18 megapixels on a 1.6x crop camera - Has Canon gone too far?

(Ray Fischer) wrote:
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
"mcdonaldREMOVE TO ACTUALLY REACH wrote:
Ray Fischer wrote:
Alfred Molon wrote:
In any case, if the human ear can only hear at most up to 20kHz, why do
you get better results when sampling at 176kHz than at 44kHz?

That one's easy: It's because sampling at 44kHz doesn't preserve the
phase relationship of high frequencies. It also can produce lower
frequency artifacts.

Both of those are factually incorrect.

They are not. Sampling a 22khz signal at 44khz exactly at the zero
crossing will provide no signal at all. Change the signal frequency
slightly and you will get low-frequency artifacts as the sampling
progressing in relation to the signal.

If the frequency is changed "slightly", in either
direction... the output will be a *difference* from 22
kHz, not a "low-frequency artifact".

The difference _will_ be a lower frequency.

But will *not* be "low frequency".

It will compared to the original signal.

Are you reduced to childish quibbling?

What you said was not true,

Of course it was true. Your'e whining becuse your OPINION is that
the resulting harmonic isn't, in your OPINION, a "low" frequency.

It isn't a harmonic either.

Go away, idiot.

As usual you are reduced to insults because your statements are
proven wrong:

As usual you resort to childish quibbling when you're proven ignorant
and wrong.

So why did you snip the definition of "harmonic" that I
provided, rather than attempt to demonstrate that there
actually is an harmonic? The reason of course is that
it isn't an harmonic.

Maybe if you read this you'll understand why your statement is
incorrect.

http://en.wikipedia.org/wiki/Nyquist...mpling_theorem

You obviously did *not* read it.

"If a function x(t) contains no frequencies higher
than B hertz, it is completely determined by giving
its ordinates at a series of points spaced 1/(2B)
seconds apart."

You didn't bother reading the section titled "Practical considerations"
whoch just happen to refute your claims.

There is nothing in the entire article that refutes what
I said.

Obviously you didn't read the article. The theorem only applies to
signals that are sampled for infinite time.

And *you* call me an idiot?

Your entire knowledge of this topic seems to come from a
Wikipedia page that you cannot understand. Please quote
where even that page says what you claim above, that
"The theorem only applies ...". When you get to the
point where it discusses "interpolation error", explain
that in terms or your above statement.

In fact, the distinction between the theorem in theory
and in practice is a distortion called interpolation
error. The existence of such a distortion does not deny
that either the theorem applies or that, as originally
stated, any signal less than 1/2 the sampling rate can
be reproduced.

--
Floyd L. Davidson http://www.apaflo.com/floyd_davidson
Ukpeagvik (Barrow, Alaska)

In article , Ray Fischer
writes
Kennedy McEwen wrote:
In article , Ray Fischer
writes
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Floyd L. Davidson wrote:

If the frequency is changed "slightly", in either
direction... the output will be a *difference* from 22
kHz, not a "low-frequency artifact".

The difference _will_ be a lower frequency.

But will *not* be "low frequency".

It will compared to the original signal.

What's the difference between 22,000Hz and 21,800Hz?

What the heck does the difference between 22kHz and 21.8kHz have to do
with this situation?

When you sample a 21,800hz signal at 44khz then you get a different
frequency related to the difference between 22khz and 21,800hz.

This seems to be the source of your confusion. No such frequency is
created!

When you sample a 21,800Hz signal at 44kHz you get a different frequency
related to the difference 44kHz and 21,800Hz, ie. the difference between
the input and the sampling rate, NOT the difference between input and
half the sampling rate!

A properly designed sampling system requires BOTH an input and an output
filter.

Consider this example: You sample a 22khx signal at 44khz. What's the
result? Well, if you happen to sample at the zero-crossings then the
result is no signal at all.

With a properly designed AA filter, 22kHz would never reach the sampler!
You seem to have missed the definition of the Nyquist criteria in the
references you cited earlier in the thread, exactly half the sampling
frequency is the critical condition at which unambiguous reproduction of
the input frequency cannot be achieved.

If you sample a 21,800hz signal then the result is similar except that
you will then sample along progressive parts of the signal. The
result will be at 200hz.

Wrong, and very easy to prove wrong. Since you appear to be unable to
follow the math in enough detail I suggest you take out pen and graph
paper and actually draw a few cycles of 21.8kHz sine wave and sample it
at 44kHz to find out what the output frequency is. That is the simplest
way to understand what is happening without need to get into the math at
all, but you will need a fairly long roll of paper since the difference
between the relevant frequencies is so large. You will end up with a
signal modulated at 400Hz, which is the beat frequency between the
original input and its alias at 22.2kHz, NOT 200Hz! Remove the higher
alias frequency by the reconstruction filter and you retrieve the
original input 21.8kHz frequency. 200Hz NEVER appears in this scenario.
400Hz does, but only if you ignore the reconstruction filter.

If both filters are infinitely steep cut-off "brick wall" designs then
it is indeed possible to unambiguously reconstruct every input frequency
up to half the sampling frequency.

But it's not just frequency that is important. Phase counts.

