calibration of magnetometer

[vreijs]

Hello kriswiner, I read you article: https://github.com/kriswiner/MPU-6050/wiki/Simple-and-Effective-Magnetometer-Calibration and I am wondering about a few things: . do you know if in the Android OS this calibration is included or does the App writers need to implement this? . if it is part of the OS, is this 'calibration' movement one need to do in some way recognised by the OS? And when recognised, it performs the calibration (both for hard and soft irons?).

I am asking this as I have some problems with measuring azi/alt. I am using a Samsung Tab2 7.0 GT-P3110 Android 4.2.2 Regardless of the App I am using (e.g. SunSurveyor or Measure Angle/Geosurvey Transit Apps), the results of azimuth and altitude measurements depend on the rotation of the device. So when pointing towards an object (Moon/Sun) with known azi/alt with the camera, the azimuth and altitude measurements changes when the device is rotated (see attached plot).

The precision both in azimuth and altitude is quite good (0.5 deg, by measuring the azi/alt several times using the same rotation of the device), which is close to the one of a normal compass/clinometer. But the accuracy is not that good! <see for definition of the word accuracy and precision here: https://en.wikipedia.org/wiki/Accuracy_and_precision >

Attached is a plot of the measured azi and alt (using SunSurveyor App, but other Apps [such as Measure Angle] have same behavior). I turned the device 90 degrees (starting with landscape, portrait, the -landscape and -portrait) and measured each time the az/alt of the Moon. You also see the average of these measurements and the ephemeris location of the Moon. So the average looks ok-ish compared to the ephemeris values calculated for the Moon.

I have now been measuring the Moon and Sun many times and I see every time the similar behavior and accuracy spread (although same variation can happen). Again the precision stays within the 0.5 degrees when doing multiple measured at the same rotation. By the way I am away from iron objects (external or on body) and also away from electricity cables, etc. I also do the so called 'calibration' of the magnetometer (I did several different calibration movements). That calibration has some effect, but normally the spread stays the same. I also compared it with a normal compass and clinometer, which gives values comparable to the ephemeris.

If you see that plot, do you think this is something in the OS, the sensor, the device and/or the App? Calculating azi/alt is of course not an easy thing. In the device or Operating systems things are being done of course, but I am wondering why I see this large inaccuracy (certainly in azi [up to 20 degrees], and smaller in alt [up to 7 degrees])?

Here is another article that described the issues around the magnetometer (and also how to overcome them [partly]). I am wondering if the calibration inside the device and/or OS is done properly? http://www.nxp.com/files/sensors/doc/app_note/AN4246.pdf

Any ideas/hint you see in this?

Thanks for any feedback.

All the best,

Victor

[vreijs]

Here are a few other sessions with such rotations (also different Apps) without calibration: with calibration

A few others with different Apps.:

[kriswiner]

Hi Victor,

This is a very interesting application of attitude estimation! I hadn't though about using it for astrometry.

I don't know the details of the tools you are using but I have some thoughts and a suggestion.

First of all, as you note, precision is not very useful if the accuracy is poor. Since sensor fusion to obtain absolute orientation (attitude estimation) is well-understood I must assume the problem is in the underlying sensors data. I have ample evidence that even with excellent sensor fusion filtering, the accuracy can be poor (OK, meaning 7 - 10 degrees in heading or attitude) if the underlying sensors are not very good. Even hard iron and limited soft iron calibration can't always cure a misbehaving magnetometer, and for some sensors nothing can help.

You can see examples of this here:

https://github.com/kriswiner/MPU-6050/wiki/Hardware-Sensor-Fusion-Solutions

and here:

https://github.com/kriswiner/EM7180_SENtral_sensor_hub/wiki/E.-Typical-Resul ts-Using-the-SENtral

The data plots show variations in azimuth and altitude of ~4 degrees standard deviation. This is not terribly bad, in fact I would say this is typical for even pretty good sensor fusion solutions out there like the MAX21000. Do you know which sensors are being used in the Samsung product?

The best orientation solution I have found is the EM7180 coupled with the MPU9250 which can provide 1 degree accuracy in pitch and roll and 2 degree accuracy in heading (for AHRS navigation). Not sure what this would translate to for astrometry.

So in short, the problem is IMHO not the apps, the fusion algorithms, or the calibration (this can always be better but can't fix poor sensors) but the underlying sensor data and there is nothing you can do about that except change sensors.

Kris -----Original Message----- From: vreijs [mailto:notifications@github.com] Sent: February 13, 2016 6:00 AM To: kriswiner/MPU-9250 Subject: [MPU-9250] calibration of magnetometer (#42)

Hello kriswiner, I read you article: https://github.com/kriswiner/MPU-6050/wiki/Simple-and-Effective-Magnetometer -Calibration and I am wondering about a few things: . do you know if in the Android OS this calibration is included or does the App writers need to implement this? . if it is part of the OS, is this 'calibration' movement one need to do in some way recognised by the OS? And when recognised, it performs the calibration (both for hard and soft irons?).

I am asking this as I have some problems with measuring azi/alt. I am using a Samsung Tab2 7.0 GT-P3110 Android 4.2.2 Regardless of the App I am using (e.g. SunSurveyor or Measure Angle Apps), the results of azimuth and altitude measurements depend on the rotation of the device. So when pointing towards an object (Moon/Sun) with known azi/alt with the camera, the azimuth and altitude measurements changes when the device is rotated (see attached plot). image18 https://cloud.githubusercontent.com/assets/17218070/13027981/3599099c-d259- 11e5-9ce8-4d2930c4545c.gif

The precision both in azimuth and altitude is quite good (0.5 deg, by measuring the azi/alt several times using the same rotation of the device), which is close to the one of a normal compass/clinometer. But the accuracy is not that good!

Attached is a plot of the measured azi and alt (using SunSurveyor App, but other Apps [such as Measure Angle] have same behavior). I turned the device 90 degrees (starting with landscape, portrait, the -landscape and -portrait) and measured each time the az/alt of the Moon. You also see the average of these measurements and the ephemeris location of the Moon. So the average looks ok-ish compared to the ephemeris values calculated for the Moon.

I have now been measuring the Moon and Sun many times and I see every time the similar behavior and accuracy spread (although same variation can happen). Again the precision stays within the 0.5 degrees when doing multiple measured at the same rotation. By the way I am away from iron objects (external or on body) and also away from electricity cables, etc. I also do the so called 'calibration' of the magnetometer (I did several different calibration movements). That calibration has some effect, but normally the spread stays the same. I also compared it with a normal compass and clinometer, which gives values comparable to the ephemeris.

If you see that plot, do you think this is something in the OS, the sensor, the device and/or the App? Calculating azi/alt is of course not an easy thing. In the device or Operating systems things are being done of course, but I am wondering why I see this large inaccuracy (certainly in azi [up to 20 degrees], and smaller in alt [up to 7 degrees])?

Here is another article that described the issues around the magnetometer (and also how to overcome them [partly]). I am wondering if the calibration inside the device and/or OS is done properly? http://www.nxp.com/files/sensors/doc/app_note/AN4246.pdf

Any ideas/hint you see in this?

Thanks for any feedback.

All the best,

Victor

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42 . https://github.com/notifications/beacon/AGY1qj5F8eb0XZsN4fRaS0VIuGaQFcsKks5 pjy5JgaJpZM4HZo42.gif

[vreijs]

Thanks Kris for your very valuable feedback. The stdev I see is 3 deg for the alt and around 8 deg for the azimuth. I think there would be no real difference for AHRS navigation and astrometry, IMHO.

I find it so interesting that the spread of the points is quite reproducable (not fully, but still)... So I was nt thinking about a statistical error but more a systematic error. And according you this could be very well in the sensor. Correct?

I plan to have a chat with the Samsung support desk, I will see what info I can get out of them (like which sensor they use in the Tab2). The AndroSensor App does not provide any info on type of magnetometer.

So I need to have a device with EM7180 coupled with the MPU925. Do you know something like that (in android/phone form)? Or do I need to use a Arduino or Raspberry?

Thanks again for your help.

All the best,

Victor

On 13 February 2016 at 17:23, Kris Winer notifications@github.com wrote:

Hi Victor,

This is a very interesting application of attitude estimation! I hadn't though about using it for astrometry.

I don't know the details of the tools you are using but I have some thoughts and a suggestion.

First of all, as you note, precision is not very useful if the accuracy is poor. Since sensor fusion to obtain absolute orientation (attitude estimation) is well-understood I must assume the problem is in the underlying sensors data. I have ample evidence that even with excellent sensor fusion filtering, the accuracy can be poor (OK, meaning 7 - 10 degrees in heading or attitude) if the underlying sensors are not very good. Even hard iron and limited soft iron calibration can't always cure a misbehaving magnetometer, and for some sensors nothing can help.

You can see examples of this here:

https://github.com/kriswiner/MPU-6050/wiki/Hardware-Sensor-Fusion-Solutions

and here:

https://github.com/kriswiner/EM7180_SENtral_sensor_hub/wiki/E.-Typical-Resul ts-Using-the-SENtral

The data plots show variations in azimuth and altitude of ~4 degrees standard deviation. This is not terribly bad, in fact I would say this is typical for even pretty good sensor fusion solutions out there like the MAX21000. Do you know which sensors are being used in the Samsung product?

The best orientation solution I have found is the EM7180 coupled with the MPU9250 which can provide 1 degree accuracy in pitch and roll and 2 degree accuracy in heading (for AHRS navigation). Not sure what this would translate to for astrometry.

