Towards a systematic presentation of spectrum estimation

It is shown how a general variational approach implies some well-known spectrum estimation methods, and also how it can generate new ones. The starting approach takes into account the unreliability of the initial autocorrelation estimates and extends the variational formulation of J.P. Burg's (1978) thesis. A data-dependent n-dimensional window method and a maximum entropy method for unreliable autocorrelations are among the newly generated methods outlined.<
>