Sample Rate Conversion

[john12ax7]

Thread took an interesting side tangent. I've only done non -audio designs, but have listened to enough converter and clock combos to hear a difference, even when you theoretically shouldn't. Perhaps simple rms jitter measurements are not enough. I also prefer normal res audio played back at higher res, this suggests the filters are inadequate.

 

[abbey road d enfer]

I also prefer normal res audio played back at higher res, this suggests the filters are inadequate.

What do you mean? How do you listen to a 44.1/16 program at e.g. 96/24?

 

[Matt Nolan]

I also prefer normal res audio played back at higher res, this suggests the filters are inadequate.

This can be seen as well as heard. I had a supposedly "pro" Tascam field recorder and if you ran it in passthrough (ie AD->DA) and scoped the output, you could see aliasing happening if you gave it frequencies just above the Nyquist rate.

 

[ccaudle]

This can be seen as well as heard. I had a supposedly "pro" Tascam field recorder and if you ran it in passthrough (ie AD->DA) and scoped the output, you could see aliasing happening if you gave it frequencies just above the Nyquist rate.

Nearly every ADC and most oversampling DAC integrated chips use what is known as half-band filters. That type of digital filter design saves coefficient storage space and is easier to compute, but results in a filter response that is only -6 dB at Nyquist frequency, so you are guaranteed to get aliasing or image leak-through if you have a lot of high frequency content.

 

[abbey road d enfer]

This can be seen as well as heard. I had a supposedly "pro" Tascam field recorder and if you ran it in passthrough (ie AD->DA) and scoped the output, you could see aliasing happening if you gave it frequencies just above the Nyquist rate.

All converters are not equal, that's for sure. Now, how did you do it? What SR, what frequency, what level...? It is notorious that, in this respect, Single Speed (44.1/48) is lacking, but running at 88.2/96 is almost totally immune of aliasing in the presenc of a "normal" audio signal, i.e. one where the content above 20kHz is very small.

 

[Matt Nolan]

All converters are not equal, that's for sure. Now, how did you do it? What SR, what frequency, what level...? It is notorious that, in this respect, Single Speed (44.1/48) is lacking, but running at 88.2/96 is almost totally immune of aliasing in the presenc of a "normal" audio signal, i.e. one where the content above 20kHz is very small.

It was a 44.1 test. There was something odd about that Tascam though. However I can't remember what it was. I think it was that it used exactly the same analogue pre-ADC filter at all sample rates.

 

[JohnRoberts]

I am old enough to remember when digital audio kinda sucked...(1970s). These days it is remarkably good.

JR

 

[iomegaman]

So I've started reading the Art of digital audio, fascinating book, he makes a compelling argument right away that digital audio is probably more accurately "numerical audio"...but as I chew on this I need an "explain it like I'm five" to the analog process...

So lets stick a microphone in front of my acoustic guitar...I hit the "A" string and it it vibrates at 440 Hz...so that is the "tone" or sound I hear...the microphone capsule picks up the air movement and it vibrates the coil at 440 Hz inducing a voltage that contains 2 distinct properties, the actual AC volt and the current?

I'm trying to wrap my head around how AMPLITUDE gets sampled...because we have the sound we hear, but that sound is at different volumes over time and often the initial volume is pretty far from the subsequent volumes (especially in something like a snare hit)...

The initial pick on the string is at 440 Hz as is the tail of the fade...so we really only have (essentially) one frequency to record but multiple volumes or amplitudes...

So in my limited understanding...we have our coil converting micro-volt at a specific current that we take a discrete sample of on a given time line based on a specific interval that is AT LEAST 2x the highest frequency we are recording...so in this case I would only need to take a sample every 881 cycles per second to capture 440 Hz accurately...but the sample would indicate a CURRENT difference between the first and second samples?

881 samples a second could contain a lot of information but the primary information in this case would be the difference in current?

Or am I mixing my electronics with my coffee here?

 

[JohnRoberts]

So I've started reading the Art of digital audio, fascinating book, he makes a compelling argument right away that digital audio is probably more accurately "numerical audio"...but as I chew on this I need an "explain it like I'm five" to the analog process...

So lets stick a microphone in front of my acoustic guitar...I hit the "A" string and it it vibrates at 440 Hz...so that is the "tone" or sound I hear...the microphone capsule picks up the air movement and it vibrates the coil at 440 Hz inducing a voltage that contains 2 distinct properties, the actual AC volt and the current?

frequency, and amplitude

I'm trying to wrap my head around how AMPLITUDE gets sampled...because we have the sound we hear, but that sound is at different volumes over time and often the initial volume is pretty far from the subsequent volumes (especially in something like a snare hit)...

magic

The initial pick on the string is at 440 Hz as is the tail of the fade...so we really only have (essentially) one frequency to record but multiple volumes or amplitudes...

So in my limited understanding...we have our coil converting micro-volt at a specific current that we take a discrete sample of on a given time line based on a specific interval that is AT LEAST 2x the highest frequency we are recording...so in this case I would only need to take a sample every 881 cycles per second to capture 440 Hz accurately...but the sample would indicate a CURRENT difference between the first and second samples?