So in short, the problem is IMHO not the apps, the fusion algorithms, or the calibration (this can always be better but can't fix poor sensors) but the underlying sensor data and there is nothing you can do about that except change sensors.

Kris -----Original Message----- From: vreijs [mailto:notifications@github.com] Sent: February 13, 2016 6:00 AM To: kriswiner/MPU-9250 Subject: [MPU-9250] calibration of magnetometer (#42)

Hello kriswiner, I read you article:

https://github.com/kriswiner/MPU-6050/wiki/Simple-and-Effective-Magnetometer -Calibration and I am wondering about a few things: . do you know if in the Android OS this calibration is included or does the App writers need to implement this? . if it is part of the OS, is this 'calibration' movement one need to do in some way recognised by the OS? And when recognised, it performs the calibration (both for hard and soft irons?).

I am asking this as I have some problems with measuring azi/alt. I am using a Samsung Tab2 7.0 GT-P3110 Android 4.2.2 Regardless of the App I am using (e.g. SunSurveyor or Measure Angle Apps), the results of azimuth and altitude measurements depend on the rotation of the device. So when pointing towards an object (Moon/Sun) with known azi/alt with the camera, the azimuth and altitude measurements changes when the device is rotated (see attached plot). image18 < https://cloud.githubusercontent.com/assets/17218070/13027981/3599099c-d259- 11e5-9ce8-4d2930c4545c.gif>

The precision both in azimuth and altitude is quite good (0.5 deg, by measuring the azi/alt several times using the same rotation of the device), which is close to the one of a normal compass/clinometer. But the accuracy is not that good!

Attached is a plot of the measured azi and alt (using SunSurveyor App, but other Apps [such as Measure Angle] have same behavior). I turned the device 90 degrees (starting with landscape, portrait, the -landscape and -portrait) and measured each time the az/alt of the Moon. You also see the average of these measurements and the ephemeris location of the Moon. So the average looks ok-ish compared to the ephemeris values calculated for the Moon.

I have now been measuring the Moon and Sun many times and I see every time the similar behavior and accuracy spread (although same variation can happen). Again the precision stays within the 0.5 degrees when doing multiple measured at the same rotation. By the way I am away from iron objects (external or on body) and also away from electricity cables, etc. I also do the so called 'calibration' of the magnetometer (I did several different calibration movements). That calibration has some effect, but normally the spread stays the same. I also compared it with a normal compass and clinometer, which gives values comparable to the ephemeris.

If you see that plot, do you think this is something in the OS, the sensor, the device and/or the App? Calculating azi/alt is of course not an easy thing. In the device or Operating systems things are being done of course, but I am wondering why I see this large inaccuracy (certainly in azi [up to 20 degrees], and smaller in alt [up to 7 degrees])?

Here is another article that described the issues around the magnetometer (and also how to overcome them [partly]). I am wondering if the calibration inside the device and/or OS is done properly? http://www.nxp.com/files/sensors/doc/app_note/AN4246.pdf

Any ideas/hint you see in this?

Thanks for any feedback.

All the best,

Victor

[vreijs]

Hello Kris, Used CPU-Z App to determine the sensors and the magnetic senor is MS-3E (YAS530) and the orientation sensor is MS-x.

Is this familiar?

All the best,

Victor

On 13 February 2016 at 17:54, Victor Reijs web.victor.reijs@gmail.com wrote:

Thanks Kris for your very valuable feedback. The stdev I see is 3 deg for the alt and around 8 deg for the azimuth. I think there would be no real difference for AHRS navigation and astrometry, IMHO.

I find it so interesting that the spread of the points is quite reproducable (not fully, but still)... So I was nt thinking about a statistical error but more a systematic error. And according you this could be very well in the sensor. Correct?

I plan to have a chat with the Samsung support desk, I will see what info I can get out of them (like which sensor they use in the Tab2). The AndroSensor App does not provide any info on type of magnetometer.

So I need to have a device with EM7180 coupled with the MPU925. Do you know something like that (in android/phone form)? Or do I need to use a Arduino or Raspberry?

Thanks again for your help.

All the best,

Victor

On 13 February 2016 at 17:23, Kris Winer notifications@github.com wrote:

Hi Victor,

This is a very interesting application of attitude estimation! I hadn't though about using it for astrometry.

I don't know the details of the tools you are using but I have some thoughts and a suggestion.

First of all, as you note, precision is not very useful if the accuracy is poor. Since sensor fusion to obtain absolute orientation (attitude estimation) is well-understood I must assume the problem is in the underlying sensors data. I have ample evidence that even with excellent sensor fusion filtering, the accuracy can be poor (OK, meaning 7 - 10 degrees in heading or attitude) if the underlying sensors are not very good. Even hard iron and limited soft iron calibration can't always cure a misbehaving magnetometer, and for some sensors nothing can help.

You can see examples of this here:

https://github.com/kriswiner/MPU-6050/wiki/Hardware-Sensor-Fusion-Solutions

and here:

https://github.com/kriswiner/EM7180_SENtral_sensor_hub/wiki/E.-Typical-Resul ts-Using-the-SENtral

The data plots show variations in azimuth and altitude of ~4 degrees standard deviation. This is not terribly bad, in fact I would say this is typical for even pretty good sensor fusion solutions out there like the MAX21000. Do you know which sensors are being used in the Samsung product?

The best orientation solution I have found is the EM7180 coupled with the MPU9250 which can provide 1 degree accuracy in pitch and roll and 2 degree accuracy in heading (for AHRS navigation). Not sure what this would translate to for astrometry.

So in short, the problem is IMHO not the apps, the fusion algorithms, or the calibration (this can always be better but can't fix poor sensors) but the underlying sensor data and there is nothing you can do about that except change sensors.

Kris -----Original Message----- From: vreijs [mailto:notifications@github.com] Sent: February 13, 2016 6:00 AM To: kriswiner/MPU-9250 Subject: [MPU-9250] calibration of magnetometer (#42)

Hello kriswiner, I read you article:

https://github.com/kriswiner/MPU-6050/wiki/Simple-and-Effective-Magnetometer -Calibration and I am wondering about a few things: . do you know if in the Android OS this calibration is included or does the App writers need to implement this? . if it is part of the OS, is this 'calibration' movement one need to do in some way recognised by the OS? And when recognised, it performs the calibration (both for hard and soft irons?).

I am asking this as I have some problems with measuring azi/alt. I am using a Samsung Tab2 7.0 GT-P3110 Android 4.2.2 Regardless of the App I am using (e.g. SunSurveyor or Measure Angle Apps), the results of azimuth and altitude measurements depend on the rotation of the device. So when pointing towards an object (Moon/Sun) with known azi/alt with the camera, the azimuth and altitude measurements changes when the device is rotated (see attached plot). image18 < https://cloud.githubusercontent.com/assets/17218070/13027981/3599099c-d259- 11e5-9ce8-4d2930c4545c.gif>

The precision both in azimuth and altitude is quite good (0.5 deg, by measuring the azi/alt several times using the same rotation of the device), which is close to the one of a normal compass/clinometer. But the accuracy is not that good!

Attached is a plot of the measured azi and alt (using SunSurveyor App, but other Apps [such as Measure Angle] have same behavior). I turned the device 90 degrees (starting with landscape, portrait, the -landscape and -portrait) and measured each time the az/alt of the Moon. You also see the average of these measurements and the ephemeris location of the Moon. So the average looks ok-ish compared to the ephemeris values calculated for the Moon.

I have now been measuring the Moon and Sun many times and I see every time the similar behavior and accuracy spread (although same variation can happen). Again the precision stays within the 0.5 degrees when doing multiple measured at the same rotation. By the way I am away from iron objects (external or on body) and also away from electricity cables, etc. I also do the so called 'calibration' of the magnetometer (I did several different calibration movements). That calibration has some effect, but normally the spread stays the same. I also compared it with a normal compass and clinometer, which gives values comparable to the ephemeris.

If you see that plot, do you think this is something in the OS, the sensor, the device and/or the App? Calculating azi/alt is of course not an easy thing. In the device or Operating systems things are being done of course, but I am wondering why I see this large inaccuracy (certainly in azi [up to 20 degrees], and smaller in alt [up to 7 degrees])?

Here is another article that described the issues around the magnetometer (and also how to overcome them [partly]). I am wondering if the calibration inside the device and/or OS is done properly? http://www.nxp.com/files/sensors/doc/app_note/AN4246.pdf

Any ideas/hint you see in this?

Thanks for any feedback.

All the best,

Victor

[kriswiner]

Hi Victor,

I use standard deviation but you are correct, the errors I see are rather systematic. Especially in the Bosch sensors but also in the ST ones. The heading precision is very good (I think his corresponds to the azimuth here) but when estimating ninety-degree turns the heading estimation can be off by ten degrees, reproducibly (I see the same in my cell phone). This is not sensor jitter but rather a systematic bias that in principle could be removed, but simple response surface recentering (hard iron calibration) and axes scaling (poor man's soft iron correction) does not correct for it.

So yes, I would recommend an EM7180+MPU9250 solution. You can solder a breakout board (https://www.tindie.com/products/onehorse/ultimate-sensor-fusion-solution/) with this solution to your favorite microcontroller. I have designed integrated solutions for UAVs that have a microcontroller and EM7180+MPU9250 on board (with motor drivers and boost converters that might not be useful to you). Here is an example:

https://forum.pjrc.com/threads/32985-Teensy-Flight-Controller

The advantage is a simple Arduino IDE can be used to write control software; I have all the code necessary to get absolute orientation from this kind of device.

