A note on the effects of narrow-band and stationary signal model assumptions on the covariance matrix of sensor-array data vectors

The authors examine the nonideality and meaning of two model assumptions, narrowbandedness and stationarity, particularly for finite energy pulse sources. Analysis of the eigenstructure of covariance matrices of vectorized sensor array time samples is a current methodology for determining multiple signal source parameters. It is shown that the intersensor correlations estimated using finite length records can only fit aspects of the stationary, narrowband model under certain conditions. The results explain peculiarities which have surfaced in previous simulations and point toward improved preprocessing procedures. These include narrowband filtering, selection of data window parameters, and preflattening (or presteering). The authors relate the findings to those in other recent literature dealing with wideband data