Detecting of Small-signal with Uncertain Frequency Based on Clustering

A improved K-mean cluster method was proposed to detect the randomly occurring small-signal with uncertain frequency,amplitude and phase in broad frequency band.The sophisticate cluster method was based on a feature parameter,which cannot embody the similarity of two or multiple feature parameters between the sample points in the cluster.The improved K-mean cluster method divided the sample points into a number of meshes on the basis of the spectral density,and drew the histogram of every mesh.By the parameters' histogram peak values,the number of cluster centers can be estimated,and these initial cluster centers were selected.According to the minimum Euclidean distance principle,the sample points in every mesh were clustered on the basis of the frequency.The sample points in the subclass have the similarity of frequency and spectral density.A comparative analysis of subclasses between the sample information and the to-be-detected information is conducted.The new subclasses are obtained which correspond to the feature values of randomly occurring small-signal.Experiment results show that randomly occurring small-signal with uncertain frequency can be recognized in a complicated environment.