The joint gaussian CEO/MAC problem and relaying

We consider the problem of agents which observe a noisy version of a source, and communicate with a central decoder (ldquoCEOrdquo) through a multiple access channel (MAC). We assume that the source, the observation noises and the channel noise are all Gaussian, and we are interested in reconstruction under a mean squared error distortion measure. The solution by separation (combination of CEO and MAC coding) is sub-optimal. When the source and channel bandwidth (BW) are equal, analog transmission is optimal. We present a scheme for non-equal BW which becomes asymptotically optimal in the limit of low observation noise. We use this scheme to improve our recently presented Rematch and Forward approach for the parallel relay network, and thus extend the achievable rate.