Electra / Tadiran Central AC

Broadlink-format IR codes for a central (ducted) Electra-family air conditioner — the kind installed as a whole-home/mini-central unit in Israel and driven by a generic wall/handheld IR remote rather than a per-room split's own remote.

The exact model is unknown (central units usually have no accessible remote/ indoor-unit model sticker). This will be updated if the model is identified. The protocol below was recovered purely by capturing and decoding the remote's IR.

Why this exists: these units are common but their remotes match no ready-made database (SmartIR's Tadiran code sets 1340–1346/1560 all failed here). Rather than hand-capture dozens of button states, the protocol was decoded from a handful of samples, which showed it has no checksum — so the entire command matrix can be generated in software. The codes and the method are both published here.

Contents

File	What
protocol.pdf	Illustrated protocol spec (color-coded bit map, field tables, worked example). Source: protocol.typ (Typst).
codes.json	Full generated code set: power toggle + 120 state frames (cool/heat × 16–30 °C × 4 fan speeds), plus the decoded protocol spec.
electra_ir.py	Self-contained codec: decode a capture, generate any state, or regenerate the whole matrix.
home-assistant/	Drop-in Home Assistant package (ac_control.yaml) + setup guide — control the AC from any Broadlink remote.* entity.
tools/capture.py	Capture frames from your remote (for a different variant, or to add modes/swing).
raw_captures/*.b64	The real learned captures the whole thing was derived from (for reproducibility).

📄 Start with protocol.pdf for the human-readable bit-field map. The section below is the same spec in text form.

The protocol

• Carrier format: Broadlink b64: raw IR (the strings work directly with python-broadlink send_data, or Home Assistant remote.send_command).
• Line coding: Manchester bi-phase, half-bit ≈ 31.7 Broadlink ticks.
• Frame: short leader → 34-bit core payload → repeated 3× back-to-back → lead-out gap.
• No checksum. Changing one setting flips only that field's bits; everything else is byte-stable.

Core bit fields (index 0 = first core bit)

Bits	Field	Encoding
0	power-toggle flag	1 = the power toggle command; 0 = a state/settings frame
2–3	mode	cool=01, heat=10 (MSB = bit 2)
4–5	fan	low=00, med=01, high=10, auto=11 (MSB = bit 4)
11–14	temperature	value = °C − 15, 4-bit MSB = bit 11 (16 °C=0001 … 30 °C=1111)

Only cool/heat and the four fan speeds were captured; other modes (dry, fan-only, auto) likely reuse the mode field with other 2-bit values.

Power behaviour (important)

Power is a single toggle: one code flips the unit on↔off (the "off" and "on" captures are byte-identical). State frames do not power the unit on — sending cool/22/high to an off unit does nothing. To control it: send the power toggle to turn on, then send state frames to set mode/temp/fan.

Usage

# Decode a capture into fields
python electra_ir.py decode raw_captures/anchor_20.b64
#   -> {"mode":"cool","fan":"high","temp":20, ...}

# Generate any state as a Broadlink base64 code
python electra_ir.py gen cool 22 high
python electra_ir.py gen heat 24 low

# Regenerate the entire matrix
python electra_ir.py matrix > codes.json

Send a code (example with python-broadlink):

import broadlink, base64
dev = broadlink.discover()[0]; dev.auth()
dev.send_data(base64.b64decode(open_code))   # b64 string decoded to bytes

Home Assistant

remote.send_command with a Broadlink remote.* entity:

service: remote.send_command
target: { entity_id: remote.bedroom_ac }
data:   { command: "b64:JgDEAF9f..." }   # any code from codes.json

Because power is a toggle and state frames don't self-power the unit, a clean HA setup is: a power-toggle button/script, plus input_number (temp) + input_select (mode, fan) helpers whose changes fire the matching state code via an automation. (This is how the author drives it; a native climate entity would need a template-climate wrapper since SmartIR assumes discrete on/off.)

How it was reverse-engineered (methodology)

1. Capture a few states with a Broadlink RM4C mini in IR-learn mode (python-broadlink enter_learning() / check_data()).
2. Decode the raw Broadlink pulses. A histogram showed marks and spaces clustering at multiples of one base unit → Manchester, not simple pulse-distance. Manchester-decoding gave a stable 34-bit core repeated 3×.
3. Diff adjacent captures. −1 °C vs +1 °C differed in exactly one small region → temperature is a localized field. Two known temperatures (20, 24) pinned the mapping: field value = °C − 15, and crucially no other bit moved → no checksum.
4. Generate & verify. With no checksum, any state can be synthesized by rewriting its field. Generated codes were validated (a) in-silico — a generated frame reproduces a real capture within IR jitter — and (b) on the physical AC.
5. Repeat for fan (bits 4–5) and mode (bits 2–3), each an isolated field.

The upshot: ~7 real captures were enough to generate a 120-code matrix, all cross-checked against the originals.

Contributing

If you have one of these units: PRs welcome to add the model number, more modes (dry/fan-only/auto), swing, or captures from a different remote variant. Drop raw .b64 captures in raw_captures/ and note the state each represents.

License

MIT — see LICENSE. Codes provided as-is; verify against your own unit.