Edge-Fitting Based Energy Detection for Cognitive Radios

摘要

Two well-known drawbacks of energy detection sensing are its poor performance at low SNR and dependency on the accurate knowledge of noise power. To overcome these challenges in the context of wideband sensing, this paper proposes a spectrum sensing algorithm, inspired from edge detection techniques used in image processing. It is an improved, detailed and extended version of the gradient based method reported earlier (Koley et al. IEEE Communications Letters 19, 391-394, 2015). It is characterized by (i) histogram based noise power estimation for dynamic threshold computation (ii) preprocessing to obtain high detection probability with fewer numbers of samples and (iii) use of edge values as decision metric. Monte Carlo simulations as well as real-time wideband-sensed data captured through a Universal Software Radio Peripheral (USRP) is used to evaluate the performance of the algorithm. Mathematical formulations for the position of "SNR wall" and detection performance at different signal bandwidths is derived. Computational complexity of the edge-fitting scheme is shown to be lower than some of the wavelet approaches.