5G技术的广泛应用、通信业务不断的发展以及物联网设备呈现爆炸式增长，导致频谱资源严重匮乏，无法满足目前对频谱资源的需求。频谱感知是认知无线电执行所有功能的基础，在认知物联网中，智能设备可以动态调整频谱使用策略以提高资源利用率。针对现有的物联网中频谱感知方法存在着在低信噪比检测性能差、协作软融合感知传输的数据量大以及硬融合检测性能不足等问题。开展的主要工作和创新点总结如下：

（1）针对认知物联网传统频谱感知方法在低信噪比环境下的频谱检测性能差以及协作频谱感知中传统硬融合感知判断误差大、软融合处理复杂等问题，本文提出基于深度学习的协作频谱感知算法，算法可分为三个模块：数据预处理、单节点频谱感知和多节点频谱感知。数据预处理模块先将次用户接收到的信号转成二维矩阵，使用归一化灰度化操作生成灰度图加快特征提取及模型的收敛速度，并减少模型学习量，降低了模型复杂度。单节点频谱感知将改进ResNeXt网络模块作为主干网络，利用其分组卷积特性和模块化设计特性，快速准确的提取频谱灰度图像特征，提高单节点在低信噪比下的检测性能。多节点频谱感知用SCN神经网络代替传统的融合中心，融合各个次用户单节点感知得到的评分向量矩阵获得最优全局判决感知结果。仿真结果显示，在信噪比为-19 dB环境下本文算法的检测性能至少提升10%，在低信噪比环境下能实现低虚警概率和高检测概率，且解决了传统硬融合误差大，软融合处理数据量大的问题。

（2）针对协作频谱感知中次用户数量增加造成的能量消耗增加以及接收信息量少的次用户参与协作频谱感知降低协作检测性能等问题，本文提出基于可靠性原则的协作频谱感知节点选择算法，同时考虑了在能量消耗和感知性能双重条件下选择感知节点，并建立节点选择系统，及时剔除可靠性下降的感知节点并筛选出可靠度高的节点参与频谱感知。通过融合中心与外界环境的信息交流得到各个次用户的性能反馈和能耗反馈，对节点进行实时评估，动态选择高可靠性节点进行协作感知，保持感知系统的高可靠性平衡。仿真结果显示，感知性能在虚警概率为0.1时检测概率高达0.99，该算法能量消耗仅为传统方法的25%。

The rapid proliferation of 5G technology, the continuous evolution of communication services, and the exponential growth of IoT devices have resulted in a severe shortage of spectrum resources, falling short of meeting current demand. Spectrum sensing serves as the cornerstone for cognitive radio to execute its functions. In the context of cognitive IoT, smart devices can dynamically adapt their spectrum utilization strategies to enhance resource efficiency. However, existing spectrum sensing methods in IoT face several challenges, including poor detection performance under low signal-to-noise ratio (SNR) conditions, excessive data transmission requirements in collaborative soft fusion sensing, and inadequate detection performance in hard fusion sensing. The primary contributions and innovations of this work are summarized as follows:

(1) To address the low performance of cognitive IoT for spectrum sensing at low Signal to Noise Ratio as well as the problems of large judgment error of traditional hard fusion methods and complex soft fusion processing in collaborative spectrum sensing. In this thesis, a collaborative spectrum sensing algorithm based on deep learning is proposed, The algorithm can be divided into three modules: data preprocessing, single-node spectrum sensing and multi-node spectrum sensing. The data preprocessing module first converts the signal received by the sub-user into a two-dimensional matrix; uses the normalization operation to reduce the gradient difference between the two-dimensional matrix data to accelerate the convergence speed of the feature extraction model; and finally utilizes the grayscaling operation to reduce the amount of model learning and reduce the complexity of the model. Single-node spectrum spectrum sensing uses the improved ResNeXt network as the backbone network to extract single-node spectrum grayscale image features quickly and accurately at a deep level using its group convolutional property and modular design feature. Multi-node spectrum sensing replaces the traditional fusion center with SCN neural network, which fuses the scoring vector matrices obtained from the single-node sensing results of each sub-user to obtain the optimal global verdict sensing results. Simulation results demonstrate that the proposed algorithm achieves a detection performance improvement of at least 10% in environments with a signal-to-noise ratio (SNR) of -19 dB. It effectively ensures low false alarm probability and high detection probability under low SNR conditions, addressing the significant error associated with traditional hard fusion methods and the extensive data processing requirements of soft fusion approaches.

