I doubt that one can calculate individual noise from pairs.
Indeed you can Click :
www.intersil.com/data/an/an177.pdf
The calculation is not difficult!
The measurement is a pairwise sample DUT noise density over frequency. Because the assumption is that the samples are statistical independend the noise power at sa specific frequency is simply the addition of the noise power of the DUT pair.
If you have n=4 samples you can make n*(n-1)/2=6 pairwise measurements. I will call the reference sample spectral noise power at a specific frequency simply a,b,c,d.
What you measure with the pairs is the sum of the noise powers
a+b
a+c
a+d
b+c
b+d
c+d
You get the noise power of the first sample reference by the following expression:
a=(1/3)*(((a+b)+(a+c)+(a+d))-(1/2)*((b+c)+(b+d)+(c+d)))
a=(1/3)*((3*a+b+c+d)-(1/2)*(2*b+2*c+2*d))
a=(1/3)*((3*a+b+c+d)-(b+c+d))
a=(1/3)*(3*a)
a=a
For the general case n the expression is:
a=1/(n-1)*(sumwith(a)-1/(n-2)*sumnot(a))
The application note from LT does not give a calculation example. It is only stated that you have to use a much better reference as a DUT partner.