On Wed, Dec 21, 2011 at 12:51:45PM -0800, Eric Carmichel wrote: > 1. Is there any preferred method of calibrating speakers used > in an Ambisonic setup?
If you have a decent set of identical speakers there is in general little reason to do this - or at least it's not a priority compared to getting other aspects (see below) right. Now your application context is not the usual one of listening for entertainment and that may change the picture. Whatever method you use to measure/EQ the speakers, beware of any form of 'extreme' or very detailed (in the F domain) EQ. For Ambisonics to work well you want the speakers to remain matched, including phase response up to a few kHz. As a measuring method I'd use a log sweep with deconvolution to an IR. Then compute the (complex) inverse response with normalisation - don't try to correct anything that would require lots of EQ. > 2. Has anyone compared or noted differences between the Virtual > Visual Microphone (VVM) software and offline processing using > MATLAB? There is no reason to use off-line processing. AMB decoding done in real time takes 1% or so of CPU load on a modern PC. If you think of speakers feeds as directional microphone patterns then these patterns should be frequency dependent. This is particular true for small diameter rigs, and related to some of the things discussed below. VVM doesn't provide this AFAIK, and I would not recommend it as an AMB decoder. > 3. I have seen discussion and articles regarding Ambisonics and > shelving filters. Any recommendations as to "best" filter settings > based on speaker-to-listener radius? The issues of number of speakers, shelf filters and near-field effect compensation, while not directly related to each other, can be understood only by looking at a bit of theory. AMB, if reproduced using the so-called 'systematic' decoding matrix, reconstructs the sound field in a limited area. The size of this area is measured in wavelengths, so it can be very large at LF and will be very small for HF. For first order, the radius of this area is around half the wavelength, and it increases for higher order. Now if both the listener's ears are within that 'area of reconstruction' all directional cues will be the same as they were in the original soundfield. But this possible only up to a few hundred Hz, and even less if you want a larger listening area. Outside the 'area of reconstruction' the sound field produced by a systematic decode (complex interference patterns, since the systematic decode will typically use nearly all speakers even for a single source) is not one that works well. So even for a single listener system above 700 Hz or so something else is needed. The optimal solution here is a decoding that optimizes the magnitude and direction of the energy vector, (the vector sum of the energies from each speaker) known as the 'max rE' decoding. Note that the crossover between the two regimes matches the frequency range where inter-aural phase differences become ambiguous and where our directional hearing switches to using amplitude differences instead. So a good decoder needs to be frequency-dependent. There are in practice two ways to implement this. The first is to use the same matrix for both regimes, and use phase- matched shelf filters on the input (B-format) signals to modify the effective matrix coefficients. This works well for regular layouts. The second one is to use phase matched crossover filters and two fully independent matrices. This is required for non-regular layouts (e.g. 5.1 and most 3-D installations which are almost never regular). Let's now look at the number of speakers. It has already been mentioned that eight speakers for horizontal-only first order is too much. This is absolutely true, you shouldn't use more than six in that case, and even that is a compromise (but a good one). To understand this you need to look at the radial dimension. As you mave away from the 'sweet spot', the soundfield reconstruction depends more and more on higher degree spherical harmonics. This is the reason why the 'area of reconstruction' is limited in the first place. The mathematical expression of this is the Fourier-Bessel sum. With a first order input obviously only the zero and first degree components can be reconstructed correctly. So what about the higher ones ? In the sound field of a 'real' source they are present. In an AMB rig they will be present by spatial aliasing. Think of your ring of speakers as samples on a periodic waveform to understand this in an intuitive way. A ring of eight speaker driven by an AMB decoder will reconstruct the (horizontal) components up to 3rd order plus one of the two 4th order ones (in exactly the same way as eight samples on a periodic waveform allows up to the 3rd harmonic and 'half' the 4th). If the input is just first order, the 2nd and 3rd order components (and their higher order aliases) are forced to zero. And in practice that is worse than allowing them to exist by aliasing. In practical terms, forcing these components to zero means that just around the 'area of reconstruction', in the region where the surpressed spatial harmonics dominate the sound field, there will be a 'gap' - a region with much reduced magnitude and rather confusing directional cues. Since the dimension of this region is expressed in wavelengths, each of the listener's ears will be in it for some part of the frequency range. Finally there is the matter of near-field compensation. The non-zero degree field components generated by an AMB rig have the same near-field structure as 'real' ones. That is, they will show a 'proximity effect' depending on the radius of the rig. This needs to be compensated, as for a close sources captured with an AMB mic any near-field effects are already present in the B-format signals. This is of particular importance of course to small radius rigs. If all speakers are at the same distance from the center the compensation can be done by specific high-pass filters acting on the B-format signal. If the speaker distances are irregular it requires these filters on the individual per order matrix outputs for each speaker. Very few AMB decoders provide both dual-band operation (or shelf filters) and near-field compensation. I'm the author of one that does (Ambdec), but since I suspect you are on Windows you can't use it. Ciao, -- FA _______________________________________________ Sursound mailing list [email protected] https://mail.music.vt.edu/mailman/listinfo/sursound
