Array and multichannel signal processing using higher order and cyclic statistics

Localization of multiple stationary sources or targets using sensor arrays is of primary interest in active/passive radar and sonar. In this work, higher-order statistics are employed in detection and location of non-Gaussian sources observed in (perhaps) colored Gaussian noise. The proposed methods enable one to handle greater number of sources than that possible using second-order statistics; additionally, they are tolerant to additive Gaussian noise. In the presence of multipath and reverberation, multichannel (non-)parametric modeling can be employed to identify such phenomena and provide information about the source signals. The proposed higher-order statistics based ARMA estimation schemes obviate the multi-minimum phase assumption imposed by correlation based algorithms. By exploiting the one-to-one correspondence between multivariate and cyclostationary processes computationally efficient algorithms are derived for estimating multichannel ARMA parameters via periodic ARMA identification. Higher-order cyclostationarity is a temporal property possessed by most communication signals due to modulation with the carrier. By exploiting this property, simple but high-resolution signal-selective algorithms are proposed for estimating the directions of multiple wideband and narrowband cyclostationary signals in a unified framework. The cyclic-domain methods apart from suppressing any type of stationary noise (of unknown covariance), are tolerant to Gaussian and transient interference. Moving targets/sources reflect or radiate nonstationary signals due to the time-varying delay. By expressing the delay as a polynomial, a novel method which exploits the cyclo-stationarity of the received signal is suggested for estimating the coefficients of the delay polynomial. Because of the cyclic-domain processing, this approach allows multiple signals and yields consistent and asymptotically optimal estimators even in the presence of (perhaps non-Gaussian) stationary noise. Applications of this theory for analysis of chirp detection and parameter estimation algorithms is studied. Simulation results show superior performance and demonstrate the feasibility of the proposed methods in array and multichannel signal processing.