I have a similar device using an nRF51822 BLE and ESP8266 wifi-enabled controller.

I am interested in the astrometry application. Do you think there is a commercial market for such a device? I mean like an instrumented finder scope which reads out absolute alt-azi on a display, etc. Maybe we could collaborate on development of such a device?

Kris

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 13, 2016 9:55 AM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Thanks Kris for your very valuable feedback. The stdev I see is 3 deg for the alt and around 8 deg for the azimuth. I think there would be no real difference for AHRS navigation and astrometry, IMHO.

I find it so interesting that the spread of the points is quite reproducable (not fully, but still)... So I was nt thinking about a statistical error but more a systematic error. And according you this could be very well in the sensor. Correct?

I plan to have a chat with the Samsung support desk, I will see what info I can get out of them (like which sensor they use in the Tab2). The AndroSensor App does not provide any info on type of magnetometer.

So I need to have a device with EM7180 coupled with the MPU925. Do you know something like that (in android/phone form)? Or do I need to use a Arduino or Raspberry?

Thanks again for your help.

All the best,

Victor

On 13 February 2016 at 17:23, Kris Winer notifications@github.com wrote:

Hi Victor,

This is a very interesting application of attitude estimation! I hadn't though about using it for astrometry.

I don't know the details of the tools you are using but I have some thoughts and a suggestion.

First of all, as you note, precision is not very useful if the accuracy is poor. Since sensor fusion to obtain absolute orientation (attitude estimation) is well-understood I must assume the problem is in the underlying sensors data. I have ample evidence that even with excellent sensor fusion filtering, the accuracy can be poor (OK, meaning 7 - 10 degrees in heading or attitude) if the underlying sensors are not very good. Even hard iron and limited soft iron calibration can't always cure a misbehaving magnetometer, and for some sensors nothing can help.

You can see examples of this here:

https://github.com/kriswiner/MPU-6050/wiki/Hardware-Sensor-Fusion-Solutions

and here:

https://github.com/kriswiner/EM7180_SENtral_sensor_hub/wiki/E.-Typical-Resul ts-Using-the-SENtral

The data plots show variations in azimuth and altitude of ~4 degrees standard deviation. This is not terribly bad, in fact I would say this is typical for even pretty good sensor fusion solutions out there like the MAX21000. Do you know which sensors are being used in the Samsung product?

The best orientation solution I have found is the EM7180 coupled with the MPU9250 which can provide 1 degree accuracy in pitch and roll and 2 degree accuracy in heading (for AHRS navigation). Not sure what this would translate to for astrometry.

So in short, the problem is IMHO not the apps, the fusion algorithms, or the calibration (this can always be better but can't fix poor sensors) but the underlying sensor data and there is nothing you can do about that except change sensors.

Kris -----Original Message----- From: vreijs [mailto:notifications@github.com] Sent: February 13, 2016 6:00 AM To: kriswiner/MPU-9250 Subject: [MPU-9250] calibration of magnetometer (#42)

Hello kriswiner, I read you article:

https://github.com/kriswiner/MPU-6050/wiki/Simple-and-Effective-Magnetometer -Calibration and I am wondering about a few things: . do you know if in the Android OS this calibration is included or does the App writers need to implement this? . if it is part of the OS, is this 'calibration' movement one need to do in some way recognised by the OS? And when recognised, it performs the calibration (both for hard and soft irons?).

I am asking this as I have some problems with measuring azi/alt. I am using a Samsung Tab2 7.0 GT-P3110 Android 4.2.2 Regardless of the App I am using (e.g. SunSurveyor or Measure Angle Apps), the results of azimuth and altitude measurements depend on the rotation of the device. So when pointing towards an object (Moon/Sun) with known azi/alt with the camera, the azimuth and altitude measurements changes when the device is rotated (see attached plot). image18 <

https://cloud.githubusercontent.com/assets/17218070/13027981/3599099c-d259- 11e5-9ce8-4d2930c4545c.gif>

The precision both in azimuth and altitude is quite good (0.5 deg, by measuring the azi/alt several times using the same rotation of the device), which is close to the one of a normal compass/clinometer. But the accuracy is not that good!

Attached is a plot of the measured azi and alt (using SunSurveyor App, but other Apps [such as Measure Angle] have same behavior). I turned the device 90 degrees (starting with landscape, portrait, the -landscape and -portrait) and measured each time the az/alt of the Moon. You also see the average of these measurements and the ephemeris location of the Moon. So the average looks ok-ish compared to the ephemeris values calculated for the Moon.

I have now been measuring the Moon and Sun many times and I see every time the similar behavior and accuracy spread (although same variation can happen). Again the precision stays within the 0.5 degrees when doing multiple measured at the same rotation. By the way I am away from iron objects (external or on body) and also away from electricity cables, etc. I also do the so called 'calibration' of the magnetometer (I did several different calibration movements). That calibration has some effect, but normally the spread stays the same. I also compared it with a normal compass and clinometer, which gives values comparable to the ephemeris.

If you see that plot, do you think this is something in the OS, the sensor, the device and/or the App? Calculating azi/alt is of course not an easy thing. In the device or Operating systems things are being done of course, but I am wondering why I see this large inaccuracy (certainly in azi [up to 20 degrees], and smaller in alt [up to 7 degrees])?

Here is another article that described the issues around the magnetometer (and also how to overcome them [partly]). I am wondering if the calibration inside the device and/or OS is done properly? http://www.nxp.com/files/sensors/doc/app_note/AN4246.pdf

Any ideas/hint you see in this?

Thanks for any feedback.

All the best,

Victor

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183713975 . https://github.com/notifications/beacon/AGY1qjzJla60mpztwc6DbriGSukmFdCoks5 pj2VhgaJpZM4HZo42.gif

[kriswiner]

No, MS probably means Measurement Specialties, but I am not familiar with the numbers. I'll poke around and see if I can find more info with this as a start.

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 13, 2016 10:06 AM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris, Used CPU-Z App to determine the sensors and the magnetic senor is MS-3E (YAS530) and the orientation sensor is MS-x.

Is this familiar?

All the best,

Victor

On 13 February 2016 at 17:54, Victor Reijs web.victor.reijs@gmail.com wrote:

Thanks Kris for your very valuable feedback. The stdev I see is 3 deg for the alt and around 8 deg for the azimuth. I think there would be no real difference for AHRS navigation and astrometry, IMHO.

I find it so interesting that the spread of the points is quite reproducable (not fully, but still)... So I was nt thinking about a statistical error but more a systematic error. And according you this could be very well in the sensor. Correct?

I plan to have a chat with the Samsung support desk, I will see what info I can get out of them (like which sensor they use in the Tab2). The AndroSensor App does not provide any info on type of magnetometer.

So I need to have a device with EM7180 coupled with the MPU925. Do you know something like that (in android/phone form)? Or do I need to use a Arduino or Raspberry?

Thanks again for your help.

All the best,

Victor

On 13 February 2016 at 17:23, Kris Winer notifications@github.com wrote:

Hi Victor,

This is a very interesting application of attitude estimation! I hadn't though about using it for astrometry.

I don't know the details of the tools you are using but I have some thoughts and a suggestion.

First of all, as you note, precision is not very useful if the accuracy is poor. Since sensor fusion to obtain absolute orientation (attitude estimation) is well-understood I must assume the problem is in the underlying sensors data. I have ample evidence that even with excellent sensor fusion filtering, the accuracy can be poor (OK, meaning 7 - 10 degrees in heading or attitude) if the underlying sensors are not very good. Even hard iron and limited soft iron calibration can't always cure a misbehaving magnetometer, and for some sensors nothing can help.

You can see examples of this here:

https://github.com/kriswiner/MPU-6050/wiki/Hardware-Sensor-Fusion-Solutions

and here:

https://github.com/kriswiner/EM7180_SENtral_sensor_hub/wiki/E.-Typical-Resul ts-Using-the-SENtral

The data plots show variations in azimuth and altitude of ~4 degrees standard deviation. This is not terribly bad, in fact I would say this is typical for even pretty good sensor fusion solutions out there like the MAX21000. Do you know which sensors are being used in the Samsung product?

The best orientation solution I have found is the EM7180 coupled with the MPU9250 which can provide 1 degree accuracy in pitch and roll and 2 degree accuracy in heading (for AHRS navigation). Not sure what this would translate to for astrometry.

So in short, the problem is IMHO not the apps, the fusion algorithms, or the calibration (this can always be better but can't fix poor sensors) but the underlying sensor data and there is nothing you can do about that except change sensors.

Kris -----Original Message----- From: vreijs [mailto:notifications@github.com] Sent: February 13, 2016 6:00 AM To: kriswiner/MPU-9250 Subject: [MPU-9250] calibration of magnetometer (#42)

Hello kriswiner, I read you article:

https://github.com/kriswiner/MPU-6050/wiki/Simple-and-Effective-Magnetometer -Calibration and I am wondering about a few things: . do you know if in the Android OS this calibration is included or does the App writers need to implement this? . if it is part of the OS, is this 'calibration' movement one need to do in some way recognised by the OS? And when recognised, it performs the calibration (both for hard and soft irons?).

