Dynamic Measurements of Wind Turbine Acoustic Signals, Employing

Portland, Oregon NOISE-CON 2011 2011 July 25-27

Dynamic measurements of wind turbine acoustic signals, employing sound quality engineering methods considering the time and frequency sensitivities of human perception

Wade Bray HEAD acoustics, Inc. Brighton, MI 48116 [email protected]

Richard James E-Coustic Solutions Okemos, MI 48805 [email protected] Our purpose is to present, in the context of the time and frequency resolutions of the human hearing receiver and the methods of sound quality engineering, some phenomena of wind turbine signals which we measured, for information, consideration and further research.

We will concentrate on low frequencies.

No conclusions regarding audibility or other physiological effects are drawn.

Further research and discussion are welcomed.

This presentation complements the published paper. 1.

Many wind turbine acoustic measurements are made over long, or very long, time scales – “macro time scale.”

Most sounds, including wind turbine signals, convey important structure, magnitude and information to the human receiver at “micro- time scale,” so should also be measured at the time/frequency resolutions of human hearing according to the well-established practices of sound quality engineering and Soundscaping.

In general acoustic measurement, different time and frequency resolutions are widely used to reveal different kinds of signal behavior. The resolutions should always include, even center on, those most relevant to the human hearing (ear/brain) system.

There is, of course, a challenge in combining macro- and micro-time- scale findings, but it is worthwhile to do so. We will suggest some possibilities. 2.

The time and frequency resolutions of human hearing; appropriate measurement criteria, and implications (low frequency): Measurement bandwidths

Compared bandwidths (on log Hz scale): 1/3-octaves (upper), critical bands (middle) and some Equivalent Rectangular Bands (ERB; calculated from 30.5 Hz to 556 Hz, lower). The 1/3-octaves chart also shows the synthesis of the lowest three critical bands in ISO 532B, DIN 45631-1991 and DIN 45631/A1 from multiple 1/3-octave bands. Impulse responses of measurement bandwidths

From top to bottom: critical bandwidth centered on 50 Hz (Bark 1, approximate LF hearing “transducer” time response); ERB on 50 Hz, 20 Hz; ANSI S1.11 1/3- octave on 50 Hz, 20 Hz. Results are identically scaled and from the same analytic signal. The time response of low-frequency human hearing (represented by the critical bandwidth Bark 1) is very short: most magnitude response occurs within about 10 milliseconds. Please see Appendices 1 and 2. Different crest factors, identical Leq

Levels vs. time of a 90 dB[SPL] synthetic band-limited low-frequency signal with controllable crest factor (pink pseudo-noise 10Hz – 100Hz): the subjective impressions and apparent loudnesses differ despite identical Leq values. Measurements are at three time-weightings: 10 ms (green, approximate low- frequency-hearing-equivalent), Fast (red, 125 ms), and 1 second (black). Bandwidth, density level, ∆f vs. ∆t

• For time-domain signals passed through band-pass filters, the measurement bandwidth affects both time and frequency resolutions. Wider bandwidth yields better time resolution but poorer frequency resolution, and vice versa. • The criteria of measuring simultaneous (ear-appropriate) frequency and time resolutions become particularly at odds with each other with low-frequency signals. • For broadband signals, wider bandwidths contain more power and hence show higher band-levels than the band-levels of subdivisions of the same frequency span into narrower bandwidths (the phenomenon of density level). Selecting a wider low-frequency hearing-appropriate measurement bandwidth preserves important hearing-appropriate time resolution, yielding a more correct representation of perceptible crest factors. Measurement in time domain vs. frequency domain

In-cabin Diesel idle, band-limited through a Bark 1 filter (critical bandwidth centered on 50 Hz). Left: level vs. time weighted by the Bark 1 impulse response (∆t ~10 ms). Right: average spectrum (FFT, ∆f = 2.69 Hz, ∆t = 372 ms.) The average values calculated in time and frequency domains, ~69 dB[SPL], are effectively identical. Loss of crest factor, and signal misrepresentation, due to narrow analysis filter

