Power bias in maximum entropy spectral analysis
PublisherUniversitat Politècnica de Barcelona. Centre de Càlcul
Rights accessOpen Access
Several algorithms have been used in maximum entropy spectral analysis. Among them, the standard Burg procedure and the forward-backward least squares method are considered. When the autoregressive model, which is implicit in these estimation methods, is used to simulate the analyzed process, the power or variance of the simulation can differ from power estimated from the signal in several orders of magnitude. This is specially dangerous in simulated studies about the maxima of certain parameters. Burg's method, although not optimal in the least squares sense, produces autoregressive models whose estimated prediction error power is consistent with the estimated total power, while least squares method sometimes do not. Suitable corrections to power bias are described and two numerical examples clear up different situations.
CitationEgozcue Rubí, Juan José; Pagès Fita, Jaume. "Power bias in maximum entropy spectral analysis". Qüestiió. 1983 , Vol. 7, Núm. 2
ISSN0210-8054 (versió paper)