MIXIT: The Network Meets the Wireless Channel Sachin Katti and Dina
Katabi (MIT)

Summary: Present opportunistic routing rejects packets that make it quite far
away, if they have very few bits corrupted.  MIXIT improves the throughput
further by using network coding at the physical layer.  MIXIT also changes the
"contract" between the layers.  In this case the phy layers contract is to
transmit bytes (not packets) to higher layer.

The network layer can forward the bytes received correctly, but if all
the bytes are transferred it leads to overlap.  Coordination is not
scalable.  MIXIT solution is using network coding of random bytes.
The non-received bytes are zeroed out in coding. 

Q & A:

Q: Does the coded byte have same size as original.

A: Yes

VV: How true is independent assumption.

A: It is mostly true; there was a study. Even if there is some
correlation, it will still work.

TA: Why not just transmit RLE of the correctly received packets.

A: It requires scheduling, which is a hard problem.  We remove the
need for scheduling.

Q: How many retransmissions per packets did you have; just for the
sake of assessing energy consumption.

A: In the paper there is an algorithm to optimize this.  But
throughput increase is clearly due to decrease in retransmits.

Q: Could you not come up with a better termination condition? 

A: Yes, this is an open question.

Q: How does your line of work reconciles with industry direction which
is more towards directional antennas and MIMO

A: For directional it does not apply, esp. if they rely on line of
sight. However for MIMO it mostly applies and can improve throughput.

Q: OFDM has huge multibyte symbols. Do your symbols have to be byte sized? 

A: Byte definition was just for exposition.