WHEN Samuel Morse tapped out his first long-distance telegraph message back
in 1844, one of his biggest problems was how to detect the signal above the
noise caused by static electricity. Today, engineers still do their best to
strip such uninvited pops and crackles from their communications. Their ideal is
a clean, clear signal as free as possible from interference.
Like Morse鈥檚 telegraph, most of life operates in a sea of background noise.
In nature, for example, crayfish and crickets have to pick out the sound of
approaching predators from a background clatter of irrelevant sounds. In the
military, where electronic devices strain to detect the telltale signs of enemy
submarines, signal engineers struggle to filter out enough of the background
cacophony for the faint signal they are looking for to come through.
But the suspicion is taking hold that those engineers are on a fool鈥檚 errand,
and that they should be leaving some crackles in. Over the past few years,
researchers have discovered that background noise鈥攁ny unwanted signal
interference, from radio buzz to television snow鈥攃an actually make it
easier to pick up faint signals on the very verge of detection.
Advertisement
Icy problem
This extraordinary idea emerged in 1981 in the equally unlikely context of an
attempt to explain what was causing the ice ages that cover much of the northern
hemisphere with ice every 100 000 years or so. Researchers had noticed that the
distance between the Earth and Sun changes on roughly the same
timescale鈥攑art of a natural wobble in the eccentricity of the Earth鈥檚
orbit. The obvious explanation is that different amounts of solar radiation
reaching the Earth at different times are periodically plunging the Earth into
an ice age, and then warming it back up again. But there is a problem: the
difference in solar radiation reaching the Earth at opposite ends of the cycle
seems much too small to account for the climate changes that occurred.
So Italian physicists Roberto Benzi, Alfonso Sutera and Angelo Vulpiani
wondered whether natural noise might be helping the signal to get through. They
suggested that the effects of the Earth鈥檚 small wobble are boosted by other,
shorter-term changes鈥攁nnual swings in the Earth鈥檚 retained heat, for
instance, or the general variability of climate. The extra noise, the
researchers argued, could be enough to boost the weak signal from the wobble,
and trigger the ice ages. They called this 鈥渟tochastic
resonance鈥濃�攕tochastic because the noise is essentially random, and
resonance because the noise resonates, or works with, the signal to maximise it
and push the Earth into a frozen state.
Neat as this idea seemed, it was never proved, and the concept of stochastic
resonance more or less disappeared from view until 1988, when physicists Bruce
McNamara, Rajarshi Roy and Kurt Wiesenfeld at the Georgia Institute of
Technology demonstrated in a laser experiment that noise really can boost an
inherently weak effect. Their experiment used a ring laser, a system in which
angled mirrors can direct laser light to travel either clockwise or
anticlockwise in a closed loop. The researchers can switch the light鈥檚 direction
using a crystal in the path of the light beam. Most of the laser light goes
straight through this crystal, but a small portion is diffracted and veers off
the main path. By generating sound waves within the crystal, the researchers
were able to change the ratio of diffracted to unaffected light. Eventually,
when the sound waves are intense enough and have just the right frequency, the
diffraction becomes so strong that the laser light changes direction.
The researchers then created a faint signal, in the form of a regular
variation in the frequency of the sound waves in the crystal. At first they made
sure that the signal was too weak to affect the direction of the laser light.
Then the team began to add background noise to the signal. As the noise level
grew, they found that the light began to switch direction, in step with the
signal. At some point the noise became too loud, and the correspondence between
the signal and the laser light disappeared.
So what exactly is happening? After all, adding noise to a system should
decrease the chances of detecting the signal you were looking for. Well, a
crucial feature of both the ring laser and the ice caps is that they are
nonlinear systems. In a linear system, a change in input produces a proportional
change in output. Suspend two identical weights from a spring, and it will
stretch twice as far as with one. Double the voltage across an electrical
resistor and you double the current.
Nonlinear systems are not like that. A change in the input (a small drop in
the amount of heat reaching the Earth, for example) can produce a
disproportionate response (a great deal more ice)鈥攁nd it is in systems
like this that stochastic resonance can operate. It is easiest to picture in
terms of a threshold below which the signal cannot be detected, or has no
effect. It鈥檚 as if the faint signal surfs on a sea of background noise, hitting
random waves that lift it over the system鈥檚 threshold.
