Exploiting partial cooperation for source and channel coding in sensor networks

A network with two sensors communicating a remote measurement to a common access point (AP) is investigated. The sensors are connected via out-of-band and finite-capacity communication links, e.g., thanks to an orthogonal radio interface, thus enabling partial cooperation. Focusing on a Gaussian model for source and observation noise and a quadratic (MSE) distortion metric, both the source coding problem (corresponding to error-free and orthogonal links to the AP), also known as the "CEO problem", and the joint source-channel coding problem over a Gaussian multiple access channel to the AP are considered. In the first case, an achievable rate-distortion tradeoff is derived that generalizes known results for the CEO problem in the absence of cooperation between the sensors. For the latter case, achievable distortions are derived with separate or joint source-channel coding. Optimality of the proposed schemes is established asymptotically with the capacity of the inter-sensor links. Moreover, it is concluded that, for both scenarios, even modest values of such capacity enable the optimal performance with full cooperation to be approached.