Subspace recombination of multiple partially adaptive beamformer outputs

In many adaptive radar applications the steering vector may not be known exactly, but may be assumed to fall within a subspace. A common theme in adaptive beamformers is to project the data into a lower dimensional space which contains the interference to be nulled, as well as the target of interest. In several applications which we describe in detail (target acceleration, sector surveillance, multipath, and wideband processing) the target of interest lies in a target subspace which spans a collection of such interference subspaces. This paper presents a general technique of recombining the outputs of several interference subspace adaptive nulling beamformers, each which contain a “piece” of the desired target. The technique is shown to be simultaneously a maximizer of the SINR across the subspaces, as well as a maximum likelihood estimator (for the cases of both known and unknown interference covariance matrices) of the target waveform, and a GLRT detector when suitably normalized. We describe the structure of the estimator for the four cases discussed above, and describe a simplified version which obviates the need to estimate the noise coherence in the span of the adaptive beamformer outputs