| Spectrum prediction is one of the key technologies in cognitive radio (CR) systems. It enables the cognitive systems to gain the usage pattern and predict the spectrum holes by analyzing the spectrum. The prediction information can be used by cognitive devices to realize intelligent spectrum sensing and dynamic spectrum access, which result in reducing the collision rate of primary and secondary users and improving the overall performance of cognitive system, and ultimately improving spectrum utilization.The existing spectrum prediction methods are usually high in complexity, low in accuracy, long in historical sequence-demanding and unsuitable for the limited energy CR nodes and fast varying channels. In this thesis, a variable order Markov model based on context tree (CT-VMM) is proposed to improve the reliability and decrease the complexity for spectrum prediction. This method adopts a variable order Markov model to mine the history channel state information and the required number of recent historical state is changing in prediction process. The CT-VMM method is more reliable and has low complexity and demand short history sequence. Besides, due to its low complexity, it can improve the prediction accuracy under non-stationary environment through updating context tree periodically. Considering the influence of spectrum sensing errors on the accuracy of prediction, an improved algorithm based on the combination of the hidden Markov model and CT-VMM is proposed. This method adopts hidden Markov model to restore the true channel state sequence and use the true channel state sequence to construct the context tree. Using the spectral data based on queuing model, it is shown that the CT-VMM is effective in stationary and non-stationary environment respectively. Besides, using the spectral data based on queuing model and discrete time Markov model respectively, the improved algorithm is proved to be effective in the case of detecting error. |
PDF Full Text Request