Measuring space-time behaviours of piano keys with Rust

栏目: IT技术 · 发布时间: 4年前

内容简介:This project was our first firmware project written in Rust. Most firmware projects for embedded devices are written in C or C++. Just like our projects such as theWe were waiting for the right opportunity to dive in. And this project seemed simple and low

This project was our first firmware project written in Rust. Most firmware projects for embedded devices are written in C or C++. Just like our projects such as the Frogwatch vibration sensor. However, Rust is slowly gaining traction as a promising alternative as it has a few important advantages. The language has a built-in dependency manager: Cargo . It guarantees memory safety at compile time and it offers zero cost high level abstractions.

We were waiting for the right opportunity to dive in. And this project seemed simple and low risk enough that it was worth the extra time investment.

The project

Measuring space-time behaviours of piano keys with Rust

Piano technology is a discipline that studies piano mechanisms and their interaction with the pianist. For example one of the most important research areas is modelling piano actions to simulate a mechanical piano action with an electronic piano action. To investigate interactions between piano keys and pianists, we have installed an array of sensors in an experimental keyboard. The sensors are optical encoders aimed at measuring space-time behaviours of the keys when pushed by a pianist or by a robotic finger with different forces.

From a functional perspective the goal of the (embedded systems) project is:

Measuring space-time behaviours of piano keys with Rust
From optical encoder to something human readable
Measuring space-time behaviours of piano keys with Rust
From optical encoder to something human readable

Translate a signal from an optical encoder to something human readable and display it on a computer.

We will first describe the hardware part of the project for some context, but you can alsodirectly to the code.

Component Selection

Microcontroller

Even though we have mostly worked with NXP LPC microcontrollers before, for this project we picked the STM32F103 . The STM32 microcontroller family seemed the most popular and best supported platform among Rustaceans. And the STM32F103 can be found on the (in)famous bluepill and blackpill boards that you can order on Aliexpress for less than the cost of a standalone STM32F103.

USB Controller and Optical Encoder

We selected the tiny AEDR-8300 optical encoder which has a resolution of 75 lines per inch. Its two outputs, A and B form a quadrature signal.

Finally the CP2102N IC will facilitate the UART to USB translation.

The electronic design

We decided to design a modular system. We needed to implement the sensors in two devices. In the first device we only need one optical encoder. In the second one we need to connect to ten encoders spread out over some distance to cover all ten piano keys for that tool. We decided to use the Dutch Polder Model and design a PCB that connects to up to five optical encoders, but make it daisy-chainable.

Measuring space-time behaviours of piano keys with Rust
Highlevel electronics design
Measuring space-time behaviours of piano keys with Rust
Highlevel electronics design

The mainboard houses one blackpil board that acts as the cpu . Measurement data will be send to the uart output connector. It will also listen to incomming transmissions on its uart input side and relay those messages to its output uart.

Additionally we designed a compatible uart->usb converter board that can connect to the same connector. This board will also supply the required. 3.3V and 5V power levels.

Fitting the pcbs in the 3d model

Checking if everything will fit together in a 3d model is always a good idea before ordering your boards.

The final product

Two weeks later we assembled the boards and fitted everything in real life.

Finally, Rust code

Since this was our first Rust project we still had many concepts to learn. I cannot effectively learn a programming language from reading books, so rather I just start with a real project and figure it out on the go.

We tried to find some other embedded Rust projects on Github to see how they where approaching things. We noticed that many projects used the concurrency framework cortex-m-rtfm . I was intrigued by this framework and how it uses the Rust macro system to create this domain specific language to define real time apps. However, We wanted to try to build the firmware without these abstractions to better understand how things work on a lower level in Rust. This approach worked fine in the beginnnig when we were only interfacing with a single optical encoder.

Instead of going over the code line by line, We’d like to touch upon some concepts that were new to us.

Peripheral registers

The most fundamental thing when working with microcontrollers is writing bits to peripheral registers. Crates.io has these very convenient auto generated Peripheral Access Crates (PACs) that provide a nice API to interface with the registers. Since these crates are generated from the SVD files, they all work more or less similar. Something that’s rarely the case in C.