I am asking this as I have some problems with measuring azi/alt. I am using a Samsung Tab2 7.0 GT-P3110 Android 4.2.2 Regardless of the App I am using (e.g. SunSurveyor or Measure Angle Apps), the results of azimuth and altitude measurements depend on the rotation of the device. So when pointing towards an object (Moon/Sun) with known azi/alt with the camera, the azimuth and altitude measurements changes when the device is rotated (see attached plot). image18 <

https://cloud.githubusercontent.com/assets/17218070/13027981/3599099c-d259- 11e5-9ce8-4d2930c4545c.gif>

The precision both in azimuth and altitude is quite good (0.5 deg, by measuring the azi/alt several times using the same rotation of the device), which is close to the one of a normal compass/clinometer. But the accuracy is not that good!

Attached is a plot of the measured azi and alt (using SunSurveyor App, but other Apps [such as Measure Angle] have same behavior). I turned the device 90 degrees (starting with landscape, portrait, the -landscape and -portrait) and measured each time the az/alt of the Moon. You also see the average of these measurements and the ephemeris location of the Moon. So the average looks ok-ish compared to the ephemeris values calculated for the Moon.

I have now been measuring the Moon and Sun many times and I see every time the similar behavior and accuracy spread (although same variation can happen). Again the precision stays within the 0.5 degrees when doing multiple measured at the same rotation. By the way I am away from iron objects (external or on body) and also away from electricity cables, etc. I also do the so called 'calibration' of the magnetometer (I did several different calibration movements). That calibration has some effect, but normally the spread stays the same. I also compared it with a normal compass and clinometer, which gives values comparable to the ephemeris.

If you see that plot, do you think this is something in the OS, the sensor, the device and/or the App? Calculating azi/alt is of course not an easy thing. In the device or Operating systems things are being done of course, but I am wondering why I see this large inaccuracy (certainly in azi [up to 20 degrees], and smaller in alt [up to 7 degrees])?

Here is another article that described the issues around the magnetometer (and also how to overcome them [partly]). I am wondering if the calibration inside the device and/or OS is done properly? http://www.nxp.com/files/sensors/doc/app_note/AN4246.pdf

Any ideas/hint you see in this?

Thanks for any feedback.

All the best,

Victor

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183716381 . https://github.com/notifications/beacon/AGY1qv04GAaQSaQm2LJhsQxOqW-scAibks5 pj2gGgaJpZM4HZo42.gif

[vreijs]

Hello Kris,

On 13 February 2016 at 18:16, Kris Winer notifications@github.com wrote:

I use standard deviation but you are correct, the errors I see are rather systematic. Especially in the Bosch sensors but also in the ST ones. The heading precision is very good (I think his corresponds to the azimuth here) but when estimating ninety-degree turns the heading estimation can be off by ten degrees, reproducibly (I see the same in my cell phone). This is not sensor jitter but rather a systematic bias that in principle could be removed, but simple response surface recentering (hard iron calibration) and axes scaling (poor man's soft iron correction) does not correct for it.

Great that you see similar behavior, that means that sensors react in a similar way. The 1 or 2 degrees you mention with the EM7180+MPU9250 is that the accuracy or the precision? So does it have systematic erratic behavior or is it more Gaussian errors?

So yes, I would recommend an EM7180+MPU9250 solution. You can solder a breakout board (https://www.tindie.com/products/onehorse/ultimate-sensor-fusion-solution/ ) with this solution to your favorite microcontroller. I have designed integrated solutions for UAVs that have a microcontroller and EM7180+MPU9250 on board (with motor drivers and boost converters that might not be useful to you). Here is an example:

https://forum.pjrc.com/threads/32985-Teensy-Flight-Controller

The advantage is a simple Arduino IDE can be used to write control software; I have all the code necessary to get absolute orientation from this kind of device.

In the past I was quite into soldering and experimenting (build my own oscilloscope in the tube times and moving to the ZX81 times: build my own speech synthesizer and software; all in Z80 assembler;-), but that is some 40 years ago. I envy the present times with all these options and controllability...

I am interested in the astrometry application. Do you think there is a commercial market for such a device?

Don't know. I know there was once (some 5 years ago?) a device that would name stars by looking into it and pointing to part of the sky. Don't know the name anymore (was build in USA), I think at that time the accuracy was not great due to the sensors. But with a 1 or 2 degrees precision (and accuracy smaller than that), that would change things. But a commercial market, I don't know! I am working in the archaeoastronomy environment: http://www.archaeocosmology.org/eng/archaeocosmology.htm and http://www.archaeocosmology.org/eng/aarde.htm (even got an MA degree in this early last year;-)

But it is quiet difficult to make profit in that environment. Ease of use is very important and that is what I like about this SunSurveyor App (Android). It has augmented reality and the paths of the Moon and Sun projected on the camera pictures (incl. equinox, and standstill events). So nice in the field. The only problem is the accuracy of the device sensors (you can't blame the App, although I have asked them to be able to offset the azi and alt)...

I mean like an instrumented finder scope which reads out absolute alt-azi on a display, etc. Maybe we could collaborate on development of such a device?

Any way, I am always interested in cooperating in this kind of projects, as I think that they are valuable (not only for archaeoastronomy). For sure I am willing to provide theoretical background and testing experience. I am willing to distribute/advertise in my environment. But again, commercially I don't know if it is viable (but perhaps I am now throwing in my own glasses;--)...

Thanks again for your interest.

All the best,

Victor

[vreijs]

See here: http://download.yamaha.com/api/asset/file/?language=ja&site=jp.yamaha.com&asset_id=46264

On 13 February 2016 at 18:19, Kris Winer notifications@github.com wrote:

No, MS probably means Measurement Specialties, but I am not familiar with the numbers. I'll poke around and see if I can find more info with this as a start.

[kriswiner]

This is the accuracy. The precision is more like a tenth of a degree or less depending on the mode.

The link I sent you has some results using this sensor fusion solution; you can judge for yourself the quality of the output.

Kris

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 13, 2016 11:02 AM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris,

On 13 February 2016 at 18:16, Kris Winer notifications@github.com wrote:

I use standard deviation but you are correct, the errors I see are rather systematic. Especially in the Bosch sensors but also in the ST ones. The heading precision is very good (I think his corresponds to the azimuth here) but when estimating ninety-degree turns the heading estimation can be off by ten degrees, reproducibly (I see the same in my cell phone). This is not sensor jitter but rather a systematic bias that in principle could be removed, but simple response surface recentering (hard iron calibration) and axes scaling (poor man's soft iron correction) does not correct for it.

Great that you see similar behavior, that means that sensors react in a similar way. The 1 or 2 degrees you mention with the EM7180+MPU9250 is that the accuracy or the precision? So does it have systematic erratic behavior or is it more Gaussian errors?

So yes, I would recommend an EM7180+MPU9250 solution. You can solder a breakout board (https://www.tindie.com/products/onehorse/ultimate-sensor-fusion-solution/ ) with this solution to your favorite microcontroller. I have designed integrated solutions for UAVs that have a microcontroller and EM7180+MPU9250 on board (with motor drivers and boost converters that might not be useful to you). Here is an example:

https://forum.pjrc.com/threads/32985-Teensy-Flight-Controller

The advantage is a simple Arduino IDE can be used to write control software; I have all the code necessary to get absolute orientation from this kind of device.

In the past I was quite into soldering and experimenting (build my own oscilloscope in the tube times and moving to the ZX81 times: build my own speech synthesizer and software; all in Z80 assembler;-), but that is some 40 years ago. I envy the present times with all these options and controllability...

I am interested in the astrometry application. Do you think there is a commercial market for such a device?

Don't know. I know there was once (some 5 years ago?) a device that would name stars by looking into it and pointing to part of the sky. Don't know the name anymore (was build in USA), I think at that time the accuracy was not great due to the sensors. But with a 1 or 2 degrees precision (and accuracy smaller than that), that would change things. But a commercial market, I don't know! I am working in the archaeoastronomy environment: http://www.archaeocosmology.org/eng/archaeocosmology.htm and http://www.archaeocosmology.org/eng/aarde.htm (even got an MA degree in this early last year;-)

But it is quiet difficult to make profit in that environment. Ease of use is very important and that is what I like about this SunSurveyor App (Android). It has augmented reality and the paths of the Moon and Sun projected on the camera pictures (incl. equinox, and standstill events). So nice in the field. The only problem is the accuracy of the device sensors (you can't blame the App, although I have asked them to be able to offset the azi and alt)...

I mean like an instrumented finder scope which reads out absolute alt-azi on a display, etc. Maybe we could collaborate on development of such a device?

Any way, I am always interested in cooperating in this kind of projects, as I think that they are valuable (not only for archaeoastronomy). For sure I am willing to provide theoretical background and testing experience. I am willing to distribute/advertise in my environment. But again, commercially I don't know if it is viable (but perhaps I am now throwing in my own glasses;--)...

Thanks again for your interest.

All the best,

Victor

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183725768 . https://github.com/notifications/beacon/AGY1qncOWLvtBQtrBkqJbL1V9AZyxChBks5 pj3UMgaJpZM4HZo42.gif

[vreijs]

Hello Kris,

Thanks indeed for that like. I think my precision is a little bit differently defined;-) So I have a camera picture and there is this (green) cross in the middle of the picture. Now I get the Moon/Sun to this center of this cross and then read the azi/alt. So my precision is including my ability to oriented the cross on the moon/sun (and of course my hands are not that stable, but similar to using an ordinary compass and clinometer;-). But the device on a tripod would be better;-) (like I do with my theodolite).

Attached is a screen shot (in the 4th quadrant of the cross you can see a tiny piece of the Moon's crescent). The large Moon and the path is the augmented reality of SunSurveyor.

