Adaptive beamforming in non-stationary environments

With moving interferers, there is a trade-off between statistical uncertainties if too few snapshots are used, and smearing of the eigenstructure of the sample covariance matrix (SCM) if too many snapshots are used. When the number of snapshots L that can be obtained before significant motion occurs is less than the number of elements M in the array, the SCM is rank-deficient. A discussion is given of the special problems in using rank-deficient SCMs. Frequently, extra information is available that is not contained in the sample covariance matrix. An estimate of the maximum number of significant eigenvalues is twice the array length divided by the acoustic wavelength. An estimate of the background level in the absence of interference is frequently available based on long-term observations. Under some circumstances the median eigenvalue of the SCM is a useful estimate of true background noise. In this paper, we examine the use of such auxiliary information to set the loading parameter of the rank-deficient SCM.