Hi all,

I've heard a lot of aliasing. What do they look like in image and
audio/speech/music with aliasing?

I did not see/hear any real aliasing stuff, so the concept of aliasing
looks abstract to me, although I know how it occurs in terms of
mathematics...

I am also wondering about the anti-aliasing filter before
downsampling...

By filtering, you lose information, right? It is hard to believe that
the loss due to anti-aliasing filtering before downsampling

is better than

the loss/distortion due to aliasing ...

Maybe it is just an engineering practice due to the fact that
sometimes we must do downsampling... so we'd better use a filter to
save some loss?

kiki wrote:

> Hi all,
> 
> I've heard a lot of aliasing. What do they look like in image and
> audio/speech/music with aliasing?
> 
> I did not see/hear any real aliasing stuff, so the concept of aliasing
> looks abstract to me, although I know how it occurs in terms of
> mathematics...
> 
> I am also wondering about the anti-aliasing filter before
> downsampling...
> 
> By filtering, you lose information, right? It is hard to believe that
> the loss due to anti-aliasing filtering before downsampling
> 
> is better than
> 
> the loss/distortion due to aliasing ...
> 
> Maybe it is just an engineering practice due to the fact that
> sometimes we must do downsampling... so we'd better use a filter to
> save some loss?

I've only noticed it in audio as a second audible tone when I was 
inputting a sine wave into the circuit.  I'm sure that if you had music 
playing on a system with aliasing you'd here a "ghost band" playing way 
out of tune.

In video it's very noticeable -- the reason you never see TV actors 
wearing fine striped shirts is because when the spatial frequency of the 
stripes approaches the color burst frequency (for vertical stripes) or 
line separation (for horizontal stripes) the stripes will merge and 
shift and generally look ugly and weird.  If you watch old westerns and 
see the stagecoach wheels rotating backwards that's due to temporal 
aliasing.

In general you if you have an idea of the properties of your input 
signal you can calculate the relative degradation in your signal due to 
aliasing vs. filtering.  Also in general, it has proven pretty easy in 
practice to implement anti-aliasing filters that do a pretty good job 
without messing up the signal nearly as much as aliasing would.

-- 

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

"Tim Wescott" <t...@wescottnospamdesign.com> wrote in message 
news:1...@corp.supernews.com...
> kiki wrote:
>
>> Hi all,
>>
>> I've heard a lot of aliasing. What do they look like in image and
>> audio/speech/music with aliasing?
>>
>> I did not see/hear any real aliasing stuff, so the concept of aliasing
>> looks abstract to me, although I know how it occurs in terms of
>> mathematics...
>>
>> I am also wondering about the anti-aliasing filter before
>> downsampling...
>>
>> By filtering, you lose information, right? It is hard to believe that
>> the loss due to anti-aliasing filtering before downsampling
>>
>> is better than
>>
>> the loss/distortion due to aliasing ...
>>
>> Maybe it is just an engineering practice due to the fact that
>> sometimes we must do downsampling... so we'd better use a filter to
>> save some loss?
>
> I've only noticed it in audio as a second audible tone when I was 
> inputting a sine wave into the circuit.  I'm sure that if you had music 
> playing on a system with aliasing you'd here a "ghost band" playing way 
> out of tune.
>
> In video it's very noticeable -- the reason you never see TV actors 
> wearing fine striped shirts is because when the spatial frequency of the 
> stripes approaches the color burst frequency (for vertical stripes) or 
> line separation (for horizontal stripes) the stripes will merge and shift 
> and generally look ugly and weird.  If you watch old westerns and see the 
> stagecoach wheels rotating backwards that's due to temporal aliasing.
>
> In general you if you have an idea of the properties of your input signal 
> you can calculate the relative degradation in your signal due to aliasing 
> vs. filtering.  Also in general, it has proven pretty easy in practice to 
> implement anti-aliasing filters that do a pretty good job without messing 
> up the signal nearly as much as aliasing would.


