Accelerometer data retrieval

From Openmoko

Jump to: navigation, search



This document describes a way to access the data provided by the accelerometers. Furthermore the format of the acquired datastream is explained in detail.

This document does not cover the physical basics of acceleration or the mathematical details on this subject.

Consult the hardware specs at the STMsite for the LIS392DL. It contains an interesting read how this unit functions.

Axis orientation

To understand the values provided by the accelerometers it is crucial to understand how the sensors are oriented. In the following I will refer to the sensors as first and second sensor. The way to access the data sources of these two will be described later on in this document.

The Z axis is pointing from the display downwards to to the back of the openmoko. This applies to both of the sensors. (Also see Definition of gmeter readings as force rather than acceleration)

Orientation of the first sensor

To explain the axis orientation I have created some images for easier understanding. Note that X and Y are in the same plane as the screen, while the Z arrow is pointing "into the screen", i.e. behind the Neo. The device for the first sensor is /dev/input/event2.

Axis orientation of the first accelerometer

Orientation of the second sensor

In contrast to the first sensor the second one is turned 45 degrees around the Z axis. See the attached image to get a clue about its orientation. Note that X and Y are in the same plane as the screen, while the Z arrow is pointing "into the screen", i.e. behind the Neo. The device for the second sensor is /dev/input/event3.

Axis orientation of the second accelerometer

Data acquisition

The information from both of the accelerometers is exported through two different input event based file mappings. These device nodes can be found at /dev/input/event2 and /dev/input/event3.

NOTE : --MicVM 19.43, 19 November 2010 (UTC) : On kernel SHR kernel the device names changed to /dev/input/event3 and /dev/input/event4 respectively.

The sensor I am refering to as the first one is event2. Therefore the second one is accessible through the event3 device.

Hint: Adding /etc/udev/rules.d/55-freerunner-accelerometer.rules with

SUBSYSTEM=="input", ATTRS{modalias}=="input:b0018v0000p0000e0000-e0,3,kra0,1,2,mlsfw", SYMLINK+="accelerometer-top"
SUBSYSTEM=="input", ATTRS{modalias}=="input:b0018v0000p0000e0000-e0,3,kra0,1,2,mlsfw", SYMLINK+="accelerometer-bottom"

will give you bit more meaningful device names. Lindi 08:16, 15 July 2009 (UTC)

These device nodes can be opened by the default filesystems calls for reading files. The data can be read from the stream as soon as the sensor measures it. Note that since the data is exported as normal files you can easily e.g. use netcat to stream the accelerometer data over bluetooth to your laptop to a control a game running on your laptop.

Data structure

To be able to use the measured data it is important to understand the format in which the data is provided. The structure of the given data is based on the kernel input event message system which exports the following data structure:

struct input_event {
struct timeval time;
__u16 type;
__u16 code;
__s32 value;

The data is written to the stream message by message. Therefore the minimal used blocksize is 128 bit (16 byte).

Lets have a look at an example readout of the accelerometer data, divided into the used sections.

|----- time ------|  |type| |code|  |-value-|
8c66 4819 721c 0006   0002   0002   03a8 0000
8c66 4819 7222 0006   0000   0000   0000 0000
8c66 4819 99e6 0006   0002   0000   0048 0000
8c66 4819 9a36 0006   0002   0001   0024 0000
8c66 4819 9a50 0006   0002   0002   0396 0000
8c66 4819 9a57 0006   0000   0000   0000 0000

Using recent kernels (>=2.6.29), you might see this instead, because accelerometers are now reporting absolute values instead of relative ones (more on that later) :

|----- time ------|  |type| |code|  |-value-|
8163 49da 6d62 000d   0000   0000   0000 0000
8163 49da 91d8 000d   0003   0000   0048 0000
8163 49da 9231 000d   0003   0001   0012 0000
8163 49da 9251 000d   0003   0002   03ba 0000
8163 49da 9270 000d   0000   0000   0000 0000
8163 49da b6cf 000d   0003   0000   0036 0000

I think the time structure does not need further explanation. A lot more interesting is the type, code and value part of every message. Lets take a closer look at these parts.

Event types

The types categorize the incoming messages. All possible types can be found in the kernel sources or include files in INCLUDEDIR/linux/input.h During the tests with the accelerometers I observed the fact that only two different message types are used.

According to linux/input.h this event is called EV_SYN. It signals the wish to syncronize. Normally this event is used in combination with code 0x00 to mark the send data complete and therefore applyable.
This event is called EV_REL and signals relative movement. It is used to transmit the acceleration the sensors encounter.

