Coupled Oscillators for Computing (2020) | Hacker News Reader

Coupled Oscillators for Computing (2020)

pubs.aip.org

46 pointsnickpsecurity15 comments

That's fascinating. I've fooled around with coupled oscillators, using 555 timers, and seen how little coupling is needed to get synchronization. And that you can synchronize at 2:1 or other ratios, although not as easily as 1:1.

The new idea here is that you can have coupled oscillators running at very different frequencies, and only the ones that are close in frequency will couple. So it's sort of like frequency-division multiplexing, or radio. In a radio receiver, there's only one time-varying voltage coming in, but in that one number is the entire spectrum. So this offers the possibility of connecting some large number of elements without a large number of interconnects.

It takes a lot of cycles to do anything with phase locked systems. Maybe. There are efficient modulation strategies.

It's a new way to think about analog computing. Not clear if it's better.

If biological brains were doing this, researchers should be seeing far more high-frequency activity than is actually observed. There's speculation that something in biological neurons ought to be going fast, because brains seem to get more done than ought to be possible with such slow signals. So far, nobody has found that. EEG measurements, even at the single neuron level, seem to be below 30 Hz. There was an article on HN a few weeks ago with speculation that something optical was going on. But it was just speculation.

yeah, if memory serves, the 2:1 coupling is how the parametron mentioned in the article works; given a reference frequency signal such as 6 gigahertz and some oscillators that can oscillate around 3 gigahertz, you have two possible frequency-locked phases that can serve as 1 and 0. one of the half-frequency oscillators serves as a phase reference for the 1 level and everything else is measured relative to it

hmm, now i see that the article explains this, but i guess i'll leave this here as a simplified summary in case that it's useful to somebody

There are a lot of high frequency electrical patterns in the brain. Individual neurons have a max frequency of less than 200hz (most less than 100hz), but volleys of neural populations can phase lock to much higher frequencies. The hair cells in the ear phase lock to sounds up to 8000hz— this is followed by phase locking in the auditory nerve and further has been followed all the way to cortical synchronization. In humans only measured up to 1200hz in cortex with EEG, but this seems to be a measurement challenge. High frequency neuroscience has a big measurement challenge, actually. Very expensive.

If you are interested in topics like coupled oscillators, synchrony and entrainment, I cowrote a paper that you may enjoy. We cite the OP (2020).

https://www.frontiersin.org/journals/neurorobotics/articles/...

thank you very much for sharing this!

How does this compare to the optical neurons? Is the visual response modulated by something other than visual wavelength stimuli?

In the retina I believe it is chemicals that selectively resonate to different optical wavelengths. Those then trigger the action potential. (Actually, it’s in reverse, iir— darkness is maximal firing rate and light inhibits the rate of firing). I’ll look it up later and post back.

Note that in the ear, the basilar membrane also affects sound perception, based on selective resonance effects, ensuring that a specific band of hair cells are most likely to phase lock to the sounds. There is both space encoding (ie where on the basilar membrane) and time encoding (phase locking to the signal).

I can confirm that all of this checks out except for the chemical "resonance" in the cones in the retina. It is fascinatingly complex how the molecules are tuned to selectively absorb energy in particular bands of the visible spectrum. Whether one would call this resonance or not will hinge on the definition (fwiw, the definition of resonance in physics can be exceedingly broad).

> Spiking neural networks use oscillators for generating the signals but do not take advantage of the nonlinear interaction between oscillators, and so most of them do not belong to OBCs as we defined them.

I don't know if I share this view. The interaction between oscillators in a SNN is indirect but certainly non-linear by way of the network elements and how they process information.

I would think of STDP like a biological form of injection locking.

in the 01950s the parametron was one of many approaches being considered for computing at microwave frequencies; it's kind of amazing that it took until the 01980s for microwave computing to realize its potential, and then by way of conventional 'level-based' combinationial logic and flip-flops rather than through oscillator-based computing

If we could compute faster we'd be able to do digital visible light interferometry, but for now we're stuck with lenses and mirrors bolted to heavy tables, and our telescopes have to be very close together.

well, i can see best around 540 terahertz, so i think you'd need digital state transition rates around a couple petahertz for digital visible light interferometry, about six orders of magnitude faster than this cellphone. that's clearly physically possible, but one or another technology might be able to compute much faster than this cellphone while topping out in the low terahertz

such speeds would likely require that most bit operations be reversible to avoid vaporizing due to landauer's limit (perhaps coincidentally also first proposed by von neumann)

We would also need an optical atomic clock at each receiver in order to have enough time resolution to correlate the signals correctly. A few optical clocks exist but they are still highly experimental.

maybe you could just send a reference light beam between the receivers over an optical fiber to provide a phase reference?

The dream would be a planet-sized optical telescope, but even a European country sized 'scope would have trouble with your idea.

people already send light beams over ocean-length optical fibers (that's how we're talking). this is achieved by the use of erbium-doped fiber amplifiers, which preserve wavelength and phase information, but even without amplification you can reach 500 km: https://phys.org/news/2023-06-scientists-km-quantum-key.html

and of course we're only talking about these rube goldberg setups because we're perversely trying to build a telescope with a giant rock in the middle of it. if we build the telescope in solar orbit, we can just shine a light through space for the phase reference

Reply on news.ycombinator.com