A rotating arrow, photographed 30 times a second.
Think of a clock hand that never stops. The grid's alternating current is like that: an arrow spinning 60 times a second. Each PMU photographs that arrow, recording its length and the angle it points at, with a GPS clock that lets meters in different cities compare snapshots taken at the same instant.
That pair (length and angle) is the phasor. It is how an operator sees the grid swing before a blackout, and how post-event analysis reconstructs what happened.
The IEEE C37.118 standard defines the total vector error: the distance between the measured arrow and the true one, divided by the length of the true one. To be compliant, that error must stay within 1%. In practice, the tip of the stored arrow may sit anywhere inside a tiny circle around the true tip, and never outside it.
359° and 1° are 2° apart.
On a ruler, 359 and 1 are far apart: 358 units. On a clock face, they sit side by side. A phase angle behaves like a clock, not a ruler. When grid frequency runs slightly above or below 60 Hz, the arrow "slips" and the angle wraps all the way around from time to time.
A generic compressor treats every channel as a number on a line. Anything "clever" it does on top of that breaks at the wrap: connecting two points, averaging, predicting the trend. The average of 359° and 1° is 180° on a line, and 0° on the circle. The specimen above shows the effect: with the same stored points, connecting them along the line creates errors of more than a hundred degrees right at the wrap.
TUBE has a circular channel (the same one it uses for a drone's heading): it predicts, compares and guarantees the tolerance by measuring distance around the circle, which is exactly what the standard measures.
Compliance with a proven answer.
Keeping months of phasors from thousands of PMUs, dozens of snapshots per second each, is a volume problem. And what is kept has regulated error. With the tolerance guaranteed per measurement, the archive can answer questions without being opened in full.
CLAMP does exactly that: for each time window it returns a range that is certain to contain the true answer, such as "did frequency leave the compliance band this hour?" or "what was the largest arrow length in this event?". It only decompresses the blocks sitting on the boundary of the question.
For an audit, the difference between "the measure probably stayed inside" and "it is proven that it stayed inside, within this margin" is the difference between a dossier and a risk. And as a bonus, the points where the signal left the tube form a map of where the grid moved, as the power grid case shows.
This is not virgin ground.
Compression tied to vector error already exists on the market: algorithms that only keep changes above a fraction of the limit reach ratios on the order of 1000 to 1 on high-rate signals. There is a specialised incumbent, and promising smaller files without measuring would be dishonest.
What we defend as the difference lies elsewhere: the tolerance guaranteed on every measurement, checked on read, not "almost always inside"; questions with proven answers on the compressed file; the circular channel for the angle; and the file signed with GIRDER, which makes forensic analysis of a disturbance tamper-proof.
The missing step, which comes before any promise: take a real PMU recording (public grid-event files exist), compress it with the circular channel, and measure size and adherence to the limit against the incumbent, the way we measured TUBE against SZ3. The closest we have measured is British grid frequency: 6 months at one reading per second, 65.8 times smaller than the original in the event window, no point outside the 10 mHz tolerance.
| Measured | Thesis | |
|---|---|---|
| Tolerance in the standard (1% vector error) | written in IEEE C37.118 | |
| TUBE circular channel | in use (drone heading) | applied to the PMU angle |
| Grid frequency, 6 months, once per second | 65.8× smaller, 0 outside tolerance | |
| Real PMU phasor vs. the incumbent | size and adherence still to measure |
The "snapshot" of alternating current: the length of the arrow and the angle it points at, at that instant.
The distance between the stored arrow and the true one, as a percentage of its length. The standard allows up to 1%.
A way of storing angles that knows 359° and 1° are neighbours.
Where this could be wrong.
Thesis, not measurement
The fit is strong on paper, but we have not yet compressed a real PMU recording or measured against the specialised incumbent.
The incumbent is good at size
With ratios on the order of 1000:1, TUBE may not win on bytes. The pitch is then the per-measurement guarantee and the proven questions.
Vector error mixes length and angle
Guaranteeing 1% overall means splitting the tolerance between the two channels. That split still has to be chosen and documented.
Regulatory duties vary
Retention and frequency-response rules differ by country and operator. The audit argument has to be checked against each client's regulator.
← Certified telemetry · stickybit.com.br
- IEEE C37.118.1: synchrophasor measurement and the 1% total vector error limit. NERC BAL-003: frequency response.
- Grid-frequency demo: National Grid ESO (UK), 1 s, Jan–Jun 2026, 15.6 million readings; Go codec.
- The previous version of this technical note, with the same content in technical language: TUBE with a circular channel and CLAMP with certified queries.