© The Author(s), Published by EDP Sciences, 2026

1 Introduction

One major feature of buildings is their ability to protect people against noise. Many countries therefore apply legal requirements with respect to different types of sound insulation in buildings. These legal requirements are often understood as minimum performance values and better noise protection is frequently stipulated. To meet the desired level of noise protection, the acoustic properties of a building are predicted from the acoustic properties of the building elements which are usually measured in laboratories. In some countries, it may be common to verify the acoustic quality of a building once it is finished by conducting field measurements.

This system of ensuring noise protection in buildings is only manageable if the uncertainty of building acoustic measurements in laboratories and in buildings is sufficiently known. There are, in principle, two methods to achieve this. The first is to use a detailed model for the measurement as a base for an uncertainty balance according to the Guide to the expression of uncertainty in measurement (GUM [1]). This approach has several advantages: the measurement effort can be focused on a reduction of the most important uncertainty contributions, the uncertainty is specific to the real situation, a small uncertainty is indicative for a high-quality measurement, to name but a few. The main problems with this approach are that an appropriate model equation is difficult to derive and that the effort for performing such an uncertainty analysis may become larger than generally accepted today.

The second method to estimate the uncertainty is to assume that measurement results in building acoustics have the same uncertainty when they are obtained according to the same standardised procedures. Then, a collection of measurement results from interlaboratory tests can be used to estimate the uncertainty. This approach has been used in building acoustics for many years. Uncertainties had already been published in ISO 140-2 [2] which has been replaced by ISO 12999-1 [3] in the meantime. When ISO 12999-1 was originally developed, it became clear that the uncertainties published in ISO 140-2 were not traceable to available publications or data bases. Therefore, a new data collection was performed. For airborne sound insulation, the results had been published in [4] and typical uncertainties were incorporated in ISO 12999-1.

Three different situations are distinguished in ISO 12999-1. Situation A are reproducibility conditions. These are conditions of measurement that include different locations (laboratories or usual buildings), operators, measuring systems, and replicate measurements on the same or similar objects. Such conditions do not require repeated measurements at the same location. So, if a measurement is replicated at each location once, the standard deviation of the measurement results from different locations is a standard deviation of reproducibility. Situation C are repeatability conditions which are conditions of measurement with the same measurement procedure, same operators, same measuring system, same location (laboratory or usual building), and replicate measurements on the same object over a short period of time. In the course of the development of ISO 12999-1 it became clear that an intermediate situation B between reproducibility and repeatability condition occurs very often in building acoustics which is termed in-situ conditions. They include the same location (laboratory or usual building), and replicate measurements on the same object by different operators using different measuring systems. As for the reproducibility conditions, an in-situ standard deviation can be determined without perfoming repeated measurements. Interlaboratory tests in building acoustics may be performed under reproducibility or in-situ conditions and may or may not include repeated measurements by each participant.

For impact noise, a very brief overview using the concept with the three described situations A, B and C was made available [5] which contained the typical uncertainties published in the current version of ISO 12999-1. The data base behind this analysis comprised interlaboratory tests on impact noise levels under repeatability conditions and under in-situ conditions. But there were no data available for impact noise levels under reproducibility conditions. Therefore, no uncertainties could be given in ISO 12999-1 for this important case in one-third octave bands whereas a value for the single-number quantity had been estimated.

In the meantime, the data base on impact noise increased significantly since several interlaboratory tests have been performed. Unfortunately, there is still no interlaboratory test for impact noise levels under reproducibility conditions available. To come to realistic uncertainties all the same, laboratories were asked to provide measured normalised impact noise levels for the reference concrete slab and for the lightweight reference floors defined in ISO 10140-5 [6]. It is believed that this data set is the closest approximation to an interlaboratory test on floor impact noise levels that is available today.

This paper gives an overview on the measured normalised or standardised impact noise levels and the impact noise reductions from interlaboratory tests. The data are analysed to derive typical uncertainties for the measurement of normalised or standardised impact noise levels and impact noise reductions. The analysis is performed for one-third octave bands and for the single-number quantities standardised in ISO 717-2 [7].

Uncertainties in this paper are quantified by empirical standard deviations for which the symbol σ is used. This is in line with the use in ISO 12999-1 and many other standards in acoustics whereas the symbol s is used for that purpose in statistical standards and textbooks.

