Last time I promised you an amplifier held permanently a hair below the point of screaming, which in a quiet enough room screams anyway, out of an ear nobody is doing anything to.
Start with the observation, because it is stranger than the theory. Seal a probe microphone into a healthy ear canal. Play nothing. Ask the person for nothing. In a large share of normal ears — reported anywhere from about 40 to 70 percent of adults, depending on how good the measurement conditions are, and more often in women than men — the microphone records a faint, steady, pure tone.
It is coming out of the head. Nobody is producing it on purpose. Your ear is making a sound, and has been the whole time.
The ear is too sharp to be sitting in fluid
In 1948 the physicist Thomas Gold worked out how faint a tone can be and still be heard, as a function of how long it lasts. From that he could back out how sharply tuned the ear’s resonators must be, and got quality factors of roughly 30 at low frequencies climbing to around 300 near 10 kHz.
A resonator that sharp loses almost no energy per cycle. But the basilar membrane is a strip of tissue immersed in fluid, and anything vibrating in fluid at that scale is heavily damped. Nothing passive in there should ring hundreds of times before dying away. Gold’s numbers missed by orders of magnitude, which usually means the picture is wrong rather than the arithmetic.
He had spent the war on radar, and he knew what engineers did when they wanted a tuned circuit sharper than its own resistance allowed: feed some of the output back into the input, in phase, so it subtracts from the damping. Push that feedback far enough and the damping goes negative — and the circuit stops waiting to be told anything and oscillates by itself. So Gold proposed the cochlea does this. Not a microphone: a microphone with a power supply, wired to its own output.
He ordered the failure in advance
Gold did not stop at the amplifier. He reasoned forward to what such a thing must do when it slips. If the feedback ever exceeds the losses, a resonant element goes self-oscillatory and we should, in his phrase, “hear a clear note” — which he thought was the real origin of much ordinary ringing in the ears. Then the sharp part: if it is genuinely a mechanical oscillation down there, some of that energy ought to radiate back out, where a sensitive enough instrument could pick it up. He said in print that this would be nearly conclusive.
Then he was ignored for thirty years. In 1978 David Kemp sealed a probe microphone into an ear canal, clicked, and recorded an echo coming back — present in healthy ears, absent in ears with cochlear damage. Within a few years he had them with no stimulus at all.
Gold had not merely predicted the amplifier. He had specified its characteristic malfunction, and nominated the malfunction as the experiment.
It is not held below the threshold
The flag is “the ear amplifies.” One step past it is where this gets good.
The natural assumption is that the ear sits just below the tipping point, and that emissions are rare failures — a few unlucky cells that crossed. That is not what the working models say. Camalet, Duke, Jülicher and Prost showed in 2000 that such a system can hold itself at the critical point unsupervised: let the control parameter drift toward instability while the cell is quiet, and let the resulting motion feed back and push it away again. The set point maintains itself — which matters, since every cell along the cochlea needs a different one.
But that feedback does not settle just short of the edge. It settles slightly past it, on the oscillating side, at whatever small amplitude the loop can hold. A healthy ear is not below the threshold. It is a hair over it, faintly ringing, always. The emissions are not a fault leaking out. They are the operating point leaking out.
Parked there, the gain does something no fixed amplifier can. Response to a tone at a cell’s own frequency rises as roughly the cube root of the force — measured near 0.4 on a living basilar membrane — so the faintest sounds get enormous gain and the loudest almost none, with no volume control anywhere. Sensitivity at the quietest levels can exceed sensitivity at the loudest by a thousandfold. Kill the animal and it vanishes at once: the response collapses and goes straight, like any other piece of wet tissue.
So the sensitivity, the sharpness and the range are not three achievements. They are one decision — run the loop slightly unstable — paid for with a machine that is never quite under control.
Two neighbours here are worth holding against it. No. 40 was a room howling through a PA — same threshold, opposite relationship to it, since that loop crosses by accident and is destroyed by crossing: the instant it succeeds at oscillating it has stopped carrying the voice. And No. 104’s flyball governor is held at a set point that is really an error. The cochlea’s set point is a catastrophe, and it is parked on the far side of it.
One caveat, because flattening it would misrepresent a live argument: there are two candidate motors — the hair bundle’s own twitchy mechanics, and prestin, the protein that makes outer hair cells shorten and lengthen with voltage — and the current reading is that mammals use both, with the division of labour still contested.
Which is a fair place to leave it. The best-understood part of this mechanism is the part where it fails.
≈
One loop I’m watching
Next: a ring of islands with nothing in the middle, tracing the outline of a mountain that sank out from under it thousands of feet ago — a shape that exists only because the thing holding it up has been falling away the entire time. Next time.