Is higher oversampling ratio better in signal fidelity in digital world?

For digitized audio signal, intuitively higher oversampling ratio, the
digitized signal is closer to original signal, therefore, higher
fidelity.

In digital work, for example, transferring digital data through wcdma/cdma
network, the digital data / digitized voice data will be oversampled, maybe
resampled, then modulated and transmitted. Will higher oversampling rato
offer better EVM, spectrum purity? I would think so, but my experiment on
EVM and ACPR seems not much difference. Can someone explain why?

thanks

>For digitized audio signal, intuitively higher oversampling ratio, the
>digitized signal is closer to original signal, therefore, higher
>fidelity.

As longer as you don't have frequency components that are
above samplerate/2 (nyquist frequency), you cannot get "closer"
to the original signal.  You can perfectly reconstruct your signal
(at least in theory). If you have components above the nyquist
frequency (before you sample the signal) you get aliasing and
once your signal is polluted you cannot improve it by oversampling
afterwards.

gr.
Anton 

"yhe" <yhe@keithley.com> wrote in message 
news:V6KdnRGLzLCE8fHZRVn-qA@giganews.com...
> For digitized audio signal, intuitively higher oversampling ratio, the
> digitized signal is closer to original signal, therefore, higher
> fidelity.
>
> In digital work, for example, transferring digital data through wcdma/cdma
> network, the digital data / digitized voice data will be oversampled, 
> maybe
> resampled, then modulated and transmitted. Will higher oversampling rato
> offer better EVM, spectrum purity? I would think so, but my experiment on
> EVM and ACPR seems not much difference. Can someone explain why?
>
> thanks

Yhe,

It's impportant to be careful with terms - although I realize that it's so 
easy to refer to digital streams as though they are signals we can hear, 
etc.

I don't think that "fidelity" applies to anything but something you can hear 
or measure.  Someone else might have another view.

Anyway, perhaps you might look at things this way:

"Fidelity" would typically include consideration of anomalies like:

- noise
- harmonic distortion
- multipath / echoes
- frequency response

Right now I can't think of anything that wouldn't fit into one of those 
categories and the last two might be lumped into one.

"noise" would include things like quantization noise caused by being 
digital.  Oversampling might help if the engergy is spread into higher 
frequencies and some of it is filtered out - but I'm not an expert on this 
topic.  Otherwise, oversampling has nothing directly to do with the 
existence of the quantization noise.

"frequency response" is affected by the sample-hold and by the 
reconstruction filtering.  Sampling at higher rates could allow both to not 
affect the highest frequencies of interest in the signal - so that is a 
really a detail.  A "properly" configured system would have acceptable 
frequency response.

"harmonic distortion" would probably have nothing to do with the sample rate 
by itself nor would "multipath / echoes".

Now, we have *no* idea what your "experiment" entails.  So, how could anyone 
explain your results?

Fred

banton wrote:

>>For digitized audio signal, intuitively higher oversampling ratio, the
>>digitized signal is closer to original signal, therefore, higher
>>fidelity.
> 
> 
> As longer as you don't have frequency components that are
> above samplerate/2 (nyquist frequency), you cannot get "closer"
> to the original signal.  You can perfectly reconstruct your signal
> (at least in theory). If you have components above the nyquist
> frequency (before you sample the signal) you get aliasing and
> once your signal is polluted you cannot improve it by oversampling
> afterwards.

If there are components above Nyquist, it can hardly be called oversampling.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

>banton wrote:
>
>>>For digitized audio signal, intuitively higher oversampling ratio, the
>>>digitized signal is closer to original signal, therefore, higher
>>>fidelity.
>> 
>> 
>> As longer as you don't have frequency components that are
>> above samplerate/2 (nyquist frequency), you cannot get "closer"
>> to the original signal.  You can perfectly reconstruct your signal
>> (at least in theory). If you have components above the nyquist
>> frequency (before you sample the signal) you get aliasing and
>> once your signal is polluted you cannot improve it by oversampling
>> afterwards.
>
>If there are components above Nyquist, it can hardly be called
oversampling.

Ok right, I just wanted to point out that if the signal to sample is below
nyquist, you can do so and reconstruct it perfectly (and oversampling makes
no difference here). Oversampling just helps in the process of AD
conversion so that your filter in the analog part is less critical because
you have more frequency headroom (so if it is not steep enough you want get
that much aliasing). But I don't claim to be an expert so I might be
missing something....

gr.
Anton
 

On Thu, 18 May 2006 16:15:18 -0500, "banton" <bantone@web.de> wrote:

>>banton wrote:
>>
>>>>For digitized audio signal, intuitively higher oversampling ratio, the
>>>>digitized signal is closer to original signal, therefore, higher
>>>>fidelity.
>>> 
>>> 
>>> As longer as you don't have frequency components that are
>>> above samplerate/2 (nyquist frequency), you cannot get "closer"
>>> to the original signal.  You can perfectly reconstruct your signal
>>> (at least in theory). If you have components above the nyquist
>>> frequency (before you sample the signal) you get aliasing and
>>> once your signal is polluted you cannot improve it by oversampling
>>> afterwards.
>>
>>If there are components above Nyquist, it can hardly be called
>oversampling.
>
>Ok right, I just wanted to point out that if the signal to sample is below
>nyquist, you can do so and reconstruct it perfectly (and oversampling makes
>no difference here). Oversampling just helps in the process of AD
>conversion so that your filter in the analog part is less critical because
>you have more frequency headroom (so if it is not steep enough you want get
>that much aliasing). But I don't claim to be an expert so I might be
>missing something....
>
>gr.
>Anton

Hi Anton,

  The "point" you were making is certainly correct.
Your original words:

   "you cannot get "closer" to the original signal"

were good.  You were trying to teach the original 
poster (yhe) a FUNDAMENTAL aspect of discrete 
signals.  (I.E., if you satisfy Nyquist, then you have 
*ALL* the information about an analog signal that you can 
possibly have.)

