHz___-0.4dB___8.6__ -18.4 fs__-1.6 dB pic
20kHz___-0.6dB___8.6__ -20.1 fs__-3.3 dB



any bright ideas, or is this just rms kind of thing, apparently the 1700b meter is “average” responding...


thanks h.


pics, arta response and scope trace of 14kHz

[Ken Newton]

The -3.16dB @ 20kHz roll-off with NOS 44.1k sample rate data is due to what's called the "zeroth order-hold" ( sample-and-hold) functioning of the resistor-ladder, or non-sigma-delta, class of DACs (PCM56, PCM1704, TDA1451, TDA1453, AD1865, etc.) utilized with NOS.

I'm not sure why you are having a disprecpency between your measurement systems, but that -3.5dB result using the M-AUDIO and ARTA 'Steps' software is the more correct of those two.

I use ARTA with an M-AUDIO Audiophile 192kHz PCI card.

[oshifis]

A true RMS meter should be used here. Alternatively, play back a pink noise test track and have a look at the FFT plot...

[hbc]

Thanks, I will try and track down a true RMS meter that can get past 50Hz and check with that. Frequency picture is pink noise IIRC.


Thanks h

[hbc]

I have now read up on "zeroth order hold" unfortunately no one told RMS meter, which also reads -0.6dB at 20kHz. but.....
I have a theory:
The meter is reading an unfiltered signal, where as the m-audio sampling @ 44.1kHz has not got a hope of capturing the higher frequencies, A quick test of comparing sine and square using an true RMS multimeter gives 2.767 V square, 2.056V for sine, 20log(2.056/2.767) gives 2.57dB, add that to the -0.6 gives -3.2 dB, I'll buy that as a possible explanation.

[DF96]

Yes, as well as the 20kHz signal you will also have a 24.1kHz image. The meter may be lumping them together.

Any DAC which uses sample-and-hold (i.e. almost all DACs) wll give this sinc frequency response, irrespective of the internal DAC technology. An oversampling system will normally compensate for it by adjusting the digital filters, and in any case will have less of it because it is working at a higher sampling rate. A simple NOS has to choose between accepting the frequency droop, or introducing extra phase shift in attempting to correct it.

[hbc]

I am not confused any more, really done it now, an amplifier with 47% distortion and its the best sounding one I have ever made (and I have made a few).

There are two versions of reality occurring here, the sanitised BS world of digital (44.1kHz sampling rate) measurement, and the real, messy world of (1970s) analogue... you take your choice. You either choose 20kHz at -3.5dB with 0.4% THD (in my case) or 20kHz at -0.5dB and 47% THD. My ears tell me the latter to be honest, and the best bit, it sounds good!!

h

[abraxalito]

Originally Posted by hbc

You either choose 20kHz at -3.5dB with 0.4% THD (in my case) or 20kHz at -0.5dB and 47% THD.

Aliasing isn't harmonic (the 'H' in THD) distortion but it is distortion nevertheless.

[hbc]

yes I do fundamentally understand that, and thanks for pointing this out. At this moment however, I am not sure I can think of a more appropriate way of expressing it, I guess THD+N would cover my back a bit better...
h

[abraxalito]

You've pointed out a lack in our current terminology. THD+N doesn't fit either because the point of 'N' here is its not related to the signal. So aliasing doesn't make it into the category of noise. Guess 'TD' is more appropriate, but its not widely accepted.