2 Impact noise levels under in-situ and repeatability conditions

2.1 Data overview

Many interlaboratory tests have been performed under in-situ conditions (Tabs. 1, 2). There are two different categories of data sources. The first comprises the sources where a citeable reference is available. These sources are specified in Table 1. Some experiments had been undertaken with the explicit aim to determine the measurement uncertainty, e.g. [8, 9, 12, 13]. Other experiments are focused on the quality control of laboratories, e.g. [10, 11, 14–20]. Measurement results are available in one-third octave bands at least between 100 Hz and 3.15 kHz. Only the results of [9] are in octave bands between 125 Hz and 2 kHz. The results from [14, 16, 18, 21] have been generated in test facilities while all the other results come from field measurements. In most experiments, each participating laboratory performed just one single measurement. Measurements have been repeated only in [10, 11, 13, 21]. The reported quantity is either the normalised or the standardised impact noise level. Usually, the cited documents do not contain the measured normalised or standardised impact noise levels. These had kindly been provided by personal communication by the authors.

The second category of data comprises the sources where no reference is available (Tab. 2). These results come from a long history of comparison measurements performed in Germany for the quality control of test houses performing building acoustic field measurements. Today, such measurement campaigns are organised in Germany by VMPA, the association of material testing institutes in cooperation with FMPA, the Leipzig Institute for Materials Research and Testing, and PTB. The results are complemented by a data set from the UK which was generated for the same purpose. Each participating test house performed one measurement in all data sets of Table 2, i.e. the measurements were not repeated. The measurement results were also kindly provided by the organisers of the comparison measurements.

All measured normalised impact noise levels from VMPA13, VMPA19B and VMPA22 are shown in Figure 1 as an example. One can clearly see that normalised impact noise levels measured by the participants cover a range of several dB for the same test object. The scatter range is larger at low frequencies. At high frequencies, some measured values from VMPA19B seem to suffer from background noise. This is because the normalised impact noise level is quite small.

Figure 1. Normalised impact noise levels measured under in-situ conditions from VMPA-comparison measurements (for details see Tab. 2).

The normalised or standardised impact noise levels from all interlaboratory tests in Tables 1 and 2 cover a very wide range (Fig. 2). It may therefore be assumed that the available data set represents the field of application of impact noise measurements sufficiently.

Figure 2. Mean values of the measured normalised or standardised impact noise levels from all interlaboratory tests of Tables 1 and 2.

2.2 Principles of data analysis

Looking at the available data, several aspects need to be considered. The first is whether the applied measurement procedures are consistent. The applicable standards which are currently valid are ISO 10140-3 [22] and ISO 16283-2 [23]. These documents were preceded by other international and national standards. Nevertheless, the basic measurement principle has not been changed. The ISO tapping machine has always been used, measurement equipment of sufficient quality has been available for all the interlaboratory measurements, number and arrangement of tapping machine and microphone positions is not changed significantly. It is therefore considered that there was no substantial change in the measurement procedure for the context of this data analysis.

Starting from this it is straightforward to calculate an in-situ standard deviation σ _situ,i for each of the N interlaboratory tests from Tables 1 and 2 according to ISO 5725-2 [24], i.e. Bessel’s correction is applied. All the calculations are performed on a dB-scale in analogy to airborne sound insulation [4]. The averaged in-situ standard deviation σ _situ is then calculated by

$$\begin{array}{c}\hfill {\sigma }_{\mathrm{situ}}=\sqrt{\frac{1}{N}\sum _{i=1}^N{\sigma }_{\mathrm{situ},i}^2}\end{array}$$(1)

With this equation, all interlaboratory tests have the same weight regardless of the number of participants. This has been decided because there are results with more than 100 participants and others with just 4 participants. With a weighted average, the results with very large numbers and the related measurement situations, i.e. these particular normalised or standardised impact noise levels and their in-situ standard deviations would dominate the average. This is not appropriate since each result from Figure 2 should have the same weight. The same decision has been made for airborne sound insulation in [4] and for absorption coefficients in [25] for the same reason.

A second question is how to handle the results VMPA19A and VMPA19B. They come from the same measurement situation with a partly separated receiving room. 34 participants of the comparison measurement used only a part of the receiving room as the relevant volume whereas 63 participants used the full volume of the receiving room. The mean impact noise levels deviate by about 1 dB but this difference depends on frequency. Since it is very questionable how to convert the results into one unified data set, both data sets are considered as individual ones.

Another aspect is how to incorporate the octave band results into the analysis in one-third octave bands. This is done by assigning the octave band standard deviation to the three one-third octave bands comprising the respective octave band.

A further question is the application of corrections prior to data processing. For one single result from [20], the A-weighting had been erroneously applied during the measurement which was corrected for the further analysis. All the other data sets provided have been used without further pretreatment and no result has been disqualified.

It is furthermore assumed that normalised and standardised impact noise levels have the same uncertainty. This is justified by the fact that the measurement of both quantities is based on the same acoustic measurands, i.e. the mean sound pressure level and the reverberation time. The determination of the volume of the receiving room which is necessary for the normalised but not for the standardised impact noise level is considered not to contribute significantly to the combined uncertainty. This is also in line with the analysis for airborne sound in [4].

