1. Scope

Keywords

Rationale

Numeric classifications defined in Standard Classifications E413, E989 and E3222 result in an integer through a complex curve fitting procedure. Having integer classifications makes sense from product classification viewpoint, however, there are a few situations where a more granular value would be useful. Those include interlaboratory studies where all the laboratories are testing the same or very similar items and a research environment where small differences in a product are being evaluated. Due to the type of rounding and curve fitting employed in E413 and E989, particularly with the 32 dB and 8 dB rules, simply modifying the rounding and the increments between itterations in the curve fitting process do not work well for obtaining a more granular classification. The technique described in this guide does a better job for the purposes described above. The technique is referred to as a Monti Carlo technique. From the original unrounded data in the form of a spectrum, a large number of similar spectra are generated by applying a scaled random number to the TL or INR in each frequency band. The classification is obtained for each of these generated spectra, each an integer classification. When the number of generated curves is large, the average of the generated classifications represents a reasonable average classification for the original spectrum when compared with other average classifications obtained using the same weighting factors. The number of generated spectra should be on the order of 1000. It is proposed in this ballot that the weighting factors added to each band of the measured spectra, be the repeatability for the laboratory in the case of research and for the test method in the case of an ILS. For example in E1414, from Precision and Bias statement, the repeatability standard deviation at 125 Hz is 1.92 and 0.61 at 1000 Hz. When generating a spectrum using the Monti Carlo technique, the 125 Hz band would be equal to the measured TL at 125 Hz plus a normally distributed random number with a standard deviation of 1.92. The 1000 Hz band would be equal to the measured TL at 1000 Hz plus a normally distributed random number with a standard deviation of 0.61.