Not all earthquakes can be used. Two main restrictions limit the acceptable magnitude range, both are a function of the domain size as well as the period range one is interested in.
Must be big enough to be measurable across the whole domain in the frequency range of interest. Small earthquakes additionally don't excite a lot of low frequency surface wave energy which further limits their use especially in the initial, long-period iterations of an inversion. This is dependent on the sites' instrumentations and noise levels and no general rule can be given and it has to be figured out by trial and error. Most studies set the minimum moment magnitude to a value between 4 and 5.
The earthquakes must be small enough so that they can be reasonably approximated by a point source. It is okay if the source mechanisms captures some of the finite fault nature. The rupture length of the earthquake should be significantly smaller than the smallest wavelength one is interested in, which is especially important for receivers close to the fault.
This is a deep topic and strongly dependent on the region and type of fault but the following figure empirically relates fault length to moment magnitude. There are no hard rules but between moment magnitudes from 6 to 7.5 the fault length approximately ranges from 10 to 100 km and depending on the chosen region and frequency range the maximum magnitude one uses for a continental scale full waveform dimension should be in that range.
Common moment magnitude vs fault length relations, constants taken from Leonard, M. (2014)
. Self-Consistent Earthquake Fault-Scaling Relations: Update and Extension to Stable Continental Strike-Slip Faults
. Bulletin of the Seismological Society of America, 104(6), 2953–2965. doi:10.1785/0120140087