Optimizing Control Overhead for Power-aware Routing in Wireless Networks

We analyze the tradeoff between the amount of signaling overhead incurred in path selection in a MANET with time-varying wireless channels and the application-level goodput and end-to-end power expended on the selected path. Here, increased overhead increases the accuracy of the linkstate estimates used in path selection but decreases the amount of bandwidth available for application use. We develop an information-theoretic, bounding approach to quantify the signaling overhead. Specifically, we investigate (i) the time granularity at which link state is sampled and communicated, and (ii) the minimum number of bits needed to encode this link state information, such that the expected power consumption within a sampling interval is minimized subject to a fixed sourcedestination goodput constraint. We formulate an optimization problem that provides a numerically computable solution to these questions, and quantitatively demonstrate that short sampling intervals incur significant overhead while long intervals fail to take advantage of the temporal correlation in link state. Additionally, we find that using a small number of bits per sample do not provide sufficient information about the network while using too many bits provide little additional information at the expense of increased overhead. Our work can be used by network operators as a tool to determine parameters such as the optimal state update frequency and the number of bits per sample.