Online DOA Estimation for Noninteger Linear Antenna Arrays in Coarray Domain

We study low complexity direction of arrival (DOA) estimation in noninteger nonuniform antenna arrays with the same number as or more uncorrelated sources than sensors. We employ the maximum entropy (ME) method to solve the matrix completion problem that arises due to having an incomplete set of lags in the coarray. In order to decrease the computational complexity associated with the determinant maximization in the ME completion method, we present a projection-free online convex optimization (OCO) based on the conditional gradient method. We then frame the problem as a sequence of DOA estimation scenarios with varying directions in which a tight bound on total regret minimization is guaranteed by the employed unsupervised learning technique. We evaluate the performance using numerical examples and demonstrate that the proposed method decreases the root mean squared error (RMSE) as the iterations increase. Furthermore, our method approaches the RMSE of the offline method, exhibiting the same saturation behavior as the CRB.