IDG-500 vs ADXRS613

[marc]

Tom,

I looked with a scope, you see glitches generated at 100uS... I assume this is the sampling rate of ADC... and each time the ADC samples, it generates an extra current on his input.
Since we cannot decrase the ADXRS impedance due to the 200uA constraint...the only solution is to add a small cap on R3... I did it and it reduced significantly (factor 5) the noise....

CCL :
- impedance as low as possible with constraint to 200uA=> R3=16K and R2=10K
- 0,1uF on R3
CU

MArc

[Tom]

Thanks for this investigation Marc!

I also added the 0.1uF cap (on a 2k2 resistor), but didn't see that much of a difference. I suppose the perfect solution would be with an opamp?
Fortunately, the noise doesn't matter that much because it appears to be white with an amplitude of about 1.5°/s (which is not much when you use it to calculate the centripetal acceleration).

[marc]

Tom,

I made some additional measurement with the oscilloscope.

* The adxr with R2=120K and R3=220K as placed originally on the board, generate a noise of =+/-100mV
* If you use R2=1K5, R3=2K2 like you did, the noise becomes +/-60mV, adding a 0.1 cap reduce it to +/-30mV... BUT you are out of the specifications of the ADXR... which should be avoided.
* If you use R2=10K, R3=15K + a 0,1cap on R3, you are compliant with the adrx specs and the noise is 30mV on a full range of 3V...

I think, it is the best we can do... the noise seems now to come from the adrx itself (and it's power supply)....

Adding an OP will make you the life easier but will not improve fundamently the thinks... according to me the issue comes from the fact that the adrx is radiometric, meaning that the noise of the power supply (5V) is directly visible in the adrx output....the fact that the servos are also connected on the same 5 volts will certainely not improve the thinks....

Anyway, the configuration as it is now, has a noise less than 1%, I think this is neglectible compared to the noise the plane frame and motors will mechanically generate...

MArc

[Tom]

I'm glad to announce that:
1) The noise did not affect performance at all
2) The ADXRS613 has little drift!
3) The new gluonpilot hardware (which will arrive in 2 weeks) will have Marc's suggestions!

[tommy_like_mytom]

Hi Tom,
I saw that you calculated angles by integral gyro rate over time. How to avoid bias drift ? Can you tell some more about quarternions and Kalman used to cope with the bias drift. And how to determine parameters of Kalman filter.

[tommy_like_mytom]

I was also thinking a magnetometer would be a nice addition. It will allow a static calibration of the yaw axis and better error management at low velocities where the gps may not be adequate. I'm just starting to look at an HMC5843 breakout which is quite small, can be remotely mounted, and only requires a four wire connection - i2c plus power. Should be quite easy to integrate in the kalman. I don't think it would be reliable on board due to the stray fields. This configuration can be wing tip or boom mounted to distance it from the bulk of the electronics - same as full scale.

Hi Mitch,
Can you tell some more about the Kalman filter used to avoid measurement errors . How to determine parameters of a Kalman filter.

[Tom]

Hi Tommy,

I won't go into detail about the quaternions, it's more or less similar.

Assume we have a noisy variable AccRollAngle from the accelerometers. It's very noise, but averaged over - let's say - 10 seconds it is correct.

IntegratedRollAngle = IntegratedRollAngle + (Gyro - GyroBias) * dt

You can adapt for bias drift in 2 ways:
1- Changing the "GyroBias" variable a little bit at a time by looking at "IntegratedRollAngle - AccRollAngle"
2- Changing the "IntegratedRollAngle" variable by comparing it with "AccRollAngle"

For the moment I use method 2. I'm planning to use Method 2 with a very slow Method 1 applied to it in the future (to adapt for temperature bias drift).

Kalman filter parameters are (1) understanding how they affect the workings of the filter and (2) trial-and-error optimizing them

[mezuzu]

Hi Tom,

I am working with IDG-500 sensor, I have many problems to calculate the angles roll, pitch and yaw.

To calculate the angles I am using the formula you say

IntegratedRollAngle = IntegratedRollAngle + (Gyro - Gyrobias) / Sensitivity

actually the problem is the bias .

What I wanted to ask is whether "IntegratedRollAngle - AccRollAngle" is in units of degrees, how to turn this remains a Volt? Gyrobias to change.

[Tom]

Mezuzu, I though you might have to start working a simpler filter first, to get the hang of it.
You can choose your units (degrees or radians), it depends on how you scale your AD-values.

[mezuzu]

Hi Tom

I am currently a Kalman filter and everything works well with few sensed data (1000) but when increasing the sample (5000), all damaged.

Works well with limited data because the calculation of bias to the average of the top 100 with the sensor data at rest.