ASHRAE OR-05-6-2-2005 What Did We Learn From ASHRAE RP-879 《我们从ASHRAE RP-879中学到了什么》.pdf

《ASHRAE OR-05-6-2-2005 What Did We Learn From ASHRAE RP-879 《我们从ASHRAE RP-879中学到了什么》.pdf》由会员分享，可在线阅读，更多相关《ASHRAE OR-05-6-2-2005 What Did We Learn From ASHRAE RP-879 《我们从ASHRAE RP-879中学到了什么》.pdf（11页珍藏版）》请在麦多课文档分享上搜索。

1、OR-05-6-2 What Did We Learn from ASHRAE RP-879? Norm Broner Member ASHRAE ABSTRACT Over the last decades, there has been an increase in the incidence of complaints of “rumble noise ”due to the excessive acoustic energy below 250 Hz in HVAC systems. ASHRAE sponsored some research to jrst document the

2、 extent and degree of low frequency noise problems and, second, to deter- mine by means ofpsycho-acoustic testing, a method of assess- ment of such noise. A seconda y goal of the research was to determine, if possible, information that could be included in the ASHRAE handbook in terms of metrics and

3、 acceptable levels. This paper reports on some of the results obtained in this study with respect to assessment metria and with respect to criteria and acceptable levels. INTRODUCTION Over the last decades, there has been an increase in the incidence of complaints of“rumb1e noise” due to the excessi

4、ve energy below 250 Hz in HVAC systems. ASHRAE therefore sponsored a research study to first document the extent and degree of low frequency noise problems and, second, to deter- mine by means of psycho-acoustic testing a method of assess- ment of such noise. The objective phase ofthe study by Brone

5、r (1994) documented over 70 samples of HVAC noise at sites in North America, Hong Kong, London, and Melbourne. It also suggested that three factors are important in determining the subjective response of people to low frequency HVAC noise. These, not necessarily in order of importance, were overall

6、level, spectral imbalance, and amplitude and temporal modu- lation effects. Psycho-acoustic testing to investigate these parameters with a goal of determining the most appropriate low frequency metrics for assessment of low frequency HVAC noise was also recommended. Phase 2 of the research involved

7、psycho-acoustic testing of subjects (Broner 2004). A secondary goal of the research was to determine, if possible, information that could be included in the ASHRAE handbook in terms of metrics and acceptable levels. Noise stimuli for use in the testing were chosen from and based on the measured stim

8、uli collected in the Phase 1 study. To assess and rate the psychological attributes of loudness and annoyance of the noise stimuli, the magnitude estimation task was used to rate the noise stimuli. Assessments of relief-on-cessation of the stimuli and un- acceptability were also collected. This pape

9、r reports on some of the results obtained in this study and includes recommen- dations in relation to criteria and acceptable levels. NOISE STIMULI Ten base stimuli were drawn from the noise recordings of over 70 HVAC noises that were collected during Phase 1 of the ASHRAE-sponsored research. Two sa

10、mples were “neutral” (spectra with no specific spectral characteristics and with slopes close to -5dB/octave), two “neutral/marginal” (spectra with slight deviations from neutral), three “rumble” (spectra with some degree ofrumble), and three “strong rumble” (spec- tra with significant low frequency

11、 energy relative to neutral). The samples were converted to a “.wav” format and additional level and temporal shaping was conducted using the Sound Forge sound processing software so that various shapes and overall SPLs were attained. A total of 60 HVAC noise stimuli were thus created from the base

12、samples. The reproduction chain included Sound Forge, an ASHLY parametric graphic equalizer (Model #SC-63), and a playback amplifier. Finally, the fully processed signal was sent to the test room speakers. A functional diagram of the test equipment is shown in Figure 1. - - Norm Broner is a building

13、 acoustics and architectural and environmental acoustics consultant at Vipac Engineers (a) spectrum when SPL was maximum, (6) spectrum when SPL was minimum. Note the sign$cant level variation (up to 20 dB) below 50 Hz. annoyance but to focus on the ratio of annoyance to loudness, i.e., the AiL ratio

14、. It was felt that this measure of subjective assessment would be sensitive to low frequency noise because low frequency noise has been known to create annoyance while not being particularly loud. Thus, we would expect a higher AiL ratio for low frequency noise than for noises with dominant energy a

15、t higher frequencies. Subjects were also asked to (I) indicate whether they would feel relief if they had been listening to the given noise stimulus all day and it had been tuned off at the end of the day and (2) judge the acceptability of each stimulus. They were also allowed to provide additional

16、comments regarding each noise stimulus. A typical response sheet is shown in Figure 5. SUBJECT DEMOGRAPHICS Subjects were generally office staff with a few students. All reported good hearing. When filling out the rating sheet, most subjects did not supply additional comments on the sounds. All subj

