Proceedings of Meetings on Acoustics

Transcription

1 Proceedings of Meetings on Acoustics Volume 19, ICA 013 Montreal Montreal, Canada - 7 June 013 Noise Session 3pNSb: Noise Barriers 3pNSb4. In-situ measurements of sound reflection and sound insulation of noise barriers: Validation by means of signal-to-noise ratio calculations Massimo Garai* and Paolo Guidorzi *Corresponding author's address: DIN - Department of Industrial Engineering, University of Bologna, Viale Risorgimento, Bologna, 40136, BO, Italy, massimo.garai@unibo.it After some years from its first release, the CEN/TS European standard for in-situ measurement of sound reflection and airborne sound insulation characteristics of noise barriers has been significantly enhanced and validated in the frame of the EU funded QUIESST proect. The procedure, based on impulse response measurements near the noise barrier and in the free field, is robust and easily applicable but much attention must be paid when: i) applying the signal subtraction technique to get the reflected signal component and ii) extracting the transmitted component, especially measuring highly insulating noise barriers. In both cases, it is essential to avoid a poor signal-to-noise ratio of the critical part of the impulse response. In the frame of the QUIESST proect specific quality criteria, applicable on site, have been introduced in order to chec and validate the result. These criteria are rigorously described here for the first time and illustrative examples are presented. Published by the Acoustical Society of America through the American Institute of Physics 013 Acoustical Society of America [DOI: / ] Received 0 Jan 013; published Jun 013 Proceedings of Meetings on Acoustics, Vol. 19, (013) Page 1

2 INTRODUCTION The sound reflection and the airborne sound insulation of noise barriers can be measured in-situ according to CEN/TS ,,3,4. The procedure, based on impulse response measurements close to the noise barrier and in the free field, is robust and easily applicable; it allows to get the results in real-time ust after the measurements, in situ or in laboratory, applying a well defined post-processing to the raw data. Some errors may anyway occur: for sound reflection, when the signal subtraction procedure 1 leaves a small residual; for airborne sound insulation, when the barrier under test is highly insulating and so the transmitted signal is very low. This ind of problems is not always easy to recognize when on site. The European standard explains the measurement procedure in details, but a criterion for validating the measurements and prevent the acquisition of possible invalid data is missing. The authors analyzed many measurements on different noise barriers, performed on outdoor test sites during an inter-laboratory test organized in the frame of the European proect QUIESST 5,6,7, having among its obectives the improvement of the CEN/TS measurement method. The analysis of this large amount of data suggested that the above mentioned measurement problems may be identified and corrected evaluating the signal-to-noise ratio (SNR) of critical parts of the impulse responses being processed. These criteria are rigorously described here for the first time and illustrative examples are presented. SOUND REFLECTION MEASUREMENTS Measurement Principle According to the QUIESST improvements to CEN/TS ,5, a loudspeaer with nown characteristics is used as sound source and a square array of 9 microphones ( microphone grid ) is used as receiver. They are placed on the same side of the noise barrier under test at prescribed distances and heights. The sound source emits a transient sound wave that travels past the measurement grid (microphone) position to the device under test and is then reflected on it (figure 1). Each microphone, being placed between the sound source and the device under test, receives both the direct sound pressure wave travelling from the sound source to the device under test and the sound pressure wave reflected (including scattering) by the device under test ( barrier measurement). The measurement is repeated in the free-field (without the noise barrier close to the loudspeaer and the microphone array, but eeping the same relative geometry) in order to measure the direct wave ( free-field measurement). The power spectra of the direct and the reflected components, give the basis for calculating the so called sound reflection index. The improved expression used in the QUIESST proect 5 to compute the reflection index RI, in one-third octave bands from 100 Hz to 5 Hz, is: RI 1 n n f 1 f r, w w r, C geo, C dir, ( f ) C gain, f g (1) where h (t) is the incident reference component of the free-field impulse response at the -th measurement point; h r, (t) is the reflected component of the impulse response taen in front of the sample under test at the -th measurement point; w (t) is the time window (Adrienne 1 temporal window) for the incident reference component of the freefield impulse response at the -th measurement point; w r, (t) is the time window (Adrienne 1 temporal window) for the reflected component at the -th measurement point; F is the symbol of the Fourier transform; is the index of the one-third octave frequency bands (between 100 Hz and 5 Hz); f is the width of the -th one-third octave frequency band; is the microphone number according to figure 1.b ( = 1,..., 9); n is the number of microphone positions on which to average (n 6; see ref. 5 ); Proceedings of Meetings on Acoustics, Vol. 19, (013) Page

