From: Jim Peters (jim_at_uazu.net)
Date: 2002-02-15 17:45:06
John Morrison wrote:
> > 'Back-tracking' is hard for some types of DSP algorithms to handle,
> So the rewind/analysis facility will only work with certain types.
Well, if you wanted to rewind with an IIR filter, you'd really need to
feed it the previous XX seconds-worth of data so that it knows where
it was. So it isn't straightforward.
> > maybe we could make 'analysis' and 'biofeedback' two
> > different parts of the program ?
>
> I'd prefer to make the one program/framework instead of wasted time
> in parallel development.
The trouble is that you might start multiplying the complexity if you
put them together into the same framework (although they could happily
live in the same program). I've read professionals talk about using
one type of display whilst recording, and then using more advanced
techniques later to see more detail. Some things are hard to do in
real-time, but are easy once the data has been recorded.
For instance, if I'm looking at brain-waves with low frequencies, to
measure the amount of a frequency at this moment *now*, I need to take
into account both past and *future* data. In a real-time system,
future data is not available yet, so all I can do is calculate the
results for some time in the past, maybe 10-20 seconds ago. After the
data has already been recorded, when you are reviewing a session for
example, there is no problem in taking future data into account.
Do you remember the displays that I posted ? That was why the
drawing-point curved off to the left for lower frequencies -- values
for 'now' often aren't available until 'later' when you are working in
real-time.
As don Juan says in one of Castaneda's books: "We think we are
perceiving things as they are now, but in fact we are always
remembering, remembering" (I'm paraphrasing it, because I can't find
the part now) -- meaning we always perceive what happened a moment or
a few moments ago. This is just the same as with the DSP analysis,
and this comes right out of the maths, too. There is no other way it
can be when you're dealing with waves.
It's a bit like when something happens in the world too fast for you,
and then only afterwards your brain figures out exactly what was what.
It's just the same -- the real-time algorithms failed, but analysis of
recorded data enabled an interpretation to be made.
It gets deeper still -- we're looking the uncertainty principle
directly in the face here, dealing with waves at these kinds of
scales. You know the idea from quantum mechanics, how you can know
the time exactly, but then the frequency is uncertain (i.e. blurred),
or the other way around -- you can know the frequency exactly, but
then you don't know exactly when it happened.
This applies here too -- you can apply a very wide window function,
which gives you great accuracy in the frequency domain, but blurs
everything a great deal in the time domain. At the other extreme,
applying a narrow window function gives you excellent time-resolution,
but unfortunately blurs the frequencies, so you can't be sure of
anything much frequency-wise.
My app will allow this window-width to be easily changed, and I expect
that you will be able to see this effect directly on the display --
either a blurring vertically or horizontally.
It's like the blurriness fits into a rectangle, and you can make it
fat and short, or thin and tall, but never thin and short at the same
time -- a bit like looking at the world through glasses that were made
wrong. I've read about techniques that allow you to use a window that
puts this 'blurriness' rectangle at an angle, allowing you to
differentiate two close-spaced falling tones, for example. However,
this is getting really advanced.
Actually, I think this is what your brain is doing when it takes a
moment or two to figure out what something is (both for vision and for
sounds) -- it is trying various different ways of analysing the data
until it finds something that makes sense.
Jim
-- Jim Peters (_)/=\~/_(_) jim_at_uazu.net (_) /=\ ~/_ (_) Uazú (_) /=\ ~/_ (_) http:// B'ham, UK (_) ____ /=\ ____ ~/_ ____ (_) uazu.net
This archive was generated by hypermail 2.1.4 : 2002-07-27 12:28:38 BST