> did you ever look through what Z really does to you *encoding* equations? > because they're always a bit spread out even vertically, not all of that >problem can be remedied. AND > The only way to really get that distance calculation right is to employ >periphony
You are undoubtedly correct. But I look at this problem more from a practical point of view. A horizontal-only presentation of a full 3D space is just wrong. But practically speaking, only occasionally do we have the opportunity to present a recording or a composition periphonically. So we must accept the compromise. The distance-related errors I don't see as being such a big deal because, especially at low order, we have to accept the summation errors which lead to a small sweet spot. But how do we perceive the space, especially when there were sources in the original recording that were out of the horizontal plane? Or, in my own recording work, when there is reverberation coming from out of the horizontal plane? This leads to a concentration of energy right at the horizon. In recent work, Aaron Heller and I have been trying to produce decoders that work 'better' than traditional decoders for arrays that only partially cover periphonic space, arrays like a hemisphere or a tilted hemisphere in a concert hall with banked seating. If the original sound source started out above the horizon but then moves downward to eventually go below the horizon, we can reproduce the direction and timbre of the sound pretty well in these arrays when the source is above. But when it dips below the horizon we have to make choices as to which wrong behavior we will allow. We could enact a strict 'no-fly' zone; sources just disappear when they dip below the horizon. Or they can 'stick' at the horizon, which tends to be the natural behavior of Ambisonic decoders. Or you can have them gradually fade away as they go lower. This is also a problem with some of the periphonic arrays that I have tried in the past. I still like the 30 degree tri-rectangle, because it fits in my listening room. But it has the fault that sources tend to stick at +30 degrees and at -30 degrees. In practice, this doesn't seem to be so much of a problem with naturally recorded program material even though there must be a concentration of reverberant energy at those two elevations. AND > advance a bit towards an understanding of higher order compatibility formats, >and in the process, of how to optimally and scalably encode ambisonics? I think that we still don't understand well enough what the perceptual effects are of the compromises we make in decoder and system design. There probably is a good deal of progress still to be made in this area. Eric Benjamin ----- Original Message ---- From: Sampo Syreeni <[email protected]> To: Surround Sound discussion group <[email protected]> Sent: Sat, June 1, 2013 6:55:04 AM Subject: Re: [Sursound] Nevaton microphones On 2013-05-31, Daniel Courville wrote: >> I always insist on recording 'Z', and then almost never end up using it... > > Not even to "look" down or up in a stereo decode? I use the Z quite often (if >not always) when recording large ensemble and the SF mic is more than 10 feet >off the floor. BTW, did you ever look through what Z really does to you *encoding* equations? Formally, in order to arrive at proper pantophony you always have to either reject Z fully or purposely subtract it from the whole B-format signal set. Otherwise, even assuming perfect coincidence, your W will have directionally aliased components from the above and the below. For the most part that isn't noticeable in e.g. concert work where you have a wide and loud array of early arrivals right in the horizontal layer. But the theoretical error easily bite even with a simple walkaround where not everything is in the horizontal plane, including close vertical room modes. That particular problem has then also been used to attac the ambisonic system as a whole. I believe I told about Christof Faller's analysis of "why ambisonic can't work", a few years back, didn't I? Which I followed live at what is now Aalto University, and then Teknillinen korkeakoulu (lit. "technical high school", formally "Helsinki Polytechnic"). How Christof saw it was much from the WTF point of view. There, if you reproduce a point source in the horizontal plane only using a horizontal array of speakers, you will get the angle of arrival right, but the normal attenuation suddenly acquires an extra 3dB/per normalized distance factor. In WFS they purposely compensate for that with their linear and rectangular arrays. But very few analyses really go into where that factor comes from, or how it could be avoided, or what it's really about. The pantophonic analysis of ambisonic doesn't go there either, even if it really, *really* should. The basic problem is that you just can't in 3D space radiate a 3D soundfield which fails to collimate in at least x dimensions, where y=3-x is the effective dimension of your array. If you use a monopole, it'll always attenuate as 1/r^2. If you use a line array or any variant of it like a circular array (any pantophonic array), it will still remain uncollimated in the third dimension. As such, power will be radiating away from a 2D array not in 2D, but in the enveloping 3D, and it will be felt within the array as a dropping off of power by distance. In WFS theory, the dropoff is steeper, because all of the speakers have to be on the horizontal plane. I believe it's ten 6dB per natural unit of length. But even in arbitrarily dense, ideal, ambisonic arrays, where some collimation from the above and the below will be taking place even with purely XY (zonal?) harmonics, because they're always a bit spread out even vertically, not all of that problem can be remedied. The only way to really get that distance calculation right is to employ periphony, so that in the limit you can reproduce true plane waves from your array. Or at least do that within your array. If you try to do anything of the sort even at infinite order, in 2D, energy will "bleed of into the third dimension" and suddenly the system reduces to a circular variant of WFS, with a single critical distance from the array which does the two-dimensional attenuation right. That much I think I know for sure. So what am I not too sure about? Well, theoretically you can also expand your soundfield in cylindrical harmonics. Instead of going with the two spherical Fourier functions and a Bessel radial part, you can go with a second kind of Bessel part and a circular Fourier part, which is then just a pure Fourier series in two neatly orthogonal coordinates, instead of spherical surface harmonic nasty one. It's mathematically given that is what happens when you expand the natural solution to the Helmholtz equation in this second coordinate system. It's going to be much easier, it's going to fit very well with the spherical surface harmonic decomposition, but the two different radial, Bessel terms won't match. If you had the kind of natural vertical line emitters available which a direct realization of the cylindrical transfrom described, ÿou'd never suffer from decollimation in the z-direction, and so your soundfields wouldn't have any decay by distance. That'd work for both WFS and pantophonic HOA at the same time. But it wouldn't work for periphonic HOA at all, because it's really difficult to build *anything* spherically symmetrical 3D stuff out of 2D symmetrical cylinders. At the same time it's pretty easy to build everything wavelike out of either planewaves (arbitrarily, via basic, rectangular Fourier theory), or point sources (under the additional far field Sommerfeld radiation condition, which takes the place of the straight forward L^2 norm when doing this kind of an integral; from the viewpoint of Kirchof-Helmholtz integrals and the pointwise Huygens principle). So, in 2D, embedded in 3D, you're really want to use basic emitters which aren't points, but lines. In 3D, within a 3D enveloping space, you¨'d want to use just points. My question to gurus Robert Green, Filippo Fazi, and perhaps both Daniel Courville and Eric Benjamin is, could it perhaps be shown that this sort of a calculus also leads to the differing constants in decoding UHJ and pantophonics? If not directly, then via some circuitous route? Because if it works out, maybe we could finally rest aside the annoying little thingie that BHJ as UHJ's horizontal variant ain't roundtrippable with B-format? That way also advance a bit towards an understanding of higher order compatibility formats, and in the process, of how to optimally and scalably encode ambisonics? -- Sampo Syreeni, aka decoy - [email protected], http://decoy.iki.fi/front +358-50-5756111, 025E D175 ABE5 027C 9494 EEB0 E090 8BA9 0509 85C2 _______________________________________________ Sursound mailing list [email protected] https://mail.music.vt.edu/mailman/listinfo/sursound _______________________________________________ Sursound mailing list [email protected] https://mail.music.vt.edu/mailman/listinfo/sursound
