Quasi-norm Kernel-based Emitter Localization

The indoor wireless signal environment is characterized by dynamic, non-deterministic signal propagation under the practical assumption of imperfect knowledge of the environment. In this paper, heterogeneous data measurements from a multi-sensor system are kernelized using standard and quasi-norms. The resulting kernel matrix is then used as input to a set of regression-based emitter localization algorithms. It is shown the use of the quasi-p-norm, p < 1, as a similarity measure, results in improved localization performance over the common p-norm such as L
2
. Additionally, the use of multiple kernels, one for each data measurement type — time-difference of arrival, received signal strength, and angle of arrival — further improved performance in comparison to a single kernel.