Most of the data sets contained also single-number quantities. To avoid any additional uncertainty which may arise from different implementations of the prescribed calculation procedure, all single-number quantities were newly calculated with the same calculation tool from the provided one-third octave band or octave band data.

2.3 Data analysis in one-third octave bands

All standard deviations under in-situ conditions for the data sets from Tables 1 and 2 are shown as grey lines in Figure 3. Each of these standard deviations therefore corresponds to one mean value in Figure 2. The in-situ standard deviation exhibits a considerable scatter between the different interlaboratory tests. The lighter green line is the average value as calculated by equation (1). The dashed darker green line is the same average but without the octave results from [9]. These octave results show a very high standard deviation which is due to the fact that measurements were performed on lean mixture screeds which were not intended to serve as a final floor topping. Tapping machines damaged the screed during operation and this occurred to a different extent with different participants in the interlaboratory test. For this reason, the octave band results are disqualified and not used in the further processing. The orange line in Figure 2 is the typical in-situ standard deviation from ISO 12999-1 [3]. It corresponds very well to the average value without octaves between 50 Hz and 3.15 kHz. At 4 kHz and 5 kHz, the average is increased very much by some very high in-situ standard deviations which belong to data sets with very small normalised/standardised impact noise levels, see Figures 1 and 2. These very low levels are affected by the background noise which then increases the in-situ standard deviation. It is not possible to define a clear criterion for this but when the three largest in-situ standard deviations at 4 kHz and 5 kHz are omitted from the averaging, the mean in-situ standard deviation approaches the value currently standardised in ISO 12999-1. So it can be concluded that the typical in-situ standard deviation given in ISO 12999-1 [3] is confirmed by this analysis under the condition that the background noise level is handled appropriately.

Figure 3. In-situ standard deviation of the measured normalised or standardised impact noise levels from all interlaboratory tests of Tables 1 and 2, average with and without octave results and in-situ standard deviation from ISO 12999-1 [3].

Besides an analysis of the in-situ standard deviation, the data from Table 1 also enables a consideration of the standard deviation of repeatability. The individual values are shown as grey lines in Figure 4. Also shown is the average calculated in analogy to equation (1), the typical value and the maximum value currently standardised in ISO 12999-1 [3]. The typical value from ISO 12999-1 is similar to the average value. Only at 5 kHz, the typical value is smaller than the average value. Here, but also at 4 kHz and 3.15 kHz, the average value is significantly increased by one very large standard deviation in each band. They are highlighted in Figure 4 by circles. A reason for the very high standard deviation is not given in the underlying reports. Nevertheless, a problem with an electrical background noise as in Figure 3 is very unlikely since this would correspond to a very stable result when repeating the measurement. It is worth mentioning here that the encircled values correspond to one participant in an interlaboratory test out of 12 or 14 participants, respectively.

Figure 4. Standard deviation of repeatability of the measured normalised or standardised impact noise levels from interlaboratory tests [10, 11, 13, 21], average with and without encircled data points and standard deviation from ISO 12999-1 [3].

The maximum standard deviation of repeatability from ISO 12999-1 is common to the measurement of impact noise levels and airborne sound insulation. This maximum standard deviation of repeatability really describes a reasonable upper limit in view of the existing results (Fig. 4) except for the already mentioned very high frequencies above 3.1 kHz. Here the typical standard deviation is even larger than the maximum standard deviation which should be adapted in a future version of ISO 12999-1. The proposed solution is to omit the three encircled data points in Figure 4 from the averaging and use the new average as the typical standard deviation in ISO 12999-1. A closer look at the data furthermore reveals that a slight reduction of the typical standard deviation of repeatability compared to the currently standardised values starting at 630 Hz seems to be appropriate. This new proposal is also shown in Figure 4.

2.4 Data analysis for single-number quantities

The single-number quantities standardised in ISO 717-2 [7] are calculated from the one-third octave or octave band results. To avoid any issues with different implementations or possible changes in the calculation procedures of different editions of ISO 717-2, all single-number quantities are calculated at PTB using an in-house code which is an exact implementation of the method of ISO 717-2 [7]. Calculation results for the in-situ standard deviation exhibit a considerable scatter (Fig. 5). Encircled data points are from source [9]. As for the results in frequency bands they are omitted. Since the reason for the deviating results have been identified (see clause 2.3), these results are not included in the further analysis.

Figure 5. In-situ standard deviation of the measured single-number quantities (dots) from all interlaboratory tests of Tables 1 and 2 and averages (lines) excluding the encircled values which are calculated from the octave band results from [9].

