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.

Work supported by the Spanish Ministry of Economy and Competitiveness under the Maria de Maeztu Units of Excellence Programme (MDM-2015-0502), TEC2015-66228-P, and the European Research Council (grant agreement 694974).