Stochastic resonance is an attractive concept because noise pervades life,
notes Peter McClintock, a physicist at the University of Lancaster. Rather than
seek impossible silence, we might learn to live with, and use, the noise around
us. Over the past few years, inspired by the demonstration that, in the ring
laser, the phenomenon really works, researchers have been hunting for stochastic
resonance in all sorts of systems鈥攁nd now they are starting to find
it.
Neurobiology was an obvious place to start. After all, sensory neurons
routinely pick out sights and sounds from a cacophony of background noise.
Perhaps, researchers reasoned, the neurons have evolved to use that noise.
Sensory neurons encode information in the form of electrical impulses, but the
way they respond to stimuli is highly nonlinear: gradual changes in sound or
light intensity don鈥檛 result in a gradual increase in perception. It is more
like pulling the trigger of a gun. Squeeze the trigger gently and nothing
happens. Only when the pressure is firm enough does the gun fire. External noise
enters this picture if you are nervously aiming the gun at an intruder. Your
hand probably shakes, adding random tremors to your weak pressure on the
trigger. The result: BOOM! The added noise has boosted your weak pressure over
the trigger鈥檚 threshold, and the gun goes off.
Twitching crayfish
In the nervous system, electrical charges collect at a neuron until they
reach a critical threshold. At that point the neuron fires, sending a message
down its body to the next neuron in line. It then resets itself to its resting
electrical state, and waits for charge to build up again. The question is, can
background noise push feeble signals over the neuron鈥檚 firing threshold?
The answer, it seems, is yes. In 1993, physicist Frank Moss and biologist Lon
Wilkens from the University of Missouri in St Louis discovered just this effect
at work in crayfish. The crayfish鈥檚 tail fan displays tiny hairs that twitch in
response to subtle water movements, like those caused by a predatory fish
swimming nearby. When a tail hair twitches, it generates an electrical impulse
along a nerve to a neural hub called a ganglion, which processes the impulse,
and tells the crayfish to swim away fast.
In their laboratory, the researchers strung a crayfish hair and its attached
nerve to a vibrating post, which was made to vibrate weakly at a frequency
similar to that caused by a swimming fish. They then added increasing levels of
the sort of random, noisy vibrations that a crayfish might encounter in the
surrounding water. Sure enough, as the noise level increased, the crayfish nerve
began to pick up and respond to the fishy signals. Eventually, as Moss and
Wilkens predicted, the noise level was so high that it began to swamp the
signal, and the benefit to the crayfish was lost.
Though this result showed how noise can help with a nice regular signal,
real-world survival can be more challenging. Rather than a hum at a single
frequency, most sensory stimuli are isolated and infrequent. Recognising this,
researchers are now conducting experiments with irregular signals.
At the spring meeting of the American Physical Society in St Louis, Missouri,
Jacob Levin, a researcher at the University of California in Berkeley, described
one such experiment. His subjects were crickets rather than crayfish, but the
principle is the same. Like crayfish, crickets use tiny hairs attached to
sensory neurons to detect movement in their surroundings. In his lab, Levin
exposed crickets to low-amplitude air currents鈥攖he signal鈥攁mid a
background of random, noisy movements. The signal was too weak to trigger the
neurons, but the background noise was occasionally high enough to make them
fire.
Using intracellular electrodes, Levin recorded the cricket鈥檚 neuronal firings
as it sensed the air waves. Correlating the neuron output with the weak air
currents, Levin could find out how well the cricket detected the faint stimuli.
Finally, he measured those responses against the background noise, checking to
see how rising noise improved the cricket鈥檚 signal reception.
As with the crayfish, noise boosted the cricket鈥檚 ability to pick up weak
signals. 鈥淭he cricket takes advantage of stochastic resonance to piggyback
important small signals on the broadband background noise it can鈥檛 avoid,鈥� says
Levin. But this only worked with noise up to a certain level; if the level of
the background noise level was set too high, it simply swamped the signal.
This raises the question of whether stochastic resonance could be useful for
boosting signals with a range of magnitudes鈥攏ot only those that peak just
below the trigger point of the sensor, but also much weaker signals that require
more noise to bring them above the threshold. For this to work, the background
noise level would have to change to suit the magnitude of the signal, something
that is unlikely to happen in the real world. But neurons can still maximise the
benefit they get from background noise, according to James Collins, a biomedical
engineer at Boston University. As he points out, neurons in living organisms do
not operate in isolation, but work as units in a network, reporting to neural
hubs such as the ganglion. The theory is that individual neurons have their own
intrinsic noise levels, depending on their cellular makeup and connections to
other neurons.
