Chaos is useful because (1) the output depends heavily on the precise input, and (2) the output spectrum is unique for each input value, so the transmissions in

Chaos is useful because (1) the output depends heavily on the precise input, and (2) the output spectrum is unique for each input value, so the transmissions interfere with exceptionally low probability. Think of them as hashing functions, mapping an input value to many output bits distributed evenly across some bit range. If you encoded the measurement as a frequency, and realizing that the frequency will have some finite bandwidth, nearby sensors with almost the same value will obscure each other. If I want to get an accurate estimate of the mean measurement, counting the number of very similar input values is critical. You can, of course, have the transmit frequency equal to some scaled version of the measurement plus a transmitter ID frequency. While this doesn’t require any dynamic negotiation to avoid interference, it does consume more bandwidth than the chaotic oscillator case.