A rope stretched from one continent to another.
Picture a tight clothesline. If someone bumps into it, the line shakes, and you feel the shake at the other end. A submarine internet cable is a line like that, more than 10,000 kilometers long, lying on the sea floor.
Google's Curie cable connects the United States to Chile. The light carrying the data inside it has an orientation, the direction in which the light wave "swings", like the rope. When something physically moves the cable (a strong ocean current or, above all, an earthquake), that orientation rotates.
The result: without installing any new equipment, the cable that already exists works as a giant seismograph. Seismologists at Caltech and engineers at Google showed this in 2021, in the journal Science.
The compressor that only speaks when it must.
TUBE is our compressor for sensor data. Instead of storing every measurement, it draws a tolerance band around the signal, agreed in advance with whoever will use the data, and follows a simple rule: while the signal stays inside the band, write nothing down. Whoever opens the file later knows the value was there, within the guaranteed margin.
Only when the signal leaves the band does the compressor store a new point and re-center the band. That silence while nothing changes is what saves space.
Here is the trick: how much the compressor has to speak is a measure of surprise. Predictable signal, silence. Unusual signal, chatter. Nobody taught TUBE to recognize an earthquake; it just notices that, suddenly, it can no longer stay quiet.
98% silence, until 15:37.
On June 23, 2020, at 15:29 UTC, a magnitude 7.4 earthquake struck Oaxaca, Mexico. The waves took about 8 minutes to reach that stretch of the cable.
We took the orientation of the light in the cable that day, the public data from the Science study, and ran it through TUBE. In the calm part of the day, the compressor was quiet 98% of the time: it stored a new point in only about 3% of the instants.
At 15:37, the orientation started rotating nonstop. The signal began leaving the band practically every second: silence dropped from 98% to 4%, and the rate of stored points jumped from about 3% to nearly 100%. In practice, that is the compressor speaking about 30 times more than on an ordinary day.
What confirms it is the earthquake, and not an equipment fault, is the timing: the jump matches the earthquake time plus the waves' travel time to the cable.
| What we looked at | Ordinary day | Waves arrive | In practice |
|---|---|---|---|
| Time the compressor stays quiet | 98% | 4% | silence practically disappears |
| Instants when it stores a new point | ~3% | ~100% | it starts writing almost every second |
| Time of the jump | – | 15:37 UTC | earthquake at 15:29 + ~8 min of wave travel |
| Training, labels, detection model | none | none | detection comes free with compression |
Detecting is not new. Detecting for free, with proof, is.
What was known. Zhan and colleagues (Science, 2021) showed that the orientation of light in the Curie cable gives away earthquakes and ocean waves. In nine months of testing they detected about 20 moderate-to-large earthquakes, and the Oaxaca one we use is among them. The signal is real and recognized by the community.
How others do it. They use purpose-built signal processing. A 2026 study on another cable (Med-Nautilus, in the Mediterranean) trained machine-learning models to tell earthquakes from noise, and the best reached about 60% accuracy.
What TUBE adds. We trained nothing and built no detector. The number of points the compressor has to store already is the detector. And because the tolerance band has a mathematical guarantee, the alarm says something checkable: "the signal provably left its normal behavior", not "the model found it odd".
Every cable already laid is a sensor waiting.
The orientation of light uses the cable as it already is, with no extra hardware, and can be measured up to 20 times per second. A warning could arrive in seconds, versus the minutes waves take to reach seismographs on land. There are other ways to use a cable as a sensor, each with its own cost:
Orientation of light (this case)
Uses the cable's normal traffic, with no new equipment, along entire transoceanic cables. This is where TUBE came in.
Acoustic sensing on the fiber (DAS)
Turns the fiber into thousands of microphones, but only reaches about 50 to 100 km from shore. It is the data in our whale case.
Phase measurement on long cables
Measures tiny variations in the light's timing along the cable (the Marra group, at the UK's National Physical Laboratory).
"SMART" cables
Dedicated sensors built into the repeaters of new cables, an ITU proposal. Requires a new cable.
The direction in which the light wave "swings" inside the fiber. A tremor in the cable makes it rotate. Specialists call it polarization.
How far the reconstructed value may drift from the measured one, agreed in advance. While the signal stays inside it, the compressor writes nothing.
The share of time the signal stays in the band and nothing needs storing. When it collapses, something unusual is happening.
Where this could be wrong.
One event, one day
We measured the day of the Oaxaca earthquake. It is the clearest case in the study, not an evaluation over the 9 months. How many false alarms would appear in months without a large earthquake, we have not measured yet.
We did not download it all
The public dataset is 4 GB. We read only the stretch covering the event day, by partial file reads.
Little space saved here
The orientation of light drifts all the time, so compressing this signal saves only modest space. The value of this case is not the space saved: it is the alarm that comes free and checkable.
Surprise is not diagnosis
The compressor says "something unusual happened, with proof". Saying it was an earthquake, and how big, still takes seismology: here, the timing matches the arrival of the waves.
← Certified telemetry · stickybit.com.br
- Zhan et al. (2021), Science, "Optical polarization–based seismic and water wave sensing on transoceanic cables" (Curie cable, US–Chile). Data: data.caltech.edu/records/50509-xhf30.
- Event: M7.4 earthquake, Oaxaca (Mexico), 2020-06-23, 15:29 UTC.
- Machine-learning detection on the same problem: Communications Earth & Environment (2026), Med-Nautilus cable, about 60% accuracy for the best model.
- Our measurement: TUBE on the orientation of light on the event day (98% silence in the calm stretch, 4% as the waves arrive). Demo: /static/demos/curie-eq.html.
- Other ways to use a cable as a sensor: DAS (Rayleigh backscatter); transoceanic phase interferometry (Marra/NPL); SMART Cables (ITU).