Seems that many audiophiles don't like CD players that use oversampling ahead of the DAC chip. I think that I have a pretty good handle on why op-amps are not good for analog audio, and why some people like vacuum tubes for analog audio, but I don't see what the issue with oversampling is. Generally oversampling is just a FIR (finite impulse response) digital filter used to filter out the sharp clock edges after the sampling frequency is boosted, in hte digital domain. To make the reconstruction analog filter after the DAC chip easier to build with less impact on the audio. Otherwise you need a brickwall low pass filter there to get rid of the ultrasonic junk that could mess up the sound in your audio amp. Such a brickwall filter would impact the sound worse than any digital oversampling processing would do I would have thought...
Okay, I suppose that the "pre-echo" a symmetrical FIR filter would produce MIGHT be audiable. It's hard to imagine how you could get a pre-echo in the real world (say a real drum hit by a real drumstick heard by real ears). But we could create a non-symmetrical FIR or maybe an IIR (infinite impulse response) filter that would have no pre-echo but some trailing "echo". And still get the frequency filtering we want. Truncation of the digital filtered words can be a problem, partly avoided with DAC chips with more bits. Proper rounding techniques help too.
I'm going to ignore the PLL jitter issue as I would just house the DAC chip in the same box the transport mechanism (the actual CD drive) resides.
I'm going to ignore the PLL jitter issue as I would just house the DAC chip in the same box the transport mechanism (the actual CD drive) resides.
Phase correction in the digital domain is trivial. If anti-imaging filters had been standardized, the CDs themselves could have had the correction "built in."
wa2ise said:
Guys, read Shannon closely . He specifies the reconstruction filter as the Sinc function, which has pre-echo. The pre-echo is required for proper reconstruction and does not make it to the actual output signal. In other words, the ringing filter results in a non-ringing analogue signal. If you don't believe this then please go through the math.
Since oversampling FIR filters (mostly) all try to approximate Sinc (and rightly so), they have to have pre-echo. The remaining question is: how good a Sinc approximation are they?
Now using ringing FIRs as anti-alias filters in an ADC is a different kettle of fish.
wa2ise said:
Yes. Do that. If you ignore that, you ignore one of the key differences between oversampled and non oversampled systems.
best
dddac said:
That is because your stimulus signal on the disc was a 'perfect' digital square, not? Like -100 +100 -100 +100 ...
Ask yourself: is such a signal allowed to exist in the 44.1kHz sampled space?
The answer to that is: no. Is it not a band-limited signal, and as such
it can not exist in a legal way in the sampled signal space. The only thing such a test does for you is to provide you with a readout of the oversampling filter's coefficients.
Same for all those impulse response tests: invalid stimulus.
Remember that a Shannon/Nyquest recording channel consists of anti-aliasing, ADC, DAC, reconstruction, and when testing one should test the whole chain, and not inject illegal signals before the DAC.
Unless one knows and understands the implications and the limitations of such a test.
dddac said:
Doede,
What remains is the result of imperfection in the filter. I won't claim you hear it or not, but the pre ringing shouldn't make it to the output when perfect implementation would be present.
Again, one of the many examples of concept versus implementation.
best
No Guido,
Sinc is a perfect reconstructor, but it too would ring when convolved with a non-limited impulse or square.
This is not about concept versus implementation, it is about understanding where the boundaries of the (digital) system under test are for any particular test.
Sinc is a perfect reconstructor, but it too would ring when convolved with a non-limited impulse or square.
This is not about concept versus implementation, it is about understanding where the boundaries of the (digital) system under test are for any particular test.
Guido and dddac,
You had best listen to Werner - he really knows this stuff.
Werner accurately explained why the first waveform has "ringing". It is because a true square wave has frequency components up to infinity. When you limit the bandwidth of the sqaure wave, you will get exactly what is seen in the first picture. The "square" wave is now approximated by only the first few harmonics... 1 kHz, 3 kHz, 5, 7, 9, 11, 13, 15, 17, and 19. The result will not, and must not look like a square wave any longer, because it is not, by the fact that the higher frequencies are gone. What you see is not ringing caused by any filter, imperfect or otherwise. It is exactly what a band-limited square wave looks like.
The fact that the second screen looks like a square wave, means that this DAC has some serious problems. It is reproducing all sorts of ultrasonic garbage that should not be there. The square edges are a symptom of the incaccuracy of this DAC, not proof of accuracy.
Try this: build a 22 kHz high pass filter and connect the outputs of your DACs to it. It should pass virtually nothing. Try to guess what you'll see on your scope with your non-OS DAC with square wave input?
You had best listen to Werner - he really knows this stuff.