(2) Aiming at the problems of increased energy consumption caused by the increase in the number of sub-users in collaborative spectrum sensing and the degradation of collaborative sensing performance caused by the participation of sub-users who receive less information in collaborative spectrum sensing. a reinforcement learning-based node selection algorithm for collaborative spectrum sensing is proposed in this thesis, which evaluates the reliability of each sub-user and filters out the nodes with high reliability to participate in collaborative spectrum sensing. Mainly through the fusion center and the external environment of continuous information interaction to obtain the performance of each sub-user feedback and energy feedback, real-time evaluation of nodes, dynamic selection of high reliability nodes to participate in collaborative perception, to maintain a dynamic balance of high-reliability perception. The experimental results show that the energy consumption is as low as 4 energy units while the detection probability is as high as 0.99 when the false alarm probability is 0.1, Which consumes only 25% of the energy of conventional methods.

[1] Syed S N, Lazaridis P I, Khan F A, et al. Deep Neural Networks for Spectrum Sensing: A Review[J]. IEEE Access, 2023, 11: 89591-89615.

[2] Muzaffar M U, Sharqi R. A review of spectrum sensing in modern cognitive radio networks[J]. Telecommunication Systems, 2024, 85(2): 347-363.

[3] Ivanov A, Mihovska A, Poulkov V, et al. Hybrid accuracy-time trade-off solution for spectrum sensing in cognitive radio networks[J]. International Journal of Mobile Network Design and Innovation, 2019, 9(1): 1-13.

[4] FCC Notice of Proposed Rulemaking FCC 04-113[R], Federal Communications Commission, May 2005.

[5] Mitola J, Maguire G Q. Cognitive Radio: Making Software Radios More Personal[J]. IEEE Personal Communications, 1999, 6(4): 13-18.

[6] Haykin S. Cognitive Radio: Brain-empowered Wireless Communications[J]. IEEE Journal on Selected Areas in Communications Dimensional Imaging, 2005, 23(2): 201-220.

[7] Abbasi M, Shahraki A, Piran M J, et al. Deep Reinforcement Learning for QoS provisioning at the MAC layer: A Survey[J]. Engineering Applications of Artificial Intelligence, 2021, 102: 104234.

[8] López-Benítez M, Casadevall F. Improved energy detection spectrum sensing for cognitive radio[J]. IET communications, 2012, 6(8): 785-796.

[9] Zhu Y, Liu J, Feng Z, et al. Sensing performance of efficient cyclostationary detector with multiple antennas in multipath fading and lognormal shadowing environments[J]. Journal of Communications and Networks, 2014, 16(2): 162-171.

[10] Yucek T, Arslan H. A survey of spectrum sensing algorithms for cognitive radio applications[J]. IEEE communications surveys & tutorials, 2009, 11(1): 116-130.

[11] Liu C, Wang J, Liu X M, et al. Maximum Eigenvalue-based Goodness-of-fit Detection for Spectrum Sensing in Cognitive Radio[J]. IEEE Transactions on Vehicular Technology, 2019, 68(8): 7747-7760.

[12] Mallick S, Das S, Ray A K. A modified three-event energy detection scheme using decision threshold optimization for sensing performance improvement in a cognitive radio system[J]. Wireless Networks, 2023, 29(6): 2747-2758.

[13] Oh H, Nam H. Energy Detection Scheme in the Presence of Burst Signals[J]. IEEE Signal Processing Letters, 2019, 26(4): 582-586.

[14] Mahajan K, Garg U. An enhancement to the existing cyclostationary feature detection in CRN[J]. Multimedia Tools and Applications, 2022, 81(26): 37087-37099.

[15] Narieda S, Cho D, Ogasawara H, et al. Derivation of sensing features for maximum cyclic autocorrelation selection based signal detection[C]//2019 IEEE 90th Vehicular Technology Conference (VTC2019-Fall). IEEE, 2019: 1-5.

