I recently worked through a series of wifi issues and had to learn a bit in the process. While I've fixed my specific problems, I still have a lingering question about co-channel interference. Specifically:
Is a minimum SNR required for co-channel interference?
Say there are two different 5Ghz networks on channel 36. The AP in each network can just barely hear the AP in the other network. Let's ignore the client signal ranges for this question to keep it simple. Any traffic picked up from the other network only has signal strength of -80 or less. For both APs, the noise level is also low, -92 or less.
Even with the poor signal, most of the traffic from the other network's AP can be heard. If you use something like tcpdump, you'd see a high number of corrupted frames from the other network's AP, but you'd still see the frames.
From what I've gathered about this situation, this should cause co-channel interference between the APs (again, please ignore the clients). Each AP must compete for the same airtime, so activity in either network will necessarily degrade the performance in both networks.
If that's true, then the various online guides for wifi placement that make it seem as those "low signal" = "low co-channel interference" are just wrong (see Netspot's guide as an example). If two AP's can hear each other, regardless of how clearly, they compete for airtime.
If that's false, then there must be a minimum signal amount required for co-channel interference. And I've somehow never come across what that is!
Update
The answer appears to be that low signal may mean low co-channel interference, but only for very low signal. Specifically:
At least -85 dBm should be considered the signal strength threshold for CCA (from Cisco's Wireless LAN planning guide, page 14)
Ron's answer of "If two wireless stations can hear each other well enough to identify 802.11 traffic, then they cause co-channel interference" is also valid, but a signal level is a more specific answer to the question. Ron's other comment of using -87 dBm as a threshold for same-channel signals at the perimeter of the network is a more-restrictive answer than Cisco's, but would certainly result in a clearer network if it's possible to achieve.
There are many more factors in wifi, of course(managed vs unmanaged interference, the range of the wireless clients, other interference, etc), but this question was specifically about the signal threshold for co-channel interference, so the most specific answer is Cisco's "-85 dBm threshold".
Best Answer
A CCI signal of -80dBm is high enough to cause problems. If the APs hear each other at around -80, then a client in between the two APs would hear both APs at a higher level. That's where co-channel interference happens.
The problem is not the APs, but the clients. CCI doesn't just corrupt frames (any noise will do that). CCI interferes with the clear channel assessment (CCA) function in each client. When a client is ready to transmit, it checks to see that the channel is clear (that no one else is transmitting). If the signal strength is strong enough that a radio can decode signals as 802.11, then it will fail the CCA test and the radio will defer transmitting. A nearby transmitter on the same channel will make the client wait for a clear channel even though the AP it's associated is ready to listen, reducing throughput.
The co-channel signal strength should be 20 dB below the AP strength at edge of the cell. Since the cell boundary is usually -67 dBm, CCI should be -87 dB or less.