Collins set out to mimic this using a computer model of such a network, and
last year he showed that only a minimum noise level鈥攏ot a changing optimal
noise range鈥攊s necessary to boost the system鈥檚 ability to detect a range
of signals. Collins and colleagues crafted a network in which neurons received a
common weak input signal, but each had a different fixed noise level. The
researchers recorded this fixed-noise system鈥檚 response to a variety of weak
input signals. They then compared those responses with a model system lacking
noise. Sure enough, the fixed-noise system detected faint signals better than
its quiet counterpart.
Noisy therapy
Collins says a little noise goes a long way because networks average the
output of many components. For a signal of any given size, the chances are that
at least some of the neurons in the network will be experiencing the optimum
noise level to boost the signal over the threshold without swamping it. So the
system can handle signals with a range of magnitudes.
Encouraged by his findings, Collins hopes to try to use neural noise
therapeutically. One target group is people suffering from movement disorders
that cause a loss in proprioception鈥攖he 鈥渟ense鈥� that makes it possible to
know what your arm is doing, for example, even when you can鈥檛 see it. In certain
people, neurons involved in this process are not as sensitive as they should be:
their firing thresholds are set too high. If someone with this disability tries
to turn, for example, they may not realise they are twisting an ankle too
far鈥攗ntil it breaks.
One solution might be to add noise to the defective neurons to bring the
signals above the neurons鈥� firing threshold. 鈥淭he challenge is to find a viable
way to introduce noise to the human system,鈥� Collins says. One option being
considered is for doctors to apply mechanical vibrations to tendons in the
ankle. If this works, it could help people suffering from a variety of
conditions involving lost sensitivity in touch or balance.
Getting the picture
Boosting faint images is another possible use of stochastic resonance being
studied by Moss, in conjunction with Enrico Simonotto, a physicist at the
University of Genoa in Italy. The researchers have been sitting volunteers in
front of TV screens, where they are shown a series of videotaped images. At
first, all they are shown is a noise-free image that is too dim to be detected.
As the video plays, noise is introduced in the form of random changes to the
shade of individual pixels. As the noise increases, Simonotto says, the
underlying picture becomes clearer, and most volunteers appear to agree on the
range of noise that is most useful for a given picture.
Simonotto now plans to tinker with the timing and position of noise added to
a picture. He suggests that noise-enhanced images may eventually improve visual
systems for pilots attempting to find their way in dark or stormy
conditions.
Electronic sensing systems may also benefit from some judiciously added
noise. Some physicists hope to use noise in electronic or superconducting
systems. Adi Bulsara, a senior scientist at the US Naval Command, Control and
Ocean Surveillance Center in San Diego, is thinking of harnessing the noise that
affects some types of magnetic detectors known as superconducting quantum
interference devices (SQUIDS). The Navy鈥檚 SQUIDS are designed to detect enemy
submarines, planes and mines by picking up the tiny changes in magnetic field
that can be caused, say, by a submarine鈥檚 power supply.
One problem with SQUIDs is that the magnetic environment they operate in is
full of noise from sources that range from lightning flashes to interference
from the Earth鈥檚 magnetic field. Bulsara noted that some SQUIDS operate as
nonlinear two-state systems just like ring lasers. Perhaps, he reasoned, you
could use the external noise to push weak signals over the detection threshold.
Another possibility, he says, is to link SQUIDs to gain an aggregate boost as
for the nervous system.
There is clearly a lot more that must be learnt about stochastic resonance
before many of the proposed applications become reality, as most researchers
will readily admit. Some, such as McClintock, fear that enthusiasts may be
overselling the phenomenon鈥檚 direct applications. 鈥淗ow many crickets鈥� lives were
saved by the fact that they used stochastic resonance?鈥� Levin asks. 鈥淲e don鈥檛
know.鈥� In fact, no one knows whether crayfish and crickets truly use noise,
still less whether researchers will ever succeed in finding useful applications
of the phenomenon. But the enthusiasts remain optimistic. Noise is everywhere,
they argue, so that chances are that nature has found a way to use it鈥攁nd
surely we will, too.