| 引用本文： |
-
谢松霖,顾志豪,王全全,等.一种基于信誉值多分类与在线学习的安全频谱感知策略[J].电讯技术，2026，66（2）： - . [点击复制]
- XIE Songlina,b,GU Zhihaob,et al.A Secure Spectrum Sensing Strategy Based on Reputation Value Multi-class Classification and Online Learning[J].，2026，66（2）： - . [点击复制]
|
|
| 摘要: |
| 针对频谱感知数据伪造（Spectrum Sensing Data Falsification，SSDF）攻击导致协作频谱感知性能下降的问题，提出了一种基于信誉值多分类与在线学习的安全频谱感知策略。首先提取次用户(Secondary User，SU)上传信号的平均能量差与平均幅度差，训练恶意次用户(Malicious Secondary User，MSU)识别模型。再通过MSU识别模型的历史判别结果，计算各SU的信誉值，然后依据信誉值将SU分为正常次用户与MSU两种，并将MSU进一步分为偶尔、时常与频繁攻击3类，针对不同种类采取相应的处理提取其上传信号的特征值，更为充分地训练主用户(Primary User，PU)状态判决模型，提升抗SSDF攻击的能力。最后，使用在线学习完成对各SU信誉值、MSU识别模型、PU状态判决模型的实时更新，并限制训练集的最大长度，来及时调整MSU的种类，提升对隐蔽性更强的潜伏性攻击的防御能力。仿真结果表明，在面对潜伏性攻击时，该方法的检测概率最高可达8443%，具有更好的抗SSDF攻击能力。 |
| 关键词: 协作频谱感知 频谱感知数据伪造（SSDF） 信誉值 在线学习 |
| DOI：10.20079/j.issn.1001-893x.241107004 |
|
| 基金项目:国家自然科学基金资助项目(62371245)；江苏省教育科学规划重点课题（B/2023/01/120） |
|
| A Secure Spectrum Sensing Strategy Based on Reputation Value Multi-class Classification and Online Learning |
| XIE Songlina,b,GU Zhihaob,WANG Quanquanb,HU Haifengb |
| (1a.Portland Institute;1b.School of Communication and Information Engineering，Nanjing University of Posts and Telecommunications，Nanjing 210003，China;2.State Radio Monitoring Center，Beijing 100037，China) |
| Abstract: |
| For the problem of spectrum sensing data falsification(SSDF) attack leading to the degradation of cooperative spectrum sensing performance，a secure spectrum sensing strategy based on multi-classification of reputation values and online learning is proposed.Firstly，the average energy difference value and the average amplitude difference value of the uploaded signal of secondary user(SU) are extracted to train the malicious secondary user(MSU) identification model.Then，the reputation value of each SU is calculated through the judgment results of the historical MSU identification model.And then SU is divided into normal secondary user and MSU according to the reputation value，and MSU is further divided into occasional，regular and frequent attacks.Corresponding processing methods are adopted for different types to extract the features of their uploaded signals，so as to train the primary user(PU) state decision model more fully and improve the ability to resist SSDF attacks.Finally，online learning is used to update the reputation value，MSU identification model and PU status decision model of each SU in real time，and the maximum length of the training set is limited to adjust the type of MSU in time，so as to improve the defense ability of the latent attack with stronger concealment.Simulation results demonstrate that the detection probability of the proposed method under latent attack is up to 84.43%，which shows that it has a better ability against SSDF attack than existing methods. |
| Key words: cooperative spectrum sensing spectrum sensing data falsification（SSDF） reputation value online learning |
《电讯技术》编辑部 