Exploring a cheapo diy spectrometry possibility.
The standard spectro meter uses eg. reflected light
bounced off a diff. grating. One modern twist is to put
this light onto a ccd, then diferent parts of ccd get
different colors, and the resultant intensity vs. color is
the
spectrum. No
moving parts.
However another poss. I am trying to grok is doing this
without a grating, instead using eg rgb leds. It would
require de convoluting a signal that is due to known
combo
of led spectra * unknown reflection spectrum. I cant
figure
if the information is 'inthere' or hopelessly smeared out.
Take for example 2 leds and vary their percent
contribution to illumination, call this percent p, freq. is
w, and the individual led spectra are led1(w) and led2(w)
.
The combined illumination is then
L(w,p)= p led1(w) + (1-p) led2(w)
Assuming the sample is 'nice' and doesnt play any
frequency shifting tricks (i believe this is quite rare and
limited eg to some crystals that can do frequency
doubling,
anyway it seems this must be an assumption of optical
spectroscopy ) then it has an unknown reflection
spectrum
R(w) to be found.
Finally assume that the detector has flat spectrum - this
can be done by normalizing the actual response, i dont
think it helps anything to include the detector sensitivity
curve in this analysis. Use two or more detectors e.g.
Rgb
values from a ccd if that helps matters, I dont think it
does.
So the problem is to use the detected signal D(p) from
the
light detector/ccd, which is a weighted average of the
intensities at each freq., to find R.
======== ============= ========== ==============
D(p)= integ over w of (L(w,p)*R(w))
= integ over w of ( p led1(w) R(w) + (1-p) led2(w)
R(w))
Knowing D(p) (and led1, led2) can R be found?
======== ============= =========== =============
Alternatively / maybe equivalently, is there only one
solution for R ( in which case numeric solution is possible
if nothing else). Use 3 leds if necessary, etc.
Differentiating wrt p does not seem to help matters. Any
maths whizzes able to shed some spectrum on this?
Take a gander at [see link] which does something close,
I
am not convinced he extracts a spectrum however.