On Friday, April 25, 2014 10:31:52 AM UTC-4, haitic...@gmail.com wrote:
> On Friday, April 18, 2014 10:34:04 AM UTC-4, gghe...@gmail.com wrote:
>
> > On Thursday, April 17, 2014 11:19:21 PM UTC-4, haitic...@gmail.com wrote:
> Hi George
> I'm a pattern recognition engineer, mainly, and we are usually trying to find
> data input which is easy to get. I am interested in weak signals from the pov
> of biomedical diagnostics and a QM effort I can't go into here.
> As Phil well knows, the NIR is the "opportunist band" for cheap data on the
> human body. The visible has too much absorption, and the far IR hard to work
> with.
>
> SO, NIR. And FTIR seems to produce better data than dispersive spectography.
> The peaks are much sharper and better defined. But Phil is leery of
mechanical
> movements, fair enough.
> I'm agnostic.
Oh, so you have a prototype up and running?
I've made a few (thrown together) interferometers, they are dang sensitive.
I'm recalling one, where once we got rid of air currents and other sources of
drift, you could then see the effect of putting a 1/4-20 screw on the optical
table the thing was built on... this bent the table enough to change path
length... OK this was built on just a 1/2 inch thick hunk of bread board.
So not that stiff.
>
>
> In my mind, the question of interest is the following: In a welter of signals
> coming from NIR reflectometry, for instance, is there a signal processing
> method, after the fact, that will pull out signals buried in the "noise." ?
>
> Yes, you can do things on the front end - polarimetry to remove skin
> reflections, time of flight, etc. And these surface measurements have
produced
> some good cancer detection approaches.
>
> So I'm looking for an "aha" moment here about some signal processing method.
And
>
> I admit, it's unlikely to be there, since so many researchers are working in
this area.
>
>
> But, already there are signs that you can extract very general information
from
> a human and detect very specific disease states. This flies in the face of
> medical diagnostic thinking, which is reductionist, and not-so-much systemic.
> What am I trying to do? George, mainly I like to do weak signal processing
and
> AI software. I like to think about lock-in amplifiers, signal averaging, and
> the like. I have spent my time mainly thinking.
>
> And I'm mainly interested in one problem: In a welter of signals, how do you
> pull out a weak signal of interest buried in there? Is there an analogy to
> bouncing a signal off the moon?
>
Well if you can't modulate or otherwise "code" your signal then it's hard.
(I guess I'm thinking in the frequency domain...)
We made a "physics package" to sell with an SRS770 FFT spectrometer.
One of the modules was a signal in the noise. A single freq. oscillator mixed
in with a big noise signal... your job was to find the signal...
There are four different signals each a decrease by about a factor of two (in
amplitude) from the previous (and at a different freq.)
So the biggest one is easy to see, you almost don't need the FFT.. just
triggering on a 'scope. It's also placed at 3kHz.. so you can put the signal
and noise into a speaker and hear the tone. (which is fun.)
The smallest signal (16 times smaller than the easy one) is almost impossible
to find. Even when I tell people where to look it's hard to find. (Limited
band width of FFT and lots of signal averaging.)
>
> One way, with classic statistical training approach, is you get data on
people
> with and without TB. (just an example only.). Then you train, say with a NN,
> to get your recognizer to indicate TB. In that way, the statistical process
> itself can filter out the "noise."
> I hope I've been able to state my case, and I feel fortunate to work with
> brilliant people like yourself and Phil, at least in the restrictions here.
"Sputter!" I'm far from brilliant, average at best.
Phil is a treasure, I too cherish his presence here.
George H.
--- SoupGate-Win32 v1.05
* Origin: you cannot sedate... all the things you hate (1:229/2)
|
