An Improved Underdetermined Blind Source Separation of Frequency Hopping Signals Based on Subspace Projection

The earlier separation methods for frequency hopping signals constrain that the number of sources must be less than that of sensors. In many practical applications, the above condition is not satisfied, so it is necessary to deal with cases when sensors are less. This paper considers the underdetermined blind separation of frequency hopping signals. The conventional algorithms for underdetermined blind separation of frequency hopping signals are assuming that the signals are sparse in time frequency domain. The mixtures need to satisfy the assumption that the active sources which contribute their energy at a point in time frequency domain are no more than that of sensors. This paper proposes an improved subspace projection method to address underdetermined blind separation problem. Firstly, it shows the mixtures model of frequency hopping signals in time domain, and also explains the mixing matrix structure. Secondly, in order to estimate the active sources at each time frequency point, Short Time Fourier Transform is exploited. Then, the effect of redundant signals is considered at each time frequency point so that a threshold value can be set. Thirdly, the number of active frequency hopping signals at each time frequency point can be estimated based on the improved subspace projection method. Simultaneously, the proposed method can identify which signals exist at each time frequency point. Finally, the time frequency representation of each frequency hopping signals can be achieved. The sources of frequency hopping signals can be obtained by Inverse Short Time Fourier Transformation. The numerical simulation results demonstrate the validity and high performance of the proposed algorithm compared to existing ones in underdetermined case.