Big time-frequency domain data mining for underdetermined BSS using density component analysis

Today's blind signal separation (BSS) processes are often controlled and supported by information systems. These systems record discrete time-frequency domain information about mixed signal during their executions. So, blind source separation problem (BSS) is transformed into data classification problem. In this paper, a novel Density Clustering algorithm (DC-algorithm) is proposed for frequency hopping signal under-determined blind source separation. Different from traditional methods, we formulate the separation problem as clustering problem, which is motivated by the fact that the mixed signal is sparse and the energy difference is as large as possible to satisfy cluster centers that are surrounded by neighbors with local lower density. In our method, we accomplish the underdetermined blind source separation by firstly computing the Short Time Fourier Transform (STFT) of each observation, secondly, formulating the separation problem as clustering problem. In this process, a new pair of cost functions are designed to improve the clustering. We verify the proposed method on several simulations. The experimental results demonstrate the effectiveness of the proposed method.