* Brandon Hale <bthaleproductions(a)gmail.com> [2021-02-26 16:27]: Have a look at the iem_ambi library, it has been available for almost 20years by now I think in some form or another.

On 2/26/21 7:26 AM, Brandon Hale wrote: ... Hi Brandon, You will need two things: - a way to design a decoder - a way to run that decoder For designing Ambisonics decoders I use ADT (the Ambisonics Decoder Toolkit, by Aaron Heller): https://bitbucket.org/ambidecodertoolbox/adt/src/master/ You will probably want to read through the references there to learn about the different design methods available, the respective tradeoffs and tricks. It is not extremely difficult but like everything else you have to learn stuff. This is definitely not a "single button, press here for decoder" solution. ADT will generate a proper Ambisonics decoder, dual band with phase matched filters, near field compensation filters for the speakers and (optional) level and delay compensation for your array. It can output ambdec configuration files (but limited to 3rd order, of course) and also Faust source code for the full decoder (which is what I currently use). Faust can be compiled to any number of plugin and whatnot formats. I have used standalone QT/Jack apps in linux and supercollider plugins - I think it can easily generate pd as well. For very irregular arrays and "weird speaker locations" there is, sorry, no magic. In some cases you might conclude that Ambisonics is not the correct solution (for very irregular arrays). I have successfully used ADT to create decoders for our concert systems (up to 32.8) and for our Stage small concert hall with a 56.8 system and Listening Room (full 3D 22.8). A LOT of work to get properly calibrated arrays but the result sounds pretty darn good (if I may say). My experience with fairly regular "dome" speaker arrays is that ALLRAD works best (Slepian decoders have the appeal of using the "right" harmonic functions but in my experience they are not good with highly irregular arrays). If your speaker coverage has "holes" then it is all about tradeoffs. You can optimize for even energy (loudness) but then you will have directional errors and/or mismatches of perceptual cues at low and high frequencies, if you minimize directional errors you will have uneven loudness. If you want Ambisonics you need decent speaker coverage for best results. You can look at the plots of meaningful parameters when designing your decoder with ADT and see what you get for different configurations. This is all with the caveat that there are limits to your perception, so what looks like an error on paper might not be significant in practical listening conditions, and as Fons rightly points out, going above order 7 does not make much sense. Some data from a composer's point of view here (Natasha Barrett): https://www.natashabarrett.org/Barrett_IRCAM_Report_v2.pdf In the paper she explores the limits of what was for her significant in terms of max ambisonics orders. My current compositional work uses mixed 5th/2nd order Ambisonics and I am quite happy with the results (in a 56.8 system that can potentially decode 6th order[*]). Hope this helps, -- Fernando [*] https://lac.linuxaudio.org/2019/doc/lopez.pdf

On 26.02.21 16:31, Brandon Hale wrote: Just a note as there is lots of decoder discussion: TASCAR has a C++ implementation of mode matching and ALLRAD methods for automatically generating decoders, validated against ADT. In case of the ALLRAD method there are minor differences because we use a different method of creating regular (virtual) layouts. TASCAR does not support NFC or dual-band decoding, thus you may argue that is does not provide true Ambisonics decoding (we typically use max-rE decoding for full bandwidth, and argue that NFC with limited low-end performance of loudspeakers and sufficient loudspeaker distance is just a wast of computational power). So in the end the HOA implementation of TASCAR is actually quite close to VBAP. There is no order limit hard-coded into the system. TASCAR can also encode 3rd oder Ambisonics with FuMa normalization without decoding (and also encode arbitrary order SN3D normalization) which can be decoded with external decoders, e.g., ambdec. For reproduction of (pseudo-)diffuse sounds we use first order Ambisonics, with loudspeaker decorrelation to avoid strong comb filter artifacts. Best, Giso

Ambdec looks nice, but I wish there wasn't a limit on order. 36 speakers may not be enough for me. Do you know of anything for beyond 7th order? I wasn't clear enough in my original email, but I am really looking for something that can go past 7th order and can do any speaker placement. On 2/26/21 11:44 AM, Chris Caudle wrote:

This is good to keep in mind. Have you ever seen the ICST plugins for max msp? Those objects have a method of putting in xyz coordinates for its ambisonic decoders. Where I work at, we use this to output ambisonics to all of our spaces at ICAT, including the Cube <https://icat.vt.edu/studios/the-cube.html>. It does work well, but I do know of colleagues who complain about how "fuzzy" the spatialization is. Do these kinds of decoders fudge the decoding to achieve output? I really want to find something like this for Linux. The reason I ask is because I want to find a Linux-based system for the Cube, which has 138 speakers arranged in a rectangular prism. I am constantly looking for a way to decode ambisonics to it without having to use max msp. I have not had much luck finding something standalone or in pd. Maybe I should learn supercollider? It can be hard to find decoders that go to 10th or 11th order. At work, I may be getting hold of 64 AVB speakers that I can arrange in a line to experiment with WFS. The great thing about this is the speakers are AVB, and not Dante. I may be able to play with these speakers on Linux. Of course, there is at least one WFS processor (is that the right term?) that I know of in max. But, I find max too annoying and locked down for real use, and of course I want to do it on Linux! Thanks for the detailed points, Brandon Hale On 2/26/21 12:17 PM, Fons Adriaensen wrote: > On Fri, Feb 26, 2021 at 11:58:28AM -0500, Brandon Hale wrote: > >> Ambdec looks nice, but I wish there wasn't a limit on order. 36 speakers may >> not be enough for me. Do you know of anything for beyond 7th order? I wasn't >> clear enough in my original email, but I am really looking for something >> that can go past 7th order and can do any speaker placement. > Ambdec is currently limited at 3rd order, but that could be increased. > Also it does not compute the actual decoder coefficients, these have > to be supplied in a preset file. > > Keep in mind that > > > 2. Higher order means more speakers. 7th order would require at > least 64, assuming they are in a regular grid, and probably a lot > more if they are not. Less speakers would mean that the information > in the higher orders can't be used, and any correctly computed > decoder will actually just ignore such information. > > > > > Ciao, >

Thanks for all of this good info Peter. I didn't know about the sursound mailing list, and will experiment with that pd library you told me about. This is all a learning experience for me. When you say 5th order 3d, wouldn't that only be 36 loudspeakers? Brandon Hale On 3/1/21 10:12 AM, Peter P. wrote:

I didn't know you could do that. Thanks for changing my life and teaching me about this Peter. Brandon Hale On 3/1/21 11:27 AM, Peter P. wrote: