Just remember, normally TKIP is not recommended. AES uses hardware encryption, TKIP is slower.
This is interesting!
I started to face 1 ... 5 reboots a day on a 433UAH fitted with R52n and later R52nM. Both cards use AR9220 chipsets so they bring AES on it and does not depend on ROS low level routines.
Those reboots started when I introduced UBNT's Nano2 Loco "M" where I already had several Nano2 Loco (not M) before.
Untill this time (last june/july I guess) I used to run only WPA2-PSK AES.
Since then until few weeks ago I was dealing with those kernel failures folowed by router reboot several times a day (and of course driving me crazy looking for the cause, going trhu cables, power supplies, shielding, grounding, damaged cards, rb firmwares, ros versions, etc).
Then folowing a test sugestion (by Janis) I changed all CPEs (& APs) encryption to WPA-PSK TKIP. Almost two days to change all CPEs at this router area (~80).
Finally I get rid of WPA2-PSK AES at CPEs and then disabled it at ROS security profiles, keeping only WPA-PSK TKIP.
Since then no more kernel failures.
Of course that if in near future I figured what's causing this (e.g it's an AR9220 issue or Nano2 Loco M chipset issue) I'll revert this security config back to WPA2-AES.
Today this RB433UAH have 3 x R52nM on 2.4 legacy mode (ch 1, 7, 13 all 10MHz BW, noise floor ~ -112dBm) and 3 sectors (one spot 2Km away and 2 slices ~12º each). Peak is 80 autheticated CPEs ... smooth, low CPU, no more reboots.
Regards;