Detection of partially correlated signals in clutter using a multichannel model-based approach

The author considers the Gaussian multichannel binary detection problem in which the signal and nonwhite clutter noise are Gaussian vector processes with unknown statistics. A generalized likelihood ratio using multichannel innovation processes is implemented via a model-based approach where the signal and clutter are assumed to be characterized by autoregressive vector processes with arbitrary temporal and cross-channel correlation. The innovations processes are obtained through linear estimation using multichannel parameter estimates. Detection performance is considered as the estimates approach steady state with increasing data block sample sizes. Results for two-channel signal and clutter noise vectors containing various temporal and cross-channel correlation are obtained using a Monte Carlo procedure. In the transient state (estimation with limited data), the detection results are considered as a function of the data sample window sizes used in the parameter estimation procedure. Furthermore, it is noted that the detection performance in the transient state is related to that of the estimator, which in turn has its own dependence upon process correlation