See Ya',
[-Rick-]
 

I'm not sure if I exactly agree with the statement of not being able to get 
closer to the original signal then simply satisfying Nyquist. It seems to me 
that if you sampled (lets say) a 100 Hz perfect sinusoidal wave form at 200 
Hz or so, that during playback (DAC process), this will appear as a very 
chunky square wave that contains tons of harmonics. It you oversampled that 
same 100 Hz sine wave at 44kHz then you have better represented the original 
sine wave and playback would contain vary little harmonic content.  The two 
methods will sound different.

It really depends on what type of audio you are sampling and the dynamics of 
the recording and playback electronics (mic, speakers etc) whether you would 
ever notice a difference.

Thomas Magma


"yhe" <yhe@keithley.com> wrote in message 
news:V6KdnRGLzLCE8fHZRVn-qA@giganews.com...
> For digitized audio signal, intuitively higher oversampling ratio, the
> digitized signal is closer to original signal, therefore, higher
> fidelity.
>
> In digital work, for example, transferring digital data through wcdma/cdma
> network, the digital data / digitized voice data will be oversampled, 
> maybe
> resampled, then modulated and transmitted. Will higher oversampling rato
> offer better EVM, spectrum purity? I would think so, but my experiment on
> EVM and ACPR seems not much difference. Can someone explain why?
>
> thanks
>
> 

yhe wrote:
> For digitized audio signal, intuitively higher oversampling ratio, the
> digitized signal is closer to original signal, therefore, higher
> fidelity.

Intuitively, yes.  Practically, maybe.

If the signal has no significant spectral content that can be aliased 
into baseband then you don't gain anything by sampling at a higher rate, 
because with good enough filters you can reconstruct the input just fine.

Oversampling, digitally filtering, and resampling does make the filter 
implementation easier.  Ditto for reconstruction.
> 
> In digital work, for example, transferring digital data through wcdma/cdma
> network, the digital data / digitized voice data will be oversampled, maybe
> resampled, then modulated and transmitted.

For any audio data through a network the audio is going to be as 
compressed as it can be made.  The cell networks use algorithms that 
emulate the human vocal tract to get the data rates down, in fact. 
Higher data rates would allow for less compression and more fidelity, 
but you have to pay for your data rate.

> Will higher oversampling rato
> offer better EVM, spectrum purity? I would think so, but my experiment on
> EVM and ACPR seems not much difference. Can someone explain why?
> 
What do you mean by EVM and ACPR?

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Posting from Google?  See http://cfaj.freeshell.org/google/

"Applied Control Theory for Embedded Systems" came out in April.
See details at http://www.wescottdesign.com/actfes/actfes.html

Thomas Magma wrote:

> I'm not sure if I exactly agree with the statement of not being able to get 
> closer to the original signal then simply satisfying Nyquist. It seems to me 
> that if you sampled (lets say) a 100 Hz perfect sinusoidal wave form at 200 
> Hz or so, that during playback (DAC process), this will appear as a very 
> chunky square wave that contains tons of harmonics. It you oversampled that 
> same 100 Hz sine wave at 44kHz then you have better represented the original 
> sine wave and playback would contain vary little harmonic content.  The two 
> methods will sound different.

The output of the DAC is not what you need to be thinking about. You
ignored the role of the reconstruction filter. If the sampling frequency
is 200 Hz, everything above 100 Hz must be removed from the analog
output. What harmonics did you mean?

The sampling frequency must be greater that twice the highest signal
component. When sampling at 200 Hz, 100 Hz can't be reliably
reconstructed. 200.001 Hz is sufficient, but only if the sampling lasts
a good part of 1000 seconds and the signal remains unchanged for that
time. The bit of oversampling that is always needed to allow practical
anti-alias and reconstruction filters also assures reasonable resolution
times, but in theoretical cases that don't call for filters, resolution
time has to be accounted for.

Jerry
-- 
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������

Thomas Magma wrote:

> I'm not sure if I exactly agree with the statement of not being able to
> get closer to the original signal then simply satisfying Nyquist. It seems
> to me that if you sampled (lets say) a 100 Hz perfect sinusoidal wave form
> at 200 Hz or so, that during playback (DAC process), this will appear as a
> very chunky square wave that contains tons of harmonics.

Sampling at 200 Hz is not sufficient for a 100Hz signal, as Jerry wrote.
Sampling at a somewhat higher rate is really sufficient. Do not make the mistake
and try to interpolate the PCM data with your eyes! This not the same as what
the reconstruction filter does. 

Not long ago a newborn analog hifi magazine in Germany did exactly this. 
The author displayed a 20 kHz signal sampled at 44.1 kHz and it looked 
like a modulated row of rectangles. He missed the point that he was 
actually looking at the addition of two frequencies: 

20 khz and its aliasing partner at 44.1-20 = 24.1 kHz. 

Sonce these two signals do not differ much in frequency it looks like a modulation,
but be assured that the reconstruction filter gets rid of the 24.1 kHz and throws 
out a perfect 20 kHz signal.

Of course, nobody uses DACs without oversampling, so the whole
argument was doomed anyway. 

bye 
Andreas
--
Andreas H�nnebeck | email: acmh@gmx.de
----- privat ---- | www  : http://www.huennebeck-online.de
Fax/Anrufbeantworter: 0721/151-284301
GPG-Key: http://www.huennebeck-online.de/public_keys/andreas.asc
PGP-Key: http://www.huennebeck-online.de/public_keys/pgp_andreas.asc