Hi !

I recently read about IOSONO's sound reproduction system (doing wave field synthesis) and I was curious about how/well this works. It' quite interesting but I guess I'm not close to understanding all the details. The idea is simple, though.



While it's more or less clear how virtual sound sources are simulated that are located outside the 'perimeter' I'm not sure how they want to create a wave field so you think that some sound is coming from a (virtual) source that's located inside the room.

Here they're explaining the principle.

I tried to simulate this on my own:
XviD AVI video (1.6 MB)
It might be me (not doing the right things) but it doen't look that convincing. (The wave field in the speaker simulation case should look the same as in the reality case). Clearly, the more speakers you have, the better the approximation ...
With 128 speakers at the invisible vertical wall in the middel it still looks bluured. Maybe delay & amplification is not enough, maybe each speaker needs a specifially filtered signal ... (?)

Comments ?


edit: I'm currently reading some papers about WFS...

Sebi

smacks of ambisonics, but extended somehow.

beats me how one would record this though. sort of sonic holography?

What a coincidence. I am currently graduating on Wave Field Synthesis (my master thesis is almost finished).

So, if you have any specific questions, let me know.

About the "blurring", it is impossible to synthesize a wave field exactly, because the distance between the speakers should be infinitely small (Huygen's principle). The greater the distance between the speakers, the more blurring. This effect is called spatial aliasing (an effect of spatial sampling -> comparable with temporal sampling).

Currently, we're working with a prototype setup (not using conventional speakers, but Multi Actuator Panels (MAPs)), where the distance between two speakers is 0.1675 meters. In the worst case the frequency up to where no aliasing occurs is equal to:

f = c/(2*dx)

with c the velocity of sound, and dx the distance between the speakers.

So, with our setup, f is approx. 1000 Hz. With normal audio signals (music, speech, etc), about 70% of the dB(B) weighted energy spectrum is below this frequency. Hence, it will not lead to severe problems in practical situations.

I can type *very* long posts on this subject (it is very interesting), but I will post my thesis here when it's in its final version.

One addition: WFS is still in a development phase, but a prototype is already up & running in a cinema in Ilmenau, Germany. This could very well be a good follow-up for Dolby surround, because in theory any virtual source can be synthesized (even "focus" sources, sources that have positions in front of the speakers instead of behind the speaker array).

Another major advantage over Dolby surround is the fact that WFS has an extended listening area, where listeners will hear the wave field like it's supposed to sound. With Dolby surround, that is only the case for a very small area (the sweet spot).

Just out of curiosity, what do you think you will do with this, career-wise?

I ask this sincerely as someone thinking of grad school 25 years after getting my B.S. in engineering.

My most marketable skill is software development that deals with markets (stocks, commodities, etc.). But that's not at all what I really like. I *like* photography, but discovered the hard way a few years ago that I did not like it when I *had* to do it. (I had freelanced professionally a few times more as favors and then began accepting gigs and, as I became busier, found I wasn't enjoying it any more at all.)

The other thing I really like is music (not as a musician due to lack of talent , but the technical side), but am averse to looking into something like what you did for a couple of reasons. The first reason is really why I asked my question. The second reason is a fear that I would have a similar experience as I had with photography (and suspect that I would really, really hate the Industry, so to speak).

Anyway - what does someone "do" with a Masters in Wave Field Synthesis?

Thanks,

Mark

heheh maybe I'll check it out ... (it's about 330 km away from my location).


With what signals do i have to feed the spakers to create a virtual sound source that is located IN the room ("focus") ? For the little test I just calculated the per-speaker delay & gain according to the distance to the virtual source. Seems to work for virtual sound sources behind speaker arrays. I suspect the proper formulas to be more complicated, though.

About spatial aliasing:
In my test it looks like the impulse response is "blurred" but nearly uniform over the room's space which I think suggests feeding the speakers with an inverse-filtered signal to compensate the aliasing thing a bit. Does this make any sense to you ?


Sebi

Actually, I'm getting my Masters in Physics. I've chosen Acoustics (Sound Control) as my "speciality", with a WFS related research as my graduation assignment.

WFS is currently developed by universities (one of them is the TU Delft in the Netherlands), so those are the only places you can work on this particular subject.

Well, what does someone graduated in physics do.. chances are that I will be getting my PhD, maybe in the field of WFS, maybe in some other acoustic related field. A lot of graduated students find work in R&D departments or engineering companies. Some even work for banks or insurance companies!

If you want to know more, PM me, to avoid offtopic chat

You should do that, I have to check it out for myself, yet. They show "conventional" 5.1 surround movies using the new WFS system. I've heard that the extended listening area is a real advantage.


For simplicity, take a virtual monopole source. For a "normal" source, behind the array, the driving filter for each speaker is:

F(w) = exp(-i*w*r/c)/r [1]

which is the well-known Greens function for monopole sources, with w (omega) the frequency (2*pi*f), c the velocity of sound, and r the distance from the virtual source to the speaker.

A focus source is basically the inverse:

F(w) = r*exp(+i*w*r/c) [2]

Note, that this filter is anti-causal. You have to add a delay to both [1] and [2] to make it causal again. This puts a limit on the distance between the source and the speaker array.

With these filters, you can calculate the driving signal for each speaker:

Y(w) = F(w)X(w) [3]

with X(w) the sound the monopole is emitting.


Yes, and unfortunately it doesn't work. Like temporal aliasing, spatial aliasing can't be compensated for. Spatial aliasing is highly place dependent. So, you can compensate for the aliasing on one exact position using inverse filtering, but the filtering will make the aliasing even worse on other positions. [edit] Spatial aliasing can be avoided as much as possible, by making spatial anti-aliasing filters. Recall temporal anti-aliasing filters: they filter out frequencies that are too high. Spatial anti-aliasing filters filter out angles that are too high. However, in practice the filtering of angles may do more damage than the aliasing effects. [/edit]

Another problem you will encounter performing WFS in the 2D plane, is the amplitude error. The WFS theory (based on Rayleigh) assumes an infinite plane with an infinite number of speakers. Today, line arrays are used, with a finite number of speakers, introducing a 1/sqrt® error in the amplitude, spatial aliasing and diffraction effects due to the limited size of the array. The amplitude error and diffraction effects can be attenuated using certain mathematical tricks.

However, in practice all these limitations are not leading to major issues. I'm pretty sure WFS has very important advantages over existing systems. You can see it as "virtual reality" for sound.

Thanks for the reply.

Does F(w) (a speaker's transfer function) always describe a delay + gain change (even in the most advanced WFS systems?) or is this a simplification of yours ?


Sebi

This is indeed the most simple equation, but it is the one you should use in simulations. More complex equations exist to synthesize other types of sources (plane waves for example), or ones that include other operations like equalization and room compensation. The last two are the ones I'm focussing on. Especially room compensation is interesting, it is used to compensate for reflections in the 2D plane (from either the walls of the room or the setup itself). [edit] Furthermore, the equation can be extended by a "2.5D Rayleigh" correction, to compensate for the amplitude error using a line of reference. It is basically a filter with a sqrt(i*w) shape. You will notice, that omitting this filter will result in a wrong energy spectrum when you use an array instead of a plane of speakers. [/edit]

In fact, the best way to calculate the driving filters for frequencies below the spatial aliasing frequency is by using matrix inversion (thus, numerical methods). That way, you minimize the error between the desired wave field and the synthesized one. The problem here is, that the matrices that have to be inverted are almost always singular and appropiate regularization methods have to be found. But that's all discussed in my thesis...

By the way, your simulation looks good. Very nice.

Hey all...been following your post. I am currently getting my masters from RPI in acoustics. For my thesis I am working on merging 3d surround sound with wave field synthesis-- I am a bit stuck on the filter design for the WFS I was hoping someone would be able to to assist me in the filter design for the system...I am planning on using BruteFIR for the real time convolution. Any help would be greatly appreciated