A traditional cooperative spectrum sensing system based on the energy features of random matrix theory does not function well when the signal-to-noise ratio is low in the Gaussian noise channel. In this paper, the poor performance of this system is improved. A novel blind cooperative spectrum sensing scheme based on new denoised mixed-features exploiting the K-Medoids algorithm has been developed. In the first place, a new mathematical formula is proposed to extract denoised mixed-features and decrease the amount of noise in the received signals of each secondary user. This formula is expressed as the square root of the sum of the squares of the third and fourth-highest order moments of skewness ($S_k$) and access kurtosis ($K_u$-3). In the fusion center, two new denoised mixed-features methods are proposed. The first method is based on calculating the Average values of ${\mathrm{S}}_k$ and ${\mathit{K}}_u$-3 of the denoised mixed features vectors, we named it the ASKU3 method. The second method is based on calculating the Maximum values of ${\mathrm{S}}_k$ and K${}_u$-3 of the denoised mixed features vectors, we named it the MSKU3 method. In comparison to the existing energy feature-based methods, the denoised mixed-features-based on highest-order moments avoid the computational complexity associated with threshold estimation. Furthermore, they mitigate the effect of noise uncertainty and improve sensing performance by reducing the overlap between clusters of the K-Medoids algorithm. Then, the K-Medoids method makes use of the newly formed methods in order to train the classifier effectively depending on whether the primary user is present or not. Finally, several energy-feature-based methods classified by the clustering algorithm are compared in this study. Moreover, the theoretical and practical performances of the denoised mixed-features methods, ASKU3 and MSKU3, are very well matched. The simulation findings show that the proposed methods have improved the effectiveness of sensing at low signal-to-noise ratio in terms of the probability of detection.

传统的基于随机矩阵理论能量特征的协同频谱感知系统在高斯噪声信道信噪比较低的情况下效果不佳。在本文中，该系统的较差性能得到了改善。已经开发了一种基于利用 K-Medoids 算法的新的去噪混合特征的新型盲协同频谱感知方案。首先，提出了一种新的数学公式来提取去噪混合特征并减少每个次用户接收信号中的噪声量。该公式表示为三阶和四阶偏斜矩的平方和的平方根（${\mathrm{小号}}_{ķ}$) 和访问峰度 (${ķ}_{你}$-3)。在融合中心，提出了两种新的去噪混合特征方法。第一种方法是基于计算平均值${\mathrm{小号}}_{ķ}$和${\mathit{ķ}}_{你}$-3的去噪混合特征向量，我们将其命名为ASKU3方法。第二种方法是基于计算最大值${\mathrm{小号}}_{ķ}$和 K${}_{你}$-3的去噪混合特征向量，我们将其命名为MSKU3方法。与现有的基于能量特征的方法相比，基于最高阶矩的去噪混合特征避免了与阈值估计相关的计算复杂性。此外，它们通过减少 K-Medoids 算法的集群之间的重叠来减轻噪声不确定性的影响并提高传感性能。然后，K-Medoids 方法利用新形成的方法，根据主用户是否存在来有效地训练分类器。最后，在本研究中比较了几种由聚类算法分类的基于能量特征的方法。此外，去噪混合特征方法的理论和实践性能，ASKU3和MSKU3，非常匹配。仿真结果表明，所提出的方法在检测概率方面提高了低信噪比传感的有效性。