The Diesel-idle time-signal passed through a Bark 1 band-pass filter (upper) and an ANSI S1.11 1/3-octave band-pass filter (lower), both centered at 50 Hz. The narrow filter reduces the crest factor, erasing short-duration level variations of the signal. Note the signal-establishment delay in the narrow-filtered result, due to the long impulse response. Cause of the “scanning” rather than “integrating” hearing sensation below approximately 50 Hz (more study required)

Impulse response of Bark 1 (upper, the lowest critical bandwidth of human hearing), and a portion of the in-cabin Diesel-idle time-signal (lower), on the same scale. The duration of the principal magnitude of the impulse response (about 10 ms), and the duration of individual time-signal peaks, are very similar. Deliberate “misuse” of modulation analysis to model low- frequency pure-tone perception

Upper: 30 Hz 85 dB[SPL] steady sine (blue) and its envelope (red), both through a 1-critical-bandwidth band-pass filter centered at 50 Hz (Bark 1). Lower: the “modulation” spectrum (based on 60 Hz due to two pressure events per period). 3. Wind turbine research: test site and local conditions (field test December 17-18, 2009 near Ubly, Michigan) • Utility: Michigan Wind 1 (46 GE 1.5 MW SLE wind turbines) – Nearest machine ~1500 feet from test location (a residence). • Weather – Wind speed <10 mph from start of test, afternoon December 17 through noon December 18; wind consistently from southeast. – Temperatures: 17 to 26 deg. F. (mean 22 deg. F.). – Relative humidity 64% – 86%; no precipitation. – Skies overcast. • Instrumentation – HEAD acoustics HMS IV artificial head binaural measurement system: • ID-equalized (Independent-of-Direction: resonances component of head- related transfer function compensated, only). • Microphones: ½-inch 54 mV/Pa. • Head wore wind muffs whose spectral influence was >2.5 kHz and compensated in equalization. • Head outside residence on tripod at approximately 5 foot elevation above ground, approximately 50 feet from residence. 105 minutes:

Peak

RMS

Crest factor in dB

(As in previous slide, lower) 105 minutes starting 10:15 PM 12/18/2009: Hearing Model spectrum vs. time at 1/3-Bark resolution (upper), 1/3-octave spectrum vs. time (lower). Much finer time-detail exists than can be seen in this 105-minute view (1 horizontal screen-resolution element = ~6 seconds. See next slide for a 1- minute-segment by the same analysis). ∆t = ~30 ms, ∆f = 1/3 Bark.

REC0002-1015-1215 (60.00-6182.00 s).Hearing Model 0.33 Bark Res. 0.50 df.Avg (). f/Hz 10k 2k 500 100 20 1000 2000 t/s 4000 5000 6000 30 40 50L/dB[SPL] 70 80 90 REC0002-1015-1215 (60.00-6182.00 s).3rdOctave vs. Time (10.0ms).Avg (). f/Hz 10k 2k 500 100 20 1000 2000 t/s 4000 5000 6000 30 40 50L/dB[SPL] 70 80 90 One minute (minute 30) showing fine time detail not resolvable due to pixel size and number in previous slide (not a matter of analysis but of representation – here, 1 horizontal resolution element = 23 ms. Longer sequences cannot fully show rapid ear-relevant time structure).

30 (1740.54-1800.56 s).Hearing Model 0.33 Bark Res. 0.50 df.Avg (). f/Hz 10k 2k 500

100

20 1750 1760 t/s 1780 1790 1800 30 40 50L/dB[SPL] 70 80 90 30 (1740.54-1800.56 s).3rdOctave vs. Time (10.0ms).Avg (). f/Hz 10k 2k 500

100

20 1750 1760 t/s 1780 1790 1800 30 40 50L/dB[SPL] 70 80 90 Minute 30, G-weighted sound pressure level vs. time at 10 ms time weighting (green), 1-second (red). Leq indicated by blue line. Audition or its likelihood is more associated with near-peak values, and pattern strength (amount of level change), than with average values. For constant Leq it also varies with crest factor.

30 (1740.54-1800.56 s).Level vs. Time (1.000).Avg (). L/dB(G)[SPL] 100 94 dB[G]

85 dB[G] 90

80 77 dB[G] 70

Average : 77.0 dB(G)[SPL] Average : 77.0 dB(G)[SPL] 40 1750 1760 t/s 1780 1790 1800 Minute 30, overall unweighted power spectral density (read leftmost ordinate scale): peak-hold (brown), average (dark blue), along with weighting curves A (red), C (light blue), G (green) over a 105 dB dynamic range. ∆f = 0.73 Hz, ∆t = 1.37 sec.