Sure does, but phase is continually changing relative to each sample for
all frequencies which are not an integer division of the sampling
frequency. That is why your absurd comment about sampling at exactly
half the sampling frequency is irrelevant - the phase in that case is
fixed relative to the samples because of the integer relationship. It
is also fixes at 11Khz and at 14.666'kHz, but when the integer is
greater than 2 samples there will always be samples at asymmetrical
points in each cycle. 21.8kHz is not an integer division of the
sampling frequency, so only a few cycles of this frequency will change
the phase enough for its exact relationship with the sampling frequency
to be irrelevant.

And
note also that there is an important limitation of the
Nyquist-Shannon sampling theorem - it's only true for an infinite
sequence of samples.

Again, that is complete nonsense. If that was indeed true then the
entire digital industry would not exist. No digital audio, CDs, iPods,
digital telephony, digital video, DVDs, satellite communications, no
digital radar etc. All of these examples require digital sampling of
inherently analogue signals, and NONE of the sampling sequences nor the
signals are infinite.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's ****ed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)

Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
"mcdonaldREMOVE TO ACTUALLY REACH wrote:
Ray Fischer wrote:
Alfred Molon wrote:
In any case, if the human ear can only hear at most up to 20kHz, why do
you get better results when sampling at 176kHz than at 44kHz?

That one's easy: It's because sampling at 44kHz doesn't preserve the
phase relationship of high frequencies. It also can produce lower
frequency artifacts.

Both of those are factually incorrect.

They are not. Sampling a 22khz signal at 44khz exactly at the zero
crossing will provide no signal at all. Change the signal frequency
slightly and you will get low-frequency artifacts as the sampling
progressing in relation to the signal.

If the frequency is changed "slightly", in either
direction... the output will be a *difference* from 22
kHz, not a "low-frequency artifact".

The difference _will_ be a lower frequency.

But will *not* be "low frequency".

It will compared to the original signal.

Are you reduced to childish quibbling?

What you said was not true,

Of course it was true. Your'e whining becuse your OPINION is that
the resulting harmonic isn't, in your OPINION, a "low" frequency.

It isn't a harmonic either.

Go away, idiot.

As usual you are reduced to insults because your statements are
proven wrong:

As usual you resort to childish quibbling when you're proven ignorant
and wrong.

So why did you snip the definition of "harmonic" that I

Because it's not relevant and you're just playing your childish games.

Maybe if you read this you'll understand why your statement is
incorrect.

http://en.wikipedia.org/wiki/Nyquist...mpling_theorem

You obviously did *not* read it.

"If a function x(t) contains no frequencies higher
than B hertz, it is completely determined by giving
its ordinates at a series of points spaced 1/(2B)
seconds apart."

You didn't bother reading the section titled "Practical considerations"
whoch just happen to refute your claims.

There is nothing in the entire article that refutes what
I said.

Obviously you didn't read the article. The theorem only applies to
signals that are sampled for infinite time.

And *you* call me an idiot?

Yep.

Your entire knowledge of this topic seems to come from a

And more of the usual bull****ting.

--
Ray Fischer

Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Kennedy McEwen wrote:
In article , Ray Fischer
writes
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Floyd L. Davidson wrote:

If the frequency is changed "slightly", in either
direction... the output will be a *difference* from 22
kHz, not a "low-frequency artifact".

The difference _will_ be a lower frequency.

But will *not* be "low frequency".

It will compared to the original signal.

What's the difference between 22,000Hz and 21,800Hz?

What the heck does the difference between 22kHz and 21.8kHz have to do
with this situation?

When you sample a 21,800hz signal at 44khz then you get a different
frequency related to the difference between 22khz and 21,800hz.

It will be 200Hz lower, but it will still be at the high
end of the bandwidth limited spectrum,

Nope. It will be 200hz.

A properly designed sampling system requires BOTH an input and an output
filter.

Consider this example: You sample a 22khx signal at 44khz. What's the
result? Well, if you happen to sample at the zero-crossings then the
result is no signal at all.

That of course requires the sample rate and the sampled
signal be phase locked.

So there is an example of a 22khz signal that is not correctly sampled
at 44khz.

While there are cases where
that happens, and it also affects sub-multiples of the
sample rate other than 2, it isn't particularly of any
significance to music and even less so for voice data.

So your claims are wrong, but that fact isn't significant, in your
opinion.

If you sample a 21,800hz signal then the result is similar except that
you will then sample along progressive parts of the signal. The
result will be at 200hz.

The result will be a 21,800 Hz signal.

And so you once again stick your head up your ass and refuse to think.

If both filters are infinitely steep cut-off "brick wall" designs then
it is indeed possible to unambiguously reconstruct every input frequency
up to half the sampling frequency.

But it's not just frequency that is important. Phase counts. And
note also that there is an important limitation of the
Nyquist-Shannon sampling theorem - it's only true for an infinite
sequence of samples.

But just how "untrue" is it for a finite set of samples?

As you admitted above, it can be quite significant.

