---
title: "CSI trios"
question: "How do C-PHY trios carry a camera’s video, and how do they compare with regular CSI?"
url: "https://nearfieldnotes.com/videos/csi-trios/"
site: "Near Field Notes"
video: "https://nearfieldnotes.com/media/csi-trios/full.mp4"
duration: "2:42"
date: "2026-11-09"
note: "https://nearfieldnotes.com/notes/csi-trios/"
---

# CSI trios (video)

How do C-PHY trios carry a camera’s video, and how do they compare with regular CSI?

2:42, with sound. [Read Note 29](https://nearfieldnotes.com/notes/csi-trios/) for the full answer, with figures, conditions and sources.

## Transcript

Same Sony camera sensor, wired two ways.

Regular CSI gets 15 frames a second at full resolution. Trios get 30.

Here's how three wires beat two.

Regular CSI runs on D-PHY. Each lane is a pair of wires.

One wire goes up while the other goes down. The receiver reads the difference, 1 bit per symbol.

The clock gets its own pair. 4 lanes plus the clock is 10 wires.

C-PHY uses three wires per lane, called a trio.

Each wire is high, mid or low, and the three are always different.

That makes six states.

The receiver compares them in pairs: A − B, B − C, C − A.

The pair holding high and low sees a big signal. The other two see half.

The three signs tell it which state it's in.

The one rule: the state must change every symbol. So from any state, there are 5 places to go.

7 symbols give 78,125 patterns. That covers every 16-bit number, with spares left for sync codes.

So a trio carries 2.28 bits per symbol on 3 wires. A pair carries 1, on 2.

And there's no clock wire. Every symbol flips at least one receiver.

The receiver clocks on whichever flips first, and ignores the rest until the next symbol.

I simulated a trio and a pair on the same 15 cm flex, at 2.5 billion symbols a second.

The pair moved 2.5 Gb/s. The trio moved 5.7, and every symbol decoded.

3 trios carry 17.1 Gb/s on 9 wires. 4 lanes carry 10, on 10.

Chips use the same 10 pins for both. In trio mode, one pin sits unused.

The catch is a smaller eye.

Some receivers only see half the swing. And the edges don't line up: they spread 76 ps, against 15 for the pair.

Even on a perfect wire, the encoding alone spreads them 48.

Is it quieter? Mostly. Regular CSI's clock pair is one steady tone.

In my sim, it stood 17 dB above the data. A trio has no tone, just spread-out noise like a data pair.

Skew on any wire still leaks common-mode noise, so phones add three-wire chokes. And newer D-PHY can drop its clock pair too.

Routing: every pair in a trio is a receiver, so all three pairs must be 100 Ω.

Route it tight like a diff pair, and the pairs split: A-B 80 Ω, A-C 94.

So it's three separate 50 Ω traces, 2.5 widths apart, matched within 2 ps.

In a flat row, A and C sit farther apart. Some designs stack the trio as a triangle across two layers.

Between trios, 160 ps is fine, because each trio carries its own clock.

Flagship phone sensors from Samsung, Sony and OmniVision offer both. The phone maker picks.

Raspberry Pi cameras stay on pairs.

On trios, that Sony sensor moves 41 Gb/s. On pairs, 18. That's 30 frames against 15.

---

Source: https://nearfieldnotes.com/videos/csi-trios/