基于k最近邻回归的频谱占用度预测

Spectrum Occupancy Prediction Based on k-Nearest Neighbor Regression

认知无线电技术可以在授权用户和非授权用户间进行频谱分配,预测模型的建立可帮助非授权用户推断频谱空洞是否可用,不仅能提升频谱利用率而且还能降低冲突率。采用理论分析、监测实验、数学建模、数据实证等方法,对频谱占用度建模理论进行了研究。针对频谱的可预测性问题,通过对数据集的分析,使用k最近邻(k NN)回归模型预测频谱的信道-场强值。基于观测数据呈现出的周期性,提出了一种针对周期性数据进行优化的k NN模型,并用其进行预测。比较了原始k NN回归模型和优化后的周期性k NN模型在测试数据上的预测精度,结果表明优化后的模型比原始的k NN模型有着更好的预测精度。

Cognitive radio technology can conduct spectrum allocation between the authorized users andsecondary users. The establishment of predication model can help secondary users infer whether the spectrumhole is available, which can both improve spectral efficiency and reduce collision rate. By means oftheoretical analysis, experiment monitoring, mathematical modeling and data demonstration, spectrum occupationmodeling theory is researched. For the predictable problems of spectrum, through the analysis of datagroup,k-Nearest Neighbour （ kNN） regression model is used to predict the channel-field value of spectrum.At the same time, based on the periodicity shown by the observation data, a kNN model is proposed tooptimize periodical data and offers predication. Then the predication accuracy is compared in test data of originalkNN regression model and optimized periodical kNN. The result shows the optimized model is of betterpredication accuracy than the original kNN model.