A Fast Soft Decision Algorithm for Cooperative Spectrum Sensing

Hidden Primary User problem caused by fading and shadowing severely affects the detection rate of the cognitive radio systems with a single spectrum sensor. Cooperative Spectrum Sensing has been introduced to tackle this problem by using the spatial diversity of spectrum sensors. It is shown that the use of soft decision algorithms in fusion center has a better performance than hard decision algorithms. The problem of soft decision based on sensor measurements perfectly matches the Machine Learning paradigm. In this brief, a novel fast soft decision algorithm is proposed based on Machine Learning theory for wideband Cooperative Spectrum Sensing, which finds a decision boundary to classify the Power Spectral Density (PSD) measurement vectors of sensors into the empty and occupied channel classes. The statistical characteristics of sensors PSD samples are employed to derive a fast solution, which outperforms the SVM-linear algorithm. By solving an optimization problem a Constant False Alarm Rate algorithm is introduced and then it is improved to the Adaptive False Alarm Rate algorithm. The proposed algorithm reduces the average training time to one tenth of the SVM-linear training time, while the detection probability is the same. The hardware implementation of the proposed algorithm is described in Verilog HDL and the corresponding simulation results are presented.