There are certain processing steps that should be performed before deconvolution. Calibration and image stacking should be done first. Hot and dead pixels should be fixed. If the camera does not produce square pixels, it is recommended that you interpolate them to a square aspect ratio before proceeding. It is recommended that any other processing functions be deferred until after deconvolution is run.
If you are using a Uniform noise model you might also try experimenting with the Uniform Noise Level. A higher noise level will reduce the convergence rate (making it easier to converge), and a lower one will increase it. For either noise model some experimentation with the Background Level may also be in order, since selecting a proper background level is essential for good results.
You can use subframe deconvolution to determine whether your settings will work. Deconvolution algorithms use Fast Fourier Transforms. This means they work only on images whose width and height are both powers of two dimensions (e.g. 256, 512, 1024, 2048…). If your image size is not of this form, it will be padded up to the next larger size. If your image is just slightly larger than a power of two, it is best to crop it slightly – this will produce better convergence and faster processing.
One problem is the tendency for deconvolution to dig round holes around stars (”donuts”). To minimize these, make sure the background level is set appropriately. For non-scientific applications, these donuts can be cosmetically eliminated using the Clone tool.