acoustic measurements-using gating techniques

17-163

3-Dimensional acoustic measurements— using gating techniques

by Henning MellerBriie! & Kjaer

1. Abstractsome experience to interpret the 3Dresults just as it requires more expe-rience to interpret a frequency re-sponse than a single number on avoltmeter.

The paper will show the influenceof various time resolutions and fre-quency resolutions as well as rec-tangular and Gaussian time weight-ings.

The subjective listening impres-sion in a room is a function of manyacoustic qualities. The more parame-ters that can be viewed at the sametime, the easier it is to evaluate theobjective qualities of the room.

This paper will therefore try to ex-pand the traditional "1 -dimen-sional" acoustic reverberation timemeasurement — first, to a "2-di-mensiona! frequency response

measurement" of the various acous-tic surfaces using a gated sine toneburst, and then to a "3-dimen-siona!" plot of how the various fre-quency responses change as func-tion of time when the reflections ar-rive. The 3D plots are obtained us-ing a gated pink noise pulse, a Real-Time Analyzer, a calculator and adigital plotter.

However, it requires, of course.

2. Principle of Gating Techniques — using Swept Tone BurstsOne of the basic problems in

room acoustic measurements has ai-ways been to determine the direc-tion of a certain reflection, andmore important, its frequency con-tent. For exarnpie, what is theacoustic influence of a certain typeof ceiling structure and surface in aconcert hali or a studio.

Gating techniques are a simple so-lution to these problems. Gating isa selective measurement in thetime domain just as frequency analy-sis is a selective measurement inthe frequency domain. A tone burstis transmitted from a loudspeakerand measured at the listening posi-tion as indicated in Fig. 1 .

Fig.1. Principle of gated measurements

The simple implementation of thismeasurement is as described in the1975 paper (Ref.1) — a swept sinethat is cut into tone bursts at thezero crossings. The instrument set-up is shown in Fig.3 and the timeand frequency spectra of the tonebursts are shown in Fig.4. The out-put of the Sine Generator 1023 ispassed through the transmitting sec-tion of the Gating System 4440and transformed into a tone burst ofadjustable length (0,1 rns to 1 s)which then is fed through thepower amplifier to the loudspeaker.The signal received by the micro-phone passes through the Measur-ing Amplifier to the receiving sec-tion of the Gating System wherethe desired portion of the signal isselected by the measuring gate,whose width and delay are also ad-justable over a wide range. A peakdetector measures the amplitude ofthe desired signal and feeds a DCvoltage proportional to this value tothe Level Recorder which is syn-chronized with the Sine Generatorfor automatic recording of fre-quency response. The peak detectorcontains a hold circuit which isreset for each new tone burst to per-mit capturing the new amplitude asthe frequency changes. To monitorthe adjustment of the Gating Sys-tem, a two-channel oscilloscope isessential.

The restriction on lower limitingfrequency is related to the size ofthe room, as is the case for ane-choic rooms. A detailed descriptionof this can be found in the 1975paper (Ref.1).

The transmission of the toneburst will (as indicated in Figs.1and 2) result in a direct received sig-nal that obviously will arrive first,followed by a number of burstscorresponding to the reflectionsfrom the various surfaces.

In the 1975 paper (Ref.1) we em-phasized the measurement of the di-rect sound that only contains thefree-field information of the loud-speaker itself, while this paper willconcentrate on the measurement ofthe reflections in order to obtain in-

Fig.2. Reflections from the environment

Fig.3. Set-up using Gating System

Fig.4. Time and frequency spectra of gated sine wave. Notice that a narrow time spectrumcorresponds to a broad frequency spectrum and vice versa

gate is adjusted so it only picks upthe information corresponding to aparticular acoustic surface.

formation about the acoustic pro-perties of the room. The measure-ment set-up is essentially the sameas in Fig.3 except the measuring

3. Measurements of Frequency Response of Acoustic surfaces— using Gating Techniques with Swept Sine Bursts

We selected an acoustic environ-ment of medium size as indicated inFigs.5, 6 and 7. The dimensionswere;

length 12,1 mwidth 8,2 mheight sloping from 2,2 m to3,4 m

Before measuring the acousticsurfaces of the room, we measuredthe "free-field" responses of theloudspeaker used. The frequency re-sponses at a distance of 2, 4 and6 m are shown in Fig.8 and the di-rectional characteristics for 1, 2, 4,8 and 16 kHz are shown in Fig.9.

Of course, the measurements ofthe surfaces must be seen relativeto the frequency response as wellas the directional characteristics ofthe loudspeaker but for some appli-cations the directional characteris-tics can be neglected.

