Distributed rate splitting in Gaussian and discrete memoryless multiple-access channels

We analyze a distributed M-user multiple-access problem in both the additive white Gaussian noise channel and the discrete memoryless channel. A system where each user independently chooses a transmission rate according to a probability distribution function is considered. Under the assumption of symmetric capacity regions, we show that the optimal transmission rate distribution is a point mass for both channel models. To implement this, we propose a scheme called distributed rate splitting. In this scheme, each real user creates the same number of virtual users via a power/rate splitting mechanism in the M-user Gaussian channel or via a random switching mechanism in the M-user discrete memoryless channel. At the receiver, all virtual users are successively decoded. Distributed rate splitting allows multiple-access communication to take place with minimal control overhead. Moreover, the rate tuple achieved by this scheme converges to the maximum equal rate point allowed by the information-theoretic bound as the number of virtual users per real user tends to infinity