3 C geo, C dir, f ) C gain, f g ) f g is the correction factor for geometrical divergence at the -th measurement point; is the correction factor for sound source directivity at the -th measurement point. is the correction factor to account for a change in the amplification settings of the loudspeaer and in the sensitivity settings of the individual microphones when changing from the free-field to the barrier measurement configurations or vice versa (if any); is the frequency range encompassing the one-third octave frequency bands between 500 Hz and Hz. In this formulation three new corrective factors are added to the usual formula described in ref. 1 : C geo, is a correction factor used to compensate the geometrical divergence at the -th measurement point and taes into account the path difference from the direct and reflected waves; C dir, ( ) is a correction factor used to compensate the difference of sound source directivity, at the -th measurement point, due to different incidence angles of direct and reflected waves on the microphones. C gain, ( ) is a correction factor used to compensate a gain mismatch (if any) of the amplification settings between the free-field and barrier measurement configurations or vice versa. More details on these correction factors can be found in ref. 5. FIGURE 1. (a) Sound source (Zircon ) and microphone array in front of a strongly non flat, sound absorbing noise barrier. (b) Numbering of the microphones on the measurement array. Improved Signal Subtraction Technique The barrier impulse response consists of a direct component, a component reflected from the surface under test and other parasitic reflections (figure.a). The direct component and the reflected component from the barrier under test must be separated for each microphone position. This is done using the signal subtraction technique: the reflected component is extracted from the overall impulse response after having removed the direct component by subtraction of an identical signal (figures.c and.d). This means that the direct sound component must be exactly nown in shape, amplitude and time delay. This can be obtained by performing a free-field measurement for each microphone using the same geometrical configuration of the loudspeaer and the measurement grid. In particular, their relative position must be ept as constant as possible. The direct component to be used in eq. (1) is extracted from the free-field measurement (figure.b). This technique allows broadening of the time window (7,9 ms with the standard geometrical setup), leading to the lower frequency limit of the woring frequency range (100 Hz one-third octave band), without having very long distances between loudspeaer, microphone and device under test. In principle, the signal subtraction technique requires the loudspeaer and microphones relative position be ept constant in order to get a perfect alignment between the impulse responses measured in front of the device under test and in the free-field for the same microphone. This may be very difficult in practice when in situ, due to placement of the equipment on an irregular terrain, small movements of the loudspeaer cone or the microphones when displacing the equipment, variations in the response of the measurement equipment due to temperature or electrical deviations occurring between the free-field and the reflected measurements, etc. Therefore it is necessary that, before performing the signal subtraction, the free-field signal is corrected for a small shift relative to the impulse response in front of the barrier under test at each microphone. Proceedings of Meetings on Acoustics, Vol. 19, (013) Page 3

4 FIGURE. Illustration of the signal subtraction technique. (a) Free-field impulse response. (b) Free-field impulse response overlapped to the barrier impulse response. (c) Barrier impulse response only. (d) Reflected component of the barrier impulse response after subtraction. Since in general the actual time shift is not equal to a multiple of the temporal sampling interval t, step wise shifting of one or more data points is inadequate. In the frame of the QUIESST proect 5, the authors introduced an accurate alignment procedure (see below); it allows the placement of the measurement grid without a rigid connection to the loudspeaer; the unavoidable misalignments between the impulse responses measured in front of the device under test and in the free-field for the same microphone may then be compensated up to 5 cm. The accurate alignment procedure is composed of the following steps: 1. For each microphone position, an impulse response measured in front of the device under test and one measured in the free-field with nominally the same geometry are compared.. The free-field impulse response is shifted with a small moving step (which is a fraction of the temporal sampling interval t between the discrete points of the acquired data, see below). 3. The sum of the squared differences between the free-field impulse response and the impulse response measured in front of the device under test is calculated in a limited interval around the first and main pea of the impulse response measured in front of the device under test. 4. The operations in and 3 are repeated until the minimum of the sum in 3 is found (least squares); the number n of moving steps needed to get this least square minimum is recorded. 5. The free-field impulse response is finally shifted with the temporal step n found in 4 and its amplitude is adusted so that the amplitude of its first (and main) pea is exactly the same of the first (and main) pea of the impulse response measured in front of the device under test. 6. The shifted and amplitude adusted free-field impulse response is subtracted from the impulse response measured in front of the device under test. This modified free-field impulse response is discarded after the subtraction; the free field impulse response used to calculate the reflection index according to eq. (1) is the original, unchanged one. The moving step in should be about 1/50 of the temporal sampling interval t between the discrete points of the acquired data (lined to the given sample rate f s by t =1/f s ). The least squares calculation in 3 is limited to about 50 discrete data points around the first and main pea of the impulse response measured in front of the device under test. The iteration in 4 is limited to discrete data points of the acquired data, i.e. 100 moving steps ; this range is centred on the first and main pea of the impulse response measured in front of the device under test, i.e. the main pea of incident component of the impulse response measured in front of the device under test. Proceedings of Meetings on Acoustics, Vol. 19, (013) Page 4