The first thing to do, in practice,is to manually sweep the frequencyup and down, change the width ofthe transmitting gate and the repeti-tion rate, and watch the scope untilsome well-defined bursts becomevisible. Then move the microphonearound in the listening area to seehow much the particular reflectionschange and select a typical andclearly recognizable position. Sucha picture is shown in Fig. 10. Thiswas actually recorded using the Dig-ital Event Recorder 7502 with a 4kmemory and then plotted on the Le-vel Recorder 2307. However, nor-mally the scope picture is sufficient.

Fig.5. The lecture room under test

Fig.6. The instrument set-up in practice

Fig.7. Schematic drawing of the lecture room

To determine which reflectionscorrespond to the various surfaces,simple geometry can be used as in-dicated in Fig.7. However, in prac-tice, it will often be easier to use anacoustic screen (this ApplicationNote will work quite well) and manu-ally positioning this between themicrophone and the various reflect-ing surfaces and watching thescope until only one of the burstsdisappears. When that happens thereflecting surface is determined.

Fig.8. "Free-field" measurements of the loudspeaker used

Fig.9. Directional characteristics of the loudspeaker used

The most prominent reflectionsfor the selected loudspeaker micro-phone position are indicated inFig. 10 using a 3130 Hz tone burstof 1,6 ms width. It can be seenthat at this frequency the "BlackBoard reflection" is slightly out ofphase with reflection from the wallbehind it — and the direct soundand the reflections from them there-fore is quite small.

However, ceiling/floor, the doorwall, the window wall, and the backwall are clearly visible. The blackboard reflection can clearly be Seen Fig.iO. The time function of the various reflections in the lecture room recorded on the Digitalby changing the frequency slightly. Event Recorder 7502 and plotted on the Level Recorder 2307

The next step is to measure the"frequency responses" correspond-ing to the various surfaces. Theseresponses must, of course, be seenas the difference between the direct(or free-field sound transmittedfrom the loudspeaker) and the ref-lected sound. This is done by manu-ally adjusting the measuring gate soit only picks up the response corre-sponding to one of the surfaces at atime, (for instance, as indicated inFig. 10 — the door wall informa-tion), and then start the automaticsine sweep. The results of thesemeasurements are shown relativeto the frequency response inFigs.11, 12, 13 and 14.

Fig.11. The reflected sound from the ceiling and floor relative to the direct sound from theloudspeaker

First of all, it is seen that most ofthe reflections on average are about10dB down, but they do not diedown at the same rate at all fre-quencies. Since the four curves aredifferent, they also indicate that thedecay rates at the various frequen-cies are different in the different di-rections.

Fig.12. The reflected sound from the door wall relative to the direct sound from the loud-speaker

The most pronounced "problem"is probably the window wall andback wall reflections at 2—3 kHz(Figs. 13 and 14) that show a signifi-cant "flutter echo" since the ref-lected signals are only approxi-mately 1 dB down after 15 ms and24 ms respectively.

Obviously, the time picture(Fig. 10) doesn't reveal the fre-quency information and the fre-quency pictures (Figs.1 1—14) don'tshow the time information. The so-lution therefore is a 3D plot of thefrequency spectra as a function oftime. This is virtually impossible todo manually but is quite easy usingmodern programmable calculators. Fig.13. The reflected sound from the window wall relative to the direct sound from the loud-

speaker

Fig.14. The reflected sound from the back wall relative to the direct sound from the loud-speaker

4. Instrument set-up for 3D MeasurementsIn recent years, there have been

several approaches in the audio in -dustry to introduce 3D plots,(Refs.2—6) but the expense so farhas been considerable. But moderninstruments and calculators con-nected via the IEC digital interfacebus make these measurements pos-sible at a reasonable price using thetraditional "2D instruments" as thehardware basis. The instrument set-up used for this application isshown in Fig. 1 5.

The Sine Generator 1023 is re-placed by the Noise Generator1405 since the frequency analysisis done by the Digital Frequency An-alyzer 2131. The first Gating Sys-tem 4440 operates as a transmit-ting gate that in this case cuts thepink noise into bursts at the zerocrossings. The width is manually ad-justable in the range 0,1 ms to1 s. The gated pink noise pulse isintroduced to the loudspeaker andpicked up by the microphone. The

Frequency Analyzer 2120 is justused as a high pass filter in order toavoid the low frequencies that makethe scope picture unstable. It alsoremoves the low frequency informa-tion which is measured with toomuch statistical uncertainty to beuseful due to the narrow measuringgate. A further discussion of the re-lation between frequency resolutionand time resolution follows in sec-tion 5.

