There's a large literature about this topic. When you read the literature, be sure to understand the difference between _detection_, which is simply detecting a response, and _estimation, which if I recall correctly means estimating the shape of the response (e.g. if you're using a FIR).
If you're _detecting_, you can have very short timing between event types and still have a fairly "efficient" design, _as long as there is sufficient (pseudo-)randomness_. So, for example, if the event types alternate exactly, (ABABAB...), it's probably not very efficient. (I haven't looked at this literature for a few months, so you're best off consulting it directly.)
There are of course lower limits on how short you can go, because of the physiological nature of the hemodynamic response. If things are _too_ short (e.g. less than 500 ms), then you start getting nonlinearities in the response. (This is also in the literature of course.)
The most efficient design for detection is a block design, but if you're careful about (pseudo-) randomness you don't have to give up too much.
|