Surface shipping already uses something similar, mounting buoyant distress/telemetry beacons which automatically deploy in an emergency (I don't know the exact mechanism, but I'd assume it's triggered by water pressure or something similar).

Some modern recorders on board aircraft can apparently do similar tricks, but I'd imagine it's more difficult to make it work reliably; the stresses involved in a plane crash are likely far greater, and I'd suspect the number of places from which a device could both survive and get free to deploy is much smaller.

Let it send a signal.

Aircraft have floating, jettisonable radio beacons (Emergency Locator Transmitters) that transmit on the international guard frequency 121.5MHz (with intentional harmonic distortion that allow reception on the military guard frequency of 243MHz). That frequency is monitored by SARSAT (Search and Rescue Satellite) in North America -- the middle of the South Atlantic may not be monitored, since the automated search notifications would not necessarily reach anyone able to do much about it. Local air control stations will attempt to determine the position, but ultimately it's usually someone with a portable directional receiver that pinpoints the actual source of the signal. (Occasionally a unit will begin transmission due to mishandling or malfunction, and I spent far too much of my time in the air force trying to sort through ejection seat packs and spare ELTs to find an overambitious backup gone wrong.) ELTs are also normally located on any floatation rafts the aircraft may be carrying.

The aircraft itself will also carry an underwater acoustic beacon (pinging at about 40KHz) or two so that underwater wreckage can be located once the general area has been established by the ELT that should have been jettisoned on impact. Both items (the ELT and the UAB) have limited battery life -- they pump out huge signals relative to their size. By convention, the UAB is located in the same general area as the FDR and CVR longitudinally.

In the case of the Air France flight incident, it would seem that the ELT did not work or was at a significant distance from the crash site by the time the search reached the position. That the black boxes (the FDR and CVR) are expected to be recoverable at this point seems to indicate that the UAB did work (it would have been detected by a sonobuoy dropped by the search airplane, which looked to me like it was equipped with anti-submarine warfare equipment, judging by the MAD boom on the tail).

Yeah, the video seems to suggest one of the first rescue aircraft on the scene was a p-3 Orion variant, of which Brazil flies 8.

It does, and I think it is easier to pick up signal of a floating object than of an object 2 miles below sea level...

Radio doesn't work underwater. At 20 meters depth, a submarine can receive a 35 byte-per-second message using a long antenna, but it cannot transmit. [ http://en.wikipedia.org/wiki/VLF ]

I meant in the situation that black box floats.

I believe they do, but maybe not whatever they may still be attached to.

Why do you think that it doesn't?

I don't know whether it does, but it would be stupid to let it sink and send a signal.

Why?

Sending a signal makes it easier to find - that's not stupid whether or not it sinks.

Sinking means that it doesn't move much. However, since it's attached to the plane, it's likely to go down with the airplane whether or it it would sink on its own, whether or not it sinks on its own is irrelevant.

Yes, but you could put a radio beacon on a floating box. This seems like a good idea, possibly for a secondary box.

Or even better, you could have them produce a sonar ping that search and rescue could listen for... which is what they already do. :-) [ http://trueslant.com/milesobrien/ ]