MPU6050 calibration using I2Cdevlib library

Other Hardware Sensors
1

Hello guys,

I decided to play with some MPU6050 I had around and I stumbled on some doubts regarding calibration.

First of all, I use Jeff Rowberg's library (link) to interface with the IMU.

So, my questions are:

1- In this library, there are methods to set offsets for sensor calibration (e.g. accelgyro.setXAccelOffset(-4336);).
What exactly do they do with the values sampled?
I have read somewhere (unfortunately can't track where) that such calibrations only sums (or substracts, depend on the signal) an offset from all digital values sampled.
So, applying a offset using these methods has the same effect of taking each sample and summing (or subtracting) the respective offset?

To give a practical situation: suppose I need to measure a vertical acceleration motion, that can be 2g up and 2g down (without accounting the presence of gravity). And that I want maximum sensitivity, therefore would like to stay in the 2g range.
So, in absolute values, I would have 1g up (2g motion minus 1g from gravity) and 3g down (2g motion plus 1g from gravity). With raw values, any motion with down acceleration higher than 1g would overflow the reading, since 1g from motion plus 1g from gravity would already reach the accelerometer maximum range.
But, if we calibrate the sensor so it outputs a 0 value when aligned with vertical axis (therefore 1g from gravity would give a 0 output), would we be able to properly measure motion accelerations higher than 1g (e.g. 1.2g, 1.3g, etc.)? Or would the output now simply overflow at 1g, i.e. would still be sensing values higher than 2g (e.g. 2.2g, 2.3 g), therefore overflowing but outputting all these values as 1g (would take the overflowed output of 2g and "subtract" 1g because of the offset given)?
This is really confusing me and I even tried to devise some experiments to determine that myself, but could not take any conclusions from the results I had.

2- Looking for a calibration sample code, I found one at I2Cdevlib forum (post with code). But in the code, the offset values inserted on the offset methods cited above are divided by a factor (8 for accelerations and 4 for gyros). My question is: where do these values come from?

I saw some questions regarding this in the post, and even asked myself, but since the post is very old, I think I will not get an answer there.

Searching in Google, I found a note from Freescale (link to note) saying that to calibrate their accelerometer, one had to divide the offset values by 4 or 2, depending on the scale range (+-2g, +-4g, +-8g), so the offset value could fit in a 8-bit register. When I read that, that seemed like a good reason for those 8 and 4 division, but after a lot of struggling to find a logic for MPU6050 scale range, I gave up.
In fact, looking at the MPU6050.h source file, the methods for gyro offset indicate they need to receive a 8 bit value, while accelerometer should receive a 16 bit value (look here). That seemed like a good clue, but I could not make much of it.

Further searching made me found a post by the creator of the calibration sample code where he said: "It is not so straigthforward to update your offset every time. In my case, I had to put my initial readings, negated (change sign) and divided by 4 (gyros) and by 7.8 (accels)." (link to post). From this, it seemed to me to be some kind of trial and error, but I may be wrong. From another post, here or on the I2Cdevlib, that I can't find anymore, I remember reading Jeff Rowberg (if I am not mistaken) commenting that the input values in the offset methods should be divided by a scaling factor, but since that wasn't the focus of the topic no further explanations where presented.

Best regards

2

Great overview and tutorial on calibration here: Tutorial: How to calibrate a compass (and accelerometer) with Arduino | Underwater Arduino Data Loggers

3

Thanks for the link jremington.

Hope I can find some insights to answer these questions in this tutorial and related links.

EDIT:

Reading a post that was linked (post), it is said that the considered accelerometer offset register had a 12-bit alignment, which justified a division by 16 (24). Here is the code of this part:

void loop()
{
compass.read();
float Xa_print, Ya_print, Za_print;

Xa_print = compass.a.x/16.0; //Acceleration data registers contain a left-aligned 12-bit number, so values should be shifted right by 4 bits (divided by 16)
Ya_print = compass.a.y/16.0;
Za_print = compass.a.z/16.0;
}

But I really can't grasp this process... Xa_print is a float, a 32-bit data type, why do I have to divide by 24 to put it on a 12-bit register? In fact, I could not understand what is this 12-bit aligned number...

Moreover, from the read of the tutorial, I understood that offsets really act on just the already sampled values, right? In other words, they can't avoid overflow due to high excitation outside the programmed accelerometer range, right? I just don't want to misunderstand something in the post.

Best regards.

4

The raw values that you get from the sensor do not need to be scaled for input to magneto. It performs internal scaling to the value you enter for Hm (the expected magnitude of the Earth's magnetic field in your vicinity).

And Hm itself is somewhat arbitrary because only the relative X and Y values of the magnetic field are used for calculating the yaw (compass) angle.

So, don't bother with the divide by 16.

5

Right! Thanks for your answer jremington, a lot is clearer now.

So, the Magneto software used for calibrating the sensor don't really need this scaling because it can treat the input so it will properly fit inside the accelerometer's register, right?

But consider a situation in which I want to implement a calibration routine inside an Arduino sketch for my MPU6050: How do I properly set/scale the calibration offsets values that will be the argument of the, for example, "accelgyro.setZAccelOffset( offset_value)" method?

My doubt is that I really don't understand what number I am supposed to put in the argument, because all these examples make a division, which I suspect should depend on the acceleration range I configure the MPU.

Also, any thoughts about how these offsets affects the accelerometer's reading? Will it avoid overflow/saturation of the acceleration measured in a given scale?

Best regards!

6

How do I properly set/scale the calibration offsets values that will be the argument of the, for example, "accelgyro.setZAccelOffset( offset_value)" method?

I don't know, because I don't use that library. 0 might be appropriate for an offset, and 1 for the scale factor.

The magneto-corrected values for the accelerometer should be fed directly to whatever part of the code uses those measurements for the orientation calculations, which will assume that the appropriate corrections have been made.

In other words, skip the library-applied corrections, if you can figure out how to do that.