Inverse Distance Weighting for Extrapolating Balloon-Directivity-Plots

Joerg Panzer NEXO [email protected] Daniele Ponteggia Audiomatica [email protected]

Presented at the AES 131th Convention, New York, 2011

This paper investigates an extrapolation for missing directivity-data in connection with Balloon-Plots. Such plots display the spherical directivity-pattern of radiators and receivers in form of contoured sound pressure levels. Normally the directivity-data are distributed evenly so that at each point of the display-sphere there would be sufcient data-points. However, there are circumstances where we want to display data that are not evenly distributed. For example, there might be only available the horizontal and vertical scans. The proposed Inverse Distance Weighting method is a means to extrapolate into these gaps. This paper explains this method and demonstrates some examples.

INTRODUCTION

The radiation characteristic of electro-acoustic devices can be measured or simulated by sampling over a sphere of a given radius around the device under test. Usually the sampling is carried out in spherical coordinates on a regular angular grid, as described in [1]. Once data are collected, the data-set is graphically represented with the help of so called balloon-plots. These plots are in the form of contour plots over a spherical surface or as surfaces in spherical coordinates where the radius is proportional to the gain of the device along the pointed direction. It may happen that, due to the complexity of measurement infrastructure or non-regular sampling over the spherical surfaces, that there is need to extrapolate data from a reduced or incomplete set of measurement points. On the sphere, therefore, there might be gaps large enough that we do not know the values, hence the usage of the term extrapolate in the sense of [2]: In mathematics, extrapolation is the process of constructing new data points. It is similar to the process of interpolation, which constructs new points between known points, but the results of extrapolations are often less meaningful, and are subject to greater uncertainty. It may also mean extension of a method, assuming similar methods will be applicable. ... Beside the gaps, dense data might be available. For 1

example a common case is the extrapolation from measurements done on horizontal and vertical scan-lines only. Because on the sphere any point is naturally surrounded by other points, we could perform any of the interpolation-schemes in order to obtain an intermediate value [3][4]. However, in this paper, we would like to draw attention to a particular scheme, which is straightforward to apply and which can be used for interpolation as well as for extrapolation problems. It seems to cope well also with anisotropic distribution of points. The approach is based on [5]. We have adapted it for spherical coordinates and we call it Inverse Distance Weighting (IDW) as described in section 2.1.2. In the rst part of this paper we investigate the extrapolation applied to a canonical test-function on an anisotropic grid. For comparisons two approaches are shown: 1) Decomposition with Spherical Harmonics. 2) The Inverse Distance Weighting. The rst part also explains the Inverse Distance Weighting. The second part demonstrates the application of the Inverse Distance Weighting method to a set of measurement curves.

TEST FUNCTION

The SincSinc-function features the directivity of a rectangular piston in an innite bafe [6]. D (kx , ky ) = sinc Lx kx 2 sinc Ly ky 2

Panzer, Ponteggia

IDW for Extapolating Balloon-Plots

Figure 1: Test-function at 5 kHz mapped on a mesh of regular polar-scan-lines with an angular resolution of 5 deg. To the right is shown the projection of the levels along the ring at 45 deg onto a polar-plot.

Figure 2: Spherical grids. with kx = k cos ()sin (), ky = k sin ()sin (), k = /c, is angular frequency, c is speed of sound, is the azimuthal and is polar angle in spherical coordinates. Lx and Ly are the dimensions of the rectangle. For all examples the test-function is used with following values: Lx =0.2 m, Ly =0.1 m, c= 343 m/s. A frequency f = 5 kHz is used. The highest angular resolution of any balloon-mesh is 5 deg. All contours display a range of 50 dB in steps of 3 dB. The SincSinc-function turns out to be a good testfunction for checking extrapolation schemes for radiation problems, because it features typical radiation patterns including marked interferences (gure 1). Note, the data-set has been rotated by 2.5 deg about the x-, y- and z-axis in order to provide a simple regularization for following analysis. With the help of the slight shift we make sure that target points do not coincide with source points and thus true interpolation is always performed. essary extrapolation. The idea is to use the SincSincfunction as an example and to calculate values on a non regular grid on the sphere. An anisotropic grid could be constructed in the following way. To the left of gure 2 we have the regular grid with 5 deg angular resolution. The mesh in the center is anisotropic and provides points at 5 deg resolution in polar direction and 30 deg resolution in azimuthal direction, hence there are wide gaps circumferential. The right plot, nally, pictures the extreme case, where there is a dense distribution of points on vertical and horizontal scan-lines but naught in between. The idea is to investigate how two selected interpolation schemes cope with the rareed distributions. Because we know how the function should appear, if sampled properly, we can compare the pictures of the balloon plot and a selected mapping of a polar-plot (along the dark arc) to the original plots as shown in gure 1. 2.1.1 Spherical Harmonics