--
Ray Fischer

Ray Fischer wrote:


(bogus sampling thoughts)

In reality, if you sample exactly at 44 kHz a 21,800 Hz signal,
with no filters, and DAC the result, you will get signals at
21,800 and 22,200 Hz. What Mr. Fischer fails to realize is that
this is a 22 kHz suppressed carrier AM modulated at 200 Hz.
There are no components at 200 or 400 Hz. What you will see
on a scope is a 400 Hz beat, of course. If you rectify the signal
you WILL get a 400 Hz component.

If you have a brick wall filter on the DAC at 21,900 Hz
(down say 80 dB at 22,200 Hz) (not easy in analog!) the
output will be at 21,800 Hz ... only.

Doug McDonald

(Ray Fischer) wrote:
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Kennedy McEwen wrote:
In article , Ray Fischer
writes
Floyd L. Davidson wrote:
(Ray Fischer) wrote:
Floyd L. Davidson wrote:

If the frequency is changed "slightly", in either
direction... the output will be a *difference* from 22
kHz, not a "low-frequency artifact".

The difference _will_ be a lower frequency.

But will *not* be "low frequency".

It will compared to the original signal.

What's the difference between 22,000Hz and 21,800Hz?

What the heck does the difference between 22kHz and 21.8kHz have to do
with this situation?

When you sample a 21,800hz signal at 44khz then you get a different
frequency related to the difference between 22khz and 21,800hz.

It will be 200Hz lower, but it will still be at the high
end of the bandwidth limited spectrum,

Nope. It will be 200hz.

If that is true the output spectrum will be inverted
from the input spectrum.

A properly designed sampling system requires BOTH an input and an output
filter.

Consider this example: You sample a 22khx signal at 44khz. What's the
result? Well, if you happen to sample at the zero-crossings then the
result is no signal at all.

That of course requires the sample rate and the sampled
signal be phase locked.

So there is an example of a 22khz signal that is not correctly sampled
at 44khz.

The theorem says *less than* half the sampling rate. 22
kHz is not less than half of 44 kHz, and is not included
in the specified range.

While there are cases where
that happens, and it also affects sub-multiples of the
sample rate other than 2, it isn't particularly of any
significance to music and even less so for voice data.

So your claims are wrong, but that fact isn't significant, in your
opinion.

The above "claims" are not wrong though.

If you sample a 21,800hz signal then the result is similar except that
you will then sample along progressive parts of the signal. The
result will be at 200hz.

The result will be a 21,800 Hz signal.

And so you once again stick your head up your ass and refuse to think.

This is beyond funny Ray. Are you really wanting to
claim that sampling inverts the spectrum?

If both filters are infinitely steep cut-off "brick wall" designs then
it is indeed possible to unambiguously reconstruct every input frequency
up to half the sampling frequency.

But it's not just frequency that is important. Phase counts. And
note also that there is an important limitation of the
Nyquist-Shannon sampling theorem - it's only true for an infinite
sequence of samples.

But just how "untrue" is it for a finite set of samples?

As you admitted above, it can be quite significant.

Except I said no such thing. I have said that it can be
*insignificantly* different, which is a detail
(interpolation error) you seem unable to appreciate. As
has been pointed out by Kennedy McEwen the entire
concept that an infinite sequence of samples is required
is obvious *claptrap* given the extensive use of finite
sequences of samples in so many fields (and the lack of
an infinite sequence in any functional example).

You probably should get someone to take you, page by
page, through Shannon's "A Mathematical Theory of
Communication" to get a better understanding of the
theory of digital data transmission.

--
Floyd L. Davidson http://www.apaflo.com/floyd_davidson
Ukpeagvik (Barrow, Alaska)

In article , "mcdonaldREMOVE TO ACTUALLY
REACH writes
Ray Fischer wrote:


(bogus sampling thoughts)

In reality, if you sample exactly at 44 kHz a 21,800 Hz signal,
with no filters, and DAC the result, you will get signals at
21,800 and 22,200 Hz. What Mr. Fischer fails to realize is that
this is a 22 kHz suppressed carrier AM modulated at 200 Hz.
There are no components at 200 or 400 Hz. What you will see
on a scope is a 400 Hz beat, of course. If you rectify the signal
you WILL get a 400 Hz component.

If you have a brick wall filter on the DAC at 21,900 Hz
(down say 80 dB at 22,200 Hz) (not easy in analog!) the
output will be at 21,800 Hz ... only.

Which is precisely where this erratic discussion came in! ;-)

The reconstruction filter, which as you note is not easy in analogue, is
vastly simplified by upsampling the digital data from 44kHz samples to,
say 88kHz, 176kHz or even 5.632MHz, then implementing the brick wall
filter digitally and then converting to analogue with a much simpler
analogue filter which can, ideally, even be a single RC network.

In short, as was stated by most of the posters at the start of this
deviant sub-thread, the reason for higher sampling frequency in digital
audio is to relax the analogue filter (both AA and reconstruction)
design constraints, not because humans can audibly resolve such high
frequencies.

Coming back on topic, neither can they resolve ridiculously high spatial
frequencies in printed images.
--
Kennedy
Yes, Socrates himself is particularly missed;
A lovely little thinker, but a bugger when he's ****ed.
Python Philosophers (replace 'nospam' with 'kennedym' when replying)