To improve probability of detection and reduce training time,this paper proposes a method of support vector machine (SVM) spectrum sensing based on data preprocessing with a log function.A minimum size of training set is selected,which is applicable with good performance in spectrum sensing.The sample sets are generated with laboratory instruments.The obtained sample sets are pre-processed with a log function to increase the mean difference between sample sets with and without primary users (PU).Experimental results show that,after pre-processing,performance of spectrum sensing is significantly improved under low SNR conditions with detection accuracy 90% or better.%为提高频谱感知灵敏度和准确率,减少训练时间,提出一种采用对数函数预处理的支持向量机频谱感知方法.通过实验平台采集出样本集,在保证频谱感知性能前提下,选取一个尺寸最小且具有适用性的训练样本集,利用对数函数对样本集进行预处理,增大主用户信号存在与不存在时样本的平均值之差.实验表明,经对数函数预处理的样本集送入支持向量机进行训练和测试,其频谱感知性能在低信噪比下有明显提高,检测率达到90%以上.
展开▼