Engines ------- An ordinary four-stroke car engine can run up to about 6000 rpm without difficulty, and run this way, it generates 50 Hz of impulses per cylinder; so a four-cylinder engine generates a 200 Hz pulse train, if the cylinders are timed evenly. If the engine is under no other load, I think a substantial portion of the engine's energy input is dissipated as sound.
Normally we do our best to muffle this sound (although I hear a Harley scooting by as I write this), but it seems that it could be put to productive use. Suppose we run our four-stroke four-cylinder engine at 1500 rpm instead; the resulting impulse train would be 50 Hz (with impulses 21 feet, 9 inches apart, in dry air at STP). Music from engines ------------------ You can control the speed of the engine to produce pulse trains at different frequencies; Asiatech played "The Saints Go Marching In" on one of their V10 Formula One engines. But the tone color is inflexible and rather ugly, and it's hard to get above 1000 Hz. (Asiatech used a V10 that could spin at 330 Hz, so they could get up to 3.3kHz.) Controlling the valve timing might be a more effective approach. I should experiment to determine how well you can simulate a sound merely by adjusting the timing of an impulse every 20ms. Music by reflecting impulse trains ---------------------------------- Now imagine that we use a directional pipe or a parabolic reflector to direct this pulse train out into open air at right angles to a listener. For 21 feet, 9 inches, we pass it along a linear array of 800 switchable reflectors, each capable of either reflecting some of the sound toward the listener, or allowing it to pass by. So we configure the reflectors into some pattern, such that some of them reflect toward our listener, and some don't, and pass an impulse over the array. Now the listener receives not a single impulse, but a sequence of impulses. If all the reflectors reflect, the sequence arrives at 40kHz; but by configuring these 800 reflectors in a particular way, we generate a very individual 20ms of sound. Each reflector is a single bit of output from a one-bit D/A converter. On 1-bit D/A converters, see Jim Thompson's "Care and Feeding of the One Bit Digital to Analog Converter" at http://www.ee.washington.edu/conselec/CE/kuhn/onebit/primer.htm for details. It says CD-quality audio requires around 164MHz on a 1-bit DAC, but that other techniques are also used in practice, and refers to Finck's 1989 article, "High Performance Stereo Bit-Stream DAC with Digital Filter," which I don't have a copy of. Uwe Beis wrote http://www.beis.de/Elektronik/DeltaSigma/DeltaSigma.html "An Introduction to Delta Sigma Converters," which says that 64 is a typical oversampling rates for audio applications, resulting in bit rates around 3 MHz. Now, if we merely run the engine to produce a sequence of impulses, a new impulse will come every 20ms, so if we leave the reflectors alone, they will repeat the same 20ms signal over and over again. This may not be the worst possible failure mode for a sound output medium, but it's not that great. But if we can change the position of each reflector within 20ms, we can generate a different 20ms of sound from each engine impulse --- so we can generate a fully custom 1-bit waveform! 20ms is pretty fast for a mechanical device, though. If each of the 4 cylinders has a separate set of 800 reflectors to pass over, then you could have 80ms to reconfigure the reflectors instead. You can get solenoids that respond that fast, and I think 80ms solenoids might be considerably cheaper than 20ms solenoids. We still have a couple of problems, though. One is that we still need 21 feet, 9 inches of reflector array at right angles to our listener, and that's large. Another is that presumably reflectors will obstruct one another's sound. A third possible improvement: a jake-brake driven by an engine might be a better source of impulses than an engine alone. If the reflectors were spaced, say, 13 inches apart, an impulse passing over them would generate a 1kHz pulse train. But if they reflected the pulse back in the direction it came rather than at right angles, they would generate a 500Hz pulse train, due to the Doppler effect --- in effect the source of the sound is zipping away from you at Mach 1. So by this simple trick you can get away with a reflector array that's only half as long, at 10 feet, 10.5 inches. You still need 800 reflectors for 40kHz, though, so you have to space the reflectors closer together, at 6 per inch instead of 3. Holes in a tube --------------- The "reflectors" might actually be merely holes in the side of a tube, like the valves on a flute or saxophone. Now, in a flute or saxophone, the purpose of the valves is to reflect sound when open and consequently change the resonant frequency or frequencies of the air cavity. This kind of resonance will probably pollute the sound from this instrument, and I'm not sure how to handle that. It might be possible to use a closed loop of tube and feed new impulses into it going around in a specific direction; this would allow sound energy not released through reflectors on one trip to be preserved rather than dissipated in some kind of a muffler. With 2" pipe, the size of the exhaust pipe on many cars with 4-cylinder engines, you can reasonably make a two-foot-diameter circle; if you run it in a helix, you'd need less than one and a half turns to reach the 11-foot length described above, and less than three turns to reach the 22-foot length described earlier. So the whole exhaust manifold with the four pipes would be about 12 turns, or two feet --- the whole thing would fit in a 2'x2'x2' cube, comparable in size to the engine itself. Some cars with engines like this use much smaller exhaust pipes. If you only wanted 8kHz, you would only need 160 valves or reflectors per exhaust pipe at 1500rpm. (More exhaust pipes doesn't let you use fewer reflectors per pipe; it just reduces their duty cycle so they can run slower.) Mechanical modulation --------------------- I pointed out that 20ms was pretty fast for a mechanical device, but of course the timing of the engine valves is tighter than that --- at 6000 rpm, a full up-and-down stroke takes only 10ms, and the valves must open and close in roughly the right part of this cycle. (Typically, if they don't, for example because the timing belt breaks, they destroy the engine forthwith.) I think engine valves are still usually driven by a camshaft rather than solenoids, judging by the "DOHC" markings I still see on new cars. Suppose you used a straight pipe, 2 inches across and 21 feet long, and cut a lengthwise slit 2 mm wide all the way down the length of it. Then you nest it snugly inside a slightly larger pipe that's free to rotate around the smaller pipe. Now 2mm-diameter holes drilled in the outer pipe will result in impulses being released from particular parts of the pipe when it's rotated to the correct position. If you rotate it by 2mm every 20ms, you repeat the same signal every 1600ms or so. Presumably you'd put a muffler on the end of the straight pipe to absorb the unused noise. Media ----- Obviously a medium with a slower speed of sound would permit a smaller apparatus. I don't know what medium that would be, though. Probably a tube filled with hot engine exhaust will have a significantly different speed of sound, but I think it is higher. Credits ------- This was inspired by Dead Media Working Note 42.1, about old Mayan "sound recordings" in the form of stone structures whose echo PSF was a 100-ms bird's chirp --- so clapping your hands near them makes them chirp back at you. It was also inspired by the recording of someone at Asiatech playing "The Saints Go Marching In" on an engine (sha1 of the mp3 is b363942f854c02f98d745787941c27298ad8b29e), by the Harleys that drive by my house, and by the folks out on the playa in 2003 who generated 5Hz impulse trains by repeatedly detonating giant columns of propane-air mixture, making noise that shakes your bones from miles away.
