As we go back in time I would expect the percentage of couplings producing viable offspring to gradually go down, until eventually it would get to zero. I doubt there is some magic cut-off point.
Of course, even horses and donkeys (extremely) rarely breed fertile mules, but we don't classify based on the exceptions. Quite a lot of taxonomy is about making straight lines out of fuzzy edges, and the Linnean system has nearly no temporal flexibility. It sucks but it's the way science rolls. I'm just trying to give the scientific answer here, that any ancestor we could breed fertile young with would be a member of our own species, by definition.
The lack of temporal flexibility is exactly why I don't see it as the last word on the issue. Paleontologists define species by the fossils' morphological characteristics not their genetic compatibility.
If we discover a lost erectus
population how sure are we that no inter-breeding could happen?
How many millions of years ago do we have to go before we reach a situation analogous to the horse/donkey example?
How far back could we go and still hybridise?
Still interesting questions that the Linnean classification is ill-equipped to answer.