Continuous order tracking

[SomptingGuy]

Does anyone have a reference (still can't type that word first time!) for this, or a method?

I have a stream of crank angle, time and sensor data. I want to track a specific order as the data arrives - don't want to go through all that resampling and FFT malarkey. I should be able to use the crank angle signal to construct the appropriate multiplier for my data, but just can't get quite get my head around doing it in transient. It's the summation bit that's bogging me down. Ok for steady-state, but transient?

- Steve

 

[GregLocock]

You need a good speed estimate so you've got a problem if the slew rate is high.

As to the rest of it, you've managed to confuse me about what you are confused about.

What I'd do is create a fake sampling interval in crank angle, resample the incoming data using interp1 (or is it 2?) to give me crank angle based data and then bandpass that directly.

Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[SomptingGuy]

This is the sort of thing I'm doing. Theta is the crank angle.

re = 2*sum(signal.*cos(order*2*pi*theta))/numpoints;
im = 2*sum(signal.*sin(order*2*pi*theta))/numpoints;

For steady-state speeds, the summations give the real and imag parts of the line at the chosen order, no filters required (assuming of course that signal is cyclic within the period).

Maybe I just need to sample and reset the summation at the end of each cycle?

- Steve

 

[GregLocock]

Yes I think it is that easy. BTW that is a DFT even if you claim it isn't!



Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[SomptingGuy]

Trying to explain something in words to someone else is always the best way to get your "stuff" sorted out in your own mind. Of course it is a DFT, just not an FFT.


- Steve

 

[GregLocock]

You are right there, my reports verge on the comprehensible these days, due mainly to thousands of answers I've typed on this board.



Cheers

Greg Locock

SIGlease see FAQ731-376 for tips on how to make the best use of Eng-Tips.

 

[SomptingGuy]

Well, I got it working. It was slightly complicated by having to do it in the time domain, using unevenly spaced data. The key was the sample and reset of the integral(s)at the cycle boundaries. I did spend several hours comparing 2nd order from this experimental model with overall computed somewhere else, wondering why they were close, but not the same...

- Steve