Patrick Dixon;131267 Wrote:
Correct - although you still need the analogue filter, it's much a much
less demanding spec.
OK, just to make sure I've understood this correctly. Say we have a
44.1kHz signal. Putting it through a non-oversampling DAS, the repeated
spectra start at 44.1kHz. If we
I think it's really a matter of semantics. Oversampling is performed
solely for the purpose of reducing the harmful effects of brickwall
filters. Upsampling is really a form of sample rate conversion, and is
necessary for systems that have asynchronous timing (i.e. input and
output clocks are
Patrick Dixon;130951 Wrote:
This is completely wrong - Upsampling/oversampling doesn't invent any
data! Interpolation is actually a filtering process which removes the
repeat spectra that are created by the upsampling/oversampling process.
This thread seems to have developed in all kinds of
I give up!
--
Patrick Dixon
www.at-tunes.co.uk
Patrick Dixon's Profile: http://forums.slimdevices.com/member.php?userid=90
View this thread: http://forums.slimdevices.com/showthread.php?t=26685
dwc;131160 Wrote:
I think all of the above hulabaloo builds a strong case for those of us
on the fringe with non-oversampling filterless DACs.
The second benefit is there is no dog poop in the yard because all the
neighborhood dogs can't handle the super-high frequency noise. :)
Pat Farrell;131224 Wrote:
Ah, well, I hate to break this to you, but according to the Neumann
site, under their entries for Historic Microphones
http://www.neumann.com/?lang=enid=hist_microphonescid=km83_publications
they say that the frequency response of the KM83 is 40 - 16K hz.
Sorry
reeve_mike wrote:
Pat Farrell;131224 Wrote:
Ah, well, I hate to break this to you, but according to the Neumann
site, under their entries for Historic Microphones
http://www.neumann.com/?lang=enid=hist_microphonescid=km83_publications
they say that the frequency response of the KM83 is 40 -
Patrick Dixon;131182 Wrote:
I give up!
I take it that this is a response to my second post in this thread.
Saying I give up doesn't contribute much. It would seem that you
believe that my understanding of what upsampling does is wrong. I'm
genuinely interested in finding out whether I've
cliveb;131259 Wrote:
I take it that this is a response to my second post in this thread.Sorry, it
wasn't aimed you particularly.
cliveb;131259 Wrote:
Further thought suggests to me that it's possible that the interpolation
may not actually generate any higher frequencies, and that
seanadams;130863 Wrote:
No, it interpolates. So you get something maybe like: 0, 0.5, 1, 1.5, 2
and so on.
What you're talking about is oversampling - just another name for
upsampling, but usually used in reference to what modern DACs do
internally. It is fundamental to how they work and
Pat Farrell;130865 Wrote:
seanadams wrote:[color=blue][color=green]So it is more than twice as
good as the SACD single bit rate of 2.82
MHz, eh? Any chance that the DAC in the Transport actually is 5.64
mHz?
Comparing sample rates for 1-bit DSD/SACD vs.
redbook/upsampled/oversampled
ezkcdude;130930 Wrote:
Namely, upsampling shifts aliasing artifacts (so-called ghost images) to
a much higher (inaudible) frequency range.Actually this is not quite correct.
Alias artifacts are 'fixed' in the signal at Analogue to Digital
conversion. Once there they can't be removed.
What
cliveb;130937 Wrote:
As the terms are typically used, it's OVERSAMPLING rather than
UPSAMPLING that shifts aliasing higher up the frequency range and makes
life easier for the filters. Pretty much every DAC on the planet does
oversampling these days (with the exception of the niche NOS ones,
cliveb;130937 Wrote:
In contrast, upsampling (as the term is generally used) involves
INVENTING additional data (usually by interpolation) in the expectation
that it will deliver improved high frequency resolution.This is completely
wrong - Upsampling/oversampling doesn't invent any
data!
Patrick Dixon;130940 Wrote:
Actually this is not quite correct.
Alias artifacts are 'fixed' in the signal at Analogue to Digital
conversion. Once there they can't be removed.
They can be shifted to a higher frequency range, so that a more
gentle reconstruction (anti-aliasing) filter can
Patrick Dixon;130953 Wrote:
Actually it is, it's just not very easy or practical!
Well, OK, it is almost impossible then..
--
P Floding
P Floding's Profile: http://forums.slimdevices.com/member.php?userid=2932
View
Patrick Dixon;130951 Wrote:
This is completely wrong - Upsampling/oversampling doesn't invent any
data! Interpolation is actually a filtering process which removes the
repeat spectra that are created by the upsampling/oversampling process.