Large and small standard deviations are observed for the full range of single-number quantities (Fig. 5). Since no general trend is observed it is appropriate to average all in-situ standard deviations according to equation (1). The averaged in-situ standard deviations are represented by lines in Figure 5. This new analysis confirms the value from ISO 12999-1 [3] for the single number quantity without a spectrum adaptation term (Tab. 3). When C_I is included, the in-situ standard deviation is slightly increased to 1.1 dB whereas an inclusion of C_{I, 50 − 2500} increases the in-situ standard deviation to 1.4 dB. This is due to the larger uncertainties at low frequencies observed in clause 2.3.

The described procedure had been applied in the past to determine all the standard deviations for single-number quantities in ISO 12999-1. As an alternative to that method one could also calculate the uncertainty of the single-number quantity from the uncertainty in one-third octave bands. Nevertheless, this would require an assumption on the correlation between the one-third octave bands which is not available. This approach is therefore not pursued in the course of this paper.

The same analysis is performed for the standard deviation of repeatability σ_r (Fig. 6). Here, the current values from ISO 12999-1 [3] of 0.5 dB and 0.6 dB for single-number quantities without and with the spectrum adaption term C_I are confirmed. There is no data set available for the calculation of the standard deviation of repeatability for single-number quantities including the spectrum adaption term C_{I, 50 − 2500}.

Figure 6. Standard deviation of repeatability σ r of the measured single-number quantities (dots) from all interlaboratory tests of Table 1 with n > 1 and averages (lines).

3 Impact noise levels under reproducibility conditions

3.1 Data overview

To the knowledge of the author no interlaboratory test on the impact noise of a floor has ever been conducted. It is to be emphasised here that such tests could be performed with any floor type as long as it is a whole floor and not just a floor covering. In lack of such interlaboratory results it was decided to use impact noise levels measured with reference floors defined in ISO 10140-5 [6]. This data set is then assumed to be obtained with nominally identical test specimens which is at least a good approximation of an interlaboratory test.

CEN/TC 126 “Buildings Acoustics” maintains a list of laboratories which may be contacted in case an interlaboratory test is performed. These laboratories and some others to which the author has personal contacts were asked whether they could provide measured normalised impact noise levels for the reference concrete slabs and for the lightweight reference floors defined in ISO 10140-5 [6]. 20 laboratories kindly provided such results. Altogether three independent results are available for the lightweight reference floor C1 and 20 independent results for the heavyweight reference floor. No results were provided for the reference floors C2 and C3. Five laboratories could also provide the loss factor measured on the heavyweight reference floor. All these results are single results without any repetition. If a lab delivered more than one result, e.g. because it uses more than on reference floor or more than one test facility for the same reference floor, one result of these was randomly selected. It is thereby ensured that all the results are statistically equal.

Whereas the lightweight reference floor C1 is defined precisely, the requirements for the heavyweight reference floor are less restrictive [6]. It should be a homogeneous concrete slab, have a uniform thickness between 100 mm and 160 mm and a size of at least 10 m². The information on the thickness, size, mass density and the volume of the receiving room was also provided by the laboratories. It could thus be verified that all provided results with the heavyweight reference floor are yielded in full compliance with the specifications of the standard.

In view of the tolerated spread of the parameters it is understandable that the normalised impact noise levels of the heavyweight reference floor are within a span of 15 dB (Fig. 7). Whereas some modal effects are observed at lower frequencies, the individual results are more smooth at higher frequencies.

Figure 7. Normalised impact noise levels of heavyweight reference floors measured in 20 laboratories and of the lightweight reference floor C1 measured in 3 laboratories.

For measurements in building acoustics, the three results for the lightweight reference floor are in excellent agreement (Fig. 7) even though all three specimens were realised independently. This is also a proof that the lightweight reference floor is well described by ISO 10140-5 [6].

3.2 Analysis for single-number quantities

To check whether systematic effects could be removed from the data set with the heavyweight reference floor, an analysis of the single-number quantities is performed. The weighted normalised impact noise levels cover a span of 9 dB but show no dependency on the surface mass (Fig. 8). This result is highly unexpected since it is basic acoustic knowledge that the impact noise level should be reduced when the surface mass is increased. When the correction for the loss factor is applied to the subset of the data where the loss factor is available, the correlation with the surface mass does not get better (Fig. 8). Furthermore, the weighted impact noise levels of the heavyweight reference element turned out to be independent of the specimen size, the volume of the receiving room and the tapping machine type.

Figure 8. Weighted normalised impact noise level as a function of the surface mass for the heavyweight reference floor with and without the correction for the loss factor.

In a next step, the standard deviations of reproducibility were calculated for the different single-number quantities. It is to be pointed out here that the whole data base is very limited, especially when considering that there are just three results for the lightweight reference floor. Nevertheless, since no larger data base is available, the analysis is performed all the same.

