As I understand it, one typically performs Fourier deconvolution for the data produced by some instrument of a linear system by:

  1. Taking the observation of some unit impulse signal $m$, with the property that every other input can be represented as the sum of scaled versions of $m$ - let's call the data of this observation $s$.

  2. Taking the observation of the phenomenon we care about, let's call this $o$.

Then, we say that to go from $m$ to $s$ the instrument has convolved with some other function and then say, via certain remarkable properties of Fourier transforms and convolutions, that $O = IS$, where $i$ is the ideal data we wish to obtain from $o$ via deconvolution and capital letters denote the Fourier transform of a function.

The other day, a friend suggested the following in a situation related to measuring the spectra of certain light sources:

Because we couldn't measure a unit impulse signal, we should instead measure a known blackbody radiation curve, $bbr$, as the data $s$. Then, he said that "via a parallel" we see that for a general input $i$ to our system and a general output $o$ associated with that input, we have $O = SI$ and then you solve for $I$ using inverse fourier transforms.

I pointed out that it was definitely not known that $bbr$ could represent every other signal in the form of the superposition of scaled versions of itself and so that there was no reason I knew of for this to work. Given that they said "it was a form of genuine physical insight" and the fact that they are generally correct, I believe them - but why is what they said true?


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.