The definition should not be taken too seriously in this context, because the data values provided by the accelerometer always represent the absolute acceleration measured at the given time. So, in newer kernels, the proper absolute movement notification is used instead :

This event is called EV_ABS and signals absolute movement. It replaces the EV_REL notifications in newer kernels, but

the information (the event codes) is the same.

Event codes

Both types of event may supply different codes. These codes can be understood as some kind of further specification about the specified data values

Synchronization event codes

The synchronization event may use quite a lot of codes, as linux/input.h shows. However the only used one seems to be the 0x00 code.

This code is referred to as SYN_REPORT. It means that the last dataset was completely transmitted. Therefore the before transmitted set of data values can be considered complete. This means if this message is received you may process the given data further.

Relative movement event codes

The amount of possible codes for this event type is quite big. The only used ones are the following ones.

REL_X - Acceleration in x direction
REL_Y - Acceleration in y direction
REL_Z - Acceleration in z direction

The X, Y and Z axis are to be understand as defined in the chapter about Axis orientation.

A typical message block

A typical message block consists of 3 messages containing the acceleration data for every of the three axis followed by a syncronization message to signal the end of the block.

The following example is such a message block with detailed explanation of its different messages and data sections.

8c66 4819 99e6 0006   0002   0000   0048 0000
|------Time-------|  EV_REL  REL_X  |-Value-|
(Measured acceleration in x axis direction of 72)

8c66 4819 9a36 0006   0002   0001   0024 0000
|------Time-------|  EV_REL  REL_Y  |-Value-|
(Measured acceleration in y axis direction of 36)

8c66 4819 9a50 0006   0002   0002   0396 0000
|------Time-------|  EV_REL  REL_Z  |-Value-|
(Measured acceleration in z axis direction of 918)

8c66 4819 9a57 0006   0000   0000        0000 0000
|------Time-------|  EV_SYN  SYN_REPORT  |-Value-|
(The transmitted data block is complete you may process the given data)

Absolute movement event codes

More information about this change may be available here :;a=commit;=b2ae84d8bd469fe81fa0a967fde9008416a78365

Although the information is the same as before, the absolute interface differs from the relative one by the fact that only changed values are reported. So if the phone was completely idle, experiencing exactly the same acceleration from one sample to the next, you will read only synchronization events (this will not happen due to noise, though).

So when you miss the value for an axis between two synchronization events, just assume the previous value still holds.

Test application

A simple test application written in ruby:

#!/usr/bin/env ruby
x = 0
y = 0
z = 0"/dev/input/event3") do |f|
  while true
    event ="llSSl")
    time = event[0] + event[1] / 1000000.0
    type = event[2]
    code = event[3]
    value = event[4]
    if type == 2 || type == 3
      x = value if code == 0
      y = value if code == 1
      z = value if code == 2
    if type == 0 && code == 0
      sum = Math::sqrt(x*x + y*y + z*z).to_i
      printf("%5d %5d %5d %5d\n", x, y, z, sum)

Equivalent version in python:

import struct
from math import sqrt

x = 0
y = 0
z = 0
secondsensorfile = "/dev/input/event3"
#int, int, short, short, int
fmt = 'iihhi'
#open file in binary mode
in_file = open(secondsensorfile,"rb")
event =
while event:
	(time1,time2, type, code, value) = \
	time = time2 / 1000.0

	if type == 2 or type == 3:
		if code == 0:
			x = value
		if code == 1:
			y = value
		if code == 2:
			z = value
	if type == 0 and code == 0:
		sum = int(sqrt(x*x + y*y + z*z))
		print time, x, y, z, sum
	event =

Equivalent in Perl


$x = 0;
$y = 0;
$z = 0;

$secondsensorfile = "/dev/input/event3";
#open file in binary mode
open FILE, $secondsensorfile;
binmode FILE;

while (read FILE, $buf,16) {
    ($t1,$t2,$type,$code,$value)=unpack "iissi",$buf;
    if ($type==2) {
	if ($code==0) {$x=$value};
	if ($code==1) {$y=$value};
	if ($code==2) {$z=$value};
    if (($type==0) && (code==0)) {
        printf "%f %05d %05d %05d\n",$t1+$t2/1000000,$x,$y,$z;

A PHP absolute version


$x = 0;
$y = 0;
$z = 0;

$secondsensorfile = "/dev/input/event3";
#open file in binary mode
$dfeed2 = fopen($secondsensorfile, "rb");

while (!feof($dfeed2)) {
  $contents = fread($dfeed2, 16);
  $arr = unpack("s4time/S1data/S1code/l1value",$contents);
  if ($arr[data]=3){
	if ($arr[code]==0) $x=$arr[value];
        if ($arr[code]==1) $y=$arr[value];
        if ($arr[code]==2) $z=$arr[value];
 print "x= $x, y= $y, z=$z \n";

Also, for the new absolute interface (Perl) :


$x = 0;
$y = 0;
$z = 0;

$secondsensorfile = "/dev/input/event3";
#open file in binary mode
open FILE, $secondsensorfile;
binmode FILE;

while (read FILE, $buf,16) {
    ($t1,$t2,$type,$code,$value)=unpack "iissi",$buf;
    if ($type==3) {
        if ($code==0) {$x=$value};
        if ($code==1) {$y=$value};
        if ($code==2) {$z=$value};
    if (($type==0) && ($code==0)) {
        printf "%f %05d %05d %05d\n",$t1+$t2/1000000,$x,$y,$z;

Read the data in C (FIXME: We're not printing times here).

