一种计算有效的Nystrom特征子空间匹配主用户频谱感知新算法

A New Computationally Efficient Nystrom Feature Subspace Matching Algorithm for the Primary User Spectrum Sensing

针对核(kernel)空间下主用户频谱感知算法存在的计算任务繁重这一共性问题,提出一种低计算复杂度的Nystrom特征子空间匹配(NSM)新算法.该算法依据数据样本维的独立同分布特性随机地选择数据子集.在高维核空间下应用Nystrom近似获得主特征向量,用以分别构建主用户特征信号与次用户接收信号的Nystrom特征子空间.以此为基础计算相应的Frobenius距离,实现主用户检测.计算机仿真结果表明:与代表性的核空间下主用户频谱感知算法相比,所提算法在保证检测性能较为理想的前提下,可将相应的计算复杂度降低近66%.

Considering the high computational burden of the previous kernel spectrum sensing methods,this paper proposes a computationally more efficient Nystrom subspace matching（ NSM） algorithm. Based on the independent identically distributed observations,the subset is randomly chosen to implement the Nystrom approximation and reconstruct the related kernel features in a high-dimensional Euclidean space. Then,the related Nystrom subspaces respectively for the primary users and the secondary users are modified,and the Frobenius range between these two subspaces can be computed to determine whether the primary users exist or not. Compared to the previous kernel subspace matching methods,the novel version reduces the computational complexity by 66% while provides almost the same detection performance. Computer simulations are conducted to evaluate the performance of the proposed algorithm.