denilsonsa / atmega8-magnetometer-usb-mouse

Introduction

This is a USB HID absolute pointing device using an ATmega8 AVR 8-bit microcontroller and a HMC5883L magnetometer. It allows the user to control the mouse pointer by moving the sensor in the air, pointing it at the desired position, somewhat similar to the Wiimote controller (although using a completely different technology).

It was developed by Denilson Figueiredo de Sá (see blog post) in the year 2011 as the final graduating project in order to obtain Bachelor's degree in Computer Science at DCC/UFRJ.

This project is mirrored at:

• https://bitbucket.org/denilsonsa/atmega8-magnetometer-usb-mouse
• https://github.com/denilsonsa/atmega8-magnetometer-usb-mouse

The full text of my thesis (written in Portuguese) is available as PDF in the download section of GitHub and BitBucket. The LaTeX source is available in the monografia/ directory (some minor tweaking might be needed in order to compile it).

Photos and videos

• Photos at Picasa.
• Video: USB Absolute Pointing Device implemented in ATmega8 using Magnetometer (work-in-progress video)
• Video: Dispositivo apontador com interface USB usando magnetômetro (final version, English subtitles available)
• Video: Moving the mouse pointer using head movements (final version)
• Video playlist: Playlist with all videos from this project (includes many work-in-progress videos)

The schematic diagram of the circuit is available at the monografia/img/ subdirectory of this repository.

How it works

The device implements USB HID and should work on any operating system (has been successfully tested on Linux, Mac OS X and Windows). It identifies itself as a keyboard and a mouse (actually, an "absolute pointing device").

It has a physical switch that selects between two modes of operation (configuration mode and mouse mode) and three push-buttons.

Upon plugging the device to the computer, the user should set the switch to configuration mode and open any simple text editor. In this mode, the device prints the configuration menu by sending (virtual) keyboard events to the computer (maybe it would be more accurate to say that it "types" the menu items, instead of printing). Two of the device buttons are used to navigate the menu items (selecting the next or the previous item), and the third button confirms the current selection.

Once in the configuration mode, the user should calibrate the "zero" from the sensor, as well as the screen corners. The calibration data is stored in the EEPROM memory of the microcontroller, and thus it will be remembered even after unplugging the device.

Upon starting the "zero" calibration, the device will start printing values from the sensor, and the user should move the sensor in all possible directions, trying to obtain the maximum and minimum values for each of the three axes (X, Y, Z). The confirm button should be pressed to finish the calibration. This calibration is required because the sensor might have a bias and thus return values that are not centered on number zero (see images zerocal_off and zerocal_on from monografia/img/ subdirectory).

After the "zero" is correctly calibrated, the user should calibrate each screen corner. The user should navigate the menu items up to Set topleft, point the sensor at top-left corner of the screen and then press the confirm button. This should be repeated for all other corners. For best results, the user should be directly in front of the screen center, and the screen should be facing either to the North or to the South direction.

The "zero" calibration should be needed only once, right after building the project. The corner calibration, on the other hand, is required anytime the user faces a different screen orientation.

After completing these two calibrations, the device is ready, and the user may switch to mouse mode. In this mode, the mouse pointer will be moved according to the movements of the sensor, and the three buttons work as mouse buttons (left, right and middle button).

The device reads the magnetic field measurements from the sensor as a 3-axis vector and applies an algorithm to convert that 3D vector into 2D screen coordinates. For details about the algorithm, read the mouseemu.c source code.

Due to the limited sensor precision and the amount of captured noise, the device applies a smoothing filter to the pointer position. This increases the perceived precision, but also introduces a slight delay in the movements.

The mode switch can also be used to pause the mouse position, as the pointer is not moved while in the configuration mode.

All the steps mentioned here can be seen in this video.

Possible improvements

• Use a microcontroller with more memory. This is needed before implementing any further improvements.

• Use the DRDY interrupt signal from the sensor in order to achieve up to 160Hz. The currently implemented method uses a 75Hz continuous measurement mode together with polling. It was implemented this way because the sensor PCB I bought from eBay did not have the DRDY line available. The PCB being sold at Love Electronics has that line.

• For best results, the user must be facing to the North or to the South direction. If, instead, the user is facing to the West or to the East, the vertical movement of the pointer is severely degraded. This happens because, in this case, the sensor rotates around the same axis as the magnetic field, and thus gives little to no change in the measurements. A solution for this problem is to attach an accelerometer as a second sensor to this device.

  • With these two sensors, the magnetometer can be used for horizontal pointer movement and the accelerometer for the vertical pointer movement.

  • These two sensors can be used together to implement a tilt-compensation (similar to this tutorial from Love Electronics).

  • A third sensor, gyroscope, can be added in order to improve precision and reduce the pointer shaking, increasing the responsiveness of the device.

• Try another magnetometer with better precision (if there is such thing).

• Try other algorithms for converting the coordinates.

• Implement wireless communication between the device and the computer.

  • It can be done by using a pair of microcontrollers: one next to the computer, talking to the USB port; and another next to the sensor. The communication between these two microcontrollers can be wireless. This solution has been done before in two other projects.

  • Or it can be done by implementing a Bluetooth HID device.

Requirements

In order to build this project, you need:

• ATmega8 or any other similar AVR microcontroller. If using a different model, some minor fine-tuning of the firmware might be neded. By the way, if you are going to buy a microcontroller, I highly recommend choosing one with more memory. Although 8KiB was enough, some parts of the firmware had to be disabled in order to fit. If you can, get a device with at least 16KiB of flash memory.
• HMC5883L 3-axis magnetometer. If you use a different sensor, be prepared to rewrite the sensor handling code.
• Other electronic components. See the circuit schematic at monografia/img/AVR-magnetometer-usb-mouse, available in SVG, PNG and PDF formats.

