Single-channel blind source separation based on cyclic spectrum estimation

In this paper, we propose a new algorithm of cyclic spectrum based on time-varying ARV (Vector Auto-regressive) model that aims to deal with the disadvantages of the cycle spectrum made by the traditional periodogram such as large estimated variance and low resolution. This paper uses the algorithm to get cyclic spectrum by building a cyclostationary time-varying ARV model for communication signal. Then the algorithm transforms linear non-stationary problems into linear time-invariant via time varying basis function. And lastly the algorithm employs covariance matrix and power spectrum theory to deduce the results. In order to test and verify the algorithm, the cyclic spectrum of 2ASK(2 Amplitude Shift Keying) and BPSK (Binary Phase Shift Keying) signals are simulated in MATLAB environment. And the results are consistent with the theoretical deduction. Finally, applying cyclic spectrum information extracted by the algorithm on cyclic spectrum domain filter to achieve the single-channel blind source separation is explored.