NOTE: For signals with steep spectral tilts rising toward LF, the A-weighting appears inappropriate. For this reason a major European automaker has long banned it for vehicle-interior measurements. Minute 30, power spectral density: peak-hold (left), average (right). Unweighted (dark blue), G-weighted (light blue), C-weighted (red), A-weighted (green). Crest factors range from 7.9 to 10.8 dB; the unweighted average slope between 5 Hz and 100 Hz (i.e., most of Bark 1) is about 10 dB/octave. Minute 30 overall spectra, peak-hold (blue) and rms (red) calculated according to the equal- loudness contours of ISO 226:2003 over their frequency range: 0 dB[EQL] (corresponding to the 0 Phon contour) approximates audibility threshold. SPL at each frequency is linked with the equal-loudness contour intersecting there, whose Phon value is assigned, assembling a complete spectrum in an “equal loudness weighting” (dB[EQL], HEAD acoustics implementation [see References]). FFT, ∆f = 2.93 Hz, ∆t = 340 milliseconds (actual peaks in Bark 1 ~60 ms).

30 (1740.54-1800.56 s).FFT (16384,50.0%,HAN).Avg (). L/dB(EQL)[SPL] 50

0 20 50 100 200 f/Hz 1000 2000 5000 10k 4. Psychoacoustic discussion • Psychoacoustic measures would be valuable in evaluating LF time-varying loudness, but LF results from available methods do not agree. • DIN 45631/A1 time-varying loudness and other ISO 532B (Zwicker method) implementations are compromised in LF time response by the “precursor” 1/3-octave filtration, making ear-equivalent LF critical-bandwidth time responses unobtainable. Crest factors are reduced. • ISO 532B/DIN 45631 have a correction factor to match pure tone loudness results of ISO 226:1987 (lower threshold than :2003 at low frequencies). ANSI S3.4-2007 has a similar factor to match ISO 226:2003. • A loudness method by W. Aures (thesis, 1984, in Zwicker institute), also ANSI S3.4-2007, sum slope and main excitations to determine overall loudness. ISO 532B/DIN, based on a graphic procedure, do not. Aures also did not use 1/3-octave filters. • The Aures method, containing a frequency-dependent threshold factor and stated by Zwicker at the time to represent the fullest implementation of his intention, shows more contribution in Bark 1 (and to some extent 2) than other methods, due mainly to this factor. • Investigation has found no listening tests below 100 Hz on record. • Listening evaluations below 100 Hz are surprisingly difficult in terms of transducer behavior and the transducer-subject interface. • Rather than “resurrecting” the little-used Aures method, its LF behavior and that of other methods should be evaluated along with listening studies, as guides toward developing a new time-varying loudness standard suitable for real-world low-frequency and VLF sounds. Average specific loudness of Minute 30 by (upper, on Bark scale) DIN 45631-1991 (blue) and “HEAD/FFT” (red, Aures [1984] modified by HEAD acoustics, 1/3-Bark resolution). (Lower) ANSI S3.4-2007 (on ERB scale).

1.25 4.93 soneGD 7.22 soneGD 1

0.75

0.5 N'/soneGD/Bark 0.25

0 0 2.5 5 7.5 10 f/Bark 15 17.5 20 22.5

1.25 6.29 sone 1

0.75

0.5 N'/sone/ERB

0.25

0 0 5 10 15 f/ERB 25 30 35 Minute 30 specific loudness vs. time by the 3 methods, with cuts showing partial loudness vs. time (upper left of each) at the center of Bark 1/ERB 2, and specific loudness at that time (upper right of each). Basilar membrane linear frequency scales (Bark; ERB). Left: DIN 45631/A1 time-varying (time domain). Middle: FFT/HEAD (1/3- Bark resolution). Right: ANSI S3.4-2007 (FFT). Cuts positioned as identically as possible in time and frequency. FFT ∆f = 5.86 Hz, ∆t = 171 milliseconds. Loudness vs. time by the same three methods, in defined bandwidths: Below 100 Hz (Bark 1) Note the steepness of the variations, especially in the left and center results. 100 – 200 Hz (Bark 2) 200 – 500 Hz 500 – 1500 Hz 1500 – 5000 Hz 5.

