Research On Dynamic Spectrum Access In Cognitive Radio Based On Massive Data Mining Via Spectrum Monitoring

As wireless communications kept developing rapidly for the recent years, demand on radio spectrum resources has dramatically increased, which forced the government to attach as much importance to radio spectrum resources as that to the natural ones such as land, mine and forest. Specially, as for the industrially hot topic with inestimable marketing prospect, the Internet of Things, which will result in a massive amount of terminals and data over-sizing today's wireless inter-communication. By then, the shortage of radio spectrum resources will become a thorny bottleneck for the Internet of Things. As spectrum resources become more and more deficient, there exists close relationship between any wireless technique and spectrum resources, only the techniques that optimize spectrum resources can be popularized, otherwise they will gradually quit the arena of history. To tackle today's conflict between the limitedness of spectrum resources and low spectrum efficiency, a new technique that opportunistically utilizes the vacant spectrum becomes urgently necessary. Based on cognitive radio, the notion of dynamic spectrum access comes into being as the future direction for wireless communications in spite of facing with challenges from conception to application. By conducting a long-term and extensive spectrum monitoring campaign, we acquire massive spectrum data, with which, we have in-depth study on spectrum usage model, multi-dimensional spectrum correlation, multi-dimensional spectrum prediction and strategy optimization for dynamic spectrum access. The main contributions are as follows:(1) Carry out a long-term spectrum monitoring activity concurrently in multiple locations with careful selection for sites and testing devicesIn order to make the best to wireless spectrum resources, the primary task is to have deep analyses on regularity and usage pattern of the current wireless spectrum, which serves as the precondition for channel estimation and dynamic band selection. Most of previous works on spectrum measurement were either single-location ones, or multi-location while non-concurrent ones without careful selection for representative sites. Most of the measurement results lie in evaluation for spectrum occupancy and conclusions are more or less uniform saying that the licensed spectrum are underutilized without further analyzing and mining the spectrum data. We propose a concurrent spectrum monitoring activity at multiple locations with consideration for sites selection. Massive spectrum data is acquired by one week's non-stop monitoring, which supports the following analytical work.(2) Have in-depth study on licensed channel model and regular pattern of spectrum usage by theory of regression analysis, correlation analysis and time series analysisAfter in-depth analysis on the massive spectrum data, Channel Vacancy Duration (CVD) and Service Congestion Rate are proposed for channel statistics, temporal, spectral, and even spatial correlation are investigated, and relationship between channel state and its history information are disclosed. All these theoretically support channel modeling, spectrum analyses and dynamic decision.(3) Propose a temporal-spectral prediction algorithm with high performance on the spectrum data and relevant results are applied to in anomaly detection for radio spectrum monitoring.Prediction algorithms for spectrum usage are studied with different training sets and testing sets and several metrics, such as accuracy rate and missing rate are quantitatively analyzed. Optimal size of training set for prediction is presented. Some of the experiment results exhibit high prediction accuracy, which imply the practicality of the prediction algorithm. In addition, by leveraging the conclusion on the correlation among the spectrum measurement data, a novel scheme for anomaly detection in radio monitoring is proposed based on Mahalanobis Distance. By simulation with the spectrum measurement data, the results show that it can efficiently detect the monitored data that are different from the ''routine" model by temporal and spectral data mining, such that spectrum manager can subtly capture anomalies and make countermove in advance. It can yet be regarded as a brand-new solution for anomaly detection.(4) Based on the previous work and relevant theory of optimization, propose several optimized sensing strategy for secondary users'dynamic spectrum access in slotted modeSecondary users'(SU) channel sensing strategies based on dynamic spectrum access in slotted transmission mode are intensively studied to improve the transmission performance of secondary users while normal data transmission of primary users can be still ensured. An optimal-stopping strategy is proposed at first to improve the conventional LBT scheme. It takes time overhead on the existing sensing techniques into account and maximizes the SUs'expected transmission rate in a single timeslot. Results show that the optimal-stopping strategy outperforms the conventional LBT scheme. Then according to the analytical results from the spectrum measurement data, the notion of spectral prediction is proposed by leveraging the conclusion that there exists high spectral correlation among real wireless spectrum. With the concept of Channel Availability Vector (CAV), spectral prediction is introduced to optimization for SU's transmission performance such that an SU can adaptively make decision on channel sensing based on current state to optimize its transmission rate. Single-channel mode and multi-channel mode are considered respectively. The results show that the prediction-based strategy further outperforms the optimal-stopping strategy. Besides, best time for channel switching during SU's transmission is studied as well, and the results show that the best time for channel switching is not bound to be when maximal interference users coexist in one channel.