Linear prediction is a method to construct FID points, which have been truncated. This often is the case in 2-dimensional NMR spectroscopy in the F1 dimension, for time saving reasons. You can gain substantial resolution using linear prediction on a truncated signal. Under normal Fourier-Transformation you have to heavily dampen the FID with linebroadening to prevent the side wiggles on the baseline of peaks ("sinc-wiggles"). The other useful application of linear predicting FID points is in the beginning of the signal. Reconstructing these first points straightens out wavy baselines.
To help you set up a linear prediction experiment, we have two macros available:
The macro "linpred" helps you to set the parameters for forward prediction. Without an argument, after answering the questions you set up the parameters for one-dimensional spectra or for the directly detected dimension (F2) in 2-dimensional spectra. Using "linpred(1)" you enter the parameters for the indirectly detected dimension in 2-dimensional spectra (F1-dimension).
The macro "lpbc" will delete the first points of your NMR signal, linear predict what these points should be and Fourier transform your spectrum. You can enter two arguments lpbc(#1,#2) where #1 allows you to set the number of points to be reconstructed. With #2 you set the number of lines you expect in the spectrum. For example lpbc(32,10) reconstructs the first 32 points assuming the spectrum has 10 lines. Originating in the response of the electronics to the rf-pulses (especially with large sweep width), the bad first points lead to baseline humps and wavy baselines. The macro lpbc helps to eliminate these problems.
Related issues:
You can use two other approaches to remove wavy baselines: Experimentally, before you take the data, you can set alfa higher (standard is 20 microseconds). The other approach helps after you have taken your data: With lsfid = #pts you can left-shift the fid by #pts and then fourier transform. Using df you can display the fid. After expanding to see the first few points, you can estimate the number of points you need to remove. In both approaches, you may have to manually phase the peaks going successively across the spectrum after the Fourier transformation.