Putting optical data in sound while you decide what to do with it | Ars Technica
The speed of light is a magnificent thing. Using light, data can, and often does, travel at the fastest pace allowed by physics. But sending data is not the only job involved in communication. The data also has to be processed and routed. For these jobs, the speed of light is a curse. If it takes you one nanosecond to decide where a bit needs to go, then that bit has already traveled 20-30cm.
In terms of silicon chips and processing, this is a bit like telling a taxi driver to turn left thirty thousand blocks too late. In effect, this means that to make a routing decision, you may have to store the data in a memory register, make the decision, and then extract it again. At the moment, this necessitates storing the information electronically, a painfully slow process. Of course, engineers know this and use clever strategies to minimize the number of times any sort of decision needs to be made.
Ultimately, what you would really like to do is slow the light down for a few nanoseconds while you perform whatever processing and routing is necessary, then let it fly away like a souped up pigeon. A group of Australian researchers have a new take on an old idea about how to get this to work.
The key idea is very simple: a bit is encoded into a pulse of light. When needed, the light pulse is converted into a pulse of sound, which travels much more slowly. A few nanoseconds later, the sound pulse is reconverted into light, which shoots off into the great beyond and carries data with it. The process of transforming light to sound and back again is achieved using light pulses, so electronic processing steps are eliminated to some extent.
Turning light into sound and back again
So, how do you make a sound wave using light? This actually happens all the time but with very low efficiencies. Let’s start by thinking about what sound waves look like. And, to make my life easier, let’s think about them traveling through a crystalline material, where all the atoms are aligned in regular arrays. The atoms are bound to each other via shared electrons. This bond is not like a fixed rod, rather it’s more like a spring. Your picture should be an array of spherical masses, all regularly spaced, with springs connecting neighbors.
If I move one of the atoms a tiny amount and let it go—I borrowed some supernatural tweezers to achieve this—it doesn’t spring back to position. Instead, it overshoots and oscillates back and forth until it finally settles back to its regular position. This motion sets the neighboring atoms in motion, and a sound wave starts to propagate through the crystal.
When light propagates through a material, it doesn’t generally do a lot to the atoms, because atoms are heavy. However, the electrons around the atomic nucleus respond mightily, because light has an electric field that oscillates. The electrons feel that electric field and are driven from place to place by it. So the electron cloud around an atom goes from spherical to egg-shaped to spherical again as the light field oscillates by. The electrons exert a force on the central (and heavy) nucleus, and occasionally, this force just happens to move the nucleus enough to set a sound wave in motion.
As with all things, energy is conserved when light generates sound, so blue photons go into the material, a small amount of sound is produced, and fewer blue photons come out, as do some green photons. The energy difference between the blue and green photons corresponds to the energy in the sound wave phonons (the sound wave equivalent of a photon); the number of phonons produced corresponds to the number of green photons produced (and the number of blue photons lost).
Sound on demand
Unfortunately, things are going to get complicated. The sound waves are typically generated in a particular range (determined by the properties of the material). This lets you play a trick. Instead of one light beam, you send in two. The two light beams have slightly different colors: the frequency difference corresponds exactly to a frequency of sound wave that the crystal likes. Now, you can generate quite a strong sound wave because the two light beams interfere and stimulate the production of phonons. In the process, the higher frequency light beam vanishes and the lower frequency light beam is amplified.
Likewise, if the sound wave is already present, the interference between the two light beams can absorb phonons, amplifying the higher frequency light beam at the expense of the lower frequency light beam.
The basic ideas here go back as far as the thirties, and they were revived with the invention of the laser. With the laser, we finally had light sources that were intense enough to really drive these processes. Nowadays, using one light beam to amplify another via the production of sound waves is used in optical communications systems to compensate for the loss of intensity over long stretches of optical fiber. But with these amplifiers, the sound waves are an unwanted byproduct.
Sound waves have traffic accidents, too
If the basic idea is so old, how come it has taken until 2016 for people to think about using these sound waves for practical purposes? There are several reasons, but the main one is bandwidth. If I have an optical pulse that is less than a nanosecond in duration, then I need about a GHz of acoustic bandwidth to replicate that pulse in a sound wave. At high acoustic frequencies, this is readily available. But at low acoustic frequencies, it is not.
And for practical reasons, high acoustic frequencies are undesirable, because they are more sensitive to imperfections in the crystal. Think of it like this: a sound wave propagates by displacing atoms from their equilibrium position, but the relative displacement between two neighboring atoms depends on the frequency. At high frequencies, neighboring atoms are displaced more, while at low frequencies the effect is reduced. But if an atom should be missing or have a different mass—precisely the sorts of things we call defects—then that distorts and scatters the sound wave.
A good analogy would be a highway. Your car can cope with small potholes and bumps in the road, but if some idiot digs a great big trench across the road, you are going to be in all sorts of bother when you hit it at 130km/hr. The difference is that the trench is large compared to the contact area of the tire on the road.
So, we have to use low-frequency sound waves and, at low frequencies, the natural bandwidth of sound production is low and cannot support modern data rates. This is where the latest development comes into play.
Guiding waves expands the bandwidth
The researchers created waveguides in glass that guide not just light waves, but also sound waves. Things are complicated enough already so I won’t really go into how this works, but the basic idea is that the guiding structure gives researchers extra knobs to turn to control the propagation of light and sound through the waveguide. That allows them to choose the waveguide properties such that the bandwidth for sound generation is vastly increased for low frequencies (from a few MHz to several GHz). Furthermore, because the sound waves are confined to the waveguide, they are still quite intense after several nanoseconds, meaning that the waveguide can be used to recover an optical pulse from the sound wave pulse.
This is exactly what the researchers did. They first showed that they could transfer a standard optical pulse to a sound pulse and then recover it as a light pulse 3.5ns later. That’s not a hugely long time, but it’s long enough to be useful.
Their next step is equally important. In the old days, information was transmitted one bit at a time by simply turning the light on and off. Now, however, both the amplitude and the phase of the light are used. This means that instead of sending a single bit per light flash, you might send four bits or even more. So, does the acoustic pulse retain the phase information? The answer is an indisputable yes. The researchers set up a scheme to compare the phase of the incoming light pulse and the recovered light pulse and found that the phase was preserved. This means that the acoustic memory system is ready for a fully modern communications system.
Will it be used, though? I have no idea, to be honest. I know a lot about optics, but I know very little about the details of network engineering. I believe that there is probably something to be gained by being able to do some of the processing work all-optically. But given the cost (in terms of communications speed) of dumb routing, I suspect that much of the gains of all-optical flow control has been negated by clever routing. That said, a toolbox can always fit a new tool.