the superiority of 96 kHz sampling Options no one mentioned aliasing

Mark <[email protected]> wrote:
>There was a lot of talk about who can/can't hear what above 20 kHz..
>but nobody brought up probably the most important issue..

Because it's not an issue any more.

>In a 44.1 system, if you feed in energy at 21 kHz, that nominally
>inaudiable energy will be alised back down to 1.05 kHz where it
>certainly is audible. It will be attenuated by the amount of
>attenuation that the anti-alias filter provides which may not be
>suffiecent to make it inaudible. A system with a higher sampling
>rate will not suffer from that defect as much becasue there is a
>larger guard band between the upper end of the audio spectrum (20 kHz)
>and the range where aliasing starts to occur.

This is true for a system that does not employ oversampling.

>When sampling at 44.1, it's all about the anti-aliasing filter
>performance. At higher sampling rates, the filter performance is
>much less critical.

This was entirely true in 1985, and it was at the time the most severe
limitation in digital systems.

However, we have a solution for that, and that solution is oversampling.
That is, we run the converter at very high speed, then downsample on the
fly in the digital domain while doing the majority of the anti-aliasing
filtering in the digital domain as well. The analogue filter can be a
single-pole filter well above the audio band, because it only needs to cut
off at half the frequency of the clock on the front end, not on the back
end.

>This is all easily measured and is not black magic and needs no golden
>ear to determine..

Absolutely, but it's a solved problem now, and it was solved pretty well
by the early 1990s. There is no need for the storage sampling rate to be
the same as the converter front end sampling rate.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."

"Mark" <[email protected]> wrote in message
news:[email protected]
> important issue..

> In a 44.1 system, if you feed in energy at 21 kHz, that
> nominally inaudible energy will be aliased back down to
> 1.05 kHz where it certainly is audible.

Sounds like you presume that the typical ADC for professional recording was
designed by first year engineering students who didn't yet know how to spell
"anti-aliasing filter". Even then, you blew the example with a freshman
mistake.

A quick tutorial:

In a 44.1 kHz system, if you feed in energy at 21 kHz, you are below the
Nyquist frequency, so it will be converted to a digital signal at 21 kHz.

If you want to talk to talk about aliasing in a 44.1 kHz system, you have to
start out with a signal that is *above* 22.05 kHz, not below it. For an
example that could potentially cause a spurious response at 1.05 KHz, you
have to start out with a signal at 23.1 kHz. Then, you still have to contend
with an anti-aliasing filter that is now 30 or so dB down.

This less-than-total rejection seems sloppy, but it is intentional. These
same manufacturers sell ADCs that have a lot more rejection above the
Nyquist frequency, but they are usually targeted at test and measurement
applications. The trade-off is more ripple in the audio band.

> It will be attenuated by the amount of attenuation that the
> anti-alias filter provides which may not be sufficient to
> make it inaudible.

True, bad things can happen with cheap converters. That's why we pay the big
bucks for professional-grade parts. I mean I'm talking like $5 each, or a
little more.

To put this into perspective, check out
http://www.cirrus.com/en/products/pro/detail/P1024.html . If you download
the data sheet and check out figure 5,

Rejection is more than 100 dB down at .58 Fs, or 25,578 Hz. and 10 dB down
at 22,491 Hz.

Arny Krueger wrote:
> "Mark" <[email protected]> wrote in message
> news:[email protected]
>> important issue..
>
>> In a 44.1 system, if you feed in energy at 21 kHz, that
>> nominally inaudible energy will be aliased back down to
>> 1.05 kHz where it certainly is audible.
>
> Sounds like you presume that the typical ADC for professional recording was
> designed by first year engineering students who didn't yet know how to spell
> "anti-aliasing filter". Even then, you blew the example with a freshman
> mistake.
>
> A quick tutorial:
>
> In a 44.1 kHz system, if you feed in energy at 21 kHz, you are below the
> Nyquist frequency, so it will be converted to a digital signal at 21 kHz.
>
> If you want to talk to talk about aliasing in a 44.1 kHz system, you have to
> start out with a signal that is *above* 22.05 kHz, not below it. For an
> example that could potentially cause a spurious response at 1.05 KHz, you
> have to start out with a signal at 23.1 kHz. Then, you still have to contend
> with an anti-aliasing filter that is now 30 or so dB down.
>