Other phenomena:

Wind turbine as musician, singing a duet: “Air and variations over a ground (sub)bass” At certain times, narrowband structures (modulated at blade-pass rate; level variation ~20 dB) were observed, not harmonically-related though following near-parallel “trajectories” in frequency. Their frequency, and the lower ones’ strength, increase with infrasound strength. FFT vs. time; ∆f 0.73 Hz, ∆t 1.37 sec: 343 seconds circa 10:30 AM, spectrum vs. time: original speed (right, 0.5 Hz – 449 Hz), 49x speed (left) with axes and analysis block size rescaled accordingly. Note the sharp spectral discontinuity at ~14 Hz (arrowed) in the original (~700 Hz in the 49x version).

Complete sound (49x speed)

Through tracking band-pass filter Minute 30: High-resolution spectral analysis (Sottek) 0.1 Hz – 500 Hz. Highest levels are in a region between about 1 and 14 Hertz with strongest content at ~2-4 Hz and a fairly steep drop above ~14 Hz (see also previous slide). The four time-regions of higher level (arrowed) are spaced at approximate 9-second intervals. Whether or not significant, this includes the 2.3 Hz vicinity associated with a half-wavelength of 77 meters, the swept diameter of this rotor disk. ∆f = 0.37 Hz, ∆t = 170 ms

500

200

100

10 f/Hz 5

0.5

0.2

0.1 1745 1750 1755 1760 1765 t/s 1775 1780 1785 1790 1795 45 50 55 60L/dB[SPL] 65 70 75 80 Tendencies toward periodicity (as on last slide) were also observed in some other one-minute time-datasets [low-pass filtered 4th order at 100 Hz, 1-second (Slow) time weighting]. Unweighted measurement; signal is dominated by <100 Hz levels. Same minute datasets as last slide: modulation spectra vs. critical bands (upper), “psychoacoustic modulation” spectrum weighted by specific loudness and modulation rate (0 to 5 Hz) (lower). In most cases, the most-perceivable or likely- perceivable blade-pass modulation was in Bark 2 (100-200 Hz).

(Signal level too low here to calculate modulation)

Blade-pass

2nd harmonic of blade-pass

Modulation rate

Signal frequency 6. Conclusions • The unusually short time-response of human hearing at low and very low frequencies is a fascinating reality for sound quality and Soundscape engineers and researchers. • An aspect of the “scan”- induced time perception mode for low- frequency sines is the possibility of the attention/pattern- recognition sensitivity of the active human processor becoming engaged strengthening a sensation, as is also likely for general low-frequency sounds with inherent rapid variations accessible to the impulse response of the lowest critical bandwidths, the “acquisition bandwidths.” • Analysis of wind turbine sounds, even above the suggested sensory-transition frequency, would benefit from consideration of the time-scale of low-frequency human hearing as a core concept around which supporting analyses of other kinds and time/frequency resolutions are made, to determine causes. 6. Conclusions

• A fascinating, complex actual and potential sound-quality panorama: Wind turbine sounds exhibit inter-varying time-structures at scales accessible to the human receiver:

– >=~60 millisecond time structure and variation even in Bark 1 – Greatest short-term level variation is in Bark 1: high and variable crest factor (~10 – ~21 dB, average ~13 dB in Bark 1). – Multi-second periodicities (~6-9 sec.) appear/disappear in Bark 1. – Modulation strengths (at blade-pass rate) change apparently independently in different frequency regions, on time-scales of ~1 second, several seconds, sub-minute, multi-minute. – Modulation strengths occasionally synchronize across signal frequency, including into Bark 1 (typically, sudden-onset: paper, Fig. 10). – Narrowband, varying tone-like phenomena sometimes occur (“air and variations over a ground [sub]bass”). – Steep LF-rise spectral slope, steepest in Bark 1 (~10 dB/octave covering ~5 octaves: signal from below 2 Hz, sometimes slope near-uniform ~5 – 100 Hz, sometimes sharp discontinuity ~14 Hz). Conclusions/suggestion

