We present a measurement-driven algorithm to map the large-scale channel losses observed between a cellular base station and any point in its coverage area. The algorithm is on-line, meaning that it operates on continuously arriving measurements. Its distinguishing features are the use of two kernel functions, suitably chosen for the problem at hand, and a simple technique to sparsify the dictionary of measurements retained in memory. Evaluations in campus and urban settings indicate that the proposed algorithm reduces, roughly in half, the prediction error of existing single-kernel and multikernel algorithms.
A clean separation of the DC component from the rest of the large-scale losses, via two suitable kernel functions, benefits the online reconstruction of large-scale loss maps. The accuracy increases with respect to the single-kernel version, but also with respect to a multikernel solution. This confirms that the number of kernels should be chosen carefully to match the structure of the problem, and adding unnecessarily many kernels may end up being detrimental. With two kernels, the performance approaches the baseline represented by more complex batch schemes, where, rather than a limited-size dictionary, all measurements must be stored. To reinforce the above insight it is worth mentioning that, in Section VI, the dual-kernel approach has been run with the parameter q (which, recall, indicates the number of recent measurement over which APSM projects at every step) set to q = 1, and yet this has sufficed to outperform the single-kernel result, which has been run with q = 20. This confirms the effectiveness of the DC kernel at tracking the map’s average, something that without a DC kernel requires a larger value of q and therefore higher complexity.
Slides are accessible in the following link
Please refer to IEEE Xplore
Click here for getting the bib file.
google-site-verification: google7abf782922b58f22.html