What study? Oh, the one you mentioned in the title but didn't mention in the body so I have to reconfigure my news reader to figure out what you're talking about. Got it.
Hosts or clients? A host is a server. A client is a user. Are you looking for the maximum number of access points (hosts) that can be packed onto a channel, or the maximum number of clients that can associate with a single access point? I don't think so.
When the number of clients associated with a single access point goes over about 20 or 30, it's considered good form to turn on CTS/RTS flow control in the access point. That effectively reduces collisions at the expense of thruput. If you are getting collisions because of buggy clients or hidden clients, then reducing the packet size by reducing the fragmentation threshold, again at the expense of performance, will help. Anyway, as long as flow control is functional, the curve is linear (aggregate bandwidth / number of clients). Without flow control, you're correct... collisions are a problem.
Well, yeah, but they belong to my clients. I'll see what I can dredge up. However you're going to be disappointed. If I take 20 active clients connected to a single access point and try to move data simultaneously among all 20 clients, I will get about 1/20th of the aggregate available bandwidth. Even without CTS/RTS flow control, the bandwidth divides fairly equally as long as everyone can hear everyone else. In the specific case where CTS/RTS flow control is disabled, and you have a number of "hidden transmitters", then you will get collisions and suffer a substantial performance loss. As I vaguely recall (i.e. not sure) the break point is about 5 hidden transmitter type clients, each trying to move traffic at the full aggregate bandwidth. At that point, some form of bandwidth management is necessary to reduce collisions back to an acceptable level.
The hidden transmiter test is an absolute PITA to simulate in the lab as each transmitter has to RF isolated from each other but is still able to communicate with the access point. I did it with a customized n-way power splitter and a mess of attenuators. I managed to get 20dB isolation between ports, which was just barely enough to perform a meaningful test. Leakage from the boxes was more than the isolation so I had to wrap everything in aluminium foil. No fun. If this is your (homework?) project, you're about to have a major construction challenge.
Doing the google thing for actual tests, I found:There are also alternative MAC layers that use polling to reduce collisions, such as Karlnet's TurboCell: