Design Considerations in Wireless Sensor Networks

Abstract

We consider three problems in the design of wireless sensor networks: crosslayer optimization, neighbor discovery, and scheduling as a method of medium access control (MAC). Cross-layer optimization will be important for sensor networks, which typically have only one or two objectives to meet. We consider a sensor network which performs decentralized detection. We devise a method in which local observations by sensors are condensed into a single bit message and forwarded to a sink node which makes a final decision. The method involves unusual interactions between the application, the routing function, and the physical layer. Neighbor discovery is useful in sensor networks whose nodes are immobile, since routing and scheduling algorithms can make good use of neighbor information. We propose an asynchronous neighbor discovery algorithm. The algorithm is probabilistic: each node obtains a list of its neighbors which is possibly incomplete. Performance is analyzed and optimal parameter settings are obtained. Scheduling deserves consideration as a MAC in sensor networks, because MACs based on contention methods waste energy in re-transmissions. We state a natural centralized scheduling problem, in which link demands are to be satisfied under signal-to-interference-and-noise-ratio (SINR) constraints, and transmit powers may be varied. We show that solving this minimum length scheduling problem is at least as hard as another problem we define, MAX-SINR-MATCHING, in the sense that if there is no polynomial-time algorithm to solve the latter then there is no polynomialtime algorithm to solve the former. We give evidence that MAX-SINR-MATCHING is a difficult problem. We add several theorems on the SINR model which exploit algebraic structure. The theorems predict what sets of links could be simultaneously activated in a wireless network and depend only on the SINR requirements of the nodes and the worst propagation loss in a network. These theorems apply to all wireless networks which can be described by SINR requirements, not only to sensor networks.