B = (m1
- m2 + m3)/2
G = (m1
+ m2 - m3)/2
R = (-m1
+ m2 + m3)/2
Clearly this has become a simple problem of linear algebra.
X = m1 + m2 + m3 = 2(R + G + B).
S(m1) = m1, S(m2) = m2, S(m3) = m3.
Error propagation gives the variance in each synthesized band as,
S(R) = (m1 + m2 + m3)/4, S(G) = (m1 + m2 + m3)/4, S(B) = (m1 + m2 + m3)/4.
Thus the noise is the same for each band. For a flat spectrum, i.e., R =G=B and m1 = m2 = m3= m, and the signal-to-noise ratio (SNR) for each synthesized band is,
SNR(R) = SNR(G) = SNR(B) = ( 1/3 m )½.
Since the signal in any one of our filters, m1, m2, or m3, the ratio of SNRs in the synthesized bands for multiplexing versus the fixed filter is,
SNR(multiplex)/SNR(fixed filter) = ( 2/3 )½ = 0.817.
If the spectrum consists of an emission line in the R filter, and G =B = 0, then
SNR(multiplex)/SNR(fixed filter) = ( 2 )½ = 1.414.
For the the pan-chromatic image in the case of a flat spectrum
SNR(multiplex)/SNR(fixed filter) = ( 2 )½ = 1.414.
SNR(multiplex)/SNR(fixed filter) = 1
and for the the pan-chromatic image
SNR(multiplex)/SNR(fixed filter) = N ½,
where N is the number of multiplexed spectral channels.
A four-port IFTS is efficient because it transmits all the light to the detectors all the time! This explains why an IFTS and a camera with a set of filters have have equivalent performance. However, the IFTS pan-chromatic image has a N½, SNR advantage over adding the individual camera filter images together. For a formal treatment and to better understand why an Fourier transform spectrometer is competitive with traditional instruments read Charles L. Bennett's contribution to the recent conference "Imaging the Universe in Three Dimensions: Astrophysics with Advanced Multi-Wavelength Imaging Devices".