GNGTS 2015 - Atti del 34° Convegno Nazionale

For these signals, the HVIP analysis identifies a very high number nR 2Hz of Rayleigh-type samples at the peak frequency of 2 Hz: it is in the order of 20000, i.e. about 20 % of the entire recording. Their analysis points out very clearly the directional polarization of the signal surf100, since more than 98% of these samples is concentrated within the azimuth bin (30°-40°) containing the correct value (37°). Increasing the noise amplitude, the discrepancies between HVIP and real H/V increase, but for snr = 3 the agreement is still good, especially for the case of directional polarization (surf100sn3). It shows an increase to 20-25% of the error in the estimate of the H/V peak value, but the estimate maximizing precision ( rms sc = 0.36) is still very close to that minimizing errors ( rms err = 0.27 and 0.22, respectively). More large discrepancies are found in case of isotropically polarized signals (surf100sn3i). Indeed the most accurate estimate, even correctly identifying the peak frequency, gives an error of 27% in the estimate of H/V peak value and rms err = 0.33, whereas the curve having minimum scatter ( rms sc = 0.51) shows a maximum at 2.25 Hz, an error of 34% in the H/V peak estimate and rms err = 0.46. In such cases nR 2Hz is in the order of 1000-1500, i.e. about 1-1.5% of the entire recording, which is more close to the classification percentages found in preliminary tests on real noise recordings, thus these cases can be considered more representative of conditions actually occurring in the field applications. In the case of directionally polarized signal, the identification of site response directivity orientation is still very clear, since a pronounced maximum, larger than 50%, is found in the polarization distribution for the 30°- 40° azimuth bin. Comparatively, HVNR values derived from Nakamura’s method provide considerably worse estimates, both in the directional case ( rms sc = 0.74) and in the isotropic one ( rms sc = 0.80 and 0.62, using Euclidean and geometric average, respectively). In general, in the HVNR curve, the peak at 2 Hz is not very clear as effect of the concomitant underestimate of the peak value and overestimate of the lower part of the H/V curve (which, on the contrary, the HVIP fits very well) If the noise amplitude is comparable to that of the Rayleigh signals, the number of samples identified as Rayleigh-type decreases and the estimates undergo a further deterioration, especially with regard to the H/V peak values, whose estimate errors reach values of 30-40% and 40-50% for the directional (surf100sn1) and isotropic (surf100sn1i) signals, respectively. However, although with underestimated peak values, some general characteristics of the H/V curve can still be recovered. Indeed, the HVIP curves show a single major peak around 2 Hz and, although underestimating the maximum by up to 40%, reproduce quite well the rest of the curves. With regard to site response directivity, even in this less favorable noise conditions, it can be recognized from the sample polarization distribution: a maximum of concentration (up to 20-30%) of polarizations around azimuths differing by not more than 10° from the actual direction of Rayleigh wave ground motion is found at almost all the examined frequencies. Comparatively, using the Nakamura’s technique, the peak at 2 Hz is practically undistinguishable from the HVNR curves. Apart from the considerable underestimate of the H/V ratio maximum, these curves systematically fail in recognizing the presence of H/V ratios < 1 at higher frequencies. Conclusions. A series of tests carried out on synthetic signals simulating ambient noise recordings and including a mix of transient Rayleigh and Love waves of known characteristics together with a casual Gaussian noise of different amplitude, showed that the analysis of instantaneous polarization allows extracting Rayleigh wave properties with a good level of approximation. For this purpose, signals are first passed through band-pass filters with different central frequency. Then, on the resulting time series, Rayleigh wave packets are identified when a minimum number of consecutive data samples are found to exceed optionally defined thresholds of angular deviation of the plane of instantaneous elliptical trajectories from verticality and of ellipse major/minor axes from horizontal/vertical direction. For each sample of these wave packets, an estimate of the instantaneous ratio H max /V can be obtained, whose averages HVIP 80 GNGTS 2015 S essione 2.2