Fig.15. The instrument set-up using gated pink noise, real-time analysis and automatic control by calculators

The second Gating System isused as an analogue measuringgate (Fig. 16) which is requiredsince the normal built-in measuringgate only measures the highest posi-tive peak that was in the signalwhile the measuring gate wasopen, in other words the gated infor-mation is only available as a DC vol-tage when only one Gating System4440 Is used; but it is available asan analogue voltage, that can be fre-quency analyzed, when two GatingSystems are used. A Gauss Multi-plier is also usable as described insection 8. This set-up can, ofcourse, be manually operated as in-dicated in Figs. 15 and 16 using themeasuring gate of the first GatingSystem as a trigger for the otherGating System. By manually chang-ing the delay of the receiving gateof the first Gating System (the trig-ger position) and thereby the posi-tion of the analogue measuring gate(the transmission gate of the sec-

ond gating system) different parts ofthe received signal can be analyzed,and it can be seen how the fre-quency spectra change with time.

However the strong feature of thesystem is that it can be automati-cally controlled using a calculator.There are many of these on the mar-ket but B & K recommend either theHP 9825A and the Digital PlotterHP 9872A or the Tektronix 4051 .These both use the IEC interfacebus that most new B & K instru-ments are equipped with. (Note: a

buffer amplifier is necessary todrive the Gating Systems from theiEC bus; or the Gating Systems canbe modified for a higher input im-pedance.)

Depending on the software used,we can now automatically controlthe start of the transmission burst(Gating System No.1) and the startof the receiving gate (Gating Sys-tem No. 2). The widths of the trans-mission gate and the receiving gateare manually adjustable on the twogating systems.

Fig.16. Setting of gates to measure the various reflections with the set-up in Fig.15

5. 3D plots of how the Frequency Spectra change with time

Fig. 17. 3 D plot of lecture room showing how the frequency spectra change with time. 0—50 ms time axis. Receiving gate 1,5 ms

Fig.18. 3 D plot of lecture room showing how the frequency spectra change with time. 0—50 ms time axis. Receiving gate 4,7 ms

tively, and they show how the fre-quency spectra change with timeover the 0 to 50 ms time span.

It is seen that the very narrowmeasuring gate (1,5 ms) used inFig.17 displays a visible wave struc-ture in the 3D spectrum travellingalong the time axis, while themeasuring gate (1 5 ms —Fig.1 9)shows a wave structure travelling

along the frequency axis. Statedanother way, a short gate givesgood time resolution while a longgate gives good frequency resolu-tion.

The "flutter echo" we noticed us-ing the simple gated sine wave(Figs. 13 and 14) is clearly visible inFig. 19 but not very obvious inFig.17 and Fig.18.

Typical results of this measure-ment are shown in Figs. 17—22.

The first three plots. Figs. 17, 18and 19 are building acoustics meas-urements using the loudspeaker-microphone position indicated inFig.5 and Fig.7. The three plots areessentially the same but with differ-ent time resolutions in the receivinggate: 1,5, 4,7 and 15 ms respec-

Fig .19. 3D plot of lecture room showing how the frequency spectra change with time. 0—SO ms time axis. Receiving gate 15 msNotice the "Flutter Echo" at 2 kHz

whenever making this kind of meas-urement. They are equally appli-cable for digital and analog filters,FFT (Ref.13), and any othermethod. The "smear" in this type ofmeasurement is a result of a law ofnature that you can't cheat.

Since pink noise, which is a ran-dom signal, was used for thesemeasurements, a certain amount ofaveraging is necessary to get reprod-ucible results. For these measure-ments, averaging was performedover 1 5 noise bursts us4ng a 4 s av-eraging time in the Digital Fre-quency Analyzer 21 3 1 .

As previously seen in Figs. 17 to22, the appearance of the 3 D spec-trum depends on the width of the re-ceiving gate. One may then ask,what gate width should be used togive the correct result. The answeris: "That depends on what informa-tion you want." if good time resolu-tion is desired, a very narrow timewindow must be used — but thishas the unavoidable by-product ofgiving very poor frequency resol-tion. On the other hand, if good fre-quency resolution is desired, thetime resolution will be reduced. Themathematical relationship is verysimple:

1T

For example, if the receiving gate is5 ms (= T), then the frequency reso-lution is 200 Hz. However, eventhough the receiving gate is 5 mslong, we can move it in muchsmaller increments — and what wesee is how a given frequency com-ponent changes as a function oftime — with a "spreading out" ofthe signal over a 5 ms interval. Ri-chard C. Heyser has called this a"time smear" (Ref.7—9). These res-trictions on time and frequency reso-lution should be kept firmly in mind

6. Early reflections of the Loudspeaker usedThe 3D technique is also appli-

cable to the so-called "Early reflec-tions" that describe how the fre-quency response changes with timefor the loudspeaker itself. In otherwords, the first few milliseconds ofthe curves previously displayed in

cially in the cabinet. !t indicateshow fast the response of the sys-tem is and how much sound of itsown the loudspeaker box adds tothe signal.

