Distance compensation with KNN/DBAP

elarrieux · July 2, 2025, 9:19am

Hi all,

As I continue to explore the capabilities/best practices of spat5.spat~, I came across the following statement in spat5.tuto-alignment-1.maxpat:

Only the angular directions (azimuth, elevation) of the loudspeaker are relevant.

This remark holds for all panning processors in Spat (e.g. spat5.pan~, spat5.spat~, spat5.pan, etc.) and for most panning techniques (vbap, angular, etc.)

However, there are some exceptions: the “knn” and “dbap” panning algorithms are fundamentally distance-based, and therefore the distance of the loudspeakers is taken into account.

What exactly does “the distance of the loudspeakers is taken into account” mean? Does it imply that alignment is not necessary for non-spherical setups when using knn/dbap? Or simply that the speaker locations are taken into account when computing the individual speaker gains?

My ultimate goal is to experiment with different panning algorithms (e.g. the various flavors of vector-based panning, knn, dbap, spcap, hoa) to find which methods provide the best results for our use case. Ideally, I’ll handle time delay, level compensation, and room correction externally, directly on the speakers using the genelec software (rather than using spat5.align~), so my concern is that I could end up accidentally correcting for these factors twice when using knn/dbap.

Thanks!
-eric

tcarpent · July 2, 2025, 3:45pm

Hello again Eric,

For vbap, angular, hoa, only the directions (i.e. azimuth and elevation angles) of the speakers and sources are used to calculate the panning gains.
These types of panning assume that the loudspeaker setup is ‘surround’ (e.g. circular array or dome), and that delay/level compensation is applied.

For knn and dbap, the panning gains are calculated according to the distance between sources and speakers. The loudspeaker setup can be arbitrary, and delay/level compensation shall not be applied.

Best,
T.

elarrieux · July 2, 2025, 8:21pm

Thanks Thibaut, that’s helpful to know.

One more follow up question: in your opinion, are there any particular panning algorithms and/or settings (e.g. decoding types/methods in the case of hoa3d) that tend to produce the “best” results, i.t.o. realism/accuracy/etc.?

I suppose it’d be helpful to know: our setup is a sphere-ish 37 speaker array (more like a cylinder) with a diameter of ~3.25 m and height of ~2.5 m with speakers on 4 levels (see attached) + one sub outside the dome. And as far as source material goes, the main use case will be diffusion of speech signals in simulated “difficult listening environments” (e.g. cocktail party problem or blended with the ambience of a train station recorded with an ambisonic mic).

