You could imagine a solid cylinder about 20 feet in diameter and "whatever" long. Maybe 20*k where 0.75 <= k <= 1.5 ... whatever. We'll say it's a cylinder (deftly crafted from computer-generated plans, with gussets and whatnot) about 20 feet high and 20 feet in diameter. You can lay it on its side if you want.

These speakers tightly (honeycomb?) line the inside, facing inwards and are all übercontrollable. You can write a function for each speaker that varies with time and takes all the other speaker data as input. Whatever... "arbitrarily controllable speakers". The driver diameter is up for consideration. Whatever you like—16" sub-woofers if you lack imagination.

With all this, and with whatever power is required, you maybe run some "sequences" that came out of a deep-learner that knows (well) fancy sound wave physics.

Or you could break out your calculus text book and spend a year designing and testing.

Things that are definitely possible:

• "pumping" air through the cylinder
• Fire a (probably fluffy) projectile using the above
• Making a cyclone
• Having all speakers participate in a harmonic fumble-pile on a point at the center of the cylinder. You could have.
  • A big whomp whomp that repeatedly smashes the point from all directions, which would probably macroscopically be around the natural frequency of that "spring" of air being pushed against.
    • Near the dew point (with added humidity if desired), you could make the center "ball" in this case pulsate between visibility and invisibility as the water gets repeatedly squeezed out and sponged up again.

Things that might be possible:

• Make the above whomp-whomp space portal move around inside the cylinder. You could imagine the whomp-whomp dew-ball (above) being moved around as all the speakers smartly adjust their whatever. parameters.
• Some arrangements of "standing wave" that do things.
  • Can you do the above pressure ball with standing waves? Can you press with a standing wave? I don't think so...but. You could imagine creating a lens to shape sound going through the cylinder. It could be like a refracting sound telescope. Or maybe not that. The waves themselves have high/low pressure areas. You could deep-learn something, surely. Combine them up... A lens implies refraction due to density differences. Is there another parameter that is easier to manipulate that affects index of (sound) refraction in air...?

I feel like both 1) making standing waves of sound in a gigantic cylinder is hard but 2) hard in a regular, crackable way.

Other shapes?

• Cube :: Yuck. Good luck.
• Sphere :: For special purpose purposes. Math might be easier? Harder? It'd be hard to "appreciate" as a mega-art kind of thing, but it could do something practical...who knows what. No, not fusion.
• You Name It :: If you think there's a mathematical advantage to some shape or other, then righton.

What about all the vibration? I mean, this thing might cannot be made out of plywood because

1. All that stuff will harmonize and introduce noise which would need to be modeled, which would be a nightmare.
2. It needs to be strong because not only is it trying to shake itself apart, it's probably under a lot of extreme forces depending on how much you are pushing (meaning most of the structure itself is pulling)

I think designing a cylinder like this that needs to cary a (dynamic) load all more or less all directed outward. You could imagine a huge cylinder floating in space with a bunch of cars driving in side along the circle: That's a bridge! (sorta) The bigger point being, we have a lot of practice with bridges.

Seems to me like, especially if you're just throwing deep learning at it, having a few unobstructive sampling points would be good. (Depending) They shouldn't interfere with operation, they only really need to measure pressure. The sensor "arms" reaching inward (as I imagine it) only need to withstand the forces. "Wind" being one, but also (surely) crazy harmonics. If you are running some program from your deep-learner and you want to troubleshoot, having expected vs actual pressure "at these points" may help. Could also have a robotic probe: A cable running through the center that is robotically controlled and kept tight(sp).

Nice modeling potential. Depending on your budget and the quality vs size of speaker you can obtain, a small physical model of the jumbo one described would probably be a practical way to simulate.