For the weighted impact noise level, the standard deviation of reproducibility σ_R is 0.93 dB for the lightweight reference floor (Tab. 4). For the concrete slab, the value is 2.08 dB and thus much larger. When only the five results are used where a loss factor is available, σ_R reduces slightly to 1.83 dB when no loss factor correction is applied. Applying the loss factor correction reduces this value further to 1.42 dB. When the same reduction is assumed to happen to the results from all 20 concrete slabs, σ_R is 1.61 dB.

All the values change slightly when the spectrum adaptation terms are included. Sometimes σ_R is increased, sometimes reduced (Tab. 4).

To come to a typical standard deviation of reproducibility, the values are usually averaged over the different specimens. The result of this averaging is given in Table 5 for two different cases. The first is that the concrete slab results are used as reported, i.e. without a correction for the loss factor. The mean σ_R is then 1.61 dB for the weighted impact noise level which is close to the estimated 1.5 dB in the current ISO 12999-1. When the assumed reduction of σ_R by correcting for the loss factor is taken into account for the concrete slabs, the mean σ_R is reduced to 1.31 dB (Tab. 5). Including the spectrum adaptation term leads only to minor changes of these values.

It is at this point difficult to decide which value to propose for a revised ISO 12999-1 since both results are calculated in a transparent and reasonable way. To come to a decision, the overall picture is considered. In general, condition (2)

$$\begin{array}{c}\hfill {\sigma }_{\mathrm{R}}>{\sigma }_{\mathrm{situ}}>{\sigma }_{\mathrm{r}}\end{array}$$(2)

should be fulfilled for all single-number quantities. A comparison between Tables 3 and 5 reveals that this is for L_n, w + C_{I, 50 − 2500} only the case when the results without the loss factor correction are taken into account. Therefore, the proposed typical σ_R-values for ISO 12999-1 are 1.6 dB, 1.7 dB and 1.7 dB for L_n, w, L_n, w + C_I and L_n, w + C_{I, 50 − 2500}.

3.3 Data analysis in one-third octave bands

The analysis in one-third octave bands is performed by calculating the standard deviation of reproducibility for the lightweight and the heavyweight reference floors where no correction is applied to the latter. Both standard deviations differ considerably (Tab. 6, Fig. 9). Up to frequencies of 1.2 kHz, the result for the heavyweight reference floor is larger than for the lightweight floor. At higher frequencies, both results are much closer.

Figure 9. Standard deviation of reproducibility for the heavyweight and lightweight reference floors, average value and smooth approximation.

It is to be discussed at this point whether different typical standard deviations for lightweight and heavyweight floors or a common average should be derived. Here, one has to keep in mind that such typical values are used by laboratories as estimates for their uncertainties when measuring the impact noise levels of all kinds of floor constructions. The author is of the opinion that using the average fits this purpose much better than using different standard deviations for heavyweight and lightweight structures. This is because the uncertainties become more generic by averaging. For instance, the lightweight floor seems to exhibit some modal behaviour at 250 Hz. This happens to a different extent in the three laboratories which leads to a peak in the standard deviation at 250 Hz (Fig. 7). So, this particular peak in the standard deviation is linked to this particular test object measured in the three particular laboratories. It is very likely that a data set with other lightweight floors or other laboratories will exhibit other particularities. A further disadvantage of using different standard deviations is that it must specified which standard deviation is to be used for which kind of floor construction. Based on today’s knowledge it is to the author’s opinion in most cases impossible to come here to an informed decision. The only exceptions to this are the measurement of the heavyweight reference floor and the lightweight reference floor C1.

So, to come to a typical general value, both results are averaged according to equation (1). Since this average is based on just two specimens it exhibits some peaks and dips as explained above (Fig. 9). If there are many results for different test objects in different laboratories available, the average value of the standard deviation automatically becomes a smooth curve. This is clearly observed in Figures 3 and 4. Lacking enough measurement data for impact noise levels under reproducibility conditions, a smooth curve is fitted to the average value which does not follow all the particularities of the average. The smooth curve is derived by defining the shape of the curve starting with a decline of −0.5 dB per one-third octave band at the lowest frequencies. This decline is then reduced until a plateau is reached between 400 Hz and 1.25 kHz. For higher frequencies, the slope is increased until it reaches a slope of 0.2 dB per one-third octave band between 2 kHz and 5 kHz. With this general shape, the whole curve is shifted, and the sum of the squared deviation to the averaged reproducibility standard deviation is minimised. It turned out that this sum does not exhibit a sharp minimum but a rather broad minimum. To be on the safe side for the typical uncertainty, the final proposal displayed in Figure 9 is found by using the minimum +0.2 dB. This smooth curve is now the proposal for the typical standard deviation of reproducibility for the normalised impact noise level.

