A beamformer based upon the random coefficient model

Applying the random coefficient model to array processing, specifically in the design of a beamformer for direction finding, is described. This model is similar to the autoregressive (AR) model, except the coefficients are allowed to change with time instead of remaining constant, thus allowing the beamformer to better model any additive noise or signal correlations in the observations. Through the use of a binary hypothesis test, it is shown that random coefficient models better fit typical array data than do AR models. A Kalman filter is presented that has the array observations as inputs and the parameters of the random coefficient model as outputs. A beamformer based on the random coefficient model is derived that is similar to the constant coefficient linear predictive beamformer. The two beamformers are compared and it is shown that the random coefficient beamformer outperforms the CCLP beamformer