Naively I would assume the closest you can get to an impulse function, aka a Dirac delta function, because it contains all frequencies at once. So, as loud and short of a sound as possible. This should give you the most information about your surroundings from listening to the echo, because different frequencies can reflect differently in a room. Also because mathematically an impulse convolved with a function returns the function, so an impulse gives you the most neutral echo with which to compare.
Autocorrelation is a delta function. Basically a random known signal, so you can disambiguate simply from background noise (look into matched filters).
I don't know - if I were actually going to make this device I'd try to figure it out by talking to people who can echolocate, people who do signal processing, scientists studying echolocation in bats and whales - etc.
