"You can control two orthognal states (i.e. make measurements along one axis to send a 1 and then change axis to send a zero) but now the other end needs to know which axis to measure along (and when) thus one needs a normal commuinications channel to pass this information along.
The other possibility is using the statistical nature itself to send information (i.e the rate of up/down), however, this requires copying the quantum state which cannot be done.
I read that and wonder, "Why not?" Firstly, let's talk about the top part where we need a communications channel to transfer two important bits of information: Which axis should we measure and at which time. It seems that both of these can be theoretically settled *before* extra-light distances are between the two subjects. Assume we have a protocol which maps axial shifts as a function of time. For example, to transmit an "A," we shift axis five times evenly distributed over a five-second period and then make no axial changes for another five second period.
For timing, each of the subjects start at the same point with a device they synchronize to the same timing sequence. A "second" is relative to the synchronization of these two devices. After they separate, rather than synchronizing them to each other, use known physical properties to augment drift due to whatever forces act upon the devices. Maybe even have them send physical sub-light waves and use the known properties of the transmission timing disparities to handle drift.
Now, on to the second paragraph, our storage of the protocol need not be tied up in the entangled photons. Rather than copying quantum state we simply have two communications devices which each only communicate one way. Since we have devices that can currently "read" the results of entanglement at a local level, these two devices can store the data at sub-light speeds.
It seems to me that the trick is developing a statistical protocol that makes error detection reliable within a certain percentage. It seems doable if we develop an appropriate statistical model for the 50% chance of change when forcing a quantum bit out of a superstate. Like all of those models you make a certain number of transmissions to enhance certainty of accuracy rather than ever expected your result to be perfect. If you find the right mix between valid and invalid, you can be relatively certain of the message.
Now, why what I have stated theoretically impossible? I know there are lots of really tough pieces to what I've outlined, but if the idea is doable, then those pieces are worth pursuing, and I want "instantaneous" communication over arbitrary distance damnit! Whittling interstellar communication down to a protocol that takes minutes to communicate a solid message is better than one that takes years.