Comment 33 for bug 544254

Unfortunately (as nearly expected) the time frame between release and the first rc of the next kernel version has the majority of changes. It looks a bit like the problem is spread between generic code and the driver itself.
The message and the frequency it occurs sounds like the code (at least) thinks the link is idle. After 30s idle time it will try to check whether the AP is still there. Then wait for half a second and when there is no response, assume the connection dropped.
So one experiment would be to try have a ping running in the background and see whether this would stop the disconnect. That would at least prove that normal traffic is detected as activity.
For hopefully more insight, I compiled a kernel with debugging turned on for the iwlagn driver. The packages are on http://people.canonical.com/~smb/lp544254
There seem to be two debug facilities one might be turned on by "echo 0xb80a >/sys/class/net/wlan0/device/debug_level" (as root). I would expect output from this to go into dmesg. The other facility adds directories under /sys/kernel/debug. The definitions are a bit cryptic so I am not sure about the exact naming. Either net/wlan0 or iwlagn or whatever. The should be a subdir called debug and in that rx_statistics and tx_statistics as well as rx and tx_queue. Maybe output of those helps too. But I would first try to see what the dmesg shows with the debug_level set.