Oops! If you want a spurious response at 1.05kHz, you need to start out
with a signal at 43.05kHz - it is a mirror round the Nyquist frequency,
remember.

d

"Don Pearce" <[email protected]> wrote in message
news:[email protected]
> Arny Krueger wrote:
>> "Mark" <[email protected]> wrote in message
>> news:[email protected]
>>> important issue..
>>
>>> In a 44.1 system, if you feed in energy at 21 kHz, that
>>> nominally inaudible energy will be aliased back down to
>>> 1.05 kHz where it certainly is audible.
>>
>> Sounds like you presume that the typical ADC for
>> professional recording was designed by first year
>> engineering students who didn't yet know how to spell
>> "anti-aliasing filter". Even then, you blew the example
>> with a freshman mistake.

>> A quick tutorial:

>> In a 44.1 kHz system, if you feed in energy at 21 kHz,
>> you are below the Nyquist frequency, so it will be
>> converted to a digital signal at 21 kHz.

>> If you want to talk to talk about aliasing in a 44.1 kHz
>> system, you have to start out with a signal that is
>> *above* 22.05 kHz, not below it. For an example that
>> could potentially cause a spurious response at 1.05 KHz,
>> you have to start out with a signal at 23.1 kHz. Then,
>> you still have to contend with an anti-aliasing filter
>> that is now 30 or so dB down.

> Oops! If you want a spurious response at 1.05kHz, you
> need to start out with a signal at 43.05kHz - it is a
> mirror round the Nyquist frequency, remember.

Right you are. My bad.

Well, this makes things just peachy. That 43.05 kHz signal is way out there
in the -100 dB plus stop band. I've actually tried to force signals this
high into the inputs of ADCs at like 10 volts RMS, and have found that their
dynamic range is surprisingly good up there.

Arny Krueger wrote:
> "Don Pearce" <[email protected]> wrote in message
> news:[email protected]
>> Arny Krueger wrote:
>>> "Mark" <[email protected]> wrote in message
>>> news:[email protected]
>>>> important issue..
>>>> In a 44.1 system, if you feed in energy at 21 kHz, that
>>>> nominally inaudible energy will be aliased back down to
>>>> 1.05 kHz where it certainly is audible.
>>> Sounds like you presume that the typical ADC for
>>> professional recording was designed by first year
>>> engineering students who didn't yet know how to spell
>>> "anti-aliasing filter". Even then, you blew the example
>>> with a freshman mistake.
>
>>> A quick tutorial:
>
>>> In a 44.1 kHz system, if you feed in energy at 21 kHz,
>>> you are below the Nyquist frequency, so it will be
>>> converted to a digital signal at 21 kHz.
>
>>> If you want to talk to talk about aliasing in a 44.1 kHz
>>> system, you have to start out with a signal that is
>>> *above* 22.05 kHz, not below it. For an example that
>>> could potentially cause a spurious response at 1.05 KHz,
>>> you have to start out with a signal at 23.1 kHz. Then,
>>> you still have to contend with an anti-aliasing filter
>>> that is now 30 or so dB down.
>
>> Oops! If you want a spurious response at 1.05kHz, you
>> need to start out with a signal at 43.05kHz - it is a
>> mirror round the Nyquist frequency, remember.
>
> Right you are. My bad.
>
> Well, this makes things just peachy. That 43.05 kHz signal is way out there
> in the -100 dB plus stop band. I've actually tried to force signals this
> high into the inputs of ADCs at like 10 volts RMS, and have found that their
> dynamic range is surprisingly good up there.
>
>
>

It is always amusing to watch a student trying to get to grips with a
sampling oscilloscope. Great high frequency response, and capable of
great measurements. Unfortunately they have no anti-alias filter, and it
takes experience plus a knowledge of what the signal should look like to
get meaningful results out of it. The funniest thing is to watch them
wind the timebase switch back and forth with a puzzled look on their
face as the same signal keeps disappearing and reappearing.

d