Interested if we could make something accurate!

All the best,

Victor

On 13 February 2016 at 19:11, Kris Winer notifications@github.com wrote:

This is the accuracy. The precision is more like a tenth of a degree or less depending on the mode.

The link I sent you has some results using this sensor fusion solution; you can judge for yourself the quality of the output.

[kriswiner]

Hi Victor,

Didn't get any attachment on the last e-mail.

Accuracy here is in absolute orientation. For navigation heading is the hard part, and two degree heading accuracy is superb for a low-cost, small footprint device.

Accuracy means when the device says it is pointed North, it is less than 2 degrees from True North, everytime, no matter the pitch and roll.

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 13, 2016 11:30 AM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris,

Thanks indeed for that like. I think my precision is a little bit differently defined;-) So I have a camera picture and there is this (green) cross in the middle of the picture. Now I get the Moon/Sun to this center of this cross and then read the azi/alt. So my precision is including my ability to oriented the cross on the moon/sun (and of course my hands are not that stable, but similar to using an ordinary compass and clinometer;-). But the device on a tripod would be better;-) (like I do with my theodolite).

Attached is a screen shot (in the 4th quadrant of the cross you can see a tiny piece of the Moon's crescent). The large Moon and the path is the augmented reality of SunSurveyor.

Interested if we could make something accurate!

All the best,

Victor

On 13 February 2016 at 19:11, Kris Winer notifications@github.com wrote:

This is the accuracy. The precision is more like a tenth of a degree or less depending on the mode.

The link I sent you has some results using this sensor fusion solution; you can judge for yourself the quality of the output.

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183729817 . https://github.com/notifications/beacon/AGY1qtK_-gmgbOfKiNlN8m-Gzq1BjdQkks5 pj3vAgaJpZM4HZo42.gif

[vreijs]

I added the picture on the github site: https://cloud.githubusercontent.com/assets/17218070/13029622/29c54132-d289-11e5-8d52-177abd3872a2.jpg

Indeed 2 degrees sounds good (4 times better then what I now have). A normal compass is close to this also, might be somewhat smaller). I assume the 2 degrees is the 68% (1 sigma) definition (assuming it is normally distributed).

All the best,

Victor

On 13 February 2016 at 20:11, Kris Winer notifications@github.com wrote:

Hi Victor,

Didn't get any attachment on the last e-mail.

Accuracy here is in absolute orientation. For navigation heading is the hard part, and two degree heading accuracy is superb for a low-cost, small footprint device.

Accuracy means when the device says it is pointed North, it is less than 2 degrees from True North, everytime, no matter the pitch and roll.

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 13, 2016 11:30 AM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris,

Thanks indeed for that like. I think my precision is a little bit differently defined;-) So I have a camera picture and there is this (green) cross in the middle of the picture. Now I get the Moon/Sun to this center of this cross and then read the azi/alt. So my precision is including my ability to oriented the cross on the moon/sun (and of course my hands are not that stable, but similar to using an ordinary compass and clinometer;-). But the device on a tripod would be better;-) (like I do with my theodolite).

Attached is a screen shot (in the 4th quadrant of the cross you can see a tiny piece of the Moon's crescent). The large Moon and the path is the augmented reality of SunSurveyor.

Interested if we could make something accurate!

All the best,

Victor

On 13 February 2016 at 19:11, Kris Winer notifications@github.com wrote:

This is the accuracy. The precision is more like a tenth of a degree or less depending on the mode.

The link I sent you has some results using this sensor fusion solution; you can judge for yourself the quality of the output.

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183729817 . < https://github.com/notifications/beacon/AGY1qtK_-gmgbOfKiNlN8m-Gzq1BjdQkks5 pj3vAgaJpZM4HZo42.gif>

— Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183746461.

[kriswiner]

We haven't dome super precise characterization yet. I think the 2 sigma accuracy is less than 2 degrees, but we still need to measure this. I wonder if a small battery powered device affixed to a finder scope would give sufficiently accurate tracking with this solution to be interesting?

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 13, 2016 12:37 PM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

I added the picture on the github site: https://cloud.githubusercontent.com/assets/17218070/13029622/29c54132-d289-1 1e5-8d52-177abd3872a2.jpg

Indeed 2 degrees sounds good (4 times better then what I now have). A normal compass is close to this also, might be somewhat smaller). I assume the 2 degrees is the 68% (1 sigma) definition (assuming it is normally distributed).

All the best,

Victor

On 13 February 2016 at 20:11, Kris Winer notifications@github.com wrote:

Hi Victor,

Didn't get any attachment on the last e-mail.

Accuracy here is in absolute orientation. For navigation heading is the hard part, and two degree heading accuracy is superb for a low-cost, small footprint device.

Accuracy means when the device says it is pointed North, it is less than 2 degrees from True North, everytime, no matter the pitch and roll.

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 13, 2016 11:30 AM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris,

Thanks indeed for that like. I think my precision is a little bit differently defined;-) So I have a camera picture and there is this (green) cross in the middle of the picture. Now I get the Moon/Sun to this center of this cross and then read the azi/alt. So my precision is including my ability to oriented the cross on the moon/sun (and of course my hands are not that stable, but similar to using an ordinary compass and clinometer;-). But the device on a tripod would be better;-) (like I do with my theodolite).

Attached is a screen shot (in the 4th quadrant of the cross you can see a tiny piece of the Moon's crescent). The large Moon and the path is the augmented reality of SunSurveyor.

Interested if we could make something accurate!

All the best,

Victor

On 13 February 2016 at 19:11, Kris Winer notifications@github.com wrote:

This is the accuracy. The precision is more like a tenth of a degree or less depending on the mode.

The link I sent you has some results using this sensor fusion solution; you can judge for yourself the quality of the output.

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183729817 . <

https://github.com/notifications/beacon/AGY1qtK_-gmgbOfKiNlN8m-Gzq1BjdQkks5 pj3vAgaJpZM4HZo42.gif>

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183746461.

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183749628 . https://github.com/notifications/beacon/AGY1qhVFGDRv79QT1C7lG23t1NzVns2Yks5 pj4tUgaJpZM4HZo42.gif

[vreijs]

Hello Kris,

On 13 February 2016 at 21:04, Kris Winer notifications@github.com wrote:

We haven't dome super precise characterization yet. I think the 2 sigma accuracy is less than 2 degrees, but we still need to measure this.

Ok, so it looks good.

I wonder if a small battery powered device affixed to a finder scope would give sufficiently accurate tracking with this solution to be interesting?

So you need a GPS, camera, a display and then overlay the location and/or name of the celestial object. The camera needs to be light sensitive and the display needs to have a good resolution. Of course it does not have to be that accurate (due to the sensor's accuracy of around 1 or 2 degrees (1 or 2 sigma)).

For my applications: determining where the sun/moon (or other celestial object) set/rise/culminates, etc) aka its path is; teh device can work at the level of a normal phone camera/screen (aka using an App like SunSurveyor, but with a better sensor). Google Sky Map is another App (although it does not augment [yet] the seen sky http://google-sky-map.en.softonic.com/android ).

I found it again: https://en.wikipedia.org/wiki/SkyScout . That web page also mentioned the advance of the tablet/ipad... Knowing your competitor ( http://www.telescopes.com/blogs/helpful-information/18931204-celestron-skyscout-review ) is always good.

All the best,

Victor

[kriswiner]

I had in mind a simple finder scope with a small electronic package attached. The electronic package would report the altitude and azimuth where pointed. If this were mounted on a simple tripod, the user could record altitude, azimuth and time and get data to track any celestial object. If the device was equipped with a button, the user could simply push it to record these data and the data could be stored on the device for later retrieval or could be immediately sent via BE or wifi to a host computer or smartphone. The electronic device would be a module that could be attached to a (most, any?) finder scope. The total cost would be tens of dollars with the accuracy discussed previously.

I don't see any use for a camera or GPS in this use case.

Kris -----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 13, 2016 2:21 PM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris,

On 13 February 2016 at 21:04, Kris Winer notifications@github.com wrote:

We haven't dome super precise characterization yet. I think the 2 sigma accuracy is less than 2 degrees, but we still need to measure this.

Ok, so it looks good.

I wonder if a small battery powered device affixed to a finder scope would give sufficiently accurate tracking with this solution to be interesting?

So you need a GPS, camera, a display and then overlay the location and/or name of the celestial object. The camera needs to be light sensitive and the display needs to have a good resolution. Of course it does not have to be that accurate (due to the sensor's accuracy of around 1 or 2 degrees (1 or 2 sigma)).

For my applications: determining where the sun/moon (or other celestial object) set/rise/culminates, etc) aka its path is; teh device can work at the level of a normal phone camera/screen (aka using an App like SunSurveyor, but with a better sensor). Google Sky Map is another App (although it does not augment [yet] the seen sky http://google-sky-map.en.softonic.com/android ).

I found it again: https://en.wikipedia.org/wiki/SkyScout . That web page also mentioned the advance of the tablet/ipad... Knowing your competitor ( http://www.telescopes.com/blogs/helpful-information/18931204-celestron-skysc out-review ) is always good.

All the best,

Victor

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183765524 . https://github.com/notifications/beacon/AGY1qsY-U_nhBBj8tLezv4JF5axBInmtks5 pj6OwgaJpZM4HZo42.gif

[vreijs]

Indeed for that idea no camera is needed. To be able to determine what celestial object is in the direction one needs to know the latitude, longitude and time (so a GPS). All the best,

Victor

P.S. Do you know what magnetometer sensor is used in the iPads? It seems that these devices are more accurate in the azimuth...

