Prediction Assisted Fast Spectrum Sensing

As the development of wireless applications, the spectrum resources become increasingly scarce. Under the fixed spectrum allocation policy, licensed bands cannot be used even though they are idle, which aggravates the shortage of spectrum resources. Cognitive Radio is considered to be the potential solution, with which Secondary users (SU) can use spectrum bands dynamically without interfering the communication of Primary users (PU). To ensure the acceptable disturbance to PU, SU must detect the potential bands before using them. As sensing time increases, the energy required for the detecting process is huge. Things will be even worse in the wide band sensing task. To solve this problem, a prediction assisted fast spectrum sensing scheme is proposed in this paper. Under this new scheme, we predict most of the channel states in a wireless service by mining the occupancy patterns of them. We select several channels as the detecting channels and others to be estimated channels. The states of detecting channels will be detected during the sensing process, while the states of estimated channels are predicted using the information of detecting channels.Channels are firstly grouped based on their correlations. Then, in each group, we combine temporal information of channels with channel correlations between channels to make an accurate prediction. Two grouping algorithms will be presented and compared in this paper. We conduct a continuous spectrum measurement of470MHz-806MHz for7days to collect real-world data and use this data to verify our scheme. The results show that our scheme can provide an efficient spectrum sensing scheme under high sensing accuracy.