The problem of parametric modeling for noisy damped sinusoidal signals from multiple channels is addressed. Utilizing the shift invariance property of the signal subspace, the number of distinct sinusoidal poles in the multiple channels is first determined. With the estimated number, the distinct frequencies and damping factors are then computed with the multi-channel weighted linear prediction method. The estimated sinusoidal poles are then matched to each channel according to the extreme value theory of distribution of random fields. Simulations are performed to show the performance advantages of the proposed multi-channel sinusoidal modeling methodology compared with existing methods.
I E E E Transactions on Signal Processing, 2013, Vol 61, Issue 15, p. 3895-3907