On 14 February 2016 at 05:16, Kris Winer notifications@github.com wrote:

I had in mind a simple finder scope with a small electronic package attached. The electronic package would report the altitude and azimuth where pointed. If this were mounted on a simple tripod, the user could record altitude, azimuth and time and get data to track any celestial object. If the device was equipped with a button, the user could simply push it to record these data and the data could be stored on the device for later retrieval or could be immediately sent via BE or wifi to a host computer or smartphone. The electronic device would be a module that could be attached to a (most, any?) finder scope. The total cost would be tens of dollars with the accuracy discussed previously.

I don't see any use for a camera or GPS in this use case.

Kris -----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 13, 2016 2:21 PM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris,

On 13 February 2016 at 21:04, Kris Winer notifications@github.com wrote:

We haven't dome super precise characterization yet. I think the 2 sigma accuracy is less than 2 degrees, but we still need to measure this.

Ok, so it looks good.

I wonder if a small battery powered device affixed to a finder scope would give sufficiently accurate tracking with this solution to be interesting?

So you need a GPS, camera, a display and then overlay the location and/or name of the celestial object. The camera needs to be light sensitive and the display needs to have a good resolution. Of course it does not have to be that accurate (due to the sensor's accuracy of around 1 or 2 degrees (1 or 2 sigma)).

For my applications: determining where the sun/moon (or other celestial object) set/rise/culminates, etc) aka its path is; teh device can work at the level of a normal phone camera/screen (aka using an App like SunSurveyor, but with a better sensor). Google Sky Map is another App (although it does not augment [yet] the seen sky http://google-sky-map.en.softonic.com/android ).

I found it again: https://en.wikipedia.org/wiki/SkyScout . That web page also mentioned the advance of the tablet/ipad... Knowing your competitor (

http://www.telescopes.com/blogs/helpful-information/18931204-celestron-skysc out-review ) is always good.

All the best,

Victor

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183765524 . < https://github.com/notifications/beacon/AGY1qsY-U_nhBBj8tLezv4JF5axBInmtks5 pj6OwgaJpZM4HZo42.gif>

— Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183822897.

[kriswiner]

I suppose a separate GPS would do. The device I have in mind would have a very good clock (2 ppm) on board which could be calibrated against a GPS time pulse. The idea is to get real time tracking information on a celestial object onto a laptop or smartphone.

I don't know what is used on the iPhone. Generally smartphone sensors are optimized for small size and ow power, not accuracy.

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 14, 2016 2:50 AM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Indeed for that idea no camera is needed. To be able to determine what celestial object is in the direction one needs to know the latitude, longitude and time (so a GPS). All the best,

Victor

P.S. Do you know what magnetometer sensor is used in the iPads? It seems that these devices are more accurate in the azimuth...

On 14 February 2016 at 05:16, Kris Winer notifications@github.com wrote:

I had in mind a simple finder scope with a small electronic package attached. The electronic package would report the altitude and azimuth where pointed. If this were mounted on a simple tripod, the user could record altitude, azimuth and time and get data to track any celestial object. If the device was equipped with a button, the user could simply push it to record these data and the data could be stored on the device for later retrieval or could be immediately sent via BE or wifi to a host computer or smartphone. The electronic device would be a module that could be attached to a (most, any?) finder scope. The total cost would be tens of dollars with the accuracy discussed previously.

I don't see any use for a camera or GPS in this use case.

Kris -----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 13, 2016 2:21 PM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris,

On 13 February 2016 at 21:04, Kris Winer notifications@github.com wrote:

We haven't dome super precise characterization yet. I think the 2 sigma accuracy is less than 2 degrees, but we still need to measure this.

Ok, so it looks good.

I wonder if a small battery powered device affixed to a finder scope would give sufficiently accurate tracking with this solution to be interesting?

So you need a GPS, camera, a display and then overlay the location and/or name of the celestial object. The camera needs to be light sensitive and the display needs to have a good resolution. Of course it does not have to be that accurate (due to the sensor's accuracy of around 1 or 2 degrees (1 or 2 sigma)).

For my applications: determining where the sun/moon (or other celestial object) set/rise/culminates, etc) aka its path is; teh device can work at the level of a normal phone camera/screen (aka using an App like SunSurveyor, but with a better sensor). Google Sky Map is another App (although it does not augment [yet] the seen sky http://google-sky-map.en.softonic.com/android ).

I found it again: https://en.wikipedia.org/wiki/SkyScout . That web page also mentioned the advance of the tablet/ipad... Knowing your competitor (

http://www.telescopes.com/blogs/helpful-information/18931204-celestron-skysc out-review ) is always good.

All the best,

Victor

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183765524 . <

https://github.com/notifications/beacon/AGY1qsY-U_nhBBj8tLezv4JF5axBInmtks5 pj6OwgaJpZM4HZo42.gif>

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183822897.

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183873184 . https://github.com/notifications/beacon/AGY1qqj0IMfcUsyGJpKLT2QC9jVZ1G6dks5 pkFN1gaJpZM4HZo42.gif

[vreijs]

Hello Kris,

The time does not need to be that accurate (seconds level is ok), but knowing longitude and latitude are essential. Here you see the SkyScout mounted on a telesope. SckyScope are offered around 80$ on ebay. All the best,

Victor

On 14 February 2016 at 18:32, Kris Winer notifications@github.com wrote:

I suppose a separate GPS would do. The device I have in mind would have a very good clock (2 ppm) on board which could be calibrated against a GPS time pulse. The idea is to get real time tracking information on a celestial object onto a laptop or smartphone.

[vreijs]

Hello Kris, In the datasheet of the MPU-9205 I don't see the heading accuracy ( http://store.invensense.com/datasheets/invensense/MPU9250REV1.0.pdf ). Do you know the specs for this (I know what you measured). For the Project Tango Tablet development kit the BMX055 has a 3 deg heading accuracy. For the Samsung Tab2 with YAS530 has a 5 deg sensitivity.

Would be great to see the official accuracy of the MPU 9205. Thanks.

All the best,

Victor

On 14 February 2016 at 18:32, Kris Winer notifications@github.com wrote:

I suppose a separate GPS would do. The device I have in mind would have a very good clock (2 ppm) on board which could be calibrated against a GPS time pulse. The idea is to get real time tracking information on a celestial object onto a laptop or smartphone.

I don't know what is used on the iPhone. Generally smartphone sensors are optimized for small size and ow power, not accuracy.

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 14, 2016 2:50 AM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Indeed for that idea no camera is needed. To be able to determine what celestial object is in the direction one needs to know the latitude, longitude and time (so a GPS). All the best,

Victor

P.S. Do you know what magnetometer sensor is used in the iPads? It seems that these devices are more accurate in the azimuth...

On 14 February 2016 at 05:16, Kris Winer notifications@github.com wrote:

I had in mind a simple finder scope with a small electronic package attached. The electronic package would report the altitude and azimuth where pointed. If this were mounted on a simple tripod, the user could record altitude, azimuth and time and get data to track any celestial object. If the device was equipped with a button, the user could simply push it to record these data and the data could be stored on the device for later retrieval or could be immediately sent via BE or wifi to a host computer or smartphone. The electronic device would be a module that could be attached to a (most, any?) finder scope. The total cost would be tens of dollars with the accuracy discussed previously.

I don't see any use for a camera or GPS in this use case.

Kris -----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 13, 2016 2:21 PM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris,

On 13 February 2016 at 21:04, Kris Winer notifications@github.com wrote:

We haven't dome super precise characterization yet. I think the 2 sigma accuracy is less than 2 degrees, but we still need to measure this.

Ok, so it looks good.

I wonder if a small battery powered device affixed to a finder scope would give sufficiently accurate tracking with this solution to be interesting?

So you need a GPS, camera, a display and then overlay the location and/or name of the celestial object. The camera needs to be light sensitive and the display needs to have a good resolution. Of course it does not have to be that accurate (due to the sensor's accuracy of around 1 or 2 degrees (1 or 2 sigma)).

For my applications: determining where the sun/moon (or other celestial object) set/rise/culminates, etc) aka its path is; teh device can work at the level of a normal phone camera/screen (aka using an App like SunSurveyor, but with a better sensor). Google Sky Map is another App (although it does not augment [yet] the seen sky http://google-sky-map.en.softonic.com/android ).

I found it again: https://en.wikipedia.org/wiki/SkyScout . That web page also mentioned the advance of the tablet/ipad... Knowing your competitor (

http://www.telescopes.com/blogs/helpful-information/18931204-celestron-skysc

out-review ) is always good.

All the best,

Victor

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183765524 . <

https://github.com/notifications/beacon/AGY1qsY-U_nhBBj8tLezv4JF5axBInmtks5 pj6OwgaJpZM4HZo42.gif>

Reply to this email directly or view it on GitHub <https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183822897 .

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183873184 . < https://github.com/notifications/beacon/AGY1qqj0IMfcUsyGJpKLT2QC9jVZ1G6dks5 pkFN1gaJpZM4HZo42.gif>

— Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183945959.

[kriswiner]

Hi Victor,

Heading accuracy really doesn't apply to a 9 DoF sensor, that's why Invensense doesn't quote it. The "accuracy" quoted for the BMX055 and YAS530 are more likely what I would call jitter or Gaussian error. I have tested the BMX055 and I can tell you the heading accuracy is more like 7 or 10 degrees RMS.