2.1

Extrapolation Example

In the following we want to demonstrate the basic problem by constructing an example, which would makes nec2

Intuitively a promising interpolation scheme should be holistic, which means it would take into account all available points in order to magically ll the gaps. Is there a function, which by nature behaves like a radiator? If

Panzer, Ponteggia

IDW for Extapolating Balloon-Plots

Figure 3: Extrapolation with Spherical Harmonics. we take many of such functions and weight each of these in such a way that the sum-total satises known-values on the sphere, then it should also give reasonable values for the unknown territory. These functions are called Spherical Harmonics or Multipole Expansion [6]. Spherical Harmonics is the natural mode-set of a sphere. In principal any regular distribution of values on 3 a sphere can be assembled from a spectrum of Spherical Harmonics, similar to Fourier and polynomial interpolation in Cartesian coordinates. The text-book implementation of modal decomposition in Spherical Harmonics works ne as long as each mode gets sufcient data-points to adjust itself. This is one of the reasons why we have added a little rotation to

Panzer, Ponteggia

IDW for Extapolating Balloon-Plots

Figure 4: Extrapolation with IDW. the original data-set, in order to avoid nodal lines of the modes to fall on peculiar scan-lines of the data-set. For 5/5 deg: The rst row of the picture-table 3 is a decomposition with 2020-modes on a regular 5 deg mesh. The decomposition compares well to the original. For 5/30 deg: The second row shows the decomposition of an anisotropic grid. The algorithm turned out to 4 be stable only up to 66 modes. Compared to the original, the balloon-contours and the polar-curve are quite different. However, for a reach into the unknown the result may be considered not too bad. First, the overall pattern is reproduced in level. Second, the Spherical Harmonics yields typical interference patterns of a radiator, which other general extrapolation schemes may not be able to

Panzer, Ponteggia

IDW for Extapolating Balloon-Plots

produce. For 5/90 deg: The modal decomposition algorithm turned out to be unstable for any mode beyond 22 (dipoles). However, the result is an omni-directional mean-value. 2.1.2 Inverse Distance Weighting

The Inverse Distance Weighting method (IDW) is a sort of mean-value forming approach for calculating a value at any point on the sphere by taking into account the contribution of all original points.
i

du zi i
i

f (P ) =

du i

f (P ) = zi |di 0 with f (P ) a value anywhere on the sphere at point P . zi is a value from the original data-set at point Di . The function di is the surface-distance on the sphere between vectors P and Di : di = arccos (P Di ) For Inverse Distance Weighting the factor u 1. Typically u = 2...5. For the above examples u = 3 is used. For interesting details see [5]. If di = 0 then we substitute the original value zi at point Di . It turned out that the algorithm works better if only points from a hemisphere are used with: P Di > 0 Thus, each data-point contributes according to the inverse of its distance. Data-points, which are close to P provide a strong inuence, and data-points further away contribute less. Naturally, the IDW provides a very stable extrapolation as long as u 1. However, as the weightfunction does not provide oscillations there are no additional interferences produced, which in turn yields the resulting values of the extrapolation often to be too optimistic. Or, in other words, the values further away from the original data-set turn out to be too high compared to the regular sampled version. Figure 4 provides levels as contours on the balloon surface and polar-plots of the Inverse Distance Weighting method. As in the previous example the original dataset is the SincSinc-function (gure 1) rotated by 2.5 deg about x-, y- and z-axis. The rotation is applied in order to trigger the IDW for all cases. All extrapolations are performed with u = 3. For 5/5 deg: The rst row shows the application of the IDW on the regular grid. Because of the rotation by 2.5 deg the IDW provides interpolated values. The results compare well to the pictures of the original function. Some spatial smoothing is present. For 5/30 deg: The second row shows the result of the IDW for original points distributed on an anisotropic 5

