embed-connectivity

Embedded Connectivity

A field-deployed device that can't reconnect is a brick. Provisioning, credential storage, retry behavior, and protocol choice shape both field reliability and fleet economics — and most of those decisions are expensive to change after launch.

When to reach for this

• Choosing a radio / protocol stack for a new device class
• Designing or debugging pairing, provisioning, or reconnect flows
• Sizing cellular data plans or cutting per-device data cost
• Planning RF certification (FCC/CE/IC, BT SIG, WFA, Matter/CSA, carrier approval)

Principles

1. Pick the radio by duty cycle and payload shape first. LoRaWAN for tiny-and-rare, BLE for paired-and-close, Thread/Matter for ecosystem-first, LTE-M/NB-IoT for field-deployed without a gateway, Wi-Fi for mains-powered. Cost and power diverge fast when you pick wrong.
2. Design for reconnect, not connect. First-time pairing is easy; week-three-in-a-basement reconnects are where fleets die. Exponential backoff with jitter, persistent connection state across reboots, captive-portal detection, and a physical "reset to pairing" control.
3. Store credentials in hardware. Secure element (ATECC608, SE050, NXP EdgeLock) or TrustZone-backed key storage — never flash-resident cleartext. Provision per-device keys at manufacturing, not at first boot.
4. Compact payloads aggressively on cellular. CBOR over JSON, delta encoding, batching. A 200-byte payload can bloat to 600+ bytes once TLS record and IP overhead are counted; use DTLS/TLS session resumption and PSM/eDRX, and size the data plan from measured bytes/day, not the application payload.
5. Instrument the comms thread. Counters for connect attempts, failures-by-reason, bytes up/down, and latency — emitted periodically even on success. Reconnect flakes are undebuggable without field telemetry.
6. Plan the cert path from day zero. BLE → BT SIG listing, Wi-Fi logo → WFA, Matter → CSA, cellular → PTCRB/GCF plus carrier approval. These are months-long pipelines; retrofitting certification is painful and expensive.

Radio selection — first cut

Radio	Best fit	Power profile	Watch out for
LoRaWAN	bytes, minutes–hours apart, km range	years on primary cells	regional duty-cycle limits; downlink is scarce
BLE	phone-paired, short range	coin-cell viable	needs a mediator (phone/gateway) for cloud reach
Thread / Matter	smart-home ecosystem play	low; mesh extends range	commissioning UX + CSA cert effort
Wi-Fi	mains-powered, high throughput	worst of the set	provisioning UX; 2.4 GHz congestion
LTE-M / NB-IoT	field-deployed, no gateway	battery-viable only with PSM/eDRX	carrier cert, roaming policy, per-MB cost
LTE Cat-1 / 4G	bandwidth-heavy field devices	mains or large battery	module + plan cost dominates BOM

Connection state machine — minimum states

PROVISIONING → CONNECTING → CONNECTED → BACKOFF (jittered exponential, capped) → OFFLINE_BUFFERING → FACTORY_RESET_PENDING. Persist the current state and backoff counter across reboots; on MQTT use LWT so the fleet sees ungraceful drops, and pick QoS per topic (telemetry QoS 0/1, commands QoS 1) rather than globally.

Pitfalls

• Retry loops without jitter — a fleet-wide outage recovery becomes a thundering herd against your broker
• Credentials provisioned at first boot over an unauthenticated channel
• Trusting "connected to AP" as "has internet" — no captive-portal/DNS check
• JSON-over-TLS telemetry on NB-IoT plans sized from payload bytes only
• Certification discovered at EVT — module pre-certs help but don't cover the end product's intentional-radiator testing

---

Related: embed-driver (radio-module bus driver), embed-rtos (comms task design), embed-ota (update transport), embed-power (radio duty-cycle vs sleep) · domain agent: embed-architect · output/ADR format: playbook-conventions