The actual interpolation process doesn't improve
ezkcdude;130954 Wrote:
They can be shifted to a higher frequency range, so that a more gentle
reconstruction (anti-aliasing) filter can be used. That is my
understanding.Your understanding is wrong I'm afraid. Once the alias is in
the
signal, it looks just like part of the original signal
PhilNYC;130961 Wrote:
if you upsample a 44.1khz data sample to 96khz, how do you not invent
new data?You're not inventing data because you're using the information that
already exists within the digital signal to create the intermediate
points. There's no more information in the signal -
P Floding;130956 Wrote:
Well, OK, it is almost impossible then..
:-)
You'll find a simple example of an analogue FIR filter in old PAL TV
sets (PAL is the analogue colour TV system in use in much of Europe
(the French of course had to be different) - it's NTSC in N America
Japan). The colour
cliveb;130937 Wrote:
In contrast, upsampling (as the term is generally used) involves
INVENTING additional data (usually by interpolation) in the expectation
that it will deliver improved high frequency resolution. But this extra
data that's invented can't ever be known to be correct. Quite
reeve_mike;130976 Wrote:
Absolutely, I forget who said it but I think the following sums up well
the use of asynchronous sample rate conversion
(it was said in the context of its use for jitter reduction but the
comment is more generally applicable):
it shifts the problem from being the
P Floding;130981 Wrote:
Upsampling has nothing to do with asynchroneous sample rate conversion.
Upsampling from 44.1K to 96K is asynchronous sample rate conversion by
definition (the input output clocks are different) ...
BTW most so called upsampling DACs (the boxes not the chips) use an
P Floding;130992 Wrote:
However, there is no jitter introduced when doing a mathematical
upsampling.
Agreed!
[And I never said that there was - apologies if my tangential reference
to ASRC for jitter reduction, as in some add-on boxes marketed, caused
confusion,
I guess I should have left
Patrick Dixon;130969 Wrote:
You're not inventing data because you're using the information that
already exists within the digital signal to create the intermediate
points. There's no more information in the signal - you're not
creating anything, you're just filtering the signal.
This is
You guys should read the data sheet for AD1896, which is Analog's ASRC.
I'm using it right now for a DAC I am building. It explains very well
the theory and implementation. AD1896 is used in practically all
upsampling DACs these days.
--
ezkcdude
SB3-Derek Shek TDA1543/CS8412 NOS DAC-MIT
Patrick Dixon;130969 Wrote:
You might like to think of it in relation to what happens at the DAC;
the digital signal is converted to an instantaneous analogue level at
the sampled points, and then held (and filtered) to give a continuous
analogue signal. But the signal between the precise
PhilNYC;130999 Wrote:
I think this is what is called oversampling
Is it?
I'm not sure, since I never really was a believer of upsampling.
As I understood it oversampling is done as part of the reconstructions
process in the DAC, wheras oversampling is a sort of lets do something
to the data
P Floding;131008 Wrote:
Is it?
I'm not sure, since I never really was a believer in upsampling (I
got turned off by all the hype).
As I understood it oversampling is done as part of the reconstructions
process in the DAC, wheras upsampling is a sort of lets do something
to the data
P Floding;131013 Wrote:
Surely, asynchronous means the clocks aren't running in synchrony, which
would not have anything to do with sample rate conversion per se? (I
have read a fair bit about sample rate conversion.)
I don't really see why a one-clock system should perform asynchronous
PhilNYC wrote:
...but if you are changing the sample rate of the data, you will need
two clocks, no? One clock runs at the original sample-rate (44.1khz)
and the second runs at the new sample rate (96khz).
or one that runs at 44.1*48 and select the proper signal samples
off a common clock.
Pat Farrell;131023 Wrote:
PhilNYC wrote:[color=blue]
or one that runs at 44.1*48 and select the proper signal samples
off a common clock.
There was a time when 44.1kHz was a challenge.
By the time SACD came out, 2.82 MHz was not a challenge.
At least if you are not trying to use a Tube
PhilNYC;130998 Wrote:
If you upsample from 44.1khz to 96khz, there are now 2.17687x more data
points than in the original sample, and only one of those data points
per second is identical to a single data point in the original sample.300 per
second, surely: 147 periods of one stream are
PhilNYC;130998 Wrote:
If you upsample from 44.1khz to 96khz, there are now 2.17687x more data
points than in the original sample, and only one of those data points
per second is identical to a single data point in the original sample.It
makes no difference, you are still not inventing data.
Patrick Dixon;131047 Wrote:
It makes no difference, you are still not inventing data.