Here again it's a question of the

Figs. 17—19. In this case (Figs.20-22) from 0—8 ms.

Early reflections are due to over-hang of the speaker, and reflec-tions, standing waves, and reson-ances in the speaker itself and espe-

Fig.2O. 3 D plot of Early Reflection of loudspeaker. Time 0—8 ms. 1 ms 9,9 kHz tone burst

Fig.21. 3 D plot of Early Reflections of loudspeaker. Time 0—8 ms. 1 ms 1,2 kHz tone burst

10

mitting signal the time resolution ismore clearly visible. It is simplyequivalent to the scope picture, butthe disadvantage is that it doesn't si-multaneously show the behaviourof the other frequencies as Fig.22where we used gated pink noise.

optimum trade-off between fre-quency and time resolution. InFig.20 we have a 1 ms 9,9 kHztone burst and a receiving gate of0,4 ms. In Fig.21 we have used a1 ms 1,2 kHz tone burst and a re-ceiving gate of 0,4 ms and in

Fig.22 we have used a 1 ms gatedpink noise pulse and still a receiv-ing gate of 0.4 ms.

Obviously when only one fre-quency (and its sidebands becauseof the gating) is used as the trans-

Fig.22 3 D plot of Early Reflections of loudspeaker. Time 0—8 ms. 1 ms pink noise tone burst

11

?. The ProgramsAs mentioned in section 4, the

modern instrumentation has madethese kinds of 3D measurementspossible at a reasonable price.

Control of the instruments maybe made either by the HP 9825A orthe Tektronix 4051 for which B & Kcan offer custom-made programs.Our software specialists are work-ing full time on this and we expectvery much from this in the future.The programs used in this paper areonly the beginning and really just in-troduce the technique.

The programs used for these plotsare made for the HP 9825A andthe HP 9872A. Fig.23 shows theprogram for control of Gating Sys-tem and input of spectra and Fig.24shows the program for 3D plot ofspectra recorded by input routine.

Fig.23. Program for control of Gating Systems and input of spectra

Fig.24. Program for 3D plots of spectra recorded by input routine

12

8. Gaussian or Rectangular Weighting?Pressure or Free-field Microphone?

There are several technical mea-suring problems involved usingpulse techniques. The most obviousproblem is the sidebands introducedusing gating technique. This prob-lem is described in greater detail inthe 1975 paper (Ref.1).

The second consideration couldbe: what kind of weighting functionshould be used — if any. This is de-scribed in detail in Bob Randall's Ap-plication Note (Ref.1O). However us-ing the 1/3 octave resolution the in-fluence of the weighting function isalmost negligible as can be seenfrom Figs.25 and 26.

Fig.25 shows the difference be-tween the rectangular weightingand the Gaussian weighting using a5 ms burst of pink noise on theloudspeaker and it is seen that fromapproximately 500 Hz and up thedifference is almost constant. Thedifference spectrum is recorded us-ing the Digital Frequency Analyzer2131 in the Fast comparing modewhile reading out to the Level Recor-der 2307. This read-out is ex-tremely fast (paper speed10 mm/sec).

Fig.26 shows the electrical re-sponse of the system using theGauss Impulse Multiplier 5623.The 200 Hz roll off is due to thehigh pass filter in 2120.

The operation of the Gauss Im-pulse Multiplier 5623 is shown inFig.27 and is described in furtherdetail in B & K Application Note 11-194 (Ref.11). Another technicalquestion that could be asked iswhat kind of microphone should beused. The directional characteristicsof the 1/2" Free-field microphone4133 show that at 20 kHz the sen-sitivity from the back (180°) is ap-proximately 10dB down and there-fore we used a Pressure micro-phone 4134 at grazing incidencewhich gives the same sensitivity inall directions in the plane of the di-aphragm and since we are just inter-ested in the difference between thedirect and the reflected sound thisseems optimum.

