What he has done is to calculate the size of the Airy disk using the standard formula d/2 = 1.22*wavelength*f-number. Using green light of 500nm wavelength gives the values in his table.
Diffraction limit for the Rayleigh criterion:
cycles/mm = 1/(1.22 * wavelength * f-number)
Converting from number of cycles per unit of size to size of each cycle:
1/cycle size = 1/(1.22 * wavelength * f-number)
Solving for f-number:
f-number = cycle size/(1.22 * wavelength)
Normally, it would take 2-pixels to resolve one cycle (Nyquist). However, because the sensors we use have pixels that are not of zero size, because they are Bayer sensors and thus require demosaicing, and because they usually have AA filters, I like to use 3 pixels per cycle, rather than 2 pixels per cycle, to compensate for the loss of resolving power from these sources.
f-number = (3 * pixel size)/(1.22 * wavelength) for the Rayleigh criterion
If we plug in your numbers (500nm light and the pixels from the 90D), we get:
f-number = (3 * 3.2 microns)/(1.22 * 0.5 microns) = 15.7 (f-number for the Rayleigh criterion on 3.2 micron pixels using 3 pixels per cycle)
If you prefer extinction instead of Rayleigh, this formula works:
f-number = (3 * pixel size)/(1.00 * wavelength) for MTF=0 (extinction)