We present and contrast certain natural metrics between power spectral density functions of discrete-time random processes. We discuss two alternative viewpoints for quantifying uncertainty and distances between statistical models. We draw analogies with the paradigm of information geometry and the Fisher information metric, and make contact with the classical literature of linear prediction, the Kullback-Leibler distance, the Itakura-Saito distance, logarithmic L2-distances, and the total variation, which have been extensively used in applications of speech processing and pattern recognition.
|Original language||English (US)|
|Title of host publication||15th Symposium on System Identification, SYSID 2009 - Preprints|
|Number of pages||2|
|State||Published - 2009|
|Event||15th IFAC Symposium on System Identification, SYSID 2009 - Saint-Malo, France|
Duration: Jul 6 2009 → Jul 8 2009
|Name||IFAC Proceedings Volumes (IFAC-PapersOnline)|
|Other||15th IFAC Symposium on System Identification, SYSID 2009|
|Period||7/6/09 → 7/8/09|
Bibliographical noteFunding Information:
★ This work was supported in part by the National Science Foundation and the Air Force Office for Scientific Research.