The current procedure for de-fringing makes use of the constancy of the period of the fringe in wave-number space to obtain the fringe amplitude; the more fringes obtained the better and we recommend that the spectrum contains at least three fringes. Note that, with our present understanding, IT IS IMPOSSIBLE TO DE-FRINGE SPECTRA WITHOUT THIS INFORMATION. In particular, there is no way of determining the true strength of isolated lines from an extended source - taken with either the grating or Fabry-Perot - unless a spectrum of the continuum is also taken.
If you are CERTAIN that your source is point-like and well isolated, and you know its infrared coordinates to within 5", you need take no action. In most cases, however, you will want your data de-fringed. If you are making a wide-range LWS01 observation, make sure that your range is sufficient to cover at very least three fringes in any one detector channel; refer to the examples in figure 3 to decide how far to go. If you are using a small spectral range, or are doing LWS02, 03 or 04 observations, you should consider concatenating an additional LWS01 scan, sampled at intervals of one spectral resolution element.
Note that if you are doing an off-source measurement (as recommended) or map you will need to characterise the fringes in all pointings (e.g., for a 2x2 raster you would need to repeat the raster with the appropriate LWS01 scan.)
The fractional accuracy 
needed in correcting for fringing is
related to the signal-to-noise ratio 
demanded on the line:
If 
independent samples are taken of the continuum spectrum, the required signal-
to-noise ratio 
in this spectrum must satisfy:
Note the discussion in section 5.4: de-fringing of spectra requires intelligent consideration of the possible nature of the source.