Bayesian information theoretic criterion for detection of number of signals in array processing

Abstract

The generalized singular value decomposition approach is taken in AR spectral estimation in order to mitigate noise effects on the estimated spectrum. A Baysian information theoretic criterion is derived to attain the optimal truncation of smaller generalized singular values and then the separation of signal subspace and noise subspace. The effectiveness of the proposed criterion is investigated in array signal processing in cases of narrowband and broadband sources in comparison with the other AIC and the MDL criteria.<>