grid. The pattern of the contours is different compared to the original balloon, however, the main features are maintained. The response obviously appears strongly smoothed as can be seen also in the polar-plot. For 5/90 deg: The last row displays the extreme case, where the original function is provided only on two scan-lines. Interestingly the balloon contours still display some of the main features of the directivity of a rectangular piston. However, as already stated, the level in the extrapolated regions turn out to be too high. This is understandable because the original function provides interference whereas the IDW does not. Compared to the Spherical Harmonic approach the IDW seems more stable and easy to adjust. During experiments the Spherical Harmonic approach often failed, needed special regularization and asked for a careful adjustment of the number of modes. The IDW on the other hand always yields a meaningful result, however, often blurred and too optimistic.

MEASUREMENT

The measurement of a loudspeaker 3D directivity, as described in [1], requires the loudspeaker impulse response to be sampled with a 5 degree resolution equi-angular spherical pattern. This means a total of 2664 measurement points [7]. Among the different approaches to acquire the 3D measurement set we may: use multiple microphones, move one or more microphones around the loudspeaker, use a single xed microphone coupled with a mechanical system to orient the loudspeaker towards a given direction or any mix of the above. It is outside the scope of this paper to enter into the details of 3D measurements. We would like to point out here the complexity of such kind of setup, both in terms of needed hardware and time. A simpler and more cost-effective alternative is to sample the sphere over only a few scan-lines. Using a single xed microphone and a single turntable (or equivalent) it is easily possible to sample the loudspeaker response over the horizontal and vertical scan-lines. This can be done by placing the loudspeaker in vertical and horizontal position on the turntable, the rotation of the turntable corresponds to an apparent rotation of the microphone around the loudspeaker. In this case the full balloon response has to be extrapolated from the available data. We will compare here the results of 5 degrees full-balloon measurements with an extrapolation made with the IDW method. Before showing the results of real world examples, we need to introduce another graphical representation of the 3D directivity in the form of a color-map as rectangular projection of the sphere: the two spherical coordinates angles and are projected on the x- and y- axes, while the sound pressure level is represented by a color shade (see gure 5). Despite the distortion introduced by such projection, this allows for a simple representation of the

Panzer, Ponteggia

IDW for Extapolating Balloon-Plots

Figure 5: Color-map plot: directivity balloon is projected on a plane.

Figure 6: Driver on exponential horn balloon at 8 kHz: (left-top) 5 degree measured, (right-top) IDW extrapolation, (bottom) absolute error.

difference between 5 degrees 3D balloon measurement and data extrapolation made with the IDW method. We will report here two different case studies: a 1 inch compression driver on a 90x60 exponential horn, and a small bookshelf 2-way sealed loudspeaker box. Throughout these examples u = 3 has been used for the IDW extrapolation and data has been smoothed by third octave bands.

same gure 6 (right-top) the balloon extrapolated from horizontal and vertical scan-lines using the IDW method is shown. The absolute error in dB between measured and data extrapolated is shown in the bottom row of gure 6 as a color-map plot. While the absolute error is not small, the overall balloon shape is not affected, as previously found with the sincsinc test function.