How do these values affect the accreditation of laboratories? Laboratories can demonstrate that the smooth standard deviation of reproducibility (Tab. 6) is a reasonable estimate for the uncertainty of the normalised impact noise level of all different kinds of test specimens. For the heavyweight reference floor and the lightweight reference floor C1, they may also use the corresponding values given in Table 6. But they can not be asked that the normalised impact noise level of their reference floor (lightweight C1 or heavyweight) shall be within certain limits to a defined normalised impact noise level since the latter is nowhere defined. The reference curves given in ISO 717-2 [7] are used for calculating the single-number values and not for defining a requirement for test facilities. It is to be pointed out that the reference floors are not designed to have the same normalised impact noise level. Their purpose is to enable the measurement of impact noise reductions of floor coverings. They serve this purpose sufficiently well as will be shown in the following clause.

4 Impact noise reduction

4.1 Available data

For the measurement of impact noise reduction according to ISO 10140-1 [26] annex H, a number of interlaboratory tests with different specimens had been performed in the past (Tab. 7). All these measurements were made on the heavyweight reference floor defined in ISO 10140-5 [6]. Mostly, PVC had been used but there are also results for a carpet, a parquet, a laminate with two different underlays and a laminate reference board with 8 different underlays available. Two of the data sets include also repeated measurements. The results from published sources are complemented with results from 10 different PVC-floorings which had been used for internal investigations at PTB.

The impact noise reduction averaged over all participants of each interlaboratory test shows the typical behaviour of locally reacting floor coverings (Fig. 10). Up to a certain cut-off frequency, the measured impact noise reduction is constant with typical values between 1 dB and 3 dB. This is caused by the change of the restitution coefficient when impacting the floor with and without covering [33]. Above the cut-off frequency which depends on the stiffness of the tested floor covering, the impact noise reduction increases with frequency. At very high frequencies and impact noise reductions, a plateau is observed which is probably caused by the background noise influence. An exception from this general shape is found for the parquet at low frequencies. Here, a negative impact noise reduction is detected which is typical for a mass-spring system at its resonance frequency. Such a behaviour would also be expected for floating floors which are unfortunately not included in the data base.

Figure 10. Average values of the impact noise reduction for each individual data set of Table 7.

4.2 Analysis in one-third octave bands

The standard deviation of reproducibility calculated from the different interlaboratory tests reveals some scatter (Fig. 11). At medium frequencies, the PVC-floorings tend to have a smaller value than the other test objects. A relatively large standard deviation of reproducibility is found for the carpet. From the very large impact noise reduction of the carpet (Fig. 10) it is clear that the carpet is relatively soft. The measured impact noise reduction then depends on the time the tapping machine is acting on the carpet. This time had not been prescribed in the relevant interlaboratory test. For the hard flooring systems like laminate, parquet and the reference board, very small air cushions under the flooring system lead to large deviations in the impact but also in the walking noise [32]. This may lead to an increased standard deviation of reproducibility compared to the PVC.

Figure 11. Standard deviation of reproducibility for all data sets of Table 7.

From these considerations, two classes of test specimens are defined, the PVC and all other specimens. The other specimens comprise laminate (L), parquet (P), carpet (C) and the reference board where the latter is a special laminate. The abbreviation LPC is therefore used for these specimens. For both classes, a separate averaging was performed (Fig. 12). It is clearly seen that both averaged standard deviations are significantly different. For both classes, smooth approximations are also derived which consist of straight sections. Also shown in Figure 12 is the standard deviation of reproducibility standardised in today’s version of ISO 12999-1. It is very close to the PVC-value, since at the time when this curve was developed, only the data from [27, 28] had been available together with some internal data from PTB. Since both classes of specimens clearly show a different behaviour, it is proposed to include the two smooth approximations in a revised ISO 12999-1.

Figure 12. Standard deviation of reproducibility, average for all data sets with laminate, parquet, carpet and reference board (LPC) from Table 7, average for all data sets with PVC from Table 7, smooth approximations and value currently standardised in ISO 12999-1.

The data base also allows to calculate a standard deviation of repeatability for PVC. These values are much smaller than the standard deviation of reproducibility (Fig. 13 compared to Fig. 11). Their average value is below one dB. From the altogether 13 standard deviations of repeatability, nine are from [30] and the remaining four from [27]. The test object used in [30] reveals a discontinuity in impact noise reduction around 800 Hz in most laboratories which seems to be difficult to be reproduced. This leads to an increased average standard deviation of repeatability at this particular frequency. Similar effects would occur at other frequencies for other test objects. The smooth approximation, which is proposed for ISO 12999-1, does not include the peak at 800 Hz since its aim is to describe the typical standard deviation of repeatability independent of the specific test object.

Figure 13. Individual standard deviations of repeatability from [27, 30], average value and smooth approximation.

