>> Is the downstream pilot tone essential for synchronisation?
Yes - its used during the sync stage.
>> Unless the pilot tone can be automatically moved,
Not that Im aware of for adsl1
The frequency is defined in the G.992.1 standards.
>> wouldn't that suggest that the pilot tone could effectively be missing
Theorising that the pilot tone can run with just 1 bit? It wont need any error correction allowances or overheads. Its a signal and afaik many routers dont even display the SNRM for that frequency. A min of 2 bits means that there must be at least 6dB of SNRM at that tone during the sync up stage or it its marked unusable. Its therefore likely that the pilot tone can operate at less.
Error correction (RS encoding) has some overheads that has to be allowed for regardless if interleaving is on or not and therefore min 2 bits for data transmission for adsl1.
This probably has something to do with s=1/2 mode and its the reason why despite adsl1 technology having bins to support up to 12Mbps, 8128 is the max sync speed.
S=1/2 mode is used in later dsl technologies such as adsl2 to get the full 12Mb.
As an aside, some (adsl1) dslams can now operate in s=1/2 mode as long as your router supports it which effectively makes RS overheads more efficient which is why you sometimes see an interleaved line sync'd at more than 7616 (I suspect these are the Fujitsu geostream dslams which are used in many maxdsl exchanges).
I dont know enough about the pilot tone to comment much further Im afraid.
------
What is entirely possible is that line over the later stages could have SNRM that goes something like 6, 6, 6, 6, 5.5, 5.5, 5.5, 5.5, 6, 6, 6dB which is what would cause that pattern towards the end.