3.1

Compression driver on exponential horn

3.2

Small 2-way loudspeaker box

The driver-horn assembly has been measured in 3D using two computer controlled turntables, with an angular resolution of 5 degrees. Since the source has a symmetry along the vertical and horizontal planes, only a quarter of a sphere has been sampled and data has been then mirrored. Figure 6 (left-top) shows the measured balloon plot (in this plot the surface is distorted in such a way that the radius of the sphere is proportional to the sound pressure level) of the driver at 8 kHz frequency band, where the effect of the rectangular aperture is clearly visible. In the 6

The 2-way box has been measured in 3D using the same setup as above. Due to the horizontal symmetry only half sphere has been sampled and then data mirrored. This loudspeaker exhibits an interference pattern between the two drivers at the crossover frequency. The interference can be seen in the 3.15 kHz one third frequency band measured balloon (gure 8 (left-top)). Figure 8 (mid-top) illustrates the balloon extrapolated from horizontal and vertical scan-lines using the IDW method. A color-map plot of the absolute error of the IDW extrapolation using the above scan lines is

Panzer, Ponteggia

IDW for Extapolating Balloon-Plots

Figure 7: Horizontal, Vertical and Equatorial Scan-Lines.

Figure 8: 2-Way loudspeaker at 3.15 kHz: (left-top) 5 degree measured, (mid-top) IDW extrapolation H+V, (right-top) IDW extrapolation H+V+E, (left-bottom) absolute error H+V, (right-bottom) absolute error H+V+E.

in gure 8 (left-bottom). Here again, while the balloon presents quite a complex behavior featuring multiple side-lobes, the overall shape is kept and side-lobes are present, albeit attenuated. The error of the IDW extrapolation can be reduced using more scan-lines. We will propose here a very simple setup to collect an additional scan-line that, in an analogy with earth science, we call equatorial (gure 7). The proposed setup uses a single turntable and a single xed microphone, as in the case of horizontal and vertical scan-lines. In order to collect the equatorial scan-line, the loudspeaker under test has to be placed on the turntable facing up (pointing towards the ceiling). The balloon plot of the data extrapolated with IDW method using horizontal, vertical and equatorial scan-line is presented in gure 8 (right-top) and the error is shown as color-map plot in the same gure (right-bottom). The presence of the equatorial data helps to reduce the er7

ror, while the required additional measurement effort is minimum.

CONCLUSIONS

This paper reports on the investigation of the application of the Inverse Distance Weighting method for extrapolating incomplete 3D directivity data. The IDW is not a substitute for the complete data-set as can clearly been seen by the investigation of the canonical SincSinc function and by the error-maps of measurements done. However, the IDW can provide rough estimates for the unknown areas even under the extreme case, where only a horizontal and vertical scan-line is available. In any case the IDW provides bounded results and is easy to apply. As most measurement setups go along scan-lines it seems benecial in this context to measure the device un-

Panzer, Ponteggia

IDW for Extapolating Balloon-Plots

der test also equatorial. Future research will investigate renements of the IDW and will look into other possibilities of extrapolating the balloon-plot.

AKNOWLEDGEMENTS

The authors sincerely acknowledge the companies NEXO S.A. (www.nexo.fr) and Audiomatica (www. audiomatica.com) for support of this paper.

References
[1] AES Standard, Sound source modeling - loudspeaker polar radiation measurements, AES562008. [2] Wikipedia, Extrapolation, https: //secure.wikimedia.org/wikipedia/ en/w/index.php?title= Extrapolation&oldid=438224623 (accessed August 8, 2011). [3] Renka, R., Interpolation of data on the surface of a sphere and triangulation and interpolation at arbitrary distributed points in the plane, ACM Trans. Math. Softw 10, 4 (Dec 1984). [4] Buss, S., Fillmore, J., Spherical averages and applications to spherical splines and interpolation, ACM Trans. Graph., 20, 2 (April 2001). [5] Shepard, D., A two-dimensional interpolation function for irregular-spaced data, ACM Proc. Nat. Conf. 1968. [6] Williams, E., Fourier acoustics, Academic Press, 1999. [7] Ponteggia, D., AN-002, Automated Balloon Measurements Using CLIO 10, http: //www.audiomatica.com/download/ appnote_002.pdf, Audiomatica, 2009