ADR-0014: One bigger cell, a bigger pebble, and continuous running on a phone charger

# ADR-0014: One bigger cell, a bigger pebble, and continuous running on a phone charger

- **Status:** accepted (2026-10-13)
- **Date:** 2026-10-13
- **Requirements touched:** P-11 (72 h stream), N-01, D-01; the v01 meeting target
- **Builds on:** ADR-0009 (light ring), ADR-0013 (kit build)

## Context

Rafa, reviewing ADR-0013's battery result (worst case 1.43 h):

> "Can we double the battery size, it's ok if it weighs more, ideally single battery tho, and make sure it works continuously if plugged into normal USBC phone charger?"

> "Feel free to change the shape / make it bigger as needed to satisfy the requirement (we need basically 3 hours of continuous play for long discussions and meetings)"

> "Don't anchor on worst case, worst case should be 1.5 hours or 2 hours is fine for the first iteration"

Three problems with the ADR-0013 power design:

- **Capacity:** 2000 mAh gave 1.4–2.8 h.
- **Plugged in:** the bq25185 board passes at most about 1 A at 4.5 V from USB. Transcribing at the high Radxa figure needs more than that, so the cell drained slowly even on a charger.
- **Heat was never modelled.** Running continuously, heat becomes the limit. The battery used to run out before the shell got hot.

## Options

| Option | For | Against |
|---|---|---|
| Two 2000 mAh cells in parallel | In stock, the pebble keeps its width | Not one battery; two protection circuits in parallel; about 9 mm thicker |
| A narrow 4000 mAh cell (104080) | The pebble keeps its width | Not sold by any DE/EU maker shop; only Amazon.com or AliExpress |
| **EREMIT 4000 mAh, 90 × 60 × 6 mm** (chosen) | In stock in DE (€9.90), JST-PH, PCM, 2 A continuous | 60 mm wide: the pebble grows |
| Soldered 333288 5000 mAh, 75 × 55 × 10 | More capacity, smaller footprint | Out of stock for about 6 weeks; the drop-in upgrade later |
| Keep the bq25185 and feed USB 5 V straight to the 5 V rail through the LM66200's second input | Fewer conversion losses on the charger | Both inputs sit near 5 V, so the diode can chatter, and the power switch would no longer cut a USB-powered device |
| **Adafruit 4755 (bq24074)** (chosen) | 1.5 A from USB with the cell supplementing above it; OUT held at 4.4 V; 1 A charge by default; safety timer off, so a 4000 mAh charge can't time out | A 41 × 33 mm board with a 12.6 mm SMT DC jack that stays on (removing it needs hot air) |

## Decision

1. **Cell:** EREMIT 3.7 V 4000 mAh High Cap. (90 × 60 × 6 mm, 74 g) in a 91 × 61 × 9.3 mm bay, with foam above and below.
2. **Charger:** Adafruit 4755 (bq24074), fed from the tail USB-C. Its defaults are right:
   - 1.0 A charge (0.25 C);
   - 1.5 A input limit;
   - cut only the THERM jumper, for RT1.
3. **Continuous on a charger:** USB runs the device first and charges the cell with what is left.
   - Any USB-C phone charger works: the tail port's 5.1 kΩ CC resistors ask for up to 3 A, and the device needs 1.5 A.
   - Plugging in or pulling out doesn't reset anything. The power path switches in microseconds and the boost holds 5 V.
4. **Shape:**
   - From 138 × 56 × 29 mm to **154 × 78 × 33 mm**, about 197 g.
   - The Radxa sits 12 mm off-centre on the wide cell. (As first published, one post still landed in the eye opening; ADR-0015 drops that post and adds a check.)
   - The charger lies on the cell at the nose end, beside the board.
   - The ring grows to **50 LEDs**, and the software splits it as before. Its brightness cap drops from 35 % to 30 %, which keeps the worst case under the ring fuse.
5. **Heat path, no fan:**
   - SoC → thermal pad → 14 × 14 × 6 mm aluminium heatsink → two layers of 8 W/mK pad → aluminium tape spreader inside the top shell. The whole shell then sheds the heat.
   - The layout keeps a 9 mm column clear above the SoC for it, which moved the charger to the nose end of the cell.
   - A fan would be recorded by the microphones and would need vents.
   - The spreader stays at least 15 mm from the NFC tag.
6. **Firmware rule:**
   - On a charger, the Radxa is held to about 2.4 W averaged over minutes, by its own thermal zone.
   - Transcription runs in bursts: it falls behind during talk and catches up in pauses, from the encrypted buffer.
   - The battery targets are 3 h at typical power and at least 2 h at the worst measured power.

## Consequences

- **Simulation (`v01/sim/power_sim.py`):**
  - typical 3.1 h of busy transcription and 3.8 h of listening per charge; worst case 2.7 h;
  - a 72 h conference on a charger ends at 100 %;
  - listening around the clock on a charger keeps the shell at about 38–41 °C in a 25 °C room.
- **Bigger and heavier:** about 197 g instead of 133 g. The pebble is now palm-sized rather than pocket-sized.
- **Still to measure:**
  - shell and cell temperature on a 4 h plugged-in recording (B-40);
  - the Radxa's real listening and ASR power (B-10);
  - whether the room-off glow and segments read as well on the longer ring (B-06).
- **The 72 h stream on battery alone** still needs v02's low-power audio MCU (B-21). Model energy efficiency is now its own research track (B-41).