There's probably no valid global answer to this question, but here are some thoughts ...
> > I might also point out that you 'should' get better solutions to the > sources you are trying to solve for, the longer your tile size is in time. > Then the fringes from the interfering sources should average out better. >
True, but only up to a point... if the antenna phases have a complex behaviour in time (which in this simulation they do), then the bigger your tile, the harder it is to fit them with a small-degree polynomial over that tile. The optimum trade-off tile size then strongly depends on your field and on your specific phase behaviour.
Cheers, Oleg