Maximum likelihood training of adaptive beamformers in distributed interference

Adaptive beamforming based on sample matrix inversion requires the availability of a secondary data set for "training" (i.e., estimating the sample covariance matrix). In nonstationary interference, adaptive beamformer performance depends greatly on how this training data is selected. Previously, the author and his colleague proposed using maximum likelihood estimation (MLE) to select training data in nonstationary environments and formulated a computationally efficient implementation ("power selected training") for a special case of rank one nonstationary interference (see Rabideau, D.J. and Steinhardt, A.O., 1996, 1999). This paper specializes our basic MLE approach to the case of distributed nonstationary interference. Computationally efficient training strategies are formulated and evaluated using recorded data from a US Navy ADS-18s antenna array. We also describe the relationship between our basic MLE approach and a subsequently published method, "NHD" (see Melvin, W.L. and Wicks, M.C., 1997).