1 It should be noted that the future of highway noise prediction includes accounting for the effects of various pavement types. At this point in time, enough information is known about the effects of pavements to understand that not accounting for the effects can lead to under- or over-predictions in sound levels. Internationally, efforts are already well underway to account for the effects of pavements.
3 It was found with examination of many EFR data points, that measured peak amplitudes are much lower than theoretical peak amplitudes. As such, to extract the most accurate EFR possible, it was assumed that the EFR value for measured old asphalt was 30000 cgs rayls; the difference between the measured and theoretical peak amplitudes for the old asphalt was determined, and this difference was then applied to all measured peak amplitudes for all other pavements in order to adjust for differences between measurements and theory. This process was termed the normalization process.
4 Theoretical EFR curves were regenerated adding the extra height of a porous layer to the source and receiver locations; this assumes the porous layer is not being seen at all acoustically, but allows determination of the frequency location for the peak. The extra height pushes the peak to a lower frequency. So, as an example, where pavement may normally have a peak at 2000 Hz, the extra height pushes the peak to 1250 Hz, as seen in Figure 14, which shows one peak for the surface pavement and one for the underlying pavement.
5 A preliminary test of dependence that does not require non-linear modeling would be to determine the correlation between the level at idle and the level at, say, 55 mph. A non-zero correlation would indicate some level of dependency. It is suggested that any data set with a correlation of 0.3 or greater at least be further investigated using the non-linear model in Equation 1.
6 The issue of confidence intervals will be discussed in detail in Section 6.
7 The t-distribution can be used to compute exact difference in the confidence interval for the 3 parameter models and 14 parameter models, however, given that the base count of 30 is a gross approximation, a detail analysis thereafter is not warranted.
8 The frequency weightings are powers of log10(f), therefore errors in the parameter estimates for higher frequencies are weighted more than errors in the parameter estimates for lower frequencies.