For example, configuring 10 pins for external interrupt to decode the quadrature encoder signals looks like this:

// configure correct pins for external interrups
afio.exticr2.exticr2().modify(|_,w| unsafe {
    w.exti5().bits(0b0000);  //PA5 1A
    w.exti6().bits(0b0000);  //PA6 2A
    w.exti7().bits(0b0000)   //PA7 3A
});
afio.exticr3.exticr3().modify(|_,w| unsafe {
    w.exti8().bits(0b0000);  //PA8 5A
    w.exti9().bits(0b0000);  //PA9 4A
    w.exti10().bits(0b0000); //PA10 4B
    w.exti11().bits(0b0000)  //PA11 3B
});
afio.exticr4.exticr4().modify(|_,w| unsafe {
    w.exti12().bits(0b0000); //PA12 2B
    w.exti15().bits(0b0001); //PB15 5B
    w.exti13().bits(0b0010)  //PC13 1B
});

// Set interrupt request masks; enable interrupts
exti.imr.modify(|_, w| {
    w.mr5().set_bit();
    w.mr6().set_bit();
    // ... some lines skipped
    w.mr13().set_bit()
});

// Set interrupt falling and rising edge triggers
exti.ftsr.modify(|_, w| {
    w.tr5().set_bit();
    w.tr6().set_bit();
    // ... some lines skipped
    w.tr13().set_bit()
});
exti.rtsr.modify(|_, w| {
    w.tr5().set_bit();
    // ... some lines skipped
    w.tr13().set_bit()
});

We write to registers using these closures.

There are two methods to write to the registers. We can use either write() or modify() . The differences is that modify performs an OR operation between the current register values and the new values in the closure, and write just writes the new value. So be careful that all the bits you do not explicitely set in the write closure will be zeroed.

For setting single bits there are usually methods available called set_bit() and clear_bit() . Those can be used immediately. When settings multiple bits using the bits(0b000) method, you need to use an unsafe{} block.

Sharing resources with interrupts

The decoding of the quadrature signals is done in interrupts. This means we need to access a few things in these interrupts.

EXTI

This is where things differ considerably from C. Since Rust will guarantee safe access to memory, we cannot just mark it volatile and access it in multiple ‘threads’. Interrupts at different priority levels can be considered different threads, but also the main loop which runs on the lowest priority is a thread. So if we want to share resources (read memory) between threads, we have to do it the Rust way.

By wrapping them in a globally available mutex we can access the data in both the main loop and interrupts. We have to do this for both the EXTI peripheral as the encoder state that we have encapsulated in the Encoder struct.

// Make external interrupt registers globally available
static INT: Mutex<RefCell<Option<EXTI>>> = Mutex::new(RefCell::new(None));

static ENCODER: Mutex<RefCell<Option<Encoder<
        gpioa::PA5<Input<PullUp>>,
        gpioc::PC13<Input<PullUp>>
        >>>> = Mutex::new(RefCell::new(None));

The types in Rust are very verbose and even the specific GPIO pins that will be used for this encoder are explicity part of the type. We also need RefCell and Option because the memory is intially empty and only initialized in main() .

Now we can use access these objects in multiple contexts with the caveat that it needs to be accessed within a critical section (interrupt free section) cortex_m::interrupt::free . This is enforced by the mutex that requires a parameter _cs: &'cs CriticalSection that can only be created in such a section. That is how we can be sure that we will not be interrupted by a higher level thread and thus can modify this memory safely.

/**
 *Ch A interrupt
 */
#[interrupt]
fn EXTI9_5() {
    encoder_isr(Channel::A);
}

/**
 *Ch B interrupt
 */
#[interrupt]
fn EXTI15_10() {
    encoder_isr(Channel::B);
}

fn encoder_isr(channel: Channel) {
    cortex_m::interrupt::free(|cs| {
        if let Some(ref mut exti) = INT.borrow(cs).borrow_mut().deref_mut() {

            let pr = exti.pr.read();

            if pr.pr5().bit_is_set() || pr.pr13().bit_is_set() {
                // Clear the interrupt flagw.
                exti.pr.write(|w| {
                    w.pr5().set_bit();
                    w.pr13().set_bit()
                });
                if let Some(ref mut encoder) = ENCODER.borrow(cs).borrow_mut().deref_mut() {
                    encoder.update(&channel);
                }
            }
        }
    });
}

This still looks manageable, yet, when we add support for five encoders and also keep time using a timer interrupt and blink a LED, our list of globals will look like this:

// Make external interrupt registers globally available
static INT: Mutex<RefCell<Option<EXTI>>> = Mutex::new(RefCell::new(None));

