To start with, you must ensure that the resolution and sensitivity are good enough in your 1D spectrum. Indeed, quality tubes and good tuning/matching are essential, but in addition I would suggest to try to further optimize the shimming manually after gradshim, using your BSMS keyboard. If your peptide is rather stable, You could also perform several 1H NMR at different temperatures/pHs in order to search what conditions give the best resolution and signal dispersion. What is your sample concentration by the way?

Then comes the NOESY experiment itself, and loads of parameters that can be adjusted. You can of course increase the number of scans (NS) and the size of FID (TD) in both dimensions. The mixing time (d8) is a very important parameter that should be optimised by doing a NOE build-up curve. The way that the data processing is done can have a big impact too on the final rendering: sometimes manual rephasing + noise filtering is necessary.

If all this doesn't help to get a nicer 2D, you should consider running a ROESY instead: for a mid-size molecule like yours, it would not be surprising that ROEs give a much better result.

It is easy to verify your shimming - add a tiny (probably 50 uM will be visible) amount of TSP or DSS - water soluble molecule with trimethyl-silyl groups. It should give a sharp peak and will give you a 0 ppm reference.

If the reference compound peak is sharp, but your peptide is not - something must be happening with the peptide - like aggregation or chemical exchange on the intermediate time scale.