[argyllcms] Re: About Hi-Res spectral mode

  • From: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Tue, 26 Oct 2010 09:40:44 +1100

Gerhard Fuernkranz wrote:

Hi Gerhard,

looking at your laser pointer spectral plots it seems that the FWHM of
the instrument's optical system is still the limiting factor [hard to
tell exactly from the diagram - but I'd guess about 20nm ? - assuming of
course that the laser is really monochromatic]

Yes, the optical system is certainly the limiting factor. The switch from
10nm to 3.3 nm sampling does seem to improve the FWHM from roughly
20nm to 10nm though (I estimated 18 and 9nm from the plot), although
this may well vary somewhat with wavelength.

The benefit of higher
sampling resolution is of course limited if the bandwidth cannot be
reduced as well. Theoretically the higher sampling resolution should
offer an opportunity do deconvolve to a smaller effective bandwidth, but
in practice a stronger deconvolution is limited either, by a too bad S/N
ratio.

I did play with this, but I didn't manage to get anything better than choosing
a straightforward lanczos2 3.3 nm sampling.

Certainly, as we see, high-res mode does reduce the effective
FWHM, but obviously it can't reduce the effective BW by the same factor
as the sampling resolution is increased. So I'd still expect a
noticeable improvement from an instrument which reports not just with
1nm resolution, but also *with 1nm FWHM*. According to the literature,
20nm BW is too much for accurate measurement of narrow-band light
spectra, the literature rather suggests to use a BW of 5nm or less (and
a corresponding resolution too, of course).

By this measure a FWHM of 9nm is still insufficient, but is at least
somewhat better than 18nm.

To get a feeling for the
impact, it's quite easy to simulate how the effective spectrum locus
moves in chromaticity space, when monochrome spectra are "broadened" to
a triangle or Gaussian with say 5nm, 10nm, 20nm, etc. FWHM [see
http://tinyurl.com/ykrms2], and it should be easy as well to compute the
corresponding colorimetric errors.

So the difference between normal an high resolution modes is roughly
the difference between the cyan and blue curves.

cheers,
        Graeme.

Other related posts: