Hi, I enjoyed your comments. I have pondered long and hard about this.
On 12 Jun 2019, at 17:00, sursound-requ...@music.vt.edu wrote: > 1. practical HOA encoding (Richard Lee) > > From: Richard Lee <rica...@justnet.com.au> > Subject: [Sursound] practical HOA encoding > Date: 12 June 2019 10:09:00 BST > To: 'Surround Sound discussion group' <sursound@music.vt.edu> > Reply-To: "rica...@justnet.com.au" <rica...@justnet.com.au> Clip > What HOA does is define the 'point' to be larger at any frequency. We > don't need a larger 'point' at LF so FOA is usually sufficient at LF. At > mid & higher frequencies, HOA encoding allows a larger 'point' to be > defined. Clip Surely you mean that the point is smaller, and direction more precise with HOA. > Back to HOA encoding. Basic ambisonic maths for artificial sound objects does not include distance modelling. Everything is assumed to lie on the surface of a a sphere of radius 1, unity, which is dimensionless. The loudspeakers are all at distance 1. If not they can be acoustically corrected by delaying nearer speakers to match the furthest. That brings up time. The time of sound arrival at the central listener is in miliseconds, which has dimension. Gerzon suggested a way to maintain the perceived volume of a sound moving inside the sphere, by increasing the W component, as the XYZ values go to zero, to compensate. Otherwise the sound would be appreciably quieter when placed at zero, getting louder as it moved away until it reached unity distance. This is a good start. To simulate things outside the unity sphere you have to start with the inverse square law. Amplitude of the sound should be 1/d to follow the power inverse square law. 1/d goes to infinity at d=0. You can simply add 1 to d. 1/d then goes to 1 at zero, and to 0.5 at distance 1. That works reasonably well, especially if you use a send to a suitable reverb using a 1/d sq law. The direct signal diminishes at a faster rate than the reverberation. Just what Chowning suggested. Doing both these leads to what seems an excessive loudness at zero in comparison with that at other positions inside the sphere. The effect of the modified inverse square law is excessive near zero. Most, if not all HAO components will go to zero at zero distance. You are left with W. That is essential, so we have to deal with inside the sphere separately from outside the sphere. We could sensibly limit the minimum distance to outside the head. The mathematics has to have a smooth cross-over at distance 1. The inverse square law can be limited to minimum 1, or scaled between zero and 1 if desired. When we come to the real world, the speakers are a distance away. This has dimensions, distance, and time of sound arrival due to the finite speed of sound. The ambisonic sphere then has to be put on something with real dimensions More distant sounds take more time to arrive. Sound moving towards or away from the listener produces the Doppler effect. You can model those by adding a delay, based on distance, to the sound object, but probably want to make it scalable per object if you wish to avoid the pitch changes. You will then be adding that delay to the acoustic delay. Some recordings will have a built in Doppler effect, and their own acoustic character. Imposing Doppler shifts on those could be very disturbing. How do we deal with moving between different size working spaces ? The speakers should still be at distance 1, though that varies in dimensions distance/time. It would be nice to make delay zero at distance 1, but negative delays are impossible, unless you can move sound objects earlier in time. The centre of the listeners head has to be at zero distance. A soundscape should not change dramatically when put in a different place, except for the acoustics of the place. Everything should be relative to 1. Delays and reverb based on real dimensions of the speaker array will change perceived size dramatically and inappropriately. I've never really resolved this. I have tried a global delay scaling, which sort of works, but hard to make automatic Ciao, Dave On 12 Jun 2019, at 17:00, sursound-requ...@music.vt.edu wrote: > 1. practical HOA encoding (Richard Lee) > > From: Richard Lee <rica...@justnet.com.au> > Subject: [Sursound] practical HOA encoding > Date: 12 June 2019 10:09:00 BST > To: 'Surround Sound discussion group' <sursound@music.vt.edu> > Reply-To: "rica...@justnet.com.au" <rica...@justnet.com.au> > > > .. a reply to stuff on the "Anyone had experience with AllRAD decoding to > physical loudspeakers?" thread where Aaron brought up the topic of NFC. > > " A practical problem is that forward NFC filters have infinite gain at DC. > Jerome Daniel's solution is to have a reference decoding distance for > Ambisonic program material, say 1 meter -- In the encoder, you encode and > then decode for speakers at 1 meter (which produces finite gains at DC). > > [1] J. Daniel, "Spatial Sound Encoding Including Near Field Effect: > Introducing Distance Coding Filters and a Viable, New Ambisonic Format > ," *Preprints > 23rd AES International Conference, Copenhagen*, 2003. > > [2] J. Daniel and S. Moreau, "Further Study of Sound Field Coding with > Higher Order Ambisonics," *Preprints from 116th AES Convention, Berlin*, > no. 6017, 2004. " > > Daniel introduced this into a shark feeding frenzy of gurus & pseudo-gurus > debating HOA encoding. I wanted to join in and introduce Yet Another > Encoding but beach bum issues intervened. Besides I was seriously > intimidated as I neber wen 2 skul. > ___________________________ > > MY SOLUTION TO LOADSA LF GAIN FOR HOA ENCODING > > Ignore it. > ___________________________ > > Making a HOA soundfield, we deal with 2 type of sources/objects. > > The first is naturally recorded stuff from a well calibrated soundfield > mike and IMHO, these are the most important. It was the incredible sense > of REALITY playing an original Calrec Soundfield recording on a properly > set up Ambisonic system that pulled many of us into the arcane world of > Ambisonia. > > As a(n Ambisonic) mike designer, you immediately come up with a serious > problem with correct Ambisonic encoding .. the huge LF boost at close > range. > > Even the First Order Ambisonic (FOA) Calrecs & TetraMics have 30dB/8ve LF > filters to alleviate their facility as seismographs, air-con, slamming > doors, distant thunder .... detectors. And you will find even more LF > filtering required at times. The pretenders w/o such filters WILL have > serious problems. > > A practical HOA mike has additional problems as it must use cancellation & > mixing of 1st & 2nd order microphone capsules + LF boost to get the high > orders which give serious S/N problems at LF. You see this with full force > in Eigenmike which uses omnis which need even more LF boost. Angelo Farina > has several reports & papers showing how useful (??) this is. > > The solution is to drastically HP filter the higher orders AND MATCH THE > PHASE OF THE UNFILTERED CHANNELS. You can do this with the method in the > appendix of BLaH3 > > http://www.aes.org/e-lib/browse.cfm?elib=14705 > > In essence, a practical HOA microphone will give you as much HOA signal AS > IS USABLE (mostly determined by S/N) while keeping the correct relationship > between the components. > ______________________ > > But you will argue, "But that means the Ambisonic order will be frequency > dependent???" > > Why is this a bad thing? Ambisonic playback is also frequency dependent. > > An Ambisonic recording is the DEFINITION OF SOUND AT A POINT. But this > POINT is frequency dependent! > > For a FOA recording, at 20Hz, an IEC listening room is a 'point'. But at > 1kHz, the 'point' is about the size of a Mk1 Human Head. > > What HOA does is define the 'point' to be larger at any frequency. We > don't need a larger 'point' at LF so FOA is usually sufficient at LF. At > mid & higher frequencies, HOA encoding allows a larger 'point' to be > defined. > > So a practical HOA mike as described above degrades in exactly the correct > fashion to be USEFUL. > > A PRACTICAL decoder can make use of as much info as is available in the > recording without worrying about EXACTLY what is available at each > frequency. (Actually this isn't strictly true but that's a whole new field > of research in itself which may lead to Yet Another HOA Encoding and/or > Decoders :) ) > ____________________________ > > Back to HOA encoding. > > The 2nd type of source/object are artificial. HOA is most useful for > shoot-them-up games and other virtual reality stuff. You pan the (mono) > bad guy to someplace in your HOA space > > BUT THIS STUFF CAN USUALLY BE SET AT AN INFINITE DISTANCE SO NO NFC IS > REQUIRED > > It's very rare that proximity adds anything to the virtual reality gaming > experience. (In practice, 'infinite distance' can be considered anywhere > outside the 'circle' of speakers.) > > If da bad guys do come closer, you can use the same drastic HP filters to > keep the LF boost within limits. ie treat the sources as though they were > being picked up by a REAL LIFE HOA MIKE. > > There is NO NEED FOR SET DISTANCES/FREQUENCIES etc. You only need to MATCH > THE PHASE. > > And usually you don't need to add NFC to the HOA signals for a realistic > effect. > > FOA (with NFC) is certainly good enough for the Spitfire flying low > overhead .. or even Peter Lennox attempting to run over Grand Vizier > Malham's Mk3A Soundfield with his motorcycle. (one of the very first > Soundfields with the correct NFC encoding) > > So in most cases, just use LF boost on the FOA signals for NFC > ____________________________ > > You can apply this to any of the generic HOA encodings. When I last looked > seriously at this, Furse-Malham was favoured up to 3rd order. I gather > some other Encoding is now 'standard'. :o > > Can some of you can enlighten me on which 'standard' has won out and is > commonly incorporated into HOA Decoders ? > _______________________________________________ Sursound mailing list Sursound@music.vt.edu https://mail.music.vt.edu/mailman/listinfo/sursound - unsubscribe here, edit account or options, view archives and so on.