17、ects seemed to be able to follow the test instructions. The tests and test subjects are described in Table 1. 662 ASHRAE Transactions: Symposia Figure 3 Sound pressure vs. time for part of Sequence C run. Final Main Study Figure 4 Subject rating sound stimuli in test room. Each subject rated 4 x 45

18、stimuli 2 1 x 4 x 2 x 15 ratings used in analysis Relief and acceptability also rated ANNOYANCE VS LOUDNESS- IS LOUDNESS THE ANSWER? 3 4 6 7 Figure 6 shows the group mean annoyance versus loud- ness for the 60 main study noise stimuli. It appears that these are highly correlated; however, it should

19、be noted that the annoyance is increasing at a faster rate than the loudness, i.e., the difference between loudness and annoyance is not constant. Clearly, the loudness alone does not account for the annoyance. Thus, it is important to seek further evidence of what is happening. A clue can be obtain

20、ed from Figure 7 that shows the NL ratios versus the A-weighted SPL for the 60 main study noise stimuli. It can be seen that the correlation is nearly nonexistent and that the NL ratios can be quite varied for very similar SPL(A) values. This implies that there are other issues of importance in dete

21、rmining the perceived annoyance apart from the loudness itself. 3 5 5 6 4 5 * 5- 3“ * d c * Table 1. Subject and Test Details 6 3 7 12 1- 2 1 x 9 ratings used in analysis 5 6 5 8 6 7 3 2 Relid I Accepmhle Co ce Yes I No I Yen I No I E I*) I 2 I3 1.1 I 18i l*l 01 I c It1 I “CI 91 I 4 I1 I:! I I 41 7

22、I8 l*l I 31 3 I5 l*l .-.I I 1.1 I 14 4 5 5 3 5 2 15 10 12 I 3 I. I5 1.1 “1 I 5 L a l*l I I , II I I I 13 I 8 I9 l*l I I*I I2 1.1 I I! I 5 6“ I“ 11 7 8 I* 5 2 3 4 14 I* 5 I I10 1.1 I I 9 I c 1.1 I I II 41 5 I7 ILI I If1 c f It1 I Figure 5 Sample subject rating form for Sequence A. ANNOYANCE PREDICTIO

23、N METRICS So, if loudness alone doesnt determine the annoyance response, can we find a noise metric that will more adequately predict annoyance? We investigated linear regression one-metric annoyance models and estimated parameters (a, and al) and the R2 statis- tic (explanation of variance) for all

24、 35 metrics calculated. The results show that for this set of data, loudness is a dominant factor, explaining over 90% of the variance of the subjects ASHRAE Transactions: Symposia 663 20 O 15 1.0 O. 5 o. o .- e -AIL = 1.0 m 2 4 O 10 20 30 40 50 LoudnecsRating Figure 6 Correlation of annoyance with

25、loudness for the 60 main study noise stimuli. response data. The results also show that level-related issues are important. The regression models were used to generate predictions of annoyance, and these are plotted versus the geometric aver- age of the subjects responses in Figure 8. Also shown in

26、Figure 8 are the predictions from using the psychoacoustic annoyance model (PA), which is a nonlinear function of the sound quality descriptors of roughness, sharpness, N5 (the loudness exceeded for 5% of the time), and fluctuation strength. The PA model predictions had to be scaled by a factor of c

27、lose to two and a half to lie on the same scale as the measured annoyance. As can be seen from the R2 statistics in Figure 8, the PA model, while giving reasonable predictions, does not in this instance capture the annoyance behavior as well as the simple regression models. As it was felt that the d

28、ifference between a measure of mean sound level and some measure of the maximum would provide an indication of the low frequency fluctuation char- acteristics, we also ran a (Cmax - A) correlation and the result was very poor (R2 value was very small). A combination of Cmax and A was also tried. Thi

29、s combination is more flexible than fixing (Cmax-A) as a metric and does do better (the linear regression model was -42.38 - O. 16 Cmax + 1.62 dBA with a R2 value of 0.87); however, this combination did not appear in the list of top models. From the results, it is clear that fluctuation at low frequ

30、en- cies is important for annoyance as the more annoying samples had large fluctuation in C but small variation in A. Basedon the above and considering the test methodology, it appears that in this research study, loudness was just too dominant in the short listening test and that the influence of o

31、ther variables on annoyance did not have time to develop fully. Future testing should therefore focus on fluctuation and spectral balance separately at different levels of loudness. 30 35 40 45 50 55 60 SPL (dBA) Figure 7 A/L ratios vs. A-weighted SPL for the 60 main study noise stimuli. * I R=W-Mod

