Abstract

Due to its opportunistic spectrum sharing capability, Cognitive Radio (CR) has been proposed as a fundamental solution to alleviate the contradiction between spectrum scarcity and inefficient utilization of licensed spectrum. In CR system, secondary users (SUs) are required to do their best to minimize the interference to primary users (PUs). Therefore, a reliable spectrum prediction program to allow the SUs to predict the occurrence of future spectrum occupancy which helps to make advanced schedule is necessary. In this paper, an efficient spectrum occupancy prediction scheme based on multilayer feedforward neural network (MFNN) is proposed, which does not require a priori knowledge of underlying distributions of the observed process. Data related to spectrum occupancy was collected through real testbeds in a dense urban area with frequency bands ranging from 410 MHz to 2700 MHz. Simulation results show that our scheme can perfectly predict the occupancy of the spectrum and the prediction results in different frequency points are illustrated to validate the predictionaccuracy of the proposed scheme.

Paper Details