Its pretty much the same as you do any other voice, but there are some issues.
How many users are you aggregating to each port on the head end? How many Erl of traffic per user? ~0.3 Erl per user for residential.
We know that IP=20 bytes of header
UDP = 8 bytes and RTP = 12 bytes, payload will depend on the codec and sample period. Add in the L2 headers for DOCSIS, I dont know them so you will have to look this up. This will tell you how many bytes each sample period will take up. Multiply by the packets per second to get the KBps. Figure out how much data speed, depending on QAMS etc… then see how many flows you can shove in. divide data speed by KB per call. Then youre done. This would be a theoretical number not including any data.
After you get max number of calls, you can figure out how many users you can aggregate to each head end port.
You would have to leave some space for the signaling of each of these calls too.