4.3 Analysis for single-number quantities

The single-number quantities ΔL _w and ΔL _{lin, 100 − 2500} defined in ISO 717-2 [7] were calculated from the one-third octave band values. Their standard deviation of reproducibility is in the range between 0.2 dB and 1.7 dB (Fig. 14). There is no systematic dependence of the standard deviation of reproducibility from the single-number quantity observed. To characterise all the values by an average value thus seems to be appropriate. As for the one-third octave band data, the values for the PVC-specimens tend to be smaller than for the other specimens. Therefore, the averaging was performed for the two classes of specimen, i.e. PVC and LPC, separately. For the PVC, average values of 0.7 dB for ΔL_w and 0.6 dB for ΔL_{lin, 100 − 2500} are yielded. For LPC, the average values are 1.2 dB for ΔL_w and 1.1 dB for ΔL_{lin, 100 − 2500}. A distinction between the two classes of specimens seems to make sense since their average values are sufficiently different.

Figure 14. Standard deviation of reproducibility of ΔLw and ΔLlin, 100 − 2500 for all data sets of Table 7.

The standard deviation of repeatability can be calculated from data sets [27, 30] and thus for PVC-specimens only. The values cover the range between about 0.1 dB and 0.5 dB and average to 0.2 dB for both single-number quantities ΔL_w and ΔL_{lin, 100 − 2500}.

5 Conclusion

The typical standard deviations for impact noise levels derived from this investigation are summarised in Figure 15. The value for the in-situ standard deviation in today’s version of ISO 12999-1 is confirmed by this investigation whereas new values are proposed for the standard deviation of reproducibility and repeatability. For comparison reasons, Figure 15 also contains all the standard deviations from ISO 140-2 [2] which are the standard deviation of reproducibility σ_R for field and laboratory measurements and the standard deviation of repeatability σ_r for laboratory measurements. The situations A, B and C defined in ISO 12999-1 had not been clearly distinguished in that standard which complicates the comparison between these values and the new averages. However, the averaged in-situ standard deviation is between the σ_R – values for field and laboratory measurements and above the σ_r – value from ISO 140-2 which seems reasonable. The new proposal for the standard deviation of repeatability is slightly larger than the value from ISO 140-2 whereas the new proposal for the standard deviation of reproducibility σ_R is considerably larger than the values from ISO 140-2.

Figure 15. Typical standard deviations for impact noise levels derived from this investigation in comparison to the values from ISO 140-2.

It is furthermore interesting to compare the typical standard deviations for impact noise levels to the values for airborne sound insulation from ISO 12999-1 (Figs. 16, 17). At medium and high frequencies, the values for impact noise level measurements tend to be slightly larger than for airborne sound insulation measurements. This can be explained by the fact that the impact noise level measurement involves an absolute measurement of the sound pressure level whereas for airborne sound insulation only a sound pressure level difference is measured. A second reason is that the excitation is realised with different tapping machines under in-situ and reproducibility conditions which adds another possibly significant uncertainty component. It is furthermore to be noticed that at low frequencies the standard deviations for impact noise levels are smaller than for airborne sound insulation. It might be argued here that the lack of modal density is less pronounced for impact sound since only the structure-borne sound field of the specimen and the airborne sound field in the receiving room are involved whereas with airborne sound insulation also the airborne sound field in the sending room has to be considered.

Figure 16. Typical standard deviations of reproducibility derived from this investigation in comparison to the values for airborne sound insulation standardised in ISO 12999-1.

Figure 17. Typical standard deviations under repeatability and in-situ conditions derived from this investigation in comparison to the values for airborne sound insulation standardised in ISO 12999-1.

The values for impact noise reduction exhibit a different behaviour (Figs. 16, 17). The standard deviation of reproducibility is very small for the PVC specimens below 500 Hz and even close to the repeatability standard deviation for airborne sound insulation and impact noise levels. At these frequencies, the locally reacting floor coverings do not change the modal behaviour of the involved airborne and structure-borne sound fields which makes the measurement of the impact noise reduction more robust with respect to measurements in different test facilities. Towards higher frequencies, the standard deviation of reproducibility of the impact noise reduction increases considerably. This increase may be caused by an insufficient reproducibility of the coupling between the covering and the base floor. For laminate, parquet, carpet (LPC), the standard deviation of reproducibility for impact noise reduction is close to the in-situ standard deviation for impact noise levels up to 200 Hz (Fig. 16). Towards higher frequencies it increases drastically and is in the order of the value which would be expected for two independent measurements of impact noise levels under reproducibility conditions. The standard deviation of repeatability for impact noise reduction for PVC specimens is very small at low frequencies (Fig. 17) and starts to increase at 400 Hz. At higher frequencies it is in the same order of magnitude as the standard deviation of repeatability for the impact noise level and airborne sound insulation.

