Hardware & electrical · topic E2 · 60 hours
MCU development (STM32)
All real-time work above 1kHz lives on an MCU. No matter how fast the host computer, it cannot do this job.
What to learn
- Clock tree, interrupt priorities, DMA: the three places that show real skill
- Common peripherals: timers and PWM, ADC sampling, CAN/FDCAN, encoder interface
- Division of labor between interrupts and tasks: what goes in an ISR, what goes in a task (RTOS scheduling details get their own lesson, E11)
- Bootloader and OTA: how to roll back a failed update
The lesson · plain words
STM32 is the robotics industry's lingua franca: motor driver boards, sensor boards, safety boards — nearly all carry an STM32. Learning it is not about memorizing registers; it is about mastering the three weapons: timers, interrupts, DMA.
Once these three are fluent, the MCU stops being a black box and becomes a tool in your hands.
Timers and PWM: the MCU's heart
The timer counts up to ARR (auto-reload value), resets to zero, repeats; the compare value CCR decides when the output flips — that is PWM. Watch the animation: bigger CCR means higher duty cycle; bigger ARR means finer resolution but lower frequency.
Motor speed control, LED breathing, servo control — all PWM. Two things must be computed: frequency (motor PWM is commonly 10–20kHz, above the audible range) and resolution (ARR decides how many levels the duty cycle can take).
Interactive figure · drag it and see
Interrupts and DMA: don't hog the CPU
Interrupts: an event arrives (timer overflow, UART data received) and the main loop pauses to run the ISR. There is only one rule: keep the ISR short — set a flag or move data, leave the heavy work to the main loop. printf inside an ISR is beginner sin number one.
DMA: peripherals and memory move data directly, with no CPU involvement at all. Continuous ADC sampling and bulk UART transfers should both use DMA. The interrupt+DMA combination takes CPU usage from 60% down to 5%.
CubeMX and HAL: the toolchain reality
CubeMX configures pins/clocks/peripherals graphically and generates init code; the HAL library provides a uniform API. Veteran engineers mock it as «slow», but it buys you productive in two weeks versus six months grinding registers. Engineering advice: start with HAL, drop down to LL/bare registers only where performance bottlenecks appear.
You must understand what it did for you: the clock tree (where the core frequency comes from via the PLL), GPIO alternate functions (which function owns a multi-function pin), interrupt priority grouping. Miss these three and every problem is a black box.
How to learn, and how to check yourself
One Nucleo board (a few dozen dollars) plus the official tutorials: GPIO → timer PWM → interrupts → DMA → UART. Then do a small project: encoder reading + PWM speed control + UART reporting, chaining all the weapons together in one go.
Pass criteria: without the tutorial, configure a 1kHz interrupt + 4-channel PWM + DMA UART; the PWM frequency and duty cycle measured on a scope match the configuration.
Three questions — do you really get it?
1. With the PWM frequency fixed, what does increasing ARR do?
- AResolution gets finer but frequency drops (unless the prescaler is raised to compensate)
- BDuty cycle gets bigger
- CNo effect
Show answer
A. Frequency = Fclk/(PSC×ARR), resolution ≈ ARR. The two are opposite ends of the same slider.
2. What is the thing you must never do inside an ISR?
- ASet a flag
- Bprintf logging / long loops of processing
- CRead one register
Show answer
B. A long ISR blocks other interrupts and jitters real-time behavior. Heavy work belongs in the main loop or a task.
3. For continuous high-speed ADC-to-memory sampling, what should you use?
- ACPU polling
- BAn interrupt per sample
- CDirect DMA transfer
Show answer
C. DMA leaves the CPU free. Interrupt-per-sample at high sampling rates eats the CPU alive.
Keep going deeper
Links without a marker stay on this site; the rest point to the Chinese edition until its translation batch lands.
How it fails in the field
- Doing long operations inside an ISR (printing, floating-point math), dragging the main loop to a crawl
- Not using DMA: multi-channel ADC sampling drops points and the data looks «randomly jittery»
You have learned enough when
The Chinese edition carries the full topic page for MCU development (STM32), including the complete outline, depth requirements per career track, and further derivations: robotguide.dev/k/E2/.