Research On Spectrum Prediction In Cognitive Radio Networks

Spectrum sensing, spectrum decision, spectrum sharing, and spectrum mobility are four major functions of the cognitive radio technology. Spectrum sensing is utilized to observe the spectrum occupancy status and recognize the sepctrum holes, while secondary users dynamically access the sepctrum holes through the regulation processes of spectrum decision, spectrum sharing, and spectrum mobility. To alleviate the processing delays, the response delays, and the energy consumption involved in these four functions, kinds of spectrum prediction methods have been proposed in literature. However, extising work ignores the cooperation among the secondary users, which could be explored to improve the spectrum prediction accuracy. Besides, none of the previous researches considers the challenge of improving the network performance through predicting more useful spectrum parameters. This thesis focuses on the spectrum prediction problem in cognitive radio networks and tackles the aforementioned challenges as follows:1. Motivated by the fact that the prediction accuracy of the local spectrum prediction performed by a single secondary user is quite limited, this thesis proposes a novel cooperative spectrum prediction scheme by exploring the cooperation among the secondary users, to improve the spectrum prediction accuracy. The contributions are summarized as follows:1) The process of coalition formation is modeled as a coalitional game and a cooperative prediction algorithm is presented to maximize the prediction accuracy of the secondary users. A coalition leader is selected within a coalition to collect the local spectrum prediction results of the coalition members and make the cooperative spectrum prediction decision through data fusion.2) The designed cooperative spectrum prediction scheme is extened to the scenario with multiple primary users and multiple secondary users. In the extended scheme, the secondary users are classified into different categories according to the task allocation technique proposed in literature. Cooperative spectrum prediction is performed within each category on a specific spectrum to reduce the complexity caused by the appearance of multiple primary users. 2. Traditional spectrum sensing technology requires the secondary user performs spectrum sensing once every time slot. However, many real world datasets demonstrate that the duration of a specific spectrum state usually lasts a number of time slots. Therefore the problem of determining the optimal spectrum sensing interval, i.e., the number of time slots during which no further spectrum sensing is needed, is well motivated. The contributions of the research under this topic are as follows:1) The probability density functions of the durations of the busy and idle states are derived based on a Hidden Markov Model charactering the spectrum occupancy status.2) Based on the derived probability density functions, a secondary user calculates the expectation of the missed transmission opportunities, the possible interference to the primary users, and the throughput of the secondary network under certain spectrum sensing interval. Then, the optimal spectrum sensing interval is determined by balancing the tradeoff among all these factors.3. Tranditionally, secondary users opportunistically access the spectrum bands on a non-interference basis. However, the quality of the spectrum bands may differ significantly and overlooking high-quality ones may drastically decrease the spectrum efficiency. Thus, this thesis tackles the challenges of spectrum quality prediction to enhance the efficiency of dynamic spectrum access. Three-fold contributions are made under this topic:1) The spectrum sensing process is modeled as a Non-Stationary Hidden Markov Model and the model parameters are estimated via Bayesian inference.2) The estimated parameters are explored to predict the expected duration of the spectrum states and the spectrum sensing accuracy (detection probability and false alarm probability) of the secondary user. Then, the predicted metrics are substituted into a novel spectrum quality evaluation metric for spectrum quality prediction.3) Each secondary user ranks the spectrum bands in a non-increasing order of the predicted spectrum quality and maintains an ordered spectrum band sequence. When a secondary user intends to access the spectrum, it sequentially senses and accesses the spectrum bands based on its spectrum band ranking.