881 samples a second could contain a lot of information but the primary information in this case would be the difference in current?

Or am I mixing my electronics with my coffee here?

Don't over think it... Imagine looking out through a screen door... all the little holes in the screen are effectively samples in x and y axis... You don't see dots you see a continous image.

Be patient there is lots to learn.

JR

 

[abbey road d enfer]

It was a 44.1 test. There was something odd about that Tascam though. However I can't remember what it was. I think it was that it used exactly the same analogue pre-ADC filter at all sample rates.

That is actually quite common, as I could see when I tested several soundcards. Audio ADC's and DAC's are optimized for ...audio. The actual frequency response is not very different when switching SR.
The advantages of using a higher SR are a little less noise, a better phase response and largely inproved anti-aliasing.

 

[Newmarket]

So I've started reading the Art of digital audio, fascinating book, he makes a compelling argument right away that digital audio is probably more accurately "numerical audio"...but as I chew on this I need an "explain it like I'm five" to the analog process...
...

You don't need to be concerned with "CURRENT".
Just the instantaneous voltages at the sample points.

 

[abbey road d enfer]

For historic and practical reasons, data converted into electric quantities are commonly voltages, very seldom currents. It could have been instant power or current, but voltages were chosen.
As a corollary, impedances have to be controlled.

 

[Dan Mills]

I would note that while a half band filter is only -6dB at Nyquist, it is usually falling like the side of a house.
What this means is that if the stopband is say 2kHz out, you alias maybe 24-26kHz down to 22-24kHz (At a 48kHz SR), still well outside the audio band.

ADCs are RF components not audio ones, especially with regards to the signal and clock inputs, it pays to treat them as such (One can view an ADC as a mixer with a more or less square wave clock and not be very wrong). In particular modern charge transfer converters take short pulses of current from the inputs, and that few nF of input cap right at the pins REALLY, REALLY matters, but also consider line length and reflections!

Anti imaging filters on the DACs are also subtle as they are again RF in some sense and knocking that down before hitting the opamps is sometimes helpful.

All of this stuff is measurable of course, and there is no magic.

 

[Brian Roth]

I'm just a dummy here but eons ago someone gave me an "AHA" moment as I was watching the waveform of a song being viewed on an oscilliscope.

The center vertical line on the graticule was the point where the signal was measured from second to second (millisecond/microsecond/whatever) into a digital number. Made sense to me.....



Bri

 

[mhelin]

I don't think aliasing in ADC is much of an issue anymore, for an example regarding the Cirrus CS5381 "the analog modulator samples the input at 6.144 MHz. The digital filter will reject signals within the stopband of the filter. However, there is no rejection for input signals which are (n 6.144 MHz) the digital passband frequency, where n=0,1,2,... Refer to Figure 24, which shows the suggested filter that will attenuate any noise energy at 6.144 MHz in addition to providing the optimum source impedance for the modulators". All audio band filtering is done in digital domain.

 

[ccaudle]

...but the sample would indicate a CURRENT difference between the first and second samples?

No, the sample just indicates the magnitude of the voltage at the time of the sample clock.

 

[Matt Nolan]

That is actually quite common, as I could see when I tested several soundcards. Audio ADC's and DAC's are optimized for ...audio. The actual frequency response is not very different when switching SR.
The advantages of using a higher SR are a little less noise, a better phase response and largely inproved anti-aliasing.

It seemed odd to me that you wouldn't change the analogue Nyquist filter before the ADC to allow less pass band ripple and better phase response when using higher sample rates when you have a more relaxed option for brick-wall-ness. Maybe they can get those filters pretty good anyhow.

And for people wanting to record bats, or samples for extreme time-stretch / slow-down manipulation, surely they'd want an option where 192ksamples/s actually has a Nyquist of around 96kHz. Maybe that's just too specialist.

 

[abbey road d enfer]

It seemed odd to me that you wouldn't change the analogue Nyquist filter before the ADC to allow less pass band ripple and better phase response when using higher sample rates when you have a more relaxed option for brick-wall-ness. Maybe they can get those filters pretty good anyhow.

And for people wanting to record bats, or samples for extreme time-stretch / slow-down manipulation, surely they'd want an option where 192ksamples/s actually has a Nyquist of around 96kHz. Maybe that's just too specialist.

These limitations pertain to rather cheap soundcards, typically below 100€. I was in the market for a soundcard that would give me about 90kHz BW at 192k SR. Several have a BW that does not exceed 40kHz.
I had poor results with a Native Instruments soundcard, where the BW is about 82kHz, but at the cost of excessive noise in the 40k-80kHz range.
I had rather good results with a Tascam soundcard, but now I'm using a Cosmos ADC and a Topping DAC with an aggregating driver, which achieves about 85kHz with excellent noise. I'm expecting the Cosmos DAC, but it is delayed.

 

[Newmarket]

I had rather good results with a Tascam soundcard, but now I'm using a Cosmos ADC and a Topping DAC with an aggregating driver, which achieves about 85kHz with excellent noise. I'm expecting the Cosmos DAC, but it is delayed.

Can you tell us what the TASCAM kit was ? Thanks

 

[abbey road d enfer]

US-2x2-HR