Using the EM7180 sensor hub with it's sensor fusion filtering and the MPU9250 sensor data as input to the sensor hub I can routinely get 2 degree or better heading accuracy, with a heading jitter of a tenth of a degree or less.

This is a far superior absolute orientation solution to any that can be achieved with either of the mags you mention.

Kris

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 17, 2016 2:35 AM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris, In the datasheet of the MPU-9205 I don't see the heading accuracy ( http://store.invensense.com/datasheets/invensense/MPU9250REV1.0.pdf ). Do you know the specs for this (I know what you measured). For the Project Tango Tablet development kit the BMX055 has a 3 deg heading accuracy. For the Samsung Tab2 with YAS530 has a 5 deg sensitivity.

Would be great to see the official accuracy of the MPU 9205. Thanks.

All the best,

Victor

On 14 February 2016 at 18:32, Kris Winer notifications@github.com wrote:

I suppose a separate GPS would do. The device I have in mind would have a very good clock (2 ppm) on board which could be calibrated against a GPS time pulse. The idea is to get real time tracking information on a celestial object onto a laptop or smartphone.

I don't know what is used on the iPhone. Generally smartphone sensors are optimized for small size and ow power, not accuracy.

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 14, 2016 2:50 AM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Indeed for that idea no camera is needed. To be able to determine what celestial object is in the direction one needs to know the latitude, longitude and time (so a GPS). All the best,

Victor

P.S. Do you know what magnetometer sensor is used in the iPads? It seems that these devices are more accurate in the azimuth...

On 14 February 2016 at 05:16, Kris Winer notifications@github.com wrote:

I had in mind a simple finder scope with a small electronic package attached. The electronic package would report the altitude and azimuth where pointed. If this were mounted on a simple tripod, the user could record altitude, azimuth and time and get data to track any celestial object. If the device was equipped with a button, the user could simply push it to record these data and the data could be stored on the device for later retrieval or could be immediately sent via BE or wifi to a host computer or smartphone. The electronic device would be a module that could be attached to a (most, any?) finder scope. The total cost would be tens of dollars with the accuracy discussed previously.

I don't see any use for a camera or GPS in this use case.

Kris -----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 13, 2016 2:21 PM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris,

On 13 February 2016 at 21:04, Kris Winer notifications@github.com wrote:

We haven't dome super precise characterization yet. I think the 2 sigma accuracy is less than 2 degrees, but we still need to measure this.

Ok, so it looks good.

I wonder if a small battery powered device affixed to a finder scope would give sufficiently accurate tracking with this solution to be interesting?

So you need a GPS, camera, a display and then overlay the location and/or name of the celestial object. The camera needs to be light sensitive and the display needs to have a good resolution. Of course it does not have to be that accurate (due to the sensor's accuracy of around 1 or 2 degrees (1 or 2 sigma)).

For my applications: determining where the sun/moon (or other celestial object) set/rise/culminates, etc) aka its path is; teh device can work at the level of a normal phone camera/screen (aka using an App like SunSurveyor, but with a better sensor). Google Sky Map is another App (although it does not augment [yet] the seen sky http://google-sky-map.en.softonic.com/android ).

I found it again: https://en.wikipedia.org/wiki/SkyScout . That web page also mentioned the advance of the tablet/ipad... Knowing your competitor (

http://www.telescopes.com/blogs/helpful-information/18931204-celestron-skysc

out-review ) is always good.

All the best,

Victor

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183765524 . <

https://github.com/notifications/beacon/AGY1qsY-U_nhBBj8tLezv4JF5axBInmtks5

pj6OwgaJpZM4HZo42.gif>

Reply to this email directly or view it on GitHub <https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183822897 .

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183873184 . <

https://github.com/notifications/beacon/AGY1qqj0IMfcUsyGJpKLT2QC9jVZ1G6dks5 pkFN1gaJpZM4HZo42.gif>

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-183945959.

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-185145473 . https://github.com/notifications/beacon/AGY1qnv3LdNxmFIVq2ShSFkEI0QFsNxDks5 plERogaJpZM4HZo42.gif

[vreijs]

Thanks Kris,

On 17 February 2016 at 16:22, Kris Winer notifications@github.com wrote:

Heading accuracy really doesn't apply to a 9 DoF sensor, that's why Invensense doesn't quote it. The "accuracy" quoted for the BMX055 and YAS530 are more likely what I would call jitter or Gaussian error. I have tested the BMX055 and I can tell you the heading accuracy is more like 7 or 10 degrees RMS.

I see value range of 24 deg (min to max). And this results (assuming the 24 degrees is twice 3 sigma) in a 4 deg (1sigma) which is close to the 5 degrees specified in YAS530.I got data from a friend who has an iPad and he sees a range of around 10 degrees (so more or less half what my Samsung does). It is always difficult to determine what people mean with accuracy/precision (I always use in my evaluations the 1 sigma value).

Using the EM7180 sensor hub with it's sensor fusion filtering and the MPU9250 sensor data as input to the sensor hub I can routinely get 2 degree or better heading accuracy, with a heading jitter of a tenth of a degree or less.

So I hope that your 2 degrees means a reduction of a factor of around 3 to 4 compared to what I see in the YAS530.... I really would need to use my workflow on the MPU-9250 to to be sure we compare like with like.

Do you know which tablets/phones have this sensor? I have also asked this question to Invensense (no answer back yet).

All the best,

Victor

[vreijs]

Seeing this datasheet (that include the AK8963, which is 'part' of the MPU-9250: http://www.invensense.com/products/motion-tracking/9-axis/mpu-9250/ ) http://www.pnicorp.com/wp-content/uploads/Sentral-MandM-Technical-Datasheet_rG.pdf the heading accuracy is 2 deg (general description and section 2.1).

On 17 February 2016 at 16:35, Victor Reijs web.victor.reijs@gmail.com wrotr

Using the EM7180 sensor hub with it's sensor fusion filtering and the

MPU9250 sensor data as input to the sensor hub I can routinely get 2 degree or better heading accuracy, with a heading jitter of a tenth of a degree or less.

So I hope that your 2 degrees means a reduction of a factor of around 3 to 4 compared to what I see in the YAS530.... I really would need to use my workflow on the MPU-9250 to to be sure we compare like with like.

[vreijs]

How can I reach you privately? Can I send a private message on GitHub? I got feedback from Invensense Tech Support on my query of heading accuracy and phone/tablet.

[kriswiner]

PM me at onehorse@earthlink.net

[vreijs]

Hello Kris,

I first thought that my measurements has a periodicity of double 'avg azi', but that was because of the formula I used in my evaluation, which resulted in an abs function (and that becomes a double 'avg.azi' function)

So reevaluated my measurements. If I take the difference of azimuths between the landscape (0 deg rot) and -landscape (180 deg rot) measurements, I get the following picture: Why this sinus behavior is seen within landscape rotation, I still don't know...

If I take the difference of azimuth for the portrait (90deg rot) and -portrait (270deg rot), I get: This is more or less a flat graph (so no dependency of the azimuth range and the avg. azimuth when using portrait mode).

So when using portrait mode it looks that the systematic error is more random (1 sigma=~5 deg),...

Still something to find out (why this sinus-behavior?).

All the best,

Victor

[kriswiner]

Hi Victor,

Your results here are pretty typical of the average sensor fusion solution available in a smartphone. The roll and pitch (corresponding to the altitude) is generally good to within a couple of degrees but the heading (corresponding to the azimuth) can be off by 7 or 8 degrees RMS or standard deviation. This is why we developed the Ultimate Sensor Fusion Solution I mentioned, where we can routinely get 1 degree accuracy in altitude and 2 degree accuracy in azimuth. You simply cannot get this performance with any commercial smartphone implementation to my knowledge.

The reason has to do mostly with the sensors themselves, such that with even the best calibration and sensor fusion filtering the limits to azimuth accuracy remain at 5 degrees (at best) but more likely 10 degrees. The root cause is non-spherical response surfaces of one or more sensors that cannot be corrected with the available calibration. It is possible to calibrate them beyond what is available natively on the tablet but there will remain an inherent error in many of the sensors chosen for their low-power, small size, and low cost in modern smart devices and tablets. In other words, the sensors are not chosen for their excellent performance.

Kris

[vreijs]

Hello Kris, Can you provide information about the accuracy/precision of your device? I am really interested to see comparable data. Remember that using my tablet I get a precision of 0.5 deg (and I repeated that precision some 200 times). So it is the measurement methodology one use to determine accuracy. If wanted I can provide guidance about the measurement methodology. I am sure that it can be better (and I also really hope so!), and that is why I ask if you can provide numbers for my measurement method (same heading and rotate the device 90 degrees) .I just want a good methodology to define accuracy and precision.

All the best,

Victor

P.S. I think that my analysis is regardless of App and/or Android/IPad. But I will now concentrate on the Iphone/IPad (although I don't have such a device).

On 23 February 2016 at 22:46, Kris Winer notifications@github.com wrote:

Hi Victor,

Your results here are pretty typical of the average sensor fusion solution available in a smartphone. The roll and pitch (corresponding to the altitude) is generally good to within a couple of degrees but the heading (corresponding to the azimuth) can be off by 7 or 8 degrees RMS or standard deviation. This is why we developed the Ultimate Sensor Fusion Solution I mentioned, where we can routinely get 1 degree accuracy in altitude and 2 degree accuracy in azimuth. You simply cannot get this performance with any commercial smartphone implementation to my knowledge.

