Capacities of time-varying multiple-access channels with side information

We determine the capacity regions for a class of timevarying multiple-access channels (TVMACs), when the underlying channel state evolves in time according to a probability law which is known to the transmitters and the receiver. Additionally, the transmitters and the receiver have access to varying degrees of channel state information (CSI) concerning the condition of the channel. Discrete-time channels with finite input, output, and state alphabets are considered first. The special case of a TVMAC, with the channel state process being a time-invariant, indecomposable, aperiodic Markov chain, shows a surprising anomaly in that imperfect transmitter CSI can cause the capacity under some distributions for the initial state to be strictly larger than that under a stationary distribution for the initial state. We also study a time-varying multiple-access fading channel with additive Gaussian noise, when various amounts of CSI are provided to the transmitters and perfect CSI is available to the receiver, and the fades are assumed to be stationary and ergodic. Implications for transmitter power control are discussed.