Empirical Coordination for Joint Source-Channel Coding

We investigate the problem of simultaneous transmission and coordination for a point-to-point communication model. The encoder and decoder choose their symbols/actions (i.e. channel input and decoder output) in order to transmit information and to coordinate themselves. This paradigm has promising repercussions for future decentralized networks. Exploiting the coordination capabilities, devices can implement decentralized policies that are coordinated. Exploiting transmission capabilities, devices choose their actions (power control, the channel allocation) in order to encode additional embedded information (channel state information, future allocations). The feasible trade-offs between transmissions and coordinations are characterized using joint probability distributions over the source and channel symbols. Encoder and decoder implement a coding scheme such that the empirical frequency of symbols are close to target joint probability distribution. We characterize the set of achievable probability distribution for a point-to-point source-channel model with strictly-causal and causal decoding. Compared to classical coding results, the coordination requirement induces a stronger information constraint. We determine optimal trade-offs between transmission and coordination by maximizing a utility function over the set of achievable probability distribution.