DAC: TDA1541 and TDA1543 question

[morpheus82]

hi guys...before to go completely crazy it's better if i ask you...
after the idea of speakers, ampli, crossover, volume control...i'm starting to think about a dac...not too much complicated...so i have read everything for a week, noting that most of the people say that it's better a non os dac(like tda1541) than an os dac, and the main problem is the jitter introduced from os type...

BUT:

tda1541 has a THD of about 0.05%

versus new type OS that have 0.0005%!!!
how is possible that it sounds better??is it really true??
i'd like to know it before to start my project...

every comment is welcome!!!

[Bernhard]

Originally posted by raikkonen

tda1541 has a THD of about 0.05%

versus new type OS that have 0.0005%!!!
how is possible that it sounds better??is it really true??
i'd like to know it before to start my project...

every comment is welcome!!!

Do not trust the numbers, use your ears, worse specs always sound better

[EC8010]

Originally posted by raikkonen
... and the main problem is the jitter introduced from os type ...

Many oversampling filters create jitter. No problem. Just reclock the data before presenting it to the DAC chip.

[morpheus82]

if this is true, where's the problem in using OS dacs???

[EC8010]

Most manufacturers' application notes ignore the jitter produced by oversampling filters. And since so many "designs" are simply rehashes of manufacturers' application notes, oversampling DACs get a bad name.

Your more serious concern is whether you want to use a multibit DAC or a noise-shaping 1 bit DAC. The TDA1541 is a multibit whereas most modern DACs are 1 bit with noise-shaping. If you choose a multibit, you can still use oversampling to avoid the need for serious filtering. Alternatively, you can take the modern audiophile SEP* approach and ignore the fact that a multibit DAC spews out a load of ultrasonic rubbish that can't be removed by a simple capacitor (as is done by 1 bit DACs).

* Someone Else's Problem

[poobah]

There seems to be a very strong relationship I've observed: The less one understands oversampling, the worse it sounds.

I am convinced that the ultrasonic rubbish, what EC8010 talks about and more, intermodulates to create audible spectra, and this rubbish is subsequently labeled as "detail".

[morpheus82]

hmm...it seems to be still an open question...
i think it's obvious that that rubbish intermodulates in some way but the point is: how much does it have effect on the sound?especially if the oversampling is at very higher frequency...?

[morpheus82]

...and otherways remains the fact that LOTS of people says that TDA1541 sounds great with a THD so high...the look for 0.0001% in their ampli and waste all in the dac...i still can't unterstand...

[EC8010]

Ah, well there are some reasonable arguments in favour of multibit DACs like the 1541. The two extremes of DACs are multibit and 1 bit. A multibit receives the entire data word and changes state just once with each wordclock to produce the correct voltage. Conversely, a 1 bit DAC can only produce a 1 or a 0, and none of the intermediate levels. So what you do is you slice the time between wordclock edges into the number of levels you require, then switch the DAC to a 1 for the right number of levels, and a 0 for the remaining levels. When you low-pass filter it, you get the same answer as you would if you'd had a multibit DAC. And because it's easy to measure time, you get very low distortion. However... 44.1kHz x 2^16 = 2.89GHz, and clocking that fast is expensive.

What practical 1 bit DACs do is to run at a much lower clock frequency, then treat errors as noise and manipulate them so that the noise is supersonic. That's not quite enough to get acceptable performance, so the next wheeze is to manipulate the in-band noise spectrum so that it isn't so perceptible to the human ear. Look carefully, and you will always find modern DACs have their S/N ratio given "A" weighted...

A good multibit DAC can produce a genuinely good S/N ratio, and that, I think, is why the 1541 is popular.

[morpheus82]

so, tell me if i've understand...
the problem is that 1bit dacs have a lot of noise in the inaudible range, but people don't want them because they simply know that there is still that noise, even if they can't hear it?
this don't seem to be a logical reason...especially if you think that the ampli cuts after 20K...
so???