You also get temporal aliasing when you move your mouse rapidly from side to 
side - you see lots of copies of the mouse pointer.

Leon 

"Tim Wescott" <t...@wescottnospamdesign.com> wrote in message
news:1...@corp.supernews.com...
> kiki wrote:
>
> > Hi all,
> >
> > I've heard a lot of aliasing. What do they look like in image and
> > audio/speech/music with aliasing?
> >
> > I did not see/hear any real aliasing stuff, so the concept of aliasing
> > looks abstract to me, although I know how it occurs in terms of
> > mathematics...
> >
> > I am also wondering about the anti-aliasing filter before
> > downsampling...
> >
> > By filtering, you lose information, right? It is hard to believe that
> > the loss due to anti-aliasing filtering before downsampling
> > is better than
> > the loss/distortion due to aliasing ...
> >
> > Maybe it is just an engineering practice due to the fact that
> > sometimes we must do downsampling... so we'd better use a filter to
> > save some loss?
>
> I've only noticed it in audio as a second audible tone when I was
> inputting a sine wave into the circuit.  I'm sure that if you had music
> playing on a system with aliasing you'd here a "ghost band" playing way
> out of tune.

It is definitely easiest to detect on a pure tone. What I find works especially
well is a swept tone, i.e. sinusoid of increasing frequency.  In that case the
alias signal will be moving in the opposite direction as the original, making it
stand out.

On "normal" wide-band audio signals, to my ears aliasing gives a harshness to
the sound, but it is usually subtle.

Regarding which loss is worse, when you sample you can't keep the high
frequencies (above Nyquist) anyway since they will alias.  So if you don't
filter them out, they will corrupt the other lower frequencies in your signal.
So basically the anti-aliasing filter is removing what you can't keep anyway in
order to avoid corrupting that which you can keep.  In those terms, it is
usually a pretty easy decision!

In article <d...@posting.google.com>,
kiki <l...@yahoo.com> wrote:
>I've heard a lot of aliasing. What do they look like in image and
>audio/speech/music with aliasing?

In audio, the distortion is easily audible if you take music with lots
of the highest frequencies you can easily hear, and downsample without
antialiasing below half that frequency.  The distortion depends on
the amount and type of high frequency sound present, sometimes like a
rasping or a non-delayed echo sound.  Take a music CD (44.1KHz samples)
with acoustic piano music or bells and try reconstructing using on every
8th sample (5.5 KHz audio) with and without antialias filtering before
downsampling.

In images, the easiest comparision is between the TV images of an old
video game (Atari or NES) and a modern video game with an ATI or NVidea
antialiasing 3D graphics chip (XBox or GameCube).  The old video games
show lots of "jaggies" and stairsteps on any tilted lines or contrasty
patterns.  The newer video games often show slightly blurry lines instead
of jaggies due to graphics antialiasing.  The same effect happens to
photos if downsampled without filtering.

>I did not see/hear any real aliasing stuff, so the concept of aliasing
>looks abstract to me, although I know how it occurs in terms of
>mathematics...
>
>I am also wondering about the anti-aliasing filter before
>downsampling...
>
>By filtering, you lose information, right? It is hard to believe that
>the loss due to anti-aliasing filtering before downsampling >
>is better than >the loss/distortion due to aliasing ...

If there was no filtering, your signal (audio or video sample rates) would
contain components from microwave background radiation, RFI and nearby
strong cellphone/radio/TV transmitters, infrared room heat gradients,
plate tectonic and earthquake rumble, etc.  Maybe even enough to totally
swamp your signal of interest.  So you usually want to lose information
outside of the range of frequencies of interest.


IMHO. YMMV.
-- 
Ron Nicholson   rhn AT nicholson DOT com   http://www.nicholson.com/rhn/ 
#include <canonical.disclaimer>        // only my own opinions, etc.