[16] Tang L, Zhao L, Jiang Y. An SVM-based feature detection scheme for spatial spectrum sensing[J]. IEEE Communications Letters, 2023, 27(8): 2132-2136.

[17] Tian J, Cheng P, Chen Z, et al. A machine learning-enabled spectrum sensing method for OFDM systems[J]. IEEE Transactions on Vehicular Technology, 2019, 68(11): 11374-11378.

[18] Patel D K, Lopez-Benitez M, Soni B, et al. Artificial neural network design for improved spectrum sensing in cognitive radio[J]. Wireless Networks, 2020, 26: 6155-6174.

[19] Chen Z B, Xu Y Q, Wang H B, et al. Deep STFT-CNN for Spectrum Sensing in Cognitive Radio[J]. IEEE Communications Letters, 2021, 25(3): 864-868.

[20] Liu C, Wang J, Liu X, et al. Deep CM-CNN for spectrum sensing in cognitive radio[J]. IEEE Journal on Selected Areas in Communications, 2019, 37(10): 2306-2321.

[21] 张孟伯, 王伦文, 冯彦卿. 基于卷积神经网络的OFDM频谱感知方法[J]. 系统工程与电子技术, 2019, 41(1): 178-186.

[22] Bouzegag Y, Teguig D, Maali A. Efficient Quantized Soft Decision for Cooperative Sensing System in Cognitive Radio[C]//2022 19th International Multi-Conference on Systems, Signals & Devices (SSD). IEEE, 2022: 2078-2083.

[23] Banerjee A, Maity S P. Jamming in eavesdropping on throughput maximization in green cognitive radio networks[J]. IEEE Transactions on Mobile Computing, 2021, 22(1): 299-310.

[24] Al-Dulaimi O M K, Vlâdeanu C, Martia A, et al. Cognitive radio using spectrum sensing by cooperative two secondary users[C]//IOP conference series: Materials science and engineering. IOP Publishing, 2021, 1094(1): 012031.

[25] Obite F, Usman A D, Okafor E. An overview of deep reinforcement learning for spectrum sensing in cognitive radio networks[J]. Digital Signal Processing, 2021, 113: 103014.

[26] 石新, 刘顺兰. 基于随机矩阵和能量双联合的协作频谱感知[J]. 杭州电子科技大学学报(自然科学版), 2021, 41(04): 1-6.

[27] 许炜阳, 李有均, 徐宏乾, 等. 基于随机矩阵非渐近谱理论的协作频谱感知算法研究[J]. 电子与信息学报, 2018, 40(1): 123-129.

[28] Zhao W, Li H, Jin M, et al. Eigenvalues-based universal spectrum sensing algorithm in cognitive radio networks[J]. IEEE Systems Journal, 2020, 15(3): 3391-3402.

[29] Shi Z, Gao W, Zhang S, et al. Machine learning-enabled cooperative spectrum sensing for non-orthogonal multiple access[J]. IEEE Transactions on Wireless Communications, 2020, 19(9): 5692-5702.

[30] Ghazizadeh E, Abbasi‐moghadam D, Nezamabadi‐pour H. An enhanced two‐phase SVM algorithm for cooperative spectrum sensing in cognitive radio networks[J]. International Journal of Communication Systems, 2019, 32(2): e3856.

[31] Tang L, Zhao L, Jiang Y. An SVM-based feature detection scheme for spatial spectrum sensing[J]. IEEE Communications Letters, 2023, 27(8): 2132-2136.

[32] Xie J, Fang J, Liu C, et al. Deep learning-based spectrum sensing in cognitive radio: A CNN-LSTM approach[J]. IEEE Communications Letters, 2020, 24(10): 2196-2200.

[33] 盖建新, 薛宪峰, 吴静谊, 等. 基于深度卷积神经网络的协作频谱感知方法[J]. 电子与信息学报, 2021, 43(10): 2911-2919.

[34] Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[J]. arXiv preprint arXiv: 1409. 1556, 2014.

[35] Krizhevsky A, Sutskever I, Hinton G E. Imagenet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90.

[36] Szegedy C, Liu W, Jia Y, et al. Going deeper with convolutions[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 1-9.

[37] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778.

[38] Huang G, Liu Z, Van Der Maaten L, et al. Densely connected convolutional networks[C]// Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 4700-4708.

[39] Dong S, Wang P, Abbas K. A survey on deep learning and its applications[J]. Computer Science Review, 2021, 40: 100379.

[40] Grissa M, Yavuz A, Hamdaoui B. Location privacy in cognitive radios with multi-server private information retrieval[J]. IEEE Transactions on Cognitive Communications and Networking, 2019, 5(4): 949-962.

[41] Guo H, Jiang W, Luo W. Linear soft combination for cooperative spectrum sensing in cognitive radio networks[J]. IEEE Communications Letters, 2017, 21(7): 1573-1576.

[42] Basha S H S, Dubey S R, Pulabaigari V, et al. Impact of fully connected layers on performance of convolutional neural networks for image classification[J]. Neurocomputing, 2020, 378: 112-119.

[43] Li Z, Liu F, Yang W, et al. A survey of convolutional neural networks: analysis, applications, and prospects[J]. IEEE transactions on neural networks and learning systems, 2021, 33(12): 6999-7019.

[44] Kiranyaz S, Avci O, Abdeljaber O, et al. 1D convolutional neural networks and applications: A survey[J]. Mechanical systems and signal processing, 2021, 151: 107398.

[45] Sony S, Dunphy K, Sadhu A, et al. A systematic review of convolutional neural network-based structural condition assessment techniques[J]. Engineering Structures, 2021, 226: 111347.

[46] 刘俊奇, 涂文轩, 祝恩. 图卷积神经网络综述[J]. 计算机工程与科学, 2023, 45(08): 1472-1481.

[47] 高玉鹏, 闫伟红, 潘新. 基于卷积神经网络与注意力机制的高光谱图像分类[J]. 光电子·激光, 2024, 35(05): 483-489.

[48] Awe O P, Deligiannis A, Lambotharan S. Spatio-temporal spectrum sensing in cognitive radio networks using beamformer-aided SVM algorithms[J]. IEEE Access, 2018, 6: 25377-25388.

[49] Bao J, Nie J, Liu C, et al. Improved blind spectrum sensing by covariance matrix Cholesky decomposition and RBF-SVM decision classification at low SNRs[J]. IEEE Access, 2019, 7: 97117-97129.

[50] 陈思吉, 王欣, 申滨. 一种基于支持向量机的认知无线电频谱感知方案[J]. 重庆邮电大学学报(自然科学版), 2019, 31(03): 313-322.

[51] Liu C, Wang J, Liu X M, et al. Deep CM-CNN for Spectrum Sensing in Cognitive Radio[J]. IEEE Journal on Selected Areas in Communications, 2019, 37(10): 2306-2321.

[52] Lee W, Kim M, Cho D H. Deep cooperative sensing: Cooperative spectrum sensing based on convolutional neural networks[J]. IEEE Transactions on Vehicular Technology, 2019, 68(3): 3005-3009.

[53] Lee W, Cho D H. Enhanced spectrum sensing scheme in cognitive radio systems with MIMO antennae[J]. IEEE transactions on vehicular technology, 2011, 60(3): 1072-1085.

[54] Li Z, Wu W, Liu X, et al. Improved cooperative spectrum sensing model based on machine learning for cognitive radio networks[J]. IET Communications, 2018, 12(19): 2485-2492.

[55] Gao J, Yi X, Zhong C, et al. Deep learning for spectrum sensing[J]. IEEE Wireless Communications Letters, 2019, 8(6): 1727-1730.

[56] 马永涛, 王煜东. 基于概率向量和机器学习的协作频谱感知[J]. 传感技术学报, 2019, 32(12): 1809-1815.

[57] 鲁华超, 赵知劲, 尚俊娜,等. 利用卷积神经网络和协方差的协作频谱感知算法[J]. 信号处理, 2019, 35(10): 1700-1707.

[58] Lee H, Noda K, Mizuno Y, et al. Distributed temperature sensing based on slope‐assisted Brillouin optical correlation‐domain reflectometry with over 10 km measurement range[J]. Electronics Letters, 2019, 55(5): 276-278.