The required software environment:

• AVR-GCC - Developed with version 4.5.3. Different versions require updating a few compiler flags at the Makefile, as the available flags change between each major version.
• AVR-Libc - Developed with version 1.7.0.
• AVRDUDE, or any other tool to write the firmware onto the device.
• Unix-like system - Developed on Gentoo Linux amd64, should work anywhere with the standard Unix tools.

Directories in this repository

The main contents of this project are in these three directories:

• firmware/ - Contains the source-code of the firmware.
• projection/ - Python code for studying different algorithms for converting the 3D vectors to 2D screen coordinates.
• monografia/ - LaTeX source of the thesis (written in Portuguese).
• apresentacao/ - LaTeX source of the presentation (written in Portuguese).

There are also some extra directories:

• html_javascript/ - Some HTML pages I used during my presentation.
• linux_usbhid_bug/ - Information about a minor bug in Linux USB HID handling.
• other_scripts/ - Some scripts to generate a graph of the firmware size over time.

How to build this project

All commands listed here assume you are inside the firmware directory (the one with Makefile and checksize).

Want a quick list of available make targets? Run make help.

Preparation

These steps only need to be done once. They are the initial setup of the project.

1. Mount the hardware on your breadboard. You can find a short description at the Hardware description comment in main.c file and a complete circuit schematic at monografia/img/AVR-magnetometer-usb-mouse, available in SVG, PNG and PDF formats.

2. Open hardwareconfig.h and check if those definitions are consistent with the hardware. Basically, just check if the USB D- and USB D+ are connected to the correct pins.

3. Open TWI_Master.h and check if TWI_TWBR value is correct. It should be updated if you use a different clock rate.

4. Open Makefile.

   1. Set AVRDUDE_PARAMS according to your AVR programmer, if you use something other than USBasp.
   2. If you use a clock other than 12MHz, update F_CPU setting.
   3. If you use a microcontroller other than ATmega8, update GCC_MCU, AVRDUDE_MCU, BOOTLOADER_ADDRESS and CHECKSIZE_CODELIMIT.
   4. Also check if the fuse bits from AVRDUDE_PARAMS_FUSE are correct.
   5. If you want to use a bootloader, set BOOTLOADER_ENABLED to 1. Make sure your device has enough space to hold the main firmware together with the bootloader.
   6. Set ENABLE_KEYBOARD, ENABLE_MOUSE and ENABLE_FULL_MENU to 1 or 0, according to what you want in the final firmware. Look at the comments in that file for detailed information.

5. Run make writefuse to write the fuse bits.

Writing the bootloader (optional)

This section is completely optional. You don't need a bootloader. It's just cool and handy, but you don't need it. Feel free to skip these steps.

This project comes with USBaspLoader. After it is written to the microcontroller, any later firmware update can be done without the need of a dedicated AVR programmer.

After the bootloader is written, if a certain condition is true (a specific button is held down) during the device boot, then the bootloader will take control and the device will identify itself as USBasp. Writing to this "virtual" USBasp will actually update the firmware, without the need of any extra hardware.

1. Did you update the Makefile as described above? Did you run make writefuse?

2. Run make clean.

3. Run make boot. This will compile the bootloader.

4. Run make writeboot. This will write the bootloader to the microcontroller. You need an AVR programmer for this step.

5. Run make clean to clean up compiled files. This is required because the compiled files from the bootloader are incompatible with the main project (and vice-versa).

All done! You don't need an AVR programmer anymore!

Writing the EEPROM (optional)

You don't need to write the EEPROM now. You can just use the firmware's built-in menus (enabled with ENABLE_KEYBOARD) to interactively update the settings stored in the EEPROM.

The EEPROM values defined in sensor.c are appropriate for my sensor. Probably your sensor will have different calibration numbers, and thus it is highly recommended to use the firmware's menus to calibrate it (at least once).

Anyway, to write the EEPROM values, just run make, followed by make writeeeprom.

Writing the main firmware

You either need an AVR programmer, or you need to start the bootloader on the microcontroller (see the section about the bootloader).

1. Run either make all or make combine.

   • make is a shortcut for make all.

   • make combine uses some special compiler flags in order to compile all files at the same time, leading to extra optimizations not possible when compiling separately. This command will not work on GCC 4.6 or newer, because the flags have changed (and, thus, they need to be updated). Read Makefile to learn more.

   • If it fails, try running make clean. The Makefile from this project is not perfect and does not list all file dependencies. It's always a good idea to run make clean whenever something fails.

2. Run make writeflash.

After you edit the firmware, you only need to redo these two steps.

Acknowledgements

• Prof. Nelson Quilula Vasconcelos, advisor for this project.
• Bruno Bottino Ferreira for the help and patience during this project.
• Marcelo Salhab Brogliato for suggesting the coordinate conversion using linear equations.
• OBJECTIVE DEVELOPMENT Software GmbH for the awesome V-USB firmware-only implementation of USB for AVR devices.
• Atmel Corporation for the AVR microcontrollers and the AVR315: TWI Master Implementation.
• Authors and contributors of all open-source and free software used during this project.
• Marcin Wichary demonstration at Google I/O 2011: The Secrets of Google Pac-Man: A Game Show, which gave me the main idea for this project.

vi:expandtab:filetype=markdown