Thanks!
-eric


----------begin_max5_patcher----------
1751.3ocyassbihCD8YmuBW7brC5B21m186XlTtvFEGMCVPAxIY1ol+8UW.A
3fHDqjMTULtb2sT2mtkZ5.s98Mq71W7Bo1a8es9aqWs522rZkhjjvpleux6T
5KGxSqUh4s+LmWv7tUypLke3QJ63tJxAtdZ.ga8ucMzWcE2dc88Mifc9TwYd
NgqlNPCUMI9uJI5IwaeJ6nmYPzLktK1+iMvNUWkdhvIU6Hrz84pA52oDJynC
Is+byMxK2NSTxHOKTVqp3jWTfyqbccII8mBcdnnnJqdB2.FJgdTh7JvW6Gfu
e+vn9.fwILDn8sFRUC7Zv2JuGn4jmHU0TQ3qS5Udokk8Hup2PjNkeTnln3aM
jnLMoPCoJxSz1wGXnlVI7IbgC4bkBMduDh85llhLRE6L0DgTgmFSREHXhvac
Y5A8fkwqV1cd4nD05LTjZQWHV8KbremiVDoOlWb3mjrdKPDN5RBixJqH0DFO
k2X7F1YjGROmy28PAiWS+WkI.jQzQ3+PiINJSIHT1++TQSyMH3XEMqfIMhAg
BI4V0I1IEnV7DzGLJIXokiLXwJCgewByZAHOWuOsRFoZ1s.aYxKJxGxxLtbx
C7F1kTF6BuHunzNyJ5wGmXr6KDLOM0bq3Tu6LSycmXQAeWc5SC817z77lsqC
m9WRYzSobBmpCAPeCScFiGqOTUjmO.uZNOMBmLwh7CjmoY7GUJp+hAg3zx1E
QdlnbF8HolOjFO8X8PJ07eoc58Hcdeyl3cbxoxbAJFJfX6AslW+Xwy0MB1tP
quCnKyd+M08y8Mf9E4.K1dn3TYQMwrosepv6ZxDdG3tTRlTM9auc8FX31fv.
wp1sw36u86LiXvVwvASJGpUtjomObqb.zzSXfQv3omwvVAgvomwHifQSOiws
Bh.SOiIFAwwaiDRHmWv1P3.o.9Fn.1BsJkIffPSLWl3QP3DykIZDMkc0EK7m
xv5hDvorLSb.fmxzLQAP3T1lIF.Rlx1LQ.HXBaCZh.PzD1FzDBfASXaPSL.F
OgsAMAAj+T1lIJffSYaA82uZe6Z3f8q1kKZv9U6xEOb+pcASFte0dhB+g6Ws
KHX39U6BBGte0tfngBFf2hCgRAuPN7.mnUwBF5arJ2EYnrJWT+PbhdYfev88
yimSYjCEmYpj4nndbFoV1.ecUr5Z500zhi6UUu8JZ0UdNRURSWs6qp30anh5
J5E1igr1frc6OyxxI6xIri5aUCRv91jJKkmpUNRDoShEvDDIunAr7hjlOPrc
ZF+gEBiUiWttA.aFZhjjeTyu..rVJbh96DCG+tYBaL.+ALl9uMghwEHrhMJs
uIAItfUSsn9XkxETBkTQMVnEYtRLE3e0XRZ5QRL.fwKPfAbBXwgKVfAcBXIR
fgfKPbgbBW.e3hMhgcDYBssHiXAtgKkEsHAVni.SnjMwKwr8QNBLw31D.m28
17afhbLPfZyo3mHoeHRLEwRRPGuabrE7LCvfhsju3KCLIWOXjeFaqzWEVrVV
wbpoHbzzceYPAb8PIFMdhfuLr.c.KhwujfB55gRhb2BHb4fErCXQnsnEztkf
qGJJ4g3kCVBcAKRiaIAlHW.SXikr..Sfua22G3mnrI+kCZb3F+JEfwKFr3xM
9UU1tQ8u6rTPCvEzHuKyBJx.cAKguiGp0+Gf4Ju6uU6cA7X3boJ.qOGtEAvb
nl.qOGtEAvbn.AaOGtEAtbpVA3RMhAbsvgweNbKBb4ZIDaVn.ys5Ir7b3VD.
C7IAL4abNPrb.oJ+PhRXjNcSXXCpzusL3DWbAXvOoai8kCLzmS19ubbg+bxI
9kiqfO9ZDsWNBp4QJod73ieAr991Wqsp+VaZhxqreuTuScu4zI.51rDYtNy1
.X9ut+Xqut+9uH+CEmNQXC6uN0DwxHuzquR+P7NZieVtmPLd19G+wc.IVc..
Wc.c8RprGPrzgfJrI4Otmot3b0g1nXS.a8PXlQp4TloyZ+loONjxMq3x60F.
yzFRrZCMDa6C4lNEQ2Ws6R47J59ybsKqeiU+t5+yi4E6SyunoMGq8Puoy3bp
40eJsSMxlEMXaQIo5Us1tl0STxyjp0+MkQ4OS4Ot96cM5oXY3dAOjH62c4zZ
NgIH1zyvq8+t2a1L7.n5KDP2M7.qcCOxZ2veaymw5J9nVCXPSaCZ6SWOqsi8
3cQ+LC9idLD9nhclnSF4ARUdwyuoOVmAJ3i87Ff89jOWE4EoYpS7waAO4hFY
o1WK7rdrRPuIDUiZXNSMhuLOUCtec9IX+DTVRNE0kb5Ut14pH7LzSqw3jhPy
PQ3O.8.uJ8nCXWb5Vjp3hS0xEmnkWeZVreRVt7TrntygsSuxM+4l+CjS1lVF
-----------end_max5_patcher-----------

tcarpent · July 3, 2025, 9:35am

Hello Eric,

Unfortunately it’s impossible to recommend a “best” approach.
Spatial Audio is entirely a matter of tradeoffs (most often, things that improve spatial accuracy tend to degrade timbral quality, and vice versa).

As a very general rule of thumb :
VBAP can generate very “point-like/precise” virtual sources, but has a small sweet spot area.
HOA can be good for ambiance soundscapes and may provide a larger listening area.
HOA decoding with energy-preserving (EPAD) and in-phase/max-re is usually a decent choice by default.
WFS is also a candidate when the speaker setup is sufficiently dense (however, the timbral quality may be problematic especially for moving sources).
For surround speaker layout like yours, I would not consider DBAP or KNN.

Also, the good news is that your setup looks relatively dense and regular, so one can expect that any technique will yield decent results.

Hope this helps,
T.

elarrieux · July 7, 2025, 7:47am

Thanks for the info, Thibaut. That’s very helpful, particularly the note about DBAP and KNN.

The current plan is to build a mixed system, with vbab/vbip/dualbandvp for specific source location and HOA for general ambiance - the next step is to test out specific settings to see what yields the best results. We’re also fortunate that the test setup is built for a single listener positioned precisely at the sweet spot, so I’m optimistic that the combination will work well.

cheers,
-eric