• To help bridge the gap between conventional macro-time measurements of wind turbine sound and the micro-time sensitivities of the human receiver, we suggest adding some measurements which track these sensitivities, such as:

– Low-frequency (<100 Hz) peak value, rms value and crest factor; for example calculated in 1-minute bins and plotted versus time – could be done in other frequency ranges as well

– (Numerical) Statistical tabulations from them such as the peak, rms and crest factor values exceeded 50%, 5% and 1% of the time

– More-detailed level and spectral views over short intervals “characteristic” or “of interest”

– Psychoacoustic measures and statistical tabulations (especially when current shortcomings regarding low frequencies are eliminated) Thank you. Appendix 1 Appendix 2 Appendix 3

(Alec N. Salt, Ph.D., private communication, with permission) [16, 17]

• Does the infrasound from wind turbines affect the inner ear?

• There is controversy whether prolonged exposure to the sounds generated by wind turbines adversely affects human health. The unweighted spectrum of wind turbine noise slowly rises with decreasing frequency, with greatest output in the 1-2 Hz range. As human hearing is insensitive to infrasound (needing over 120 dB SPL to detect 2 Hz) it is claimed that infrasound generated by wind turbines is below threshold and therefore cannot affect people. The inner hair cells (IHC) of the cochlea, through which hearing is mediated, are velocity-sensitive and insensitive to low frequency sounds. The outer hair cells (OHC), in contrast, are displacement-sensitive and respond to infrasonic frequencies at levels up to 40 dB below those that are heard. A review found the G- weighted noise levels generated by wind turbines with upwind rotors to be approximately 70 dB[G]. This is substantially below the threshold for hearing infrasound which is 95 dB[G] but is above the calculated level for OHC stimulation of 60 dB[G]. This suggests that most wind turbines will be producing an unheard stimulation of OHC. Whether this is conveyed to the brain by type II afferent fibers or influences other aspects of sound perception is not known. Listeners find the so- called amplitude modulation of higher frequency sounds (described as blade “swish” or “thump”) highly annoying. This could represent either a modulation of audible sounds (as detected by a sound level meter) or a biological modulation caused by variation of OHC gain as operating point is biased by the infrasound. Cochlear responses to infrasound also depend on audible input, with audible tones suppressing cochlear microphonic responses to infrasound in animals. These findings demonstrate that the response of the inner ear to infrasound is complex and needs to be understood in more detail before it can be concluded that the ear cannot be affected by wind turbine noise.

• This work was supported by research grant RO1 DC01368 from NIDCD/NIH. Appendix 4

• Auditory demonstration of perceived modulation of a known steady signal due to action of the outer hair cells in response to a simultaneous signal (operating point of IHC changed by OHC):

• The signal: a steady broadband random pink noise and a 103 Hz modulated tone. Below a particular sound pressure level, the perception is of a steady noise and modulated tone. At higher sound pressure levels, the perception changes to synchronous modulation of both the tone and noise. This effect can be experienced in the rear cabin of an aircraft with fuselage-mounted engines.

• A possibility worthy of further research with test subjects is to determine whether very low frequencies or infrasound could “dither” or modify audition of higher frequencies as suggested by Salt (Appendix 3). Acknowledgments

• The authors would like to thank colleagues (of author Bray) Dr.-Ing. Roland Sottek and Georg Caspary, André Fiebig, and Prof. Dr.-Ing. Klaus Genuit for support of this avenue of investigation, many productive discussions, software tools, and initial listening tests regarding the perception of time structure from low-frequency pure tones.

• The authors also thank Prof. Dr. Brigitte Schulte-Fortkamp for her encouragement of this effort, and Charles Ebbing and Dr. Malcolm Swinbanks for productive discussions.

