NMR software can erase t1 noise for you. Try ACD labs NMR proccessor. It's free for academics.

You can try a different pulse program that has pulse field gradient. And if MW < 300, a 1.5mg/ml sample is > 5mM. You can dilute your sample at least 50 times and still get good signals. Maybe 1D NOE experiment also can be tried. Just for your reference.

Diluting your sample is a very good idea; 100 mg/ml with MW < 300 is a good concentration to run a carbon experiment, but not a noesy. 10 mg should be OK and if you look for a weak noe, never run less than 16 scans. I would try too shorter mixing times, maybe 500 ms or 700ms.

I'm not a big fan of 2D NOESY for small molecules in the extreme narrowing region. I find that the gradient selective pulse 1D NOESY experiments (available from the "Selective" panel in Topspin) are much better and can detect NOEs down to a small fraction of 1%. When you get to slightly larger molecules, at the cross-over between extreme narrowing and spin diffusion, then the corresponding 1D ROESY experiments (especially the modification by Julien Furrer J. Nat. Prod. 72, 1437, 2009). work very well.

If you really need the 2D experiment, then check for things that tend to cause T1 noise: vibration, room temperature stability, proper PID parameters in the VT unit, good sample, proper lock parameters, etc. I'm suspicious of processing algorithms, including symmetrisation, for reducing T1 noise.

Also, I think your mixing time might be a bit long. It's always a competition between NOE build-up and relaxation. I find that about 500 ms. tends to be a bit better for most things.

100 mg is probably a bit to high for concentration at this MW (10 to 20 mg would be fine), and it helps to degas your sample to ensure maximum NOE by reducing relaxation caused by dissolved oxygen.

There are some tricks to optimize your spectrum once recorded but I think you should concentrate first on running an experiment with a better signal to noise ratio. A solution of 10 mg/ml should be sufficient wrt signal to noise. A mixing timne of 1.5 sec is much too long. For my experience you can use 250 – 500 msec as an initial delay, again depending on molecule and the interaction you want to observe. You can always optimize the delay but in this range you should be able to record a spectrum with a good signal to noise.

However, depending on your solvent, field strength and your molecule, you need to choose between a NOESY or ROESY, since the NOE effect will disappear and no cross-peaks will be observed in a NOESY spectrum when w*tc=1.

Back to your problem. You can do three simple things; 1) run the experiment with a mixing time of 500 msec. 1) try a ROESY using a 500 msec spinlock 2) change the tumbling rate by changing solvent, e.g. CDCl3 <-> DMSO.

Good luck

ps; like one of my friends always tells me when I am struggling to record good data: it is easier to optimize a signal then to find a signal ;-)