This paper summarizes numerically derived theoretical estimates of normalized random and normalized bias errors for one-third octave band filters applied to autoregressive processes of order two. It demonstrates that one-third octave band estimates of such processes may have substantial bias errors depending upon (1) frequency location of a true spectra peak relative to the center frequency of the one-third octave band filter and (2) half-power bandwidth of the true spectra. One-third octave band processing error is contrasted with narrowband processing error. It concludes that one-third octave band processing should be avoided whenever external information would indicate that the underlying time history data may have a narrowband character.

This content is only available as a PDF.
You do not currently have access to this content.