Title:On sensing capacity of sensor networks for the class of linear observation, fixed SNR models

Abstract: Motivated by many practical scenarios, in this paper we study fundamental limits to performance of sensor networks for a class of linear sensing models under a fixed SNR regime. Under a linear observation model the information about a signal $\bX \in \Real^n$ is acquired as projections onto a set of $m$ vectors. We extend this set-up in several ways to address problems of detection, classification and estimation that arise in sensor network applications where the signal $\bX$ exhibits sparsity and adopt a Bayesian perspective. We identify a projection as a measurement made at a sensor. In the context of fixed SNR regime we normalize the vectors to have a unit norm and then scale the system by SNR. Each component of the corresponding projection vector can then be thought of as contributing towards the \emph{sensing diversity} per sensor. For such a set-up we propose asymptotic upper and lower bounds on the probability of error in reconstruction subject to an average distortion criteria. Using these bounds, we provide asymptotic upper and lower bounds on the ratio $C(d_0) = \dfrac{n}{m}$ such that reconstruction to within a certain average distortion level $d_0$ is feasible. Bounds on $C(d_0)$ are then interpreted as providing \emph{scaling laws} for the minimal number of sensors required for reliable monitoring. Our main contribution in this direction is that we isolate the effect of signal sparsity, diversity and SNR on the performance and show that for a fixed SNR regime (a) $C(d_0)$ goes to zero as sparsity of the signal goes to zero. This means that relatively more sensors are required to monitor very sparse events, (b) low diversity implies lower $C(d_0)$. Thus the bounds presented in this paper serve as guidelines for designing efficient sensor network architectures.