Automotive EMI influence on a MEMS IMU, what can I do ?

Hi.

High frequency noise does not imply inaccuracy.

I doubt you have to bother with that.

There I see only something similar to your conclusion that EMI is the problem. Nothing you have stated, beyond your obvious assumption, makes me think so.

Have you evaluated the signal(s) in question after applying adequate low-pass filtering to suppress that noise and avoid aliasing at the ADC? If not, that may be all you need. If so, why are you concerned about the high frequency noise?

I agree with you, however, the heading being constant within reason (car is driven in straight line on a nice road at low speed 5 mph), the shift in the heading reading is much larger than what would be expected by taking into account the vibration induced by the car frame onto the IMU. I wish it were just that.

well I guess I am making this statement because of the following: - when the car engine is not running, the IMU gives the right heading.

- when the engine is running but the car is not moving, moving the IMU away from the dashboard (like in the back seat) make the high frequency

reading of the IMU heading go back to normal.

- when the car is driven in straight line, the same high frequency "noise" seems to come back in full force with amplitude beyond what would be expected from usual high frequency noise as stated above. In other words, the heading information is just absolutely not reliable.

One of the ways to do low-pass filtering is to couple the heading information with the acceleration and rate of rotation of the car. It does absolutely great when the car is not driving and the engine is running and the IMU is in the back seat. However, those same type of high frequency readings come back in full force (with amplitude beyond my expectation) when the car is driven on a nice road at low speed. Even an average of this last signal does not even out to the mean heading expected.

Jake.

That conclusion is not supported by the evidence given.

I do not understand that statement. My interpretation is that you have not taken steps to avoid aliasing.

If aliasing has already occured, averaging may not be enough to straighten it out even if there was nothing other than a high frequency signal added to an uncorrupted low frequency signal.

From what you have stated, it is possible that high frequency noise is getting into the device circuits through the power supply or through space (EM coupling). It would certainly be interesting to see what the HF noise looks like if you add filtering to the supply and put the device into a shielded enclosure. If those have any effect on the average value of the signal, evaluated with appropriate anti-aliasing for any sampled signal, then that would indicate that the noise you worry about is, in fact, a problem.

But at this point, I see no reason to believe you have anything but additive noise that simply has not been appropriately filtered prior to sampling. Can you provide evidence to the contrary?

Larry,

The unit gives heading information, it is pretty easy to see when the heading "noise" is just not right. I mean it goes beyond the noise amplitude provided by the device manufacturer. I am not doing the filtering the unit is. It uses information about the movement of the unit (acceleration, rate of rotation) through most probably some type of kalman filter in order to get rid of the high frequency noise of the heading information.

With regards to the power supply, because we felt the issue might come from it, we connected the IMU directly to a set of individual batteries (AA if I recall). The computer from which the readings are gathered is a laptop that is disconnected from the car power supply (no inverter is used). So you would propose for me to shield the device, should I use a particular material to do just that ? remember I am a newbie.

With regards to your last statement, I am not quite sure I could argue the contrary. One piece of evidence, albeit a not very technical one, is that the manufacturer seems to be having the same problem with other people than just us and do not seem to have a good answer for the time being on this issue. As i said before, their filtering does well when the car is not running or when the engine is on (and the unit is far from the engine). This would be my only evidence.

Jake.

The device manufacturer can only specify what his product produces. You may observe additional noise that is induced in your environment because of conductor geometry outside of the device.

What I am questioning here, and what you seem to not yet understand, is whether the noise you see is anything more than a high frequency signal added to a perfectly usable lower frequency signal. If that is the case, the solution is very simple: Filter off the frequency components that were added.

I urge you to go beyond "just not right" language. To me, and most other readers here, it means no more than "different from what I would expect or wish to see". The problem with such vague words is that nobody can tell whether it is your expectation or the hardware that needs to be modified.

The unit is undoubtedly not filtering noise that may be induced in wiring outside of the unit.

Your observations either confirm that the noise is picked up external to the unit, or show that the above conjecture is false. Which is it?

For an experiment, I would get the unit and the acquisition system (whatever leads to the ADC from which you are pulling data) into a heavy metal box. It hardly matters that you completely suppress induced noise. All you have to do is suppress it enough to show whether or not it has any effect on properly processed IMU output. Of course, to show that, you would have to have processed IMU output properly. At this point, I have serious doubts that has been done.

That other people see noise in an automotive environment demonstrates nothing that is not already common knowledge. That the device manufacturer does not seem (to you) to have a good answer may mean something, but there are so many things it could mean as to mean nothing.

I still see no reason to believe you have anything but additive noise that simply has not been appropriately filtered prior to sampling. Your answers to questions on this issue make that seem more likely now than when I first inquired about it.

To get more concrete about that: What is the cutoff frequency of *your* anti-alias filter? (not the one that may or may not be in the IMU) What is the sample rate at which your ADC reads IMU output?

Larry,

"...To get more concrete about that: What is the cutoff frequency of *your* anti-alias filter? (not the one that may or may not be in the IMU) What is the sample rate at which your ADC reads IMU output? .."

I don't know.

With regards to the sample rate of my ADC for the IMU output, I guess I am not quite sure this is relevant. In a MEMS device, the ADC is part of the MEMS device, so in effect the reading is already the converted reading from the analog of the tiny-micrometer-cantilevers. In effect, when I read data from the IMU, it already has gone through an ADC step inside the IMU itself.

with regards to your previous comment:

"...For an experiment, I would get the unit and the acquisition system (whatever leads to the ADC from which you are pulling data) into a heavy metal box. It hardly matters that you completely suppress induced noise. All you have to do is suppress it enough to show whether or not it has any effect on properly processed IMU output. Of course, to show that, you would have to have processed IMU output properly. At this point, I have serious doubts that has been done. .."

I'll do that as I mentioned earlier. Thank you very much for taking the time. I'll report on it a little bit later.

Jake.

Could you tell me what IMU you are using ?

I'm interested with your experiment since I'm testing a magnetic sensor for my project : openavionics.sourceforge.net

In message , None really writes

I have been involved in the design of such things (probably not yours though:). Could be due to acoustic noise as accelerometers are really potential microphones. Wool is a good acoustic isolator. Vibrating gyros and some accelerometers can use ac activation at khz is this the source of your noise? I vote for vibration.

Thank you very much Doug.

Doug,

Can you confirm that you have designing MEMS accelerometers ?

Thanks in advance,

Jake.

Is the sensor suseptable to magnetic fields?

Mark

It has to! since part of the sensor provides magnetic heading.

Jake.

Maybe they just don't find you all that attractive. ;-P

So *that's* why sheep take no notice when I yell at them :-)

-- "Electricity is of two kinds, positive and negative. The difference is, I presume, that one comes a little more expensive, but is more durable; the other is a cheaper thing, but the moths get into it." (Stephen Leacock)

In message , None really writes

Yes

I suspect you're talking at cross-purposes.

I think Larry intends that yould actually *filter* the signals in hardware to filter out the HF noise.

I think you are interpreting 'filtering' as doing something in software.

No amount of tinkering in software will fix a dirty transducer signal.

Graham

Ok. But for a MEMS device, I would have to have a whole silicon fab facility to do the filtering in the hardware. Thanks for the insight.

Jake.

Yes, I have noticed that the headings information were going crazy when we would hit a bump (z g-jitters).

Thank you Doug.

Jake.