IoT Networking and Device Connectivity¶
Deep Dive · Embedded Systems
Choose a network from the message contract.
Builds on Embedded Software once you are comfortable with MCU buses, interrupts, and RTOS basics. This sub-layer shifts from board-local communication like SPI/UART/I2C/CAN to networked embedded systems that must join, secure, and maintain real deployments.
Why This Sub-Layer Exists¶
Point-to-point buses teach you how one MCU talks to one peripheral. IoT protocols teach you how many devices form a network, recover from node loss, conserve power, and still present a clean IP-facing software model to gateways and cloud services.
This matters because modern embedded products rarely stop at "sensor attached to MCU." A production device usually needs secure onboarding, field updates, a mesh or star network, and a way to bridge low-power radios to Linux hosts, mobile apps, or cloud APIs.
What You Study Here¶
OpenThread¶
OpenThread is the best first protocol in this sub-layer because it forces you to connect real embedded constraints with real networking ideas. You need to understand IEEE 802.15.4 radios, low-power scheduling, IPv6, packet compression, host/MCU splits, and border routers all at once.
It also creates a direct bridge to later roadmap work:
- MCU / RTOS path: run the Thread stack directly on an SoC and reason about tasks, timers, radio drivers, and power states.
- Linux / host path: use a radio co-processor (RCP) over UART or SPI and let a Linux host run the higher Thread stack and border-router logic.
Start here:
Zigbee¶
Zigbee is the next useful protocol after OpenThread because it teaches a different embedded networking philosophy on top of the same low-power radio family. Instead of being IP-first, Zigbee is much more device-model-first: endpoints, clusters, bindings, trust-center behavior, and role selection shape the firmware architecture.
It also creates a direct bridge to later roadmap work:
- MCU / RTOS path: build coordinator, router, or end-device firmware directly on an SoC such as ESP32-C6.
- Linux / host path: use a Zigbee Network Co-Processor (NCP) or gateway-style design and let a stronger host manage control logic.
Then study:
Why IoT Protocols Matter for AI Hardware Engineers¶
AI systems at the edge are rarely isolated. They live inside products with commissioning flows, low-power sensor networks, battery constraints, gateways, and secure remote management.
If you can reason about Thread, RCP/NCP splits, border routers, and low-power IPv6 networking, you are much better prepared to build systems where MCUs, Linux hosts, accelerators, and cloud services all cooperate instead of existing as disconnected demos.
Choose a network from the message contract¶
Define payload size, reporting period, acceptable age, discovery/onboarding, outage behavior, and battery budget before choosing Thread or Zigbee. A sensor report can tolerate a different delivery policy from an actuator command. Neither successful pairing nor a mesh topology proves a deadline will be met.
Create a protocol-neutral report with device ID, sequence, timestamp or age estimate, units, and validity. Map it to each stack and document what the gateway must translate. Compare retries, reconnect time, and stale-message handling using the same arrival schedule. Keep camera frames and other high-bandwidth streams in a separate bandwidth budget.
Carry forward: Protocol choice with a common data/freshness schema. Continue with the next course and reuse that evidence.