Hi Bernard,
"firewall" filters? I thought they were called "brickwall" filters because they abruptly truncate the HF above 22 kHz and introduce all kinds of phase anamolies in the process.
The ultrasonic hash filterless NOS produces actually gets attenuated naturally by our mechanical auditory system (ie, middle and inner ears) - this is known. The hash is plainly measurable but it probably does not contribute negatively to our perception of sound - I have never heard the filterless NOS design that I owned sound harch up on top - just the opposite of most OS designs. The HF was more like live music than most OS designs.
The extra samples introduced in the high frequencies by linear interpolation is distortion, but I'm not sure if this is perceptible or any worse than the inevitable few HF sample points present in the original data. At least phase is preserved, which I think is more important and audible than getting a perfect 16 kHz sine wave. Something OS and filtering can not do.
Why concentrate on frequency spectrum that contributes little to our perception of sound?
"firewall" filters? I thought they were called "brickwall" filters because they abruptly truncate the HF above 22 kHz and introduce all kinds of phase anamolies in the process.
The ultrasonic hash filterless NOS produces actually gets attenuated naturally by our mechanical auditory system (ie, middle and inner ears) - this is known. The hash is plainly measurable but it probably does not contribute negatively to our perception of sound - I have never heard the filterless NOS design that I owned sound harch up on top - just the opposite of most OS designs. The HF was more like live music than most OS designs.
The extra samples introduced in the high frequencies by linear interpolation is distortion, but I'm not sure if this is perceptible or any worse than the inevitable few HF sample points present in the original data. At least phase is preserved, which I think is more important and audible than getting a perfect 16 kHz sine wave. Something OS and filtering can not do.
Why concentrate on frequency spectrum that contributes little to our perception of sound?
Hi Bernhard,
I second poobah's comment.
MGH,
-Chris
I second poobah's comment.
MGH,
This is really going to depend heavily on the amplifier design you are running. The ultrasonic noise may cause all kinds of distortion in an amplifier. Now if your amp is perfect, I doubt the tweeter likes it very much. However, with the perfect amp and speakers, you could enjoy an evening in the late summer forest, mosquito free.
Frequency response is a large component to how we perceive sound. It really is. Honest. I do take your point that there are other matters to consider, but now you have done a 180 and are going too far the other way.
-Chris
Guys,
Hell... go back to one of my first posts on this thread.
Remember something here... oversampling, by whatever method, is a means of cheating in a sense. The motivation is to avoid the cost of building an 8th order (or more) filter with good caps and good op-amps. It requires precise capacitors... a hunk of silicon with an FIR routine is cheaper... way cheaper.
I think the OS gets a bad rap because of jitter... it is NOT rounding errors in 25 bit processors... really.
Analog filters get a bad rap because of delay and/or phase. But you can't hear either. Don't construe this to mean that phase and delay don't matter where things like crossovers are concerned... they do matter because a single signal is being split into several and recombined in the air. In the case of a single signal... you can't hear it. Your ear is a spectrum analyzer; it doesn't track phase. Again, do the math. You only have to move your head 3/8 of an inch and all the "phase" info is destroyed. (1140 *12) / (20,000 *2) = 0.3 something inches.
We all got screwed here... face it. The distance between 20 kHz and 22.05 kHz was just too small to pull off a low pass filter with 100 dB stop band attentuation. Get mad at Phillips and Sony all you want... but you won't get past Nyquist and Shannon. If the sampling frequency had been chosen just a bit higher, maybe 30 kHz or so, this would all be easy.
I have looked through Guido Tent's stuff; looks pretty good to me; with proper reasoning behind things.
On a final note, if you don't understand the math behind it all; DON'T take the word of someone else who doesn't understand it either. You would not go to a doctor for bread or a baker for surgery...
or.... would you?
Hell... go back to one of my first posts on this thread.
Remember something here... oversampling, by whatever method, is a means of cheating in a sense. The motivation is to avoid the cost of building an 8th order (or more) filter with good caps and good op-amps. It requires precise capacitors... a hunk of silicon with an FIR routine is cheaper... way cheaper.
I think the OS gets a bad rap because of jitter... it is NOT rounding errors in 25 bit processors... really.
Analog filters get a bad rap because of delay and/or phase. But you can't hear either. Don't construe this to mean that phase and delay don't matter where things like crossovers are concerned... they do matter because a single signal is being split into several and recombined in the air. In the case of a single signal... you can't hear it. Your ear is a spectrum analyzer; it doesn't track phase. Again, do the math. You only have to move your head 3/8 of an inch and all the "phase" info is destroyed. (1140 *12) / (20,000 *2) = 0.3 something inches.
We all got screwed here... face it. The distance between 20 kHz and 22.05 kHz was just too small to pull off a low pass filter with 100 dB stop band attentuation. Get mad at Phillips and Sony all you want... but you won't get past Nyquist and Shannon. If the sampling frequency had been chosen just a bit higher, maybe 30 kHz or so, this would all be easy.
I have looked through Guido Tent's stuff; looks pretty good to me; with proper reasoning behind things.
On a final note, if you don't understand the math behind it all; DON'T take the word of someone else who doesn't understand it either. You would not go to a doctor for bread or a baker for surgery...
or.... would you?