Werner accurately explained why the first waveform has "ringing". It is because a true square wave has frequency components up to infinity. When you limit the bandwidth of the sqaure wave, you will get exactly what is seen in the first picture. The "square" wave is now approximated by only the first few harmonics... 1 kHz, 3 kHz, 5, 7, 9, 11, 13, 15, 17, and 19. The result will not, and must not look like a square wave any longer, because it is not, by the fact that the higher frequencies are gone. What you see is not ringing caused by any filter, imperfect or otherwise. It is exactly what a band-limited square wave looks like.
The fact that the second screen looks like a square wave, means that this DAC has some serious problems. It is reproducing all sorts of ultrasonic garbage that should not be there. The square edges are a symptom of the incaccuracy of this DAC, not proof of accuracy.
Try this: build a 22 kHz high pass filter and connect the outputs of your DACs to it. It should pass virtually nothing. Try to guess what you'll see on your scope with your non-OS DAC with square wave input?
Mac,
forget about that square wave. Next thing that happens is that someone pops in and brings Gibbs with him. No need for that.
Point is that the impulse response of digital is not that of the DAC. It is that of the ADC/DAC and all related filters together. Only if then ringing persists is there any problem.
BTW, I have nothing against non-oversampling DACs. Only against some of the arguments against oversampling DACs.
forget about that square wave. Next thing that happens is that someone pops in and brings Gibbs with him. No need for that.
Point is that the impulse response of digital is not that of the DAC. It is that of the ADC/DAC and all related filters together. Only if then ringing persists is there any problem.
BTW, I have nothing against non-oversampling DACs. Only against some of the arguments against oversampling DACs.
Werner said:
Hi
What I tried to explain was that the amount of ringing depends on the purity of the reconstructor.
best
macboy said:
I agree on the point of the squarewave as invalid input... So that answered my question a few topics ago 🙂 Still wondering if the preringing will be really gone when a valid transient (so no > 22kHz components) in the music ocures 😕 Any one did some testing and have pictures ?
Hoever, your point on "Try this" is not explaining anything as I am sure that if I built a 22kHz HP filter after my DAC output and feed it with the 1kHz (invalid) squarewave, I will se may be ringing, depending on the filter Q, by I will certainly see no pre-ringing. In "real world" analog filters there is still no prediction of transients .......
Never the less, we are probably looking at the wrong stuff to explain differences between the both sorts of DACs....
So now we are back where we started ?? Is that a kind of FIR as well ? 😎
regards
doede
Guys, have you ever imagined/checked how a sampled 22050hz sinus signal looks like ? It's a digital squarewave !
Have you ever checked how a properly filtered/bandwidth limited squarewave looks like ? That's not ringing what you see, that's EXACTLY what a filtered squarewave looks like !
Really interesting would be to check a 19+20khz signal through a OS and non-OS DAC. The OS-dac recreates this signal perfectly... (checked with scope and 96khz soundcard) But i have no non-OS...
I suspect that the guys prefering the nonOS simply love to listen to IMD... (Very common with poweramps)
Mike
Have you ever checked how a properly filtered/bandwidth limited squarewave looks like ? That's not ringing what you see, that's EXACTLY what a filtered squarewave looks like !
Really interesting would be to check a 19+20khz signal through a OS and non-OS DAC. The OS-dac recreates this signal perfectly... (checked with scope and 96khz soundcard) But i have no non-OS...
I suspect that the guys prefering the nonOS simply love to listen to IMD... (Very common with poweramps)
Mike
The saving graces of non-OS are
1) typical spectral contents of music fall off above 3kHz
2) the 1/44100 sec aperture of a non-OS multibit DAC implies a low-pass filter with interesting properties in the band from 44.1kHz-3kHz to 44.1kHz+3kHz.
So there is probably not more IMD than what you get with, say, a good moving coil cartridge ;-)
1) typical spectral contents of music fall off above 3kHz
2) the 1/44100 sec aperture of a non-OS multibit DAC implies a low-pass filter with interesting properties in the band from 44.1kHz-3kHz to 44.1kHz+3kHz.
So there is probably not more IMD than what you get with, say, a good moving coil cartridge ;-)
MikeB said:
Maybe, will measure. At least they bennefit from lower jitter. Funny to see that no-one realises this as to be the major advantage of NOS.
best
Werner said:
Hmm, i thought a dac should reproduce as accurate as possible ?
Some instruments have very high freq content, but you're right, typically there is not much content above ~3khz...
By playing this testsignal (19+20khz) through a non-os dac you immediately hear if there is a problem, if a 1khz sine becomes audible, the signal is distorted... If nothing is audible, i am wrong !
Would be interesting to see the result with a scope !
Mike
- Status
- Not open for further replies.