Robust Spectrum Sensing In Cognitive Radio Networks

The harmony coexistence of cognitive radio systems with licensed system requiresthe secondary users the capability of interference-awareness, i.e., knowing which spectrumbands are occupied by primary users, i.e., the legacy users. Spectrum sensing thus is a keyenabling module, which usually models the sensing process as a hypothesis testing problemassuming known probability distribution function (pdf) of the received signal strength underdifferent hypotheses. However, an unrealistic assumption regarding the pdf easily leads tounreliable detection probability.In this paper, we study the sensing performance considering the uncertainty of pdf inhypothesis testing, i.e., the actual distribution function of the received signal strength is notknown. We formulate the optimal detection in a robust optimization problem, by definingthe uncertainty set for pdf using its first and second moments.Main work contains two parts:In the first part, performance analysis for single user detection is conducted. We presenta scenario approximation approach to determine the decision threshold, and then investigatethe lower and upper bounds of the detection probability subject to distribution uncertainty.Furthermore, we demonstrate the attainability of these bounds by extremal distributions.In the second part, single user detection is extended to multi-user cooperative sensing.We propose a decision threshold adapting algorithm that improves the worst-case detectionperformance while maintaining the same false alarm probability as prescribed. Thealgorithm requires solving a series of semi-definite programs, which are implemented bySeDuMi, and validated through simulation results.