There is no exact minimum because it depends on a number of factors like the heterozygosity of the gene, the insert sizes of the library, and the quality of the data.
We have observed optimal results when there are at least 15,000 reads per locus. If you sequence 5 loci per sample this means 75,000 reads per sample. If you have less reads, NGSengine may still obtain an unambiguous typing result, depending on your sample.
In general, more reads will lead to better phasing and less ambiguities.
The read depth is the amount of reads that cover one single position. We have observed optimal results when the read depth throughout the locus is at least 500 reads. A lower read depth can also produce optimal results, depending on the sample.