Spectral Analytic Methods for the Estimation of Number of Signals and Directions of Arrival

Abstract : Consider the model x(t) = As(t) + n(t), t = 1,...,N where at time t , x(t) is a p-complex vector of observations, s(t) is a q-complex vector of unobservable signals emitted from q sources and n(t) is a noise p-complex vector variable. The matrix A has a special structure with its k-th column a sub k of the form a' sub k = (1, exp(j;(omega sub 0)(Tau sub k)),..., exp(-j;(omega sub 0)(p-10 (tau sub K) where j = sr rt(-1), rk = 1/c Delta is theta k, theata sub k being the direction of arrival of the signal from the k-th source. The problem is the estimation of q (the number of sources) and Theta sub i's (the direction of arrival). The paper presents an information theoretic criterion to decide on q, the number of sources, and a spectral analytic method to estimate the tau sub i's. The proposed method is straight forward and has some advantages over the search methods proposed in the literature for the latter problem. For instance, our method works even when the signals are coherent, i.e., the covariance matrix Gamma of signal is singular, whereas the MUSIC algorithm generally used in searching for the estimates of tau sub k is not applicable. Keywords: Akaike information criterion, Direction of arrival, General information criterion, Signal processing.