#include <stdio.h>
#include <time.h>
#include <sys/times.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
#include <stdint.h>
#include <assert.h>

struct input_event {
    struct timeval time;
    uint16_t type;
    uint16_t code;
    int32_t value;

int read_all(int fd, char *buf, int count)
    int n_read = 0;
    while (n_read != count) {
        int result = read(fd, buf + n_read, count - n_read);
        if (result < 0)
            return result;
        else if (result == 0)
            return n_read;
        n_read += result;
    return n_read;

int main(int argc, char *argv[])
    int fd;
    struct input_event ev;

    assert(16 == sizeof(struct input_event));

    if (argc != 2) {
        fprintf(stderr, "missing /dev/input/XXX\n");
        return 1;

    while (1) {
        if ((fd = open(argv[1], O_RDONLY)) == -1) {
            return 1;
        int ret = read_all(fd, (char *) &ev, sizeof(struct input_event));
        if (ret != sizeof(struct input_event)) {
            fprintf(stderr, "ret == %d\n", ret);
            return 1;
        printf("type:%u code:%u value:%d\n", ev.type, ev.code, ev.value);
        usleep(500000);  /* 2 reads per second, you might want to read more */

    return 0;

And in Guile Scheme. (This one is a complete autorotation program.)

;; Open device file for the second accelerometer (whose axes are
;; aligned with the screen).  Must make Guile's internal buffer small
;; (16), otherwise it reads too far ahead and this program gets mostly
;; out of date data.
(define e (open-file "/dev/input/event3"
(setvbuf e _IOFBF 16)

;; Uniform vector for reading data from device file.  Must match
;; sizeof(struct input_event), otherwise the read can get an 'invalid
;; argument' error.
(define v (make-u16vector 8))

(define (current-acceleration)
  (let loop ((x #f) (y #f) (z #f))
    (uniform-vector-read! v e)
    (let ((code (u16vector-ref v 4))
	  (axis (u16vector-ref v 5))
	  (meas (+ (u16vector-ref v 6)
		   (* 65536 (u16vector-ref v 7)))))
      (if (>= meas (expt 2 31))
	  (set! meas (- meas (expt 2 32))))
      (case code
	((2 3)
	 (case axis
	   ((0) (loop meas y z))
	   ((1) (loop x meas z))
	   ((2) (loop x y meas))
	   (else (loop x y z))))
	 (if (and x y z)
	     (list x y z)
	     (loop x y z)))))))

(define (orientation x y z)
  (cond ((and (< y -200)
	      (> (- y) x y))
	((and (> x 200)
	      (< (- x) y x))
	((and (> y 200)
	      (< (- y) x y))
	((and (< x -200)
	      (> (- x) y x))
	(else #f)))

(define (main)
  (let loop ((ori #f))
    (let ((new-ori
	   (apply orientation
      (and ori
	   (not (= ori new-ori))
	   (system* "xrandr"
		    (number->string new-ori)))
      (sleep 1)
      (loop (or new-ori ori)))))


The /sys interface

To get the sampling rate of the first accelerometer lis302dl.1 :

  1. cat /sys/devices/platform/lis302dl.1/sample_rate


To set the sampling rate :

  1. echo 400 > /sys/devices/platform/lis302dl.1/sample_rate
  2. cat /sys/devices/platform/lis302dl.1/sample_rate


OM2008.12 has too big threshold for some games and applications (value=18). You can use sysfs to change it. Use command like this one:

  1. echo 10 > /sys/devices/platform/lis302dl.1/threshold

The other one is lis302dl.2. See GTA02 sysfs#Accelerometers for the other options.

NOTE : --Garthps 15:38, 14 December 2010 (UTC) : the interface has changed in newer kernels. Now it is in :




NOTE2 : --Boudewijn 10:42, 11 August 2011 (UTC): It might have changed once more; I could not find it using om-gta02 #1 Fri Jun 10 18:09:48 CEST 2011 armv4tl, but found it in:


(following the path suggested by sysfs for 2.6.28)

See also

External Links

Personal tools