The reason has to do mostly with the sensors themselves, such that with even the best calibration and sensor fusion filtering the limits to azimuth accuracy remain at 5 degrees (at best) but more likely 10 degrees. The root cause is non-spherical response surfaces of one or more sensors that cannot be corrected with the available calibration. It is possible to calibrate them beyond what is available natively on the tablet but there will remain an inherent error in many of the sensors chosen for their low-power, small size, and low cost in modern smart devices and tablets. In other words, the sensors are not chosen for their excellent performance.

— Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-187948121.

[kriswiner]

Hi Victor,

You can see the results of simple testing here:

https://github.com/kriswiner/EM7180_SENtral_sensor_hub/wiki/E.-Typical-Results-Using-the-SENtral

Kris

[vreijs]

Hello Kris, In that link you rotate the device while rotating along the z-axis Correct? I am rotating along the heading axis. So the heading alt/az should not change, as the heading is always he same. but the device is rotated in steps of 90 deg. Do you know how I measure it?

On 24 February 2016 at 01:03, Kris Winer notifications@github.com wrote:

Hi Victor,

You can see the results of simple testing here:

https://github.com/kriswiner/EM7180_SENtral_sensor_hub/wiki/E.-Typical-Results-Using-the-SENtral

Kris

— Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-187995007.

[kriswiner]

Hi Victor,

I do rotate about the z-axis mostly, but at the end of the first test, I rotate in pitch by 45 degrees, then yaw by 45 degrees to show that the heading remains more or less fixed. This show the orientation solution is mostly free of cross axis effects. Tablets will fail this test.

Kris

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 23, 2016 10:41 PM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris, In that link you rotate the device while rotating along the z-axis Correct? I am rotating along the heading axis. So the heading alt/az should not change, as the heading is always he same. but the device is rotated in steps of 90 deg. Do you know how I measure it?

On 24 February 2016 at 01:03, Kris Winer notifications@github.com wrote:

Hi Victor,

You can see the results of simple testing here:

https://github.com/kriswiner/EM7180_SENtral_sensor_hub/wiki/E.-Typical-Resul ts-Using-the-SENtral

Kris

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-187995007.

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-188104581 . https://github.com/notifications/beacon/AGY1qnj2gGCQJDxgCSNouhQ8o1Yw6AUSks5 pnUfrgaJpZM4HZo42.gif

[vreijs]

You did this yaw of 45 by hand. What if you have a stable surface say a slanted table (at 45 degrees). And you put your device there and then turn it in steps of 90 degrees around the heading axis. I am interested what you get there. I will also do that with my tablet (now I used the Sun/Moon), but using a slanted surface is much more stable of course (my hand held variation gives some 0.5 degree precision). But I will be interested to see what the accuracy will be (now only find a slated outside surface without iron parts in it).

Thanks for your patience.

All the best,

Victor

On 24 February 2016 at 07:50, Kris Winer notifications@github.com wrote:

Hi Victor,

I do rotate about the z-axis mostly, but at the end of the first test, I rotate in pitch by 45 degrees, then yaw by 45 degrees to show that the heading remains more or less fixed. This show the orientation solution is mostly free of cross axis effects. Tablets will fail this test.

Kris

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 23, 2016 10:41 PM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris, In that link you rotate the device while rotating along the z-axis Correct? I am rotating along the heading axis. So the heading alt/az should not change, as the heading is always he same. but the device is rotated in steps of 90 deg. Do you know how I measure it?

On 24 February 2016 at 01:03, Kris Winer notifications@github.com wrote:

Hi Victor,

You can see the results of simple testing here:

https://github.com/kriswiner/EM7180_SENtral_sensor_hub/wiki/E.-Typical-Resul ts-Using-the-SENtral

Kris

Reply to this email directly or view it on GitHub <https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-187995007 .

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-188104581 . < https://github.com/notifications/beacon/AGY1qnj2gGCQJDxgCSNouhQ8o1Yw6AUSks5 pnUfrgaJpZM4HZo42.gif>

— Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-188129849.

[vreijs]

Will do this experiment tomorrow (I devised the setup in my garden;-). I think the sinus behavior I see between measured azimuth and real azimuth, could be very well linked to were I started my original quest: to understand soft iron compensation. The sensor/MD you are using compensates for hard and soft iron effect. I am really wondering if my tablet is doing the soft iron one (i asked this several weeks ago, but never got a good answer). I think it is not doing that. What I now see, is very well related to that effect I think. I now can reduce my workflow and repeat these measurements also for the Air2, iPhone5 and iPhone 6s plus to see how/what they do.

I wish I could program my own hard/soft iron compensation methods on my Android table... I also have the idea that my tablet ('s sensor) does calibration constantly; and thus it gets uncalibrated when doing erratic movements.

All the best,

Victor

On 24 February 2016 at 08:55, Victor Reijs web.victor.reijs@gmail.com wrote:

You did this yaw of 45 by hand. What if you have a stable surface say a slanted table (at 45 degrees). And you put your device there and then turn it in steps of 90 degrees around the heading axis. I am interested what you get there. I will also do that with my tablet (now I used the Sun/Moon), but using a slanted surface is much more stable of course (my hand held variation gives some 0.5 degree precision). But I will be interested to see what the accuracy will be (now only find a slated outside surface without iron parts in it).

[kriswiner]

The math for hard iron is pretty simple, and you can even make a crude correction for soft iron. There is an algorithm published as part of the mag calibration article I wrote and whose link I forwarded to you. How to code this into an Android device I do not know...

[vreijs]

I have this link already some time, which I thought was quite overarching: http://www.nxp.com/files/sensors/doc/app_note/AN4246.pdf I always though that this was included in Android, but it seems to be more a part of the motion device aggregator (if present in the device).

[vreijs]

Hello Kris, Just looked at turning my device 90 degrees, and like you said there is considerable change between these readings: a standard deviation of around 6 degrees around the 90 degrees. So close to the stand. dev seen in my measurements when rotating. I assume this is still due to the lacl of soft iron compensation.

All the best,

Victor

P.S. still need to get my hands on Air2, 6splus and 5 (but might be over a few weeks due to a lot of travel on my side).

On 25 February 2016 at 20:43, Kris Winer notifications@github.com wrote:

The math for hard iron is pretty simple, and you can even make a crude correction for soft iron. There is an algorithm published as part of the mag calibration article I wrote and whose link I forwarded to you. How to code this into an Android device I do not know...

— Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-188975960.

[kriswiner]

Partly, although many mags have relatively incurable bias in their response surface. Meaning that the response surface shape changes depending on orientation making it very difficult to compensate for even with good soft iron algorithms.

The technical term for this behavior is "crappy sensors".

Kris

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 27, 2016 10:46 AM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

Hello Kris, Just looked at turning my device 90 degrees, and like you said there is considerable change between these readings: a standard deviation of around 6 degrees around the 90 degrees. So close to the stand. dev seen in my measurements when rotating. I assume this is still due to the lacl of soft iron compensation.

All the best,

Victor

P.S. still need to get my hands on Air2, 6splus and 5 (but might be over a few weeks due to a lot of travel on my side).

On 25 February 2016 at 20:43, Kris Winer notifications@github.com wrote:

The math for hard iron is pretty simple, and you can even make a crude correction for soft iron. There is an algorithm published as part of the mag calibration article I wrote and whose link I forwarded to you. How to code this into an Android device I do not know...

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-188975960.

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-189700753 . https://github.com/notifications/beacon/AGY1qg57tYWwVix0DMiRCW4tlOYq2OReks5 poeZTgaJpZM4HZo42.gif

[vreijs]

I wish yur sensor/MD processor combination was implemented in some phone/tablet.

On 27 February 2016 at 19:10, Kris Winer notifications@github.com wrote:

The technical term for this behavior is "crappy sensors".

[kriswiner]

It's in the Sony smart watch.

But my suggestion was to design a purpose-built device for doing what you want to do. Phones and tablets are convenient, yes, but they will never do any one thing extremely well, they do many things well enough.

If you have interest in this please let me know.

Kris

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 28, 2016 12:39 AM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

I wish yur sensor/MD processor combination was implemented in some phone/tablet.

On 27 February 2016 at 19:10, Kris Winer notifications@github.com wrote:

The technical term for this behavior is "crappy sensors".

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-189819766 . https://github.com/notifications/beacon/AGY1qhM1fy3llk5wJynthYrWzKfKP_9eks5 poqm1gaJpZM4HZo42.gif

[vreijs]

For me the augmented reality is essential (so projecting path of celestial object) on reality. Like sunsurveyor http://www.sunsurveyor.com/ On 28 Feb 2016 17:39, "Kris Winer" notifications@github.com wrote:

It's in the Sony smart watch.

But my suggestion was to design a purpose-built device for doing what you want to do. Phones and tablets are convenient, yes, but they will never do any one thing extremely well, they do many things well enough.

If you have interest in this please let me know.

Kris

-----Original Message----- From: Victor Reijs [mailto:notifications@github.com] Sent: February 28, 2016 12:39 AM To: kriswiner/MPU-9250 Cc: Kris Winer Subject: Re: [MPU-9250] calibration of magnetometer (#42)

I wish yur sensor/MD processor combination was implemented in some phone/tablet.

On 27 February 2016 at 19:10, Kris Winer notifications@github.com wrote:

The technical term for this behavior is "crappy sensors".

Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-189819766 . < https://github.com/notifications/beacon/AGY1qhM1fy3llk5wJynthYrWzKfKP_9eks5 poqm1gaJpZM4HZo42.gif>

— Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/42#issuecomment-189912651.