Bregman divergence based sensor selections for spectrum sensing

Sensor selection is to pick out an appropriate subset of active sensors for reliable collaborative sensing. Naturally, the selected sensors should be as uncorrelated as possible to have more independent sensing outputs for information fusion. In this paper, various uncorrelation metrics are unified by the concept of Bregman divergence. The sensor selections are then systematically formulated as NP-hard integer programs. Unlike commonly used exhaustive enumeration, heuristic searches or simple relaxation of discrete constraints with inherent drawbacks, this paper recasts them into a continuous d.c. (difference of two convex functions) program under convex constraints. Accordingly, an efficient iterative optimization procedure is tailored for locating the optimal solution. Simulation results show its superior performances in comparison with other existing sensor selections.