32、all Nmn R2 -0.92- Model 2. Nmn Amax 40 45 50 Figure 8 The predictive performance of the annoyance models. Largest improvement is om the one- metric to the two-metric model. Psychoacoustic annoyance (PA) model performance is also shown. DISCUSSION The results as described above initially seem to indi

33、cate that the loudness-based metrics are very appropriate for predicting and assessing sounds with a significant low frequency characteristic. However, it is now clear that for the low frequency noise stimuli used in this study and typical of HVAC sounds in many occupied spaces, the subjects could n

34、ot adequately distinguish between loudness and annoyance based on the ten-second stimulus duration. The earlier pilot study conducted in this phase of the research seemed to initially indicate that a ten-second stimulus duration would be adequate to allow subjects to differentiate between loudness 6

35、64 ASH RAE Transactions: Symposia and annoyance. On review, the data seem to show that the adaptation rates for loudness and annoyance are different at low sensation levels and that they diverge with time. Further, it seems that a ten-second stimulus duration is not adequate in allowing for the diff

36、erentiation that begins to occur only after 30-60 minutes stimulus duration. On the other hand, it appears that more ?tonal? stimuli do allow subjects to differentiate between annoyance and loudness much more readily. For HVAC sounds with low frequency content, it would seem that the subjects cannot

37、 differentiate between loudness and annoy- ance based on a ten-second sample duration and that a much longer stimulus duration (30-60 minutes minimum) is required for the effect of low frequency noise to manifest itself. Empirically, we know that low frequency sounds of the type used in this researc

38、h study do create significant annoy- ance in the field. We know that the loudness is only one issue in the response and that the spectral character and level and time fluctuations are significant in resulting in an annoyance response. The use of RMS ?average? sound pressure levels in metrics simply

39、does not work. If we are ever to develop a meaningful and acceptably reliable rating methodology for low frequency noise, then the process must in some way take into account the time-amplitude statistics of the low frequency noise. What is needed to improve the low frequency noise assessment capabil

40、ity is to combine some measure of loudness with measures of spectral balance and signal level fluctuation and also a measure that takes into account the rate of change of the signal level (i.e., the time period over which the level fluctuation is occurring). However, based on the data at hand, none

41、of the existing metrics that quantify level and time fluc- tuation and spectral balance appear to be the right choice to include in the annoyance model. Also, the perception ofloud- ness does not appear to be fully modelled by N5 (Zwicker loudness exceeded 5% of the time) for the low frequency fluc-

42、 tuating sounds in this study, though they are highly correlated (R2 = 0.91). With the exception ofthe psychoacoustic annoy- ance metric (which is a function of N5, sharpness, fluctuation strength, and roughness), we had focused on linear regression models of metrics to predict annoyance. Note that

43、the PA model does not include any measure of ?tonality,? which is known to also contribute to annoyance. It may be interesting to look at subgroups of the sounds to see if different annoyance models might work better for different subsets. This would indicate whether an underlying nonlinear model of

44、 metrics is more appropriate to predict annoyance rather than the linear regression models that we used in our analysis. CRITERIA FOR HVAC An important issue that was addressed was whether any information could be gleaned from the results of this research in terms of the design guidelines for HVAC-r

45、elated back- ground sound in rooms quoted in Chapter 47, ?Sound and Vibration Control,? in the ASHRAE Handbook-Applications volume. Table 2 shows the subjective ratings including the A/L ratios versus the RC and QAI values for the noise stimuli in all four sequences. The RC room criterion curves wer

46、e developed by Blazier (1 98 1) and considered the average level of HVAC noise in the frequency range for good speech communication and also provided some information about the presence of a spectrum imbalance that might be annoying to the listener. The Quality Assessment Index (QAI) was developed b

47、y Blazier (1 995) based on the Phase 1 results by Broner (1 994). The QAI provided a method for determining the magnitude of the spec- trum imbalance in a given spectrum and is calculated over three frequency ranges (low, mid, and high) (see also Blazier 1997 and ASHRAE 2003). It can be seen that A/

48、L ratios above 1.2 occur when the QAI exceeds 25. When the RC value is low e., less than 25) and the QAI is high (i.e., greater than 25), then the AiL ratio is increased. Thus, one cannot just look at the absolute RC value alone. One has to also look at the sound quality in terms of the QAI. Table 3

49、 shows the rank ordering of the 60 stimuli in order of descending A/L ratio. The stimuli at the top of the table were judged more annoying than loud and are seen to be based on the strong rumble and rumble source noise stimuli. The least annoying stimuli at the bottom of the table were seen to be based on neutral, marginal neutral, or rumble stimuli with low RC and QAI values. Note that the NC and NCB ratings are not included in this table. The NC curves only extend down to the 63 Hz octave- band, thus not encompassing the full range of frequencies of interes