Based on the documentation of NMRpipe it appears that the SOL function uses time domain deconvolution.

When I was graduate student, I had written my own program for NMR data processing and I used time domain deconvolution for solvent suppression. I observed that the **length of the window used for deconvolution deconvolution** has a dramatic effect on the quality of the solvent suppression. Based on this, and a quick look at the NMRpipe documentation, I would expect that a substantial improvement of solvent suppression can be obtained in your spectrum if you lower the value of flenpts parameter associated with the SOL function. In addition, I would expect that changing the fshape parameter to select for sinusoidal weighting (instead of the default boxcar) would be helpful.

My interpretation of your artefact is as follows (though I have not seen the code for NMRpipe): the deconvolution algorithm calculates the average over the chosen window and subtracts it out from the current fid. The problem is that, this is not possible for the first few data-points; the window terminates at zero time, and the data for calculating the average is not available for the first few points. Therefore, the subtracted fid contains a solvent artifact contributed from the first few data points. There are several ways to overcome this problem:

Decrease the window size for the first few data points, keeping the chosen value of flenpts for the remaining data points.

Predict values of the fid for negative times by using extrapolation. Then use these values for calculating the required averages for the deconvolution. The second method, I believe, would be better.

If the problem lies with incorrect adjustment of offset, then the observation of the data after the FT in the first dimension will reveal that the positive lobe of the non-suppressed water is consistently either on the right or on the left. However, I suspect, that you will find that in some of the increments the positive lobe is on the left and in others it is on the right.

This solvent suppression method may also fail if there are problems with magnetic field homogeneity, temperature homogeneity (due to RF induced heating) or radiation damping.