• Author Bray offers special acknowledgement to electroacoustician Edward M. Long, whose short-impulse-response low-frequency loudspeaker technology (double integrator driving sealed enclosure system) was part of his “low-frequency awakening,” revealing short- time variations irreproducible with conventional loudspeaker methods. References 1. Schafer, Raymond Murray; The Tuning of the World, University of Pennsylvania Press, 1977; ISBN 081221109X 2. Acoustical Society of America, Interdisciplinary Hot Topics in Acoustics, 2010 3. Krahe, D.; Why is sharp-limited low-frequency noise extremely annoying? JASA 123(5):3569, May 2008 4. HEAD acoustics 2010 December: W. Bray (US), R. Sottek, A. Fiebig (Germany), private communications 5. Bray, Wade; Sensation and measurement of low and very low frequency time-varying sounds in accordance with the very short impulse response of low-frequency human hearing, Proceedings, Society of Automotive Engineers Noise and Vibration Conference 2011, May 2011, Grand Rapids, Michigan 6. Bray, Wade; Perceptions, and measurement implications, of time structure in low-frequency sounds, Proceedings, Noise-Con 2011, July 2011, Portland, Oregon 7. http://www.head-acoustics.de/downloads/eng/application_notes/Equalization_brochure.pdf 8. Bray, Wade R.; Binaural measurements in an Information Technology acoustic program, Proceedings, Internoise 2009, August, 2009, Ottawa, Ontario, Canada 9. Zwicker, Eberhard and Hugo Fastl; Psychoacoustics: Facts and Models, Springer-Verlag, Germany, (1999) 10. Acoustics – Method for calculating loudness level, International Standard ISO 532B:1975, International Organization for Standardization [ISO], Geneva, Switzerland (1975) 11. DIN 45631 (1991), “Berechnung des Lautstärkepegels und der Lautheit aus dem Geräuschspektrum – Verfahren nach E. Zwicker,” Deutsche Institut für Normung e. V., (1991) 12. DIN 45631/A1: (German): Berechnung des Lautstärkepegels und der Lautheit aus dem Geräuschspektrum - Verfahren nach E. Zwicker - Änderung 1: Berechnung der Lautheit zeitvarianter Geräusche (English): Calculation of loudness level and loudness from the sound spectrum - Zwicker method - Amendment 1: Calculation of the loudness of time- variant sound, Beuth Verlag GmbH, (2010-03) 13. Ott, Henry; Noise Reduction Techniques in Electronic Systems, New York, John Wiley and Sons, 1976, 1998 14. Bray, Wade; Crest factor and short-duration transients: influence of environmental background, event duration and measurement time weightings, Proceedings, InterNoise 2006, Honolulu, Hawaii December 2006 15. Genuit, Klaus; A New Approach to Objective Determination of Noise Quality Based on Relative Parameters (Proc. InterNoise 1996, Liverpool, U.K.) 16. Salt, Alec N. and Timothy E. Hullar (Department of Otolaryngology, Washington University School of Medicine, St. Louis, Missouri); Responses of the ear to low-frequency sound, infrasound and wind turbines, Hearing Research 268 (2010) 12-21, Elsevier 17. Salt, Alec N.; private communication (presented with permission as Appendix 3 of this paper) 18. Sottek, Roland; Modelle zur Signalverarbeitung im menschlichen Gehör (dissertation, RWTH Aachen, 1993) 19. W. Aures, Berechnungsverfahren für den Wohlklang beliebiger Schallsignale. Ein Beitrag zur gehörbezogenen Schallanalyse, (dissertation, TU München, 1984) 20. Acoustics – Normal equal-loudness-level contours, International Standard ISO 226:2003, International Organization for Standardization [ISO], 2nd ed., Geneva, Switzerland, (2002)) 21. (discovered after the current paper written) Møller, H. and C. S. Pedersen; Hearing at Low and Infrasonic Frequencies, Noise & Health 2004, 6;23, 37-57 22. Bray, Wade and Georg Caspary; The dB[EQL}: an alternative sound pressure weighting according to the equal loudness contours of ISO 226:2003, Proceedings, NoiseCon 2010, April, 2010, Baltimore, Maryland (From paper, Fig. 10) specific loudness vs. time, “HEAD/FFT” (W. Aures method), log Hz scale. Arrow indicates sudden blade-pass deceleration, and modulation synchronism to low signal frequencies. Perception: impulsive bass “thumps/booms” starting between one pass and the next and lasting for approximately 9 seconds. Similar events occasionally occurred on single passes, then returned to previous condition (see circa 2300 seconds).