5 In order to shift the free-field impulse response in n moving steps, n, with considerably smaller than the temporal sampling interval t between the discrete points of the acquired data, the following procedure is applied. a. The free-field impulse response is Fourier transformed in the frequency domain and its phase is changed by multiplying it with a frequency dependent factor exp(ifn). b. The resulting phase corrected Fourier transform is inverse transformed to generate the shifted free-field impulse response in the time domain which then can be used for signal subtraction. h h expi i, shifted ( 1 t) F F fn, () This approach is very different from that proposed by Robinson and Xiang 8, who used data oversampling and further optimization by means of a genetic algorithm, and allows to get equally excellent results (see figure.d) in a very short time. Validation Criterion for the Signal Subtraction Technique As the goal of the operation is to remove the incident component of the impulse response (the direct sound ), leaving only the reflected one, following Robinson and Xiang 8 the signal subtraction effectiveness can be measured by the decibel level reduction in the direct sound from the measurement to the result. Specifically, the sum of the energy within 0,5 ms of either side of the first and main pea of direct sound can be compared before and after subtraction to find the effective reduction. Eq. (3) defines the reduction factor R sub. where h,ff (t) h,res (t) R sub t p, 0,5 t p, 0,5 10lg t p, 0, 5 t p, 0,5 ms ms ms ms h h, FF, RES t t dt db dt is the incident reference component of the free-field impulse response at the -th measurement point (before the signal subtraction); is the residual incident reference component of the impulse response taen in front of the sample under test at the -th measurement point (after the signal subtraction); t p, is the time instant where the first pea of the incident component of the impulse response at the - th measurement point is located (before the signal subtraction); see figure. (3) Rsub (b) Rsub (a) FIGURE 3. Example of application of the validation criterion, eq. (3) on the same noise barrier. (a) Subtracted impulse response with a strong residual. (b) Subtracted impulse response with a small residual. (c) Reduction factors for cases (a) and (b). Proceedings of Meetings on Acoustics, Vol. 19, (013) Page 5

6 A reduction factor R sub equal to the pea to noise ratio of the measurement can be considered a complete subtraction, since this would leave nothing in the area of the direct sound except the bacground noise. Practical experience indicates that a value of R sub < 10 db should be considered a warning that the signal subtraction is not perfect. Figure 3 shows an example of application of the validation criterion, eq. (3), on the same noise barrier. For measurement (a) R sub < 10 db at microphones 7, 8 and 9 and thus it cannot be considered valid. AIRBORNE SOUND INSULATION MEASUREMENTS Measurement Principle According to the QUIESST improvements to CEN/TS ,5, a loudspeaer with nown characteristics is used for the measurement as sound source and a square grid of 9 microphones is used as receiver. They are placed on the opposite sides of the noise barrier under test at prescribed distances and heights (figure 4). The sound source emits a transient sound wave that travels toward the device under test and is partly reflected, partly transmitted and partly diffracted by it. The microphone array placed on the other side of the noise barrier receives both the transmitted sound pressure wave travelling from the sound source through the device under test, and the sound pressure wave diffracted by the top edge of the noise barrier ( barrier measurement). The measurement is repeated in the free-field (without the noise barrier between the loudspeaer and the microphone array but eeping the same relative geometry) in order to measure the direct wave ( free-field measurement). The power spectra of the direct wave and the transmitted wave, give the basis for calculating the so called sound insulation index SI. FIGURE 4. Airborne sound insulation measurement in situ on a strongly non flat barrier sample (QUIESST proect 6,7, Valladolid test site). The improved expression used in the QUIESST proect 5 to compute the sound insulation index SI, in one-third octave bands from 100 Hz to 5 Hz, is: SI 1 10 lg n n f 1 f t, t t w w t, t t db (4) where h (t) is the incident reference component of the free-field impulse response, measured at the -th microphone of the grid; h t, (t) is the transmitted component of the impulse response, measured at the -th microphone of the grid; F is the symbol of Fourier transform; is microphone identifier in the grid ( = 1 to 9); Proceedings of Meetings on Acoustics, Vol. 19, (013) Page 6

