An improved Capon estimator for HF DF using limited sample data sets

A novel method of improving the performance of Capon´s (1969) minimum variance estimator (MVE), is proposed which is entitled “loaded Capon”. The method is applied is to direction finding of ionospherically propagated signals, specifically to the case where relatively few data samples are available for formation of the covariance matrix. It is shown that as the number of samples reduces the performance of the MVE degrades. The degradation is shown to be caused by the spread of the noise subspace eigenvalues of the covariance matrix resulting from a non-infinite sample support. The new method overcomes this problem by loading the leading diagonal of the covariance matrix with a fraction of the total power contained within the covariance matrix. The consequence of diagonally loading the covariance matrix is to add a constant to each eigenvalue, which has the effect of reducing the spread in the noise subspace eigenvalues and therefore reduces how much the noise subspace corrupts the output. The loading level is chosen to be considerably below the signal subspace eigenvalues, thus minimizing any corruption of the signal subspace´s contribution to the output spectra. The performance gain of the new method is investigated using both simulated data and real data for a 354 km HF path recorded with a multichannel HF DF system. For the majority of the real data sets, the MVE fails to correctly resolve DOA, whereas the loaded Capon fluctuates closely around it. The same improvement in performance for loaded Capon is found for the simulated data sets, particularly for the more limited sample data