Robust energy detection for cognitive radio

Spectrum sensing is the fundamental technology in cognitive radio to learn the radio environment. Energy detection is one of the simplest method for spectrum sensing and has the best performance in theory for signals without special features. However, it is well-known that energy detection is vulnerable to noise uncertainty. There is still no good approach to solve the problem. On the other hand, robust hypothesis testing has been known for a long time, which is a general paradigm to deal with uncertain signal and noise. In this paper, we use this paradigm to tackle the noise uncertainty problem. It is shown that the noise uncertainty can be formed as an ϵ-contamination model. Then it is proved that the robust hypothesis testing turns to robust energy detection for independent signal samples with Gaussian distribution. Methods are found for calculating the parameters in the testing. Simulations show that the robust energy detection achieves better average performance than the energy detection on uncertain noise environment.