kriswiner
commented
8 years ago Hi Ray,
The Madgwick (and Mahony) filters are iterative, so ideally we would like 4 or 5 iterations on the filter solution for every new data sample. This means if you run at a 200 Hz data rate, you want to run the sensor fusion at ~1 kHz or so. It is possible to force this rate but I just use a Teensy 3.X at 72 MHz and I get it "for free", meaning that in my sketch I just query whatever fusion rate I get and it turns out to be > 1 kHz, which is fine. The way the code is set up, I loop through and check every time whether there is new data for each sensor. I so great, I use it in the next Madgwick call. If not, I just call Madgwick with the old data and iterate again. The code in practice will iterate on the same data 6 or 8 times which will produce a converged solution. This is the advantage of using fast processor. With a 16 MHz AVR processor running the fusion at 200 Hz you will not get a converged solution, ever, unless you run with < 50 Hz data rates, which you can do. But if your motion is "faster" than 50 Hz, you will experience latency. The other parameter is the bandwidth. Generally, you want the bandwidth to be 10 - 20% of the sample rate. For 200 Hz sample rates, I usually use 40 Hz bandwidth (low-pass filter). This smooths out transient and jitter to be expected with a 16-bit ADC. If you run with a sample rate of 50 Hz, your bandwidth has to be ~10 Hz, which is kind of low. Lastly, the sensors have to be sufficiently calibrated, as you know. The calibration algorithm you show looks OK to me. How are you collecting and averaging the mag bias?
Kris
-----Original Message----- From: RayBreslin [mailto:notifications@github.com] Sent: October 3, 2015 10:02 AM To: kriswiner/MPU-9250 Subject: [MPU-9250] Madgwick-Filter-ODR / Filtered-Magnetometer-Data (#23)
Hey Kris,
after some nicely spent vacation time I am now back to put all my effort into my little sensor-project and I do have a few more questions I would like to discuss with you.
I pretty much got your MPU9250-template set up for a CSR1010-MPU9250-combo-board and a smartphone as receiver/mc. For some reason I am not seeing any nice and stable roll, pitch and yaw angles which freaks me out...
Ok I have to admit that through my never ending researches for calibration approaches, madgwick fixes (s-parameters, inv-sqrt, ...) and other things I might have made too many adjustments and tweaks along the way! So now its "almost" back to the start, taking your work as baseline and take it on from there (again)...
Now to the questions:
- How is the function-call-frequency of your loop()-routine determined? Is there a start-skript somewhere or does it get called automatically? In other words how do you get/calculate a filter ODR of 145 hz?
- If I am sampling my sensor data at 50 hz continuously, what kind of filter ODR would you propose? Should it always be at least 200 hz? (Currently doing just 50 hz when the data is coming in, which doesnt seem to do that well...)
- What readout-frequency you use for your magnetometer? If I saw it correctly its always just 8 hz instead of 100 hz - only due to the readout-latency?
- How does all the calibration data from the magnetometer calibration needs to be applied to the raw sensor data?
Heres my approach:
scaledData[Tag.MAGNETOMETER][i] = (rawData[Tag.MAGNETOMETER][i] magResolution * currCalibrationData.magSensitivity[i] - currCalibrationData.magBias[i] * magResolution currCalibrationData.magSensitivity[i]) * currCalibrationData.magScale[i]; Legend:
i ~ x-, y-, z-axis
currCalibrationData.magSensitivity[i] = (((rawSensitivityAdjustmentValue[i]
- 128.0f) * 0.5f) / 128.0f + 1.0f);
magResolution = 10.0f * 4912.0f / 32760.0f;
I would very much appreciate your help!
Thanks in advance
Ray
Reply to this email directly or view it on GitHub https://github.com/kriswiner/MPU-9250/issues/23 . https://github.com/notifications/beacon/AGY1qo_6D5OLDM54sf6PVJeBGKSoQ56Mks5 o4AGZgaJpZM4GIc1_.gif
RayBreslin
Hey Kris,
after some nicely spent vacation time I am now back to put all my effort into my little sensor-project and I do have a few more questions I would like to discuss with you.
I pretty much got your MPU9250-template set up for a CSR1010-MPU9250-combo-board and a smartphone as receiver/mc. For some reason I am not seeing any nice and stable roll, pitch and yaw angles which freaks me out...
Ok I have to admit that through my never ending researches for calibration approaches, madgwick fixes (s-parameters, inv-sqrt, ...) and other things I might have made too many adjustments and tweaks along the way! So now its „almost“ back to the start, taking your work as baseline and take it on from there (again)...
Now to the questions:
Heres my approach:
I would very much appreciate your help! Thanks in advance Ray