7 f w (t) w t, (t) n = 9 is the index of the one-third octave frequency bands from 100 Hz to 5 Hz; is the width of the -th one-third octave frequency band; is the time window (Adrienne 1 temporal window) for the incident reference component of the free-field impulse response at the -th microphone position; is the time window (Adrienne 1 temporal window) for the transmitted reference component of the free-field impulse response at the -th microphone position; is the total number of microphones in the grid. Validation Criterion for Airborne Sound Insulation Measurements The example in figure 5.c shows the second problem addressed in this wor: the noise barrier airborne sound insulation is so high that the amplitude of the transmitted impulse response component (inside the Adrienne analysis window) is of the same order of magnitude or below that of the bacground noise. Any noise coming from sound sources around the test site was almost completely reected since measurement were done using an MLS signal; thus the bacground noise floor in the impulse responses is mainly due to the measurement system itself. Considering the MLS theory 9,10,11 and provided that a sufficiently long pseudorandom sequence is used, the only parts of the impulse response not containing useful data are the end tail and the initial few milliseconds of the measurement. Since some software or hardware systems use a long sequence for the measurement but save only the initial part of the impulse response where meaningful data are placed, in this wor the bacground noise has been evaluated considering the very beginning of the impulse response, corresponding to the travel time of the sound waves from the loudspeaer to the microphone(s). This time interval is clearly visible in the free-field measurement example in figure 5(a) from t = 0 ms to around t = 4 ms, corresponding to a distance of 1,5 m plus the barrier thicness. The same initial clean time interval is found in the barrier measurements; in case of heavy noise barriers, as in figure 5.c, the transmitted wave is not visible, but the same considerations apply. FIGURE 5. Time interval selections for bacground noise (A) and signal (B). Example of application on free-field (a) and barrier (b, c) measurements. A possible way to chec if a measurement on an highly insulating noise barrier is valid consists in the evaluation of the ratio between the transmitted signal energy and the bacground noise energy. In order to evaluate this signalto-noise ratio (SNR) for each impulse response, a time interval including only the bacground noise and a time interval including the transmitted signal must be identified. By definition, in this wor the time interval for the evaluation of the bacground noise has been chosen from t = 0 ms to t = 3,5 ms and the time interval for the evaluation of the (transmitted) signal has been chosen from t = 3,5 ms to t = 7,0 ms. These time intervals are shown in figure 5 mared as A for the bacground noise and B for the transmitted signal, respectively. The 3,5 ms time marer may be shifted in case of different geometrical configurations. These 3,5 ms wide data windows allow to obtain meaningful frequency data starting from the one-third octave band of 400 Hz, so that the SNR is evaluated in the one-third octave bands from 400 Hz to 5 Hz: SNR SI, lg 37 6f 6f t t w w signal, noise, t t db (5) Proceedings of Meetings on Acoustics, Vol. 19, (013) Page 7

8 where: h (t) is either the free-field or the barrier impulse response, measured at the -th microphone position; F is the symbol of Fourier transform; is microphone identifier among the nine in the grid ( = 1 to 9); is the index of the one-third octave frequency bands between band nr. 6 (400 Hz) and band nr. 37 (5 Hz); f is the width of the -th one-third octave frequency band; w signal, (t) is the time window for the signal evaluation of the impulse response, conventionally chosen equal to 1 from 3,5 ms to 7,0 ms, equal to 0 elsewhere; w noise, (t) is the time window for the bacground noise evaluation of the impulse response, conventionally chosen equal to 1 from 0,0 ms to 3,5 ms, equal to 0 elsewhere. The proposed SNR is denoted as SNR SI, in order to recall that it is tailored for SI measurements and specific for each microphone ( = 1,.., 9). It is essentially the db ratio of the energy contained in the time interval from 3,5 ms to 7,0 ms divided by the energy contained in the time interval from 0 ms to 3,5 ms, in the one-third octave frequency bands from 400 Hz to 5 Hz. A global single number value can be computed too, as the logarithmic average of the 9 microphones values; it is denoted SNR SI. A different approach could have been followed, computing directly a SNR in the time domain, but the method presented here allows to compute also, if required, the SNR SI,, in each one-third octave band (in the above mentioned frequency range), simply avoiding the summation on the frequency bands in eq. (5), and thus obtaining additional useful information as a function of frequency. FIGURE 6. Case study 1. (a) SI measured on a heavy concrete noise barrier. (b) Associated SNR SI, according to eq. (5). FIGURE 7. Case study. (a) SI measured on a lightweight noise barrier. (b) Associated SNR SI, according to eq. (5). Figure 6.a shows three measurements on the same heavy concrete noise barrier using different measuring equipment setups (case study 1). Measurement nr. and 3 give quite similar results (which can be considered as correct values), while measurement 1 doesn t, especially at low frequencies. Figure 6.b shows the corresponding Proceedings of Meetings on Acoustics, Vol. 19, (013) Page 8

