| 摘要: |
| 无线频谱地图是实现复杂电磁环境可视化表征的有效方法。针对空间稀疏采样及多种噪声干扰下的二维平面频谱地图重构，提出了一种基于自适应阈值与块项张量分解(Adaptive Threshold Block-Term Tensor Decomposition，ATBTD)的重构算法。该算法在F范数损失函数中引入自适应阈值机制以抑制脉冲噪声，并在块连续上界极小化框架(Block Successive Upper Bound Minimization，BSUM)下结合投影梯度法(Projected Gradient，PG)实现对因子矩阵的高效求解。为降低计算复杂度，进一步提出快速自适应阈值块项张量分解（Fast Adaptive Threshold Block-Term Tensor Decomposition，F-ATBTD）算法替代传统奇异值分解，并引入加速梯度法提升收敛效率。与基准算法相比，F-ATBTD在保持重构精度的同时显著减少了计算开销。仿真结果表明，在低信噪比条件下，ATBTD和F-ATBTD在高斯噪声场景中的归一化均方误差较基准方法降低约15%，在混合噪声场景中降低约30%。 |
| 关键词: 频谱地图重构 自适应阈值 块项张量分解 复杂噪声场景 噪声鲁棒性 |
| DOI：10.20079/j.issn.1001-893x.250908003 |
|
| 基金项目:湖南省自然科学基金项目（2024JJ8024） |
|
| Robust Wireless Spectrum Map Reconstruction via Local Low-rank Tensor Completion |
| SHEN Bin,YANG Yuyou,ZHANG Min,DENG Xueshuang |
| (1.School of Communications and Information Engineering，Chongqing University of Posts and Telecommunications，Chongqing 400065，China;2.College of Information and Communication Engineering，Hunan Post and Telecommunication College，Changsha 410015，China) |
| Abstract: |
| Wireless spectrum maps serve as an effective method for the visual characterization of complex electromagnetic environments.To address the reconstruction of two-dimensional planar spectrum maps under spatially sparse sampling and various noise interferences，a reconstruction algorithm based on adaptive threshold block-term tensor decomposition (ATBTD) is proposed.The algorithm introduces an adaptive thresholding mechanism into the Frobenius norm loss function to suppress impulsive noise，and efficiently solves the factor matrices under the block successive upper-bound minimization (BSUM) framework combined with the projected gradient (PG) method.Furthermore，to reduce computational complexity，a fast adaptive threshold block-term tensor decomposition (F-ATBTD) algorithm is proposed to replace traditional singular value decomposition，and an accelerated gradient method is incorporated to enhance convergence efficiency.Compared with baseline algorithms，F-ATBTD significantly reduces computational overhead while maintaining reconstruction accuracy.Simulation results show that under low signal-to-noise ratio conditions，ATBTD and F-ATBTD achieve approximately 15% lower normalized mean square error than baseline methods in Gaussian noise scenarios，and about 30% lower in mixed-noise scenarios. |
| Key words: spectrum map reconstruction adaptive thresholding block-term tensor decomposition complex noise environments noise robustness |
《电讯技术》编辑部 