A k-generation approach to wireless fingerprinting for position estimation

Traditional approaches to pattern based positioning have proposed schemes specifically tailored for the area of interest in which they are employed. Many, when applied to a generic search environment, may not scale computationally due to either a very large tracking area, very fine granularity in the database, or both. Here, a novel framework for fingerprinting that utilizes a k-generation structure is proposed. This significantly reduces the required number of computations while maintaining comparable accuracy to the traditionally used single-generation approach. This is the key enabler that allows generalization of our framework to any search space. We validate this approach through simulation and show that the a multigeneration approach indeed provides computational gains while simultaneously maintaining accuracy. Second, we show how a natural grouping of constituent members can improve accuracy. We use graph spectral partitioning to demonstrate this effect. The end result is a framework that can be generalized to any size search space with an arbitrary level of granularity.