9 SNR SI,, computed according to eq. () for each of the nine microphone positions. The SNR SI, values of measurement 1 are very low and for some microphones even lower than 0 db (meaning that the bacground noise is higher than the transmitted signal). The SNR SI, values of measurements and 3 are higher than 10 db for all microphones. Figure 7.a shows three measurements on a lightweight noise barrier, again obtained using different measuring equipment setups (case study ). In this case the three SI curves are in a very good agreement. The corresponding SNR SI, plotted in figure 7.b are all greater than 15 db. It is worth noting that in this case the SNR SI, values obtained with different measuring equipment setups show a similar behavior, with higher values for microphones 4 and 7: this is probably due to a sound insulation leaage nearby of the mentioned microphones positions. CONCLUSION The CEN/TS method for measuring in-situ the sound reflection and the airborne sound insulation of noise barriers has been greatly improved in the frame of the QUIESST proect. In particular, in this wor two criteria to quicly validate the acquired impulse responses during field measurements have been presented. Both are based on the evaluation of the signal-to-noise ratio (SNR) of critical parts of the impulse responses being processed. The criterion to chec sound reflection measurements is based on the wor of Robinson and Xiang 8 with a different and faster signal subtraction procedure. The criterion to chec airborne sound insulation measurements is presented here for the first time. These criteria have been tested on the data coming from the inter-laboratory test done in the frame of the European proect QUIESST. Some examples have been presented here. ACKNOWLEDGEMENT The QUIESST research is an EU funded proect under the Seventh Framewor Program (FP7-SST-008-RTD- 1 N ). REFERENCES 1. CEN/TS Road traffic noise reducing devices Test method for determining the acoustic performance Part 5: Intrinsic characteristics - In situ values of airborne sound reflection and airborne sound insulation (Comité Européen de Normalization, Geneva, 003).. M. Gara P. Guidorz European methodology for testing the airborne sound insulation characteristics of noise barriers in situ: experimental verification and comparison with laboratory data, J. Acoust. Soc. Am. 108, (000). 3. Watts, P. Morgan, Measurement of airborne sound insulation of timber noise barriers: comparison of in situ method CEN/TS with laboratory method EN 1793-, Appl. Acoust. 68, (007). 4. M. Gara P. Guidorz In situ measurements of the intrinsic characteristics of the acoustic barriers installed along a new high speed railway line, Noise Control Eng. J. 56, (008). 5. M. Gara Noise reducing devices acting on airborne sound propagation Test method for determining the acoustic performance Intrinsic characteristics In situ values of sound reflection under direct sound field conditions, QUIESST Deliverable D (Last viewed 10 Jan. 013). 6. M. Gara Noise reducing devices acting on airborne sound propagation Test method for determining the acoustic performance Intrinsic characteristics In situ values of airborne sound insulation under direct sound field conditions, QUIESST Deliverable D (Last viewed 10 Jan. 013). 7. M. Gara Inter-laboratory test to assess the uncertainty of the new measurement methods for determining the in situ values of sound reflection and airborne sound insulation of noise reducing devices under direct sound field conditions, QUIESST Deliverable D3.5. ( (Last viewed 10 Jan. 013). 8. P. Robinson, N. Xiang, On the subtraction method for in-situ reflection and diffusion coefficient measurements, J. Acoust. Soc. Am. 17, EL M. Rife D., Modulation Transfer Function Measurement with Maximum Length Sequences, J. Audio Eng. Soc., 40, (199). 10. J. Borish, Self-contained crosscorrelation program for maximum-length sequences, J. Audio Eng. Soc. 33, (1985). 11. M. Vorländer, M. Kob, Practical aspects of MLS measurements in building acoustics, Appl. Acoust., 54, (1997). Proceedings of Meetings on Acoustics, Vol. 19, (013) Page 9