An overview on the proposed standard deviations for single number quantities is given in Table 8 in comparison to airborne sound insulation. For this comparison, the spectrum adaptation terms with the upper frequency limit of 3150 Hz are chosen since this gives the best match with the frequency ranges for C_I. The standard deviations for the impact noise levels are larger than for airborne sound insulation without a spectrum adaptation term and with the spectrum adaptation term C (Tab. 8). But when the spectrum adaptation term C_tr is used, standard deviations for airborne sound insulation are getting larger and sometimes reach a similar magnitude as for impact noise levels. This is understandable in view of Figure 16. The low frequencies are more dominant with C_tr and therefore the standard deviation increases. ISO 140-2 [2] did contain the general hint, that the standard deviation of repeatability is 0.4 dB, and the standard deviation of reproducibility is between 0.4 dB and 1.1 dB. These values were related to all single-number quantities measured in laboratories for airborne sound insulation and impact noise levels. Also in view of these values without any distinction between situations A, B and C, the new proposals seem to be reasonable.

Acknowledgments

I thank VMPA for a long and fruitful cooperation by which the access to extraordinary data bases on interlaboratory tests is assured. I further express my gratitude to all the laboratories that performed measurements and to all staff members from PTB’s working group “Applied Acoustics” and, in particular, to Martin Schmelzer for the careful review of the manuscript. I also like to mention the assistance of Anatoli Valodchenka in the analysis of the concrete slab data. Finally, I’m very grateful to the following persons that provided data by personal communication (in alphabetic order): I. Alexe, G. Bauerfeind, W. Beentjes, S. Bono, R. Brot, A. Cucchi, A. Dijckmans, G. Eßer, H. Ferk, E. Gerretsen, C. Guigou Carter, S. Güven, R. Hall, V. Hongisto, J. Lang, C. Lechner, P. Meistring, M. Merillas Fernández, E. Müller, B. Nusser, A. Petošić, B. Sass, J. Scheck, A. Schiavi, M. Schneider, M. Scrimali, C. Scrosati, C. Simmons, A. Smalls, L. Sondergaard, D. Sprinz, A. Terskan, A. Warnock, A. Worch, B. Zeitler.

Conflicts of interest

The author declares no conflict of interest.

Data availability statement

Data are available on request from the author.

References

All Tables

All Figures

	Figure 1. Normalised impact noise levels measured under in-situ conditions from VMPA-comparison measurements (for details see Tab. 2).
In the text

	Figure 2. Mean values of the measured normalised or standardised impact noise levels from all interlaboratory tests of Tables 1 and 2.
In the text

	Figure 3. In-situ standard deviation of the measured normalised or standardised impact noise levels from all interlaboratory tests of Tables 1 and 2, average with and without octave results and in-situ standard deviation from ISO 12999-1 [3].
In the text

	Figure 4. Standard deviation of repeatability of the measured normalised or standardised impact noise levels from interlaboratory tests [10, 11, 13, 21], average with and without encircled data points and standard deviation from ISO 12999-1 [3].
In the text

	Figure 5. In-situ standard deviation of the measured single-number quantities (dots) from all interlaboratory tests of Tables 1 and 2 and averages (lines) excluding the encircled values which are calculated from the octave band results from [9].
In the text

	Figure 6. Standard deviation of repeatability σ r of the measured single-number quantities (dots) from all interlaboratory tests of Table 1 with n > 1 and averages (lines).
In the text

	Figure 7. Normalised impact noise levels of heavyweight reference floors measured in 20 laboratories and of the lightweight reference floor C1 measured in 3 laboratories.
In the text

	Figure 8. Weighted normalised impact noise level as a function of the surface mass for the heavyweight reference floor with and without the correction for the loss factor.
In the text

	Figure 9. Standard deviation of reproducibility for the heavyweight and lightweight reference floors, average value and smooth approximation.
In the text

	Figure 10. Average values of the impact noise reduction for each individual data set of Table 7.
In the text

	Figure 11. Standard deviation of reproducibility for all data sets of Table 7.
In the text

	Figure 12. Standard deviation of reproducibility, average for all data sets with laminate, parquet, carpet and reference board (LPC) from Table 7, average for all data sets with PVC from Table 7, smooth approximations and value currently standardised in ISO 12999-1.
In the text

	Figure 13. Individual standard deviations of repeatability from [27, 30], average value and smooth approximation.
In the text

	Figure 14. Standard deviation of reproducibility of ΔLw and ΔLlin, 100 − 2500 for all data sets of Table 7.
In the text

	Figure 15. Typical standard deviations for impact noise levels derived from this investigation in comparison to the values from ISO 140-2.
In the text

	Figure 16. Typical standard deviations of reproducibility derived from this investigation in comparison to the values for airborne sound insulation standardised in ISO 12999-1.
In the text

	Figure 17. Typical standard deviations under repeatability and in-situ conditions derived from this investigation in comparison to the values for airborne sound insulation standardised in ISO 12999-1.
In the text