Discarded, understood. Sorry, I don’t understand the retransmission part. Does ATM or AAL5 have retransmissions in the protocol?
If that is not the case, are you referring to L2 ReTx using a different protocol, such as G.INP?
This is merely my own understanding of things and please tell me if I have got things wrong: In G.INP - and presumably PhyR too - a DTU is a certain number of whole suitable protocol-specific fragments and what a fragment is depends on the upper layer in use; if ATM then a fragment is a cell including ATM header, or if PTM with 64/65B framing then a fragment is a 65 byte unit - see G.992.3 Annex N. Either way, a DTU is a string of whole data units according to the layer in use above, it does the right thing™ and makes sure that packet boundaries are aligned with inter-DTU boundaries so for example in the case of ATM never transmits a DTU containing 3.5 cells and the last half in the next DTU.
By definition, G.INP DTUs are the units retransmitted in G.INP and in the case of ATM the DTU will be some number of whole ATM cells. I expect the number will be > 1 so in that case each DTU and therefore each unit retransmitted will contain some n > 1 cells and if so there will be no way of retransmitting a single ATM cell in G.INP.