Power, spatio-temporal bandwidth, and distortion in large sensor networks

For a class of sensor networks, the task is to monitor an underlying physical phenomenon over space and time through an imperfect observation process. The sensors can communicate back to a central data collector over a noisy channel. The key parameters in such a setting are the fidelity (or distortion) at which the underlying physical phenomenon can be estimated by the data collector, and the cost of operating the sensor network. This is a network joint source-channel communication problem, involving both compression and communication. It is well known that these two tasks may not be addressed separately without sacrificing optimality, and the optimal performance is generally unknown. This paper presents a lower bound on the best achievable end-to-end distortion as a function of the number of sensors, their total transmit power, the number of degrees of freedom of the underlying source process, and the spatio-temporal communication bandwidth. Particular coding schemes are studied, and it is shown that in some cases, the lower bound is tight in a scaling-law sense. By contrast, it is shown that the standard practice of separating source from channel coding may incur an exponential penalty in terms of communication resources, as a function of the number of sensors. Hence, such code designs effectively prevent scalability. Finally, it is outlined how the results extend to cases involving missing synchronization and channel fading.