Fig.25. Difference spectrum between loudspeaker response using Rectangular and Gaussianweightings respectively. Recorded on the Digital Frequency Analyzer 2131 in the fastcomparing mode

Fig.26. Electrical response of the system using Gaussian window

Fig.27. The operation of the Gauss Impulse Multiplier S623

13

9. Transmission and Reflection of a Conventional Door

The gating technique is also veryuseful for measurement of trans-mission and reflection properties ofdoors, walls, etc. even as functionof angle. The measuring gate is ad-justed so it only picks up the reflec-tion. As shown in Figs.28—31 wemeasured a conventional door asboth on-axis and 30° off-axis.Fig.28 shows that the attenuation30° off is approximately 25 dB butslightly worse at higher frequen-cies. Fig.29 shows that the samedoor "on axis" has the. lowest atten-uation in the critical range 3 kHz.Here it is only 1 8dB.

Fig.28. Spectra of conventional door open and closed. 30° off axis

Fig.29. Spectra of conventional door open and closed. On axis

Fig.30. The open door set-up in practice (30° off axis) for measurement of transmission asfunction of frequency

14

Fig.31. Reflections as functions of frequency can also be measured using gating techniques

10. Conclusion and ApplicationsThis paper has tried to introduce

the 3D technique in acoustic andelectro-acoustic applications usingthe new relatively inexpensive calcu-lators with instruments having anIEC interface.

The building acoustic applicationwhich is especially emphasized inthis paper is useful in concert halls,cinemas, theatres, studios, etc.

The application determining ab-sorption and reflection coefficientsas a function of frequency and

angle is another important buildingacoustics parameter.

The early reflections of loudspeak-ers in 3D is also an extremely im-portant parameter in loudspeaker de-sign, but generally the 3D programcan be used for all kinds of plots in-dicating relations between 3 par-ameters. A procedure could, for in-stance, be introduced for 3D dis-plays of Harmonic, Difference Fre-quency and Intermodulation curvesmeasured with the Heterodyne Anal-yzer 2010, the Distortion Measure-

ment Control Unit 1902 and a sim-ple external A /D converter (Ref.12).3 D sound power could also be mea-sured and the iist is almost endless.

We will conclude the paper byshowing two other possibilities us-ing the 3D plots. Fig.32 shows the3D reverberation time of the lectureroom and Fig.33 shows a 1/12 oc-tave 3D display of the room. Thereare many other possibilities in thefuture.

15

Fig.32. a) Numerical values of the reverberation time in the lecture roomb) 3 D plot of reverberation time in the lecture room

Fig.33. 1/12 octave plot of the decay of the lecture room

16

Ref.1Henning Molier and Carsten Thom-senEiectro Acoustic free-field measure-ments in ordinary rooms — usinggating techniquesAES paper New York 1 975 orB & K Application Note 1 5—107

Ref.2Ole Metier SorensenThree-dimensional output fromThird-octave Frequency analysisB & K Application Note 1 2—1 20

Ref.3J.M. BermanLoudspeaker evaluation using digi-tal techniquesAES paper London 1 975

Ref.4L.R. Fincham, KEFLoudspeaker system simulation us-ing digital techniquesAES paper London 1975

Ref.5Nomoto, iwahara and Onoye, JVC,JapanA technique for observing loud-

speaker wave front propagationAES paper, Zurich 1976

Ref.SSuzuki, Morii and Matsumura, Pion-neer, JapanThree-dimensional displays for dem-onstrating transient characteristicsof loudspeakersAES paper. New York 1 976

Ref.7Richard C. HeyserLoudspeaker Phase Characteristicsand Time Delay Distortion: Part 1January 1 969, Part 2 April 1969Journal of the Audio EngineeringSociety (AES)

Ref.8Richard C. HeyserDetermination of Loudspeaker Sig-nal Arrival Times: Part 1 October1971, Part 2 November 1971, Part3 December 1971Journal of the Audio EngineeringSociety (AES)

Ref.9Richard C. HeyserCommunications

Journal of the Audio EngineeringSociety (AES)January 1 976

Ref.1OR.B. RandallFrequency Analysis of Stationary,Non-stationary and Transient sig-nalsB & K Application Note 14—165

Ref.11Instantaneous Spectrum Analysiswith the Real-Time 1/3 octave Anal-yzer Type 3347B & K Application Note 11 — 1 94

Ref.12Carsten Thomsen and HenningMailerSwept measurements of harmonic,difference and intermodulation dis-tortionB & K Application Note 1 5—098

Ref.13.Bob RandallApplications of B & K Equipment toFrequency AnalysisB & K Technical Book 1 977