Within the limit that any filtering in D/D or D/A conversion is
'guessing' the original analog signal between two adjacent sample
points
- who is to say that it indeed was the smooth transition that the
filter
I think there needs to be made a distinction between inventing data
and creation of artifactual data. Clearly, the first one is not part of
the design of upsamplers or oversamplers. Maybe marketers make it seem
that way, but we know better, right? As for the second, it is
inevitable that any
Patrick Dixon;131047 Wrote:
It makes no difference, you are still not inventing data. I've spent
the best part of 25yrs designing equipment that sample rate converts,
filters and interpolates so I know a little about it!
Then can you explain it to me? :-) How do you go from 44,100 data
tom permutt;131046 Wrote:
300 per second, surely: 147 periods of one stream are precisely 320 of
the other.
Still...out of 96000 data points, that's not a lot...
--
PhilNYC
Sonic Spirits Inc.
http://www.sonicspirits.com
PhilNYC;131061 Wrote:
Then can you explain it to me? :-) How do you go from 44,100 data
points to 96,000 data points and not create data that did not exist
before?OK, so how do you go from 44,100 data points to an infinite number -
which is what you do when you D to A Convert - and not
reeve_mike;131056 Wrote:
Within the limit that any filtering in D/D or D/A conversion is
'guessing' the original analog signal between two adjacent sample
points
- who is to say that it indeed was the smooth transition that the
filter yields ...
You're missing the point, the
Patrick Dixon;131085 Wrote:
OK, so how do you go from 44,100 data points to an infinite number -
which is what you do when you D to A Convert - and not 'create' data?
That's very different. In the case of D-to-A conversion, you are
essentially decoding something that was encoded using the
Pat Farrell;131091 Wrote:
Patrick Dixon wrote:[color=blue]Right. Mike doesn't understand (or
appears to not understand) the work
of Shannon and Nyquist. All of the digital sampling work is based on
their theories.
Nyquist showed that sampling at twice the bandwidth allows
PhilNYC wrote:
Pat Farrell;131091 Wrote:
Nyquist showed that sampling at twice the bandwidth allows
reconstruction. That is why the RedBook spec uses 44.1 kHz.
For decades, the hfi world used a bandwidth of 20 hz to 20kHz
as the limits of human hearing. Sampling at 44.1kHz allows
a little
Sorry if it disappoints but I know well the work of Claude Shannon and
Harry Nyquist ...
What I was trying to contribute was that the samples on the CD do not
faithfully represent the musical waveform,
they only represent a version of it filtered at 20.5KHz, which seemed
to be relevant at the
Chapter 3 of Analog Device's Data Conversion Handbook (by Walt Kester)
discusses this:
Kester Wrote:
The basic concept of an oversampling/interpolating DAC is shown in
Figure 3.30. The Nbits
of input data are received at a rate of fs. The digital interpolation
filter is clocked at
an
reeve_mike wrote:
Sorry if it disappoints but I know well the work of Claude Shannon and
Harry Nyquist ...
Opps, sorry.
What I was trying to contribute was that the samples on the CD do not
faithfully represent the musical waveform,
they only represent a version of it filtered at 20.5KHz,
WOWnow thats what I call a response. Thanks for all the info.
--
Walleyefisher
Walleyefisher's Profile: http://forums.slimdevices.com/member.php?userid=7122
View this thread:
Pat Farrell;131153 Wrote:
reeve_mike wrote:
What I was trying to contribute was that the samples on the CD do
not
faithfully represent the musical waveform,
they only represent a version of it filtered at 20.5KHz, which
seemed
to be relevant at the time but now I can't remember why
I think all of the above hulabaloo builds a strong case for those of us
on the fringe with non-oversampling filterless DACs.
The second benefit is there is no dog poop in the yard because all the
neighborhood dogs can't handle the super-high frequency noise. :)
non-os dacs, keeping it real.
reeve_mike wrote:
Agreed in general, but I've used some nice vintage Neumann mics that go
way up high ...
This is way off topic, but which ones? And how vintage?
Most of the classic Neumann's like the U87 or M50 fall off pretty
seriously. Now my KM184's go up high, but they aren't vintage.
I don't understand the premise of your question.
Take a signal that looks like 0, 2, 3
then upsample it to 0,0,0,0, 2,2,2,2, 3,3,3,3 at four times the rate.
How does this allow the DAC to do anything differently?
No, it interpolates. So you get something maybe like: 0, 0.5, 1, 1.5, 2
seanadams wrote:
then upsample it to 0,0,0,0, 2,2,2,2, 3,3,3,3 at four times the rate.
No, it interpolates. So you get something maybe like: 0, 0.5, 1, 1.5, 2
and so on.
Thanks for the clarification.
What you're talking about is oversampling - just another name for
upsampling, but usually
52 matches
Mail list logo
