Spectrum map retrieval using cognitive radio network tomography

Cognitive radio network tomography obtains and infers information for networking cognitive radios by various way over simple measurements. Beyond information of locating the unused spectrum or detecting primary users' signal, advanced network functions such as multi-hop cooperative routing can be easily facilitated by radio resource distribution with location information, known as spectrum map. However, it is infeasible and inefficient to place a large number of cooperative sensors persistently constructing the spectrum map in all frequency bands, such as the appearance of sensing-transmission transition. Borrowing the concept of cooperative spectrum sensing to exploit the spatial diversity, cognitive radio network tomography inferring from cooperative sensors exhibits its significance in making up the unavailability of direct measurement. In this paper, we present that the inherent correlation in sensors' historic observations can provide critical information to overcome the deficiency of conventional cooperative sensing. We propose a two-step cooperative learning procedure to attain the spatial correlations. Moreover, subset selection determining the dominant sensors is used to lower information exchange among sensors and optimize learning complexity. To infer and consummate the spectrum map in a highly dynamic environment, we apply random set theory instead of conventional inference methods. Since we can retrieve the spectrum map in the locations that sensors becomes inactive, our proposed scheme can reduce the number of required sensors while providing complete spectrum map information to boost advanced network functions, which is critical to dynamic and heterogeneous network environments such as cognitive radio networks or machine-to-machine networks.