Aspects of HDD Tone Measurements: Sound Power Microphones as Bystander Positions and Autocorrelation Analysis

This paper describes how tone content has long been known to be important in hard disk drive (HDD) sound quality. The tone-to-noise ratio (TNR) has been in use for many years, while the prominence ratio (PR) is more recent. While a system-level test using a binaural head at the system operator position is the most reflective of the end-user experience, HDD-level testing is still conducted for development and specification purposes. The HDD itself has no operator position, and even the bystander positions are nebulous since the HDD can be mounted in any orientation inside a system. The current wording in ECMA-74 (§8.6.2) specifies four bystander positions 1.5 m above the floor. These microphone positions are separate from the microphones that are used to measure sound power in a hemi-anechoic chamber, typically 10 microphones in a fixed hemisphere or a lesser number on a rotating boom. Given the rather arbitrary nature of the bystander positions, this paper investigates the use of the existing sound power microphones as bystander positions to increase measurement efficiency and spatial sampling. Whatever microphone positions are used as the bystander positions, the spectra gathered will be evaluated using the TNR and/or PR methods. The PR method is more amenable to automated tone analysis compared to the TNR method since PR does not require exact identification of peak center frequency or determination of peak width. This is because PR is based on summing the total mean square value in the lower, middle, and upper critical bands, thus reducing the effect of fine detail in the spectrum. Unless the spectrum is highly structured, a small shift in center frequency will generally have a small effect on the PR value. If the only concern is whether any peak exceeds the prominence threshold, such small changes are negligible; if a peak is near the prominence threshold a detailed analysis centered exactly on the peak in question can always be conducted for a more precise answer. Consequently, PR is sometimes simply calculated at intervals with no attempt at peak identification. Unfortunately, if the spectrum contains a peak which is approximately one critical band wide and the PR is evaluated near the center of this peak, it can be identified as highly prominent even though the peak does not represent a prominent discrete tone. This paper will show an analysis using autocorrelation of a spectrogram to determine the time-domain behavior of such a candidate peak and verify the applicability of PR for that peak.


  • English

Media Info

  • Media Type: CD-ROM
  • Features: Figures; References; Tables;
  • Pagination: pp 827-833
  • Monograph Title: Noise-Con 04. The 2004 National Conference on Noise Control Engineering

Subject/Index Terms

Filing Info

  • Accession Number: 01054380
  • Record Type: Publication
  • Files: TRIS, TRB
  • Created Date: Jul 26 2007 1:02PM