Chris,
"Frequency response is a large component to how we perceive sound. It really is. Honest. I do take your point that there are other matters to consider, but now you have done a 180 and are going too far the other way."
When I said frequency spectrum, I meant ultrasonic hash >20kHz produced by filterless NOS. How does hash above 20kHz significantly improve our perception of sound quality? May be you misunderstood.
"Frequency response is a large component to how we perceive sound. It really is. Honest. I do take your point that there are other matters to consider, but now you have done a 180 and are going too far the other way."
When I said frequency spectrum, I meant ultrasonic hash >20kHz produced by filterless NOS. How does hash above 20kHz significantly improve our perception of sound quality? May be you misunderstood.
Poobah,
"On a final note, if you don't understand the math behind it all; DON'T take the word of someone else who doesn't understand it either. You would not go to a doctor for bread or a baker for surgery...or.... would you?"
Actually I may have gone to a barber - they were the surgeons of their time.
However, the very engineers who understood the math produced some of the most atrocious sounding digital playback. I'm not sure if you're implying EC doesn't understand the math. I think he does. I trust someone who actually designs his product thru critical listening than those who produce a product blindly following Nyquist or Shannons theorem.
"On a final note, if you don't understand the math behind it all; DON'T take the word of someone else who doesn't understand it either. You would not go to a doctor for bread or a baker for surgery...or.... would you?"
Actually I may have gone to a barber - they were the surgeons of their time.
However, the very engineers who understood the math produced some of the most atrocious sounding digital playback. I'm not sure if you're implying EC doesn't understand the math. I think he does. I trust someone who actually designs his product thru critical listening than those who produce a product blindly following Nyquist or Shannons theorem.
Hi John,
I have a question about reclocking that you discussed in post #501.
You state that the MCLK should be used to reclock the I2S signals as it is in sync with the other signals involved. This I totally understand.
Now, what would happen if a free runnning clock was used at the DAC for triggering the reclocking? The ideal scenario is to use this free running clock and send it back to the CD player.
But, I think I need to use a 11 or 16 Mhz MCLK for my CDP, and I wonder if things would be less sensitive to jitter if I could use a lower clock frequency at the DAC end (more like 4 or 8Mhz).
I suppose the 16MHz could be divided down but I assume that will add no benefit.
So, basically I think my question is, how crucial is it for the CLK (not necessarilly MCLK - for arguments sake) and BCK be derived from the same clock, *if* it is possible for jitter to be reduced?
Cheers,
Phil
I have a question about reclocking that you discussed in post #501.
You state that the MCLK should be used to reclock the I2S signals as it is in sync with the other signals involved. This I totally understand.
Now, what would happen if a free runnning clock was used at the DAC for triggering the reclocking? The ideal scenario is to use this free running clock and send it back to the CD player.
But, I think I need to use a 11 or 16 Mhz MCLK for my CDP, and I wonder if things would be less sensitive to jitter if I could use a lower clock frequency at the DAC end (more like 4 or 8Mhz).
I suppose the 16MHz could be divided down but I assume that will add no benefit.
So, basically I think my question is, how crucial is it for the CLK (not necessarilly MCLK - for arguments sake) and BCK be derived from the same clock, *if* it is possible for jitter to be reduced?
Cheers,
Phil
[jack mode]
Anybody has a schematic for a nine pole elliptical filter for non os ?
Could cascade three of those ?
http://eecs.oregonstate.edu/~batten/pcbclass/appnotes/AB-026.pdf#search="+antialiasing+filter+gic"
[/jack mode]
Anybody has a schematic for a nine pole elliptical filter for non os ?
Could cascade three of those ?
http://eecs.oregonstate.edu/~batten/pcbclass/appnotes/AB-026.pdf#search="+antialiasing+filter+gic"
[/jack mode]
then 0,25dB would be ok 
Another variation of linear interpolation came to my mind, did not think it through...
It could use a 8x os filter to produce "good" new samples.
A decade counter distributes the samples to 8 DACs in a ring configuration.
The result is a data update rate of 44,1 kHz for each DAC like in non os, and overlap of the DACs is the same as with linear interpolation but it uses true ( interpolated )samples.
Any good ?
It could use a 8x os filter to produce "good" new samples.
A decade counter distributes the samples to 8 DACs in a ring configuration.
The result is a data update rate of 44,1 kHz for each DAC like in non os, and overlap of the DACs is the same as with linear interpolation but it uses true ( interpolated )samples.
Any good ?
Ummm, Poobah
don't we need 44,1 / 20 ?
There is nothing in between.
http://webench.national.com/appinfo/webench/filters/design_requirements.cgi
don't we need 44,1 / 20 ?
There is nothing in between.
http://webench.national.com/appinfo/webench/filters/design_requirements.cgi
Sorry Bernhard,
For audio we 2 "brickwall" filters with pass and stop at 20 & 22.05 kHz.
We need one when we record or digitize the music; and one when we play it back.
In the case of playback of a 20 kHz sine, there is an image freq. that appears at 24.1 kHz. Also called a mirror because the "mirror" is at 22.05.
For audio we 2 "brickwall" filters with pass and stop at 20 & 22.05 kHz.
We need one when we record or digitize the music; and one when we play it back.
In the case of playback of a 20 kHz sine, there is an image freq. that appears at 24.1 kHz. Also called a mirror because the "mirror" is at 22.05.