// Make our LED globally available
static LED: Mutex<RefCell<Option<gpiob::PB12<Output<PushPull>>>>> = Mutex::new(RefCell::new(None));

static TIMER_UP: Mutex<RefCell<Option<timer::Timer<stm32f1xx_hal::pac::TIM1>>>> = Mutex::new(RefCell::new(None));

static TIME_MS: CSCounter<u32> = CSCounter(UnsafeCell::new(0));

static ENCODER1: Mutex<RefCell<Option<Encoder<
        gpioa::PA5<Input<PullUp>>,
        gpioc::PC13<Input<PullUp>>,
        gpiob::PB8<Output<PushPull>>
        >>>> = Mutex::new(RefCell::new(None));

static ENCODER2: Mutex<RefCell<Option<Encoder<
        gpioa::PA6<Input<PullUp>>,
        gpioa::PA12<Input<PullUp>>,
        gpioa::PA4<Output<PushPull>>
        >>>> = Mutex::new(RefCell::new(None));

static ENCODER3: Mutex<RefCell<Option<Encoder<
        gpioa::PA7<Input<PullUp>>,
        gpioa::PA11<Input<PullUp>>,
        gpiob::PB0<Output<PushPull>>
        >>>> = Mutex::new(RefCell::new(None));

static ENCODER4: Mutex<RefCell<Option<Encoder<
        gpioa::PA9<Input<PullUp>>,
        gpioa::PA10<Input<PullUp>>,
        gpiob::PB1<Output<PushPull>>
        >>>> = Mutex::new(RefCell::new(None));

static ENCODER5: Mutex<RefCell<Option<Encoder<
        gpioa::PA8<Input<PullUp>>,
        gpiob::PB15<Input<PullUp>>,
        gpiob::PB14<Output<PushPull>>
        >>>> = Mutex::new(RefCell::new(None));

This is quickly becomming ugly. You would think you can just put these encoders in an array. However they have different types since they are not using the same GPIO pins. And if you put them in a tuple, you will still need to type out the full types.

Turns out that cortex-m-rtfm is designed to hide all this boilerplate code. I am not sure if there is a better or cleaner way to do this without the RTFM framework.

Compiler optimizations

Rust has zero cost abstractions . That means you can work on a relativley high abstraction level without hurting performance.

What we found out the hard way, is that this is only true if you turn on compiler optimizations (which makes sense of course). This is the case for both cpu cycles as memory usage.

Without optimizations on, the stm32f103 struggles to keep up with high uart baudrates. Especially since it does not have a FIFO buffer.

We also found out that we quickly run out of memory when we were adding buffers for the additional encoders.

For example, consider the following two examples. The first one consumes twice as much stack memory if you do not enable optimizations.

// Example 1

// Initialize encoder struct
let encoder1 = Encoder::new(pin_a5, pin_a10, ch1_led);

// Move encoder into global mutex
cortex_m::interrupt::free(|cs| {
	*ENCODER1.borrow(cs).borrow_mut() = Some(encoder1);
});
// Example 2

//Initialize encoder directly in global mutexes
cortex_m::interrupt::free(|cs| {
    *ENCODER1.borrow(cs).borrow_mut() = Some(Encoder::new(pin_a5, pin_c13, ch1_led));
});

So to turn on optimizations during development you can add the following section to your Cargo.toml

[profile.dev]
debug = true
lto = false
opt-level = 1 # or higher

RTFM after all

Besides getting used to the general Rust syntax, these were the issues that cost us the most time to figure out or debug. In the end we rewrote the firmware using the cortex-m-rtfm framework to get rid of all the boiler plate. It also allows for easier management of priorities and resources for all the different threads/interrupts. The full code can be found on Github .

The most interesting thing about Rust so far is that once you’re done fighting with the compiler it mostly just works. Also working with a proper dependency manager is like a breath of fresh air.


以上就是本文的全部内容,希望对大家的学习有所帮助,也希望大家多多支持 码农网

查看所有标签

猜你喜欢:

本站部分资源来源于网络,本站转载出于传递更多信息之目的,版权归原作者或者来源机构所有,如转载稿涉及版权问题,请联系我们

数据结构

数据结构

殷人昆 / 清华大学 / 2007-6 / 39.00元

《数据结构》(第2版)“数据结构”是计算机专业的核心课程,是从事计算机软件开发和应用人员必备的专业基础。随着计算机的日益普及,“数据结构”课程也在不断地发展。《数据结构》(第2版)按照清华大学计算机系本科“数据结构”大纲的要求,从面向对象的概念、对象类设计的风格和数据结构的层次开始,从线性结构到非线性结构,从简单到复杂,深入地讨论了各种数据结构内在的逻辑关系及其在计算机中的实现方式和使用。此外,对......一起来看看 《数据结构》 这本书的介绍吧!

在线进制转换器
在线进制转换器

各进制数互转换器

Markdown 在线编辑器
Markdown 在线编辑器

Markdown 在线编辑器

正则表达式在线测试
正则表达式在线测试

正则表达式在线测试