基于离散小波变换和最大熵模糊聚类的频谱感知技术
Spectrum Sensing Based on Discrete Wavelet Transform and Maximum Entropy Fuzzy Clustering

近年来，频谱感知技术在有效分配频谱方面具有重要作用而备受关注，但传统的频谱感知算法存在受噪声影响大，检测性能差和复杂度高的问题。因此本文提出一种基于离散小波变换和最大熵模糊聚类的频谱感知算法。首先对多天线的接收信号进行等增益合并，再采用离散小波变换将信号分解来提取相应的细节信号，小波重构后的特征向量作为最大熵模糊聚类的输入进行训练得到聚类分类器，最后利用此分类器对未知信号进行检测，从而实现频谱感知。聚类算法用于频谱感知，避免了复杂的阈值计算。本文仿真对比了K-Means、模糊聚类等传统聚类算法并对其散点图可视化。结果表明，本文所提算法检测性能优于传统算法，感知准确度更高。提取信号的小波特征性能优于提取信号特征值，且降低噪声敏感对信号产生的影响，提高聚类准确性。此外，最大熵聚类算法受噪声影响更小，因此在低信噪比条件下，提升效果更突出。
In recent years, spectrum sensing technology has attracted much attention because of its important role in effectively allocating spectrum, but the traditional spectrum sensing algorithms are still challenged by the presence of heavy noise influence, poor detection performance and high complexity. Therefore, this paper proposes a spectrum sensing algorithm based on discrete wavelet transform and maximum entropy fuzzy clustering. First, the received signals of multiple antennas are equal gain merged, thereafter the discrete wavelet transform is used to decompose the signals to extract the corresponding detailed signals, and the eigenvector after wavelet reconstruction is used as input of the maximum entropy fuzzy clustering for training to obtain a clustering classifier. Finally this classifier is utilized to detect the unknown signal to achieve spectrum sensing. The clustering algorithm is used for spectrum sensing, avoiding complex threshold calculation. In this paper, the traditional clustering algorithms such as K-Means and Fuzzy Clustering were compared through simulation and their scatter plot was visualized. The results show that the detection performance of the proposed algorithm is better than that of the traditional algorithms, with higher perceptual accuracy. The performance of extracting wavelet features of the signal outperforms that of extracting signal eigenvalues, and the influence of noise sensitivity on the signal is reduced, which improves the accuracy of clustering. In addition, the maximum entropy clustering algorithm is less affected by noise, so the improvement effect is more prominent under the condition of low signal-to-noise ratio.