基于FLOM和LSTM的频谱感知算法  

Spectrum sensing algorithm based on FLOM and LSTM

作　　者：赵韵雪 朱晓梅 曹秀俐 ZHAO Yun-xue;ZHU Xiao-mei;CAO Xiu-li(College of Computer Science and Technology,Nanjing Tech University,Nanjing 211816,China)

机构地区：[1]南京工业大学计算机科学与技术学院,江苏南京211816

出　　处：《计算机工程与设计》2023年第6期1656-1664,共9页Computer Engineering and Design

基　　金：国家自然科学基金项目(61501223);江苏省自然科学基金项目(KYCX21_1134)。

摘　　要：为解决现有频谱感知检测方法在Alpha噪声中性能下降的问题,提出一种基于FLOM(fractional low order moment,FLOM)和LSTM(long short-term memory,LSTM)神经网络的频谱感知算法。利用分数低阶矩在解决非高斯噪声下感知性能退化的强大能力以及长短期记忆神经网络在解决时序特性问题上的强大处理能力,设计一个频谱感知算法。不同于现有的基于能量和协方差矩阵等二阶统计量的频谱感知,利用FLOM对数据进行分数低阶预处理后,LSTM通过提取分数低阶协方差矩阵的特征进行决策。仿真结果表明,该算法比传统的频谱感知算法具有更高的检测概率。在低信噪比下,基于分数低阶矩阵感知的LSTM检测方案的检测概率比其它基于数据驱动的检测方法改善了至少15%。To solve the problem that the performance of the existing spectrum sensing detection method decreases in Alpha noise,a spectrum sensing algorithm based on FLOM(fractional low order moment)and LSTM(long short-term memory)neural network was proposed.A spectrum sensing algorithm was designed using the powerful ability of fractional low-order moments to solve the degradation of sensing performance under non-Gaussian noise and the powerful processing ability of long-term and short-term memory neural networks to solve the problem of timing characteristics.Different from the existing spectrum sensing based on second-order statistics such as energy and covariance matrix,LSTM made decisions by extracting the features of fractional low-order covariance matrix.Simulation results show that the proposed algorithm has higher detection probability than the traditional spectrum sensing algorithm.Moreover,under low SNR,the detection probability of LSTM detection s cheme based on fractional low-order matrix sensing is improved by at least 15%compared with other data-driven detection methods.

关 键 词：认知无线电 频谱感知 分数低阶矩 LSTM神经网络 非高斯噪声 分数低阶